Multidisciplinary rehabilitation for older people with hip fractures

Abstract

Background

Hip fracture is a major cause of morbidity and mortality in older people, and its impact on society is substantial. After surgery, people require rehabilitation to help them recover. Multidisciplinary rehabilitation is where rehabilitation is delivered by a multidisciplinary team, supervised by a geriatrician, rehabilitation physician or other appropriate physician. This is an update of a Cochrane Review first published in 2009.

Objectives

To assess the effects of multidisciplinary rehabilitation, in either inpatient or ambulatory care settings, for older people with hip fracture.

Search methods

We searched the Cochrane Bone, Joint and Muscle Trauma Group Specialised Register, CENTRAL, MEDLINE and Embase (October 2020), and two trials registers (November 2019).

Selection criteria

We included randomised and quasi‐randomised trials of post‐surgical care using multidisciplinary rehabilitation of older people (aged 65 years or over) with hip fracture. The primary outcome – 'poor outcome' – was a composite of mortality and decline in residential status at long‐term (generally one year) follow‐up. The other 'critical' outcomes were health‐related quality of life, mortality, dependency in activities of daily living, mobility, and related pain.

Data collection and analysis

Pairs of review authors independently performed study selection, assessed risk of bias and extracted data. We pooled data where appropriate and used GRADE for assessing the certainty of evidence for each outcome.

Main results

The 28 included trials involved 5351 older (mean ages ranged from 76.5 to 87 years), usually female, participants who had undergone hip fracture surgery. There was substantial clinical heterogeneity in the trial interventions and populations. Most trials had unclear or high risk of bias for one or more items, such as blinding‐related performance and detection biases. We summarise the findings for three comparisons below.

Inpatient rehabilitation: multidisciplinary rehabilitation versus 'usual care'
Multidisciplinary rehabilitation was provided primarily in an inpatient setting in 20 trials.

Multidisciplinary rehabilitation probably results in fewer cases of 'poor outcome' (death or deterioration in residential status, generally requiring institutional care) at 6 to 12 months' follow‐up (risk ratio (RR) 0.88, 95% confidence interval (CI) 0.80 to 0.98; 13 studies, 3036 participants; moderate‐certainty evidence). Based on an illustrative risk of 347 people with hip fracture with poor outcome in 1000 people followed up between 6 and 12 months, this equates to 41 (95% CI 7 to 69) fewer people with poor outcome after multidisciplinary rehabilitation. Expressed in terms of numbers needed to treat for an additional harmful outcome (NNTH), 25 patients (95% CI 15 to 100) would need to be treated to avoid one 'poor outcome'. Subgroup analysis by type of multidisciplinary rehabilitation intervention showed no evidence of subgroup differences.

Multidisciplinary rehabilitation may result in fewer deaths in hospital but the confidence interval does not exclude a small increase in the number of deaths (RR 0.77, 95% CI 0.58 to 1.04; 11 studies, 2455 participants; low‐certainty evidence). A similar finding applies at 4 to 12 months' follow‐up (RR 0.91, 95% CI 0.80 to 1.05; 18 studies, 3973 participants; low‐certainty evidence). Multidisciplinary rehabilitation may result in fewer people with poorer mobility at 6 to 12 months' follow‐up (RR 0.83, 95% CI 0.71 to 0.98; 5 studies, 1085 participants; low‐certainty evidence).

Due to very low‐certainty evidence, we have little confidence in the findings for marginally better quality of life after multidisciplinary rehabilitation (1 study). The same applies to the mixed findings of some or no difference from multidisciplinary rehabilitation on dependence in activities of daily living at 1 to 4 months' follow‐up (measured in various ways by 11 studies), or at 6 to 12 months' follow‐up (13 studies). Long‐term hip‐related pain was not reported.

Ambulatory setting: supported discharge and multidisciplinary home rehabilitation versus 'usual care'

Three trials tested this comparison in 377 people mainly living at home. Due to very low‐certainty evidence, we have very little confidence in the findings of little to no between‐group difference in poor outcome (death or move to a higher level of care or inability to walk) at one year (3 studies); quality of life at one year (1 study); in mortality at 4 or 12 months (2 studies); in independence in personal activities of daily living (1 study); in moving permanently to a higher level of care (2 studies) or being unable to walk (2 studies). Long‐term hip‐related pain was not reported.

One trial tested this comparison in 240 nursing home residents. There is low‐certainty evidence that there may be no or minimal between‐group differences at 12 months in 'poor outcome' defined as dead or unable to walk; or in mortality at 4 months or 12 months. Due to very low‐certainty evidence, we have very little confidence in the findings of no between‐group differences in dependency at 4 weeks or at 12 months, or in quality of life, inability to walk or pain at 12 months.

Authors' conclusions

In a hospital inpatient setting, there is moderate‐certainty evidence that rehabilitation after hip fracture surgery, when delivered by a multidisciplinary team and supervised by an appropriate medical specialist, results in fewer cases of 'poor outcome' (death or deterioration in residential status). There is low‐certainty evidence that multidisciplinary rehabilitation may result in fewer deaths in hospital and at 4 to 12 months; however, it may also result in slightly more. There is low‐certainty evidence that multidisciplinary rehabilitation may reduce the numbers of people with poorer mobility at 12 months. No conclusions can be drawn on other outcomes, for which the evidence is of very low certainty.

The generally very low‐certainty evidence available for supported discharge and multidisciplinary home rehabilitation means that we are very uncertain whether the findings of little or no difference for all outcomes between the intervention and usual care is true.

Given the prevalent clinical emphasis on early discharge, we suggest that research is best orientated towards early supported discharge and identifying the components of multidisciplinary inpatient rehabilitation to optimise patient recovery within hospital and the components of multidisciplinary rehabilitation, including social care, subsequent to hospital discharge.

Plain language summary

Are mixed‐speciality (multidisciplinary) teams the best way to help older people recover from hip fracture?

Key messages

‐ Compared with usual care, care provided by a team of healthcare professionals with different specialities (multidisciplinary rehabilitation teams) led by a geriatrician or other medical specialist may help more older people in hospital to recover after a hip fracture.

‐ We don’t know if multidisciplinary rehabilitation is better than usual care after hospital discharge because there is insufficient evidence.

‐ Future research should aim to identify the best treatments provided by the mix of specialties that make up multidisciplinary teams, so people can leave hospital early and be supported in the community.

How are hip fractures treated?

Hip fractures are common but serious injuries in older people. About a third of people with hip fractures die within a year of their injury. People with hip fractures may also have other conditions that slow their recovery. Many do not regain the mobility and independence they had before the fracture, and may need residential care in a nursing home afterwards.

Usual care for people with hip fracture is surgery, followed by therapy on the hospital ward to restore mobility and basic functions of daily living, such as bathing and dressing. This may involve people from other departments in the hospital. However, people with hip fractures need help with a range of activities, and may suffer mental as well as physical problems after surgery. So, a better approach to help their recovery, or ‘rehabilitation’, may be to involve a team of people with expertise from different areas, or ‘disciplines’. They develop a rehabilitation plan for each person, depending on their needs. This ‘multidisciplinary rehabilitation’ team, which is typically led by a doctor specialising in the care of older people or other medical specialist in rehabilitation, could also include other doctors, nurses, physiotherapists, dieticians, social workers and mental health specialists. Multidisciplinary rehabilitation could take place in hospital, in acute (short‐stay treatment) or rehabilitation wards, or in a person's home residence. Some types of multidisciplinary rehabilitation involve early discharge from hospital with support at home.

What did we want to find out?

We wanted to know if multidisciplinary rehabilitation helps people recover better from hip fracture than usual care. We were particularly interested in how many people had a ‘poor outcome’ at up to a year after surgery; that is, whether they died or were more dependent such that they needed to be cared for in a nursing home. We also looked at:

‐ health‐related quality of life;
‐ deaths;
‐ need for help with daily activities;
‐ mobility; and
‐ pain.

What did we do?

We searched for studies that compared multidisciplinary rehabilitation with usual care in older people after hip fracture surgery.

We compared and summarised the results of the studies and rated our confidence in the evidence, based on factors such as study methods and sizes.

What did we find?

We found 28 studies with 5351 older people who’d had hip fracture surgery. They were aged on average from 76.5 to 87 years and most were women.

Main results

Multidisciplinary rehabilitation after surgery compared with usual care in hospital (20 studies):

‐ probably results in fewer cases of 'poor outcome' at 6 to 12 months than usual care. Of 1000 people with hip fracture who received usual care, 347 would have a ‘poor outcome’ between 6 and 12 months after surgery; 41 fewer people (12% of 347) would have a ‘poor outcome’ with multidisciplinary rehabilitation;

‐ may reduce the risk of death in hospital, and in the longer term for up to 1 year.

‐ may result in fewer people with poorer mobility at 6 to 12 months.

We don’t know if multidisciplinary rehabilitation improves or worsens quality of life and activities of daily living. There was no evidence on long‐term, hip‐related pain.

Multidisciplinary rehabilitation after surgery compared with usual care at home (3 studies):

We don’t know if multidisciplinary rehabilitation makes any difference to:

‐ a ‘poor outcome’ at 1 year;

‐ quality of life at 1 year;

‐ deaths at 4 or 12 months;

‐ whether people need help with daily living activities;

‐ moving into a nursing home; or

‐ mobility.

There was no evidence on long‐term, hip‐related pain.

Multidisciplinary rehabilitation after surgery compared with usual care for nursing home residents (1 study) may make no difference to:

‐ a 'poor outcome' (died or unable to walk) at 12 months; or

‐ deaths at 4 or 12 months.

We don’t know if multidisciplinary rehabilitation makes a difference to:

‐ whether people need help with daily living activities at 4 weeks or 12 months;

‐ quality of life, mobility or pain at 12 months.

What are the limitations of the evidence?

We are moderately confident in the evidence about ‘poor outcome’ for people in hospital but much less confident in all the other evidence because less was available and all the studies had weaknesses that could affect the reliability of their results. 

How up to date is this evidence?

The evidence is up to date to October 2020.

Summary of findings

Summary of findings 1. Multidisciplinary inpatient rehabilitation versus usual care.

Multidisciplinary inpatient rehabilitation compared with usual care for older people with hip fracture
Patient or population: older people undergoing rehabilitation soon after surgically treated hip fracture Settings: inpatient care – thus, before hospital discharge. Rehabilitation extended to care in the community in some trials Intervention: multidisciplinary inpatient rehabilitation Comparison: usual care
Outcomes	Illustrative comparative risks* (95% CI)		Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Usual care	Intervention
'Poor outcome' (dead or deterioration in residential status, generally requiring institutional care) Follow‐up: 6 to 12 months	347 per 1000a	306 per 1000 (278 to 340)	RR 0.88 (0.80 to 0.98)	3036 participants (13 studies)	⊕⊕⊕⊝ Moderateb	Overall, 969 participants (31.9%) had a 'poor outcome'.
Quality of life: EQ‐5D (0: dead to 1: best quality)c Follow‐up: 4 months	The mean EQ‐5D score in the usual care group was 0.46	The mean EQ‐5D score was 0.08 higher in the intervention group (0.03 to 0.1 higher)	‐	347 participants (1 study)d	⊕⊝⊝⊝ Very lowe	The trial found a similar though slightly reduced difference at 12 months (MD 0.07, 95% CI 0.02 to 0.12; 337 participants). The CIs of both results included the MCID of 0.08
Mortality (at discharge) Follow‐up: up to hospital discharge	75 per 1000a	58 per 1000 (44 to 78)	RR 0.77 (0.58 to 1.04)	2455 participants (11 studies)	⊕⊝⊝⊝ Lowf	There were 168 deaths (6.8%) in total
Mortality (end of scheduled follow‐up) Follow‐up: 4 to 12 months	182 per 1000a	166 per 1000 (146 to 192)	RR 0.91 (0.80 to 1.05)	3973 participants (18 studies)	⊕⊕⊝⊝ Lowf	There were 682 deaths (17.2%) in total
Greater dependence in personal activities of daily livingg Follow‐up: 1 to 4 months	558 per 1000a	486 per 1000 (424 to 553)	RR 0.87, (0.76 to 0.99)	754 participants (4 studies)	⊕⊝⊝⊝ Very lowh	Evidence from 7 other studies: Three studies providing data for the Barthel Index (in two, a Chinese version was used) found higher (better) ADL scores in the intervention group. However, the clinical importance of the results is uncertain and the difference may only be slight. Four other studies, which provided incomplete or no data, reported no evidence of between‐group differences in ADL.
Greater dependence in personal activities of daily livingg Follow‐up: 6 to 12 months	723 per 1000a See Comment	463 per 1000 (368 to 587)	RR 0.64 (0.51 to 0.81)	238 participants (2 studies)	⊕⊝⊝⊝ Very lowh	Evidence from 11 other studies: The RRs from two other studies reporting related outcomes (incomplete recovery of ADL and mobility; non‐recovery of independent self‐care ability) favoured the intervention but the 95% CI in both cases crossed the line of no effect. Four studies providing data for the Barthel Index (in two, a Chinese version was used) found higher (better) ADL scores in the intervention group. However, the clinical importance of the results is uncertain and the difference may only be slight. One study reported no between‐group difference in the losses in ADL as measured by the Katz Index. Four other studies, which provided incomplete or no data for analysis, reported no evidence of between‐group differences in ADL.
Greater dependency in mobility (e.g. requiring help by another person) Follow‐up: 6 to 12 months	403 per 1000a	335 per 1000 (287 to 395)	RR 0.83 (0.71 to 0.98)	1085 participants (5 studies)	⊕⊕⊝⊝ Lowi	Another study (193 participants) found no significant between‐group difference in mobility assessed using the Short Physical Performance Battery (SPPB) tool at 12 months.
Hip‐related pain Follow‐up: 12 months	See Comment			‐	‐	Pain data specific to injury location were available only for a subgroup of one trial. However, we considered the results were unreliable given the major imbalance in the numbers available in the two groups.
Delirium in hospitalj	509 per 1000a	433 per 1000 (377 to 499)	RR 0.85 (0.74 to 0.98)	980 participants (4 studies)	⊕⊕⊝⊝ Lowi	Only delirium or confusion documented as a complication was considered for this outcome. There were 538 (45%) cases in total. A fifth trial (199 participants) reported significantly fewer participants of the intervention had post‐operative delirium (reported P = 0.003).
*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). ADL: activities of daily living; CI: confidence interval; MCID: minimal clinically important difference; MD: mean difference; RR: risk ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: the true effect is likely to be substantially different from the estimate of effect.

^aThe estimate was the median control group risk across studies (or control group risk if just one study).
^bEvidence downgraded one level for serious risk of bias.
^cEQ‐5D scores are anchored at 1 (full health) and 0 (a state as bad as being dead); negative values are possible and represent health states regarded as worse than being dead. Those who have died are given a zero.
^dTwo other trials, based in Taiwan, reported separate data for 8 domains of a Taiwan version of the SF‐36. The results at 12 months from 391 participants for all domains except bodily pain favoured the intervention (very low‐certainty evidence downgraded one level for serious risk of bias and two levels for very serious indirectness, reflecting the minimal care provided after hospital discharge and the unusually high scores that may reflect a difference in the population compared with elsewhere).
^e Evidence downgraded one level for serious risk of bias, two levels for very serious imprecision (one study only, wide confidence interval includes no clinically important effect).
^fEvidence downgraded one level for serious risk of bias and one level for serious imprecision as the confidence interval crosses the line of no effect.
^gThis was reported based on a diverse set of outcome measures: featured is survivors with greater dependency in ADL scores relative to pre‐fracture ADL scores (e.g. the Barthel Index and the Katz Index).
^hEvidence downgraded one level for serious risk of bias, one level for serious imprecision and one level for serious inconsistency; the last reflecting a general view of the variation of effect in the results from the group of trials reporting this outcome using different and incompatible measures.
ⁱEvidence downgraded one level for serious risk of bias and one level for serious imprecision (wide confidence interval).
^jThis outcome is listed as a critical outcome in the protocol for the Cochrane Programme Grant on hip fracture management.

Summary of findings 2. Supported discharge and multidisciplinary home rehabilitation versus usual care: mainly for people living in their own homes (ambulatory setting).

Supported discharge and multidisciplinary home rehabilitation compared with usual care for older people, mainly living in their own homes, with hip fracture
Patient or population: older people, mainly living in the community, undergoing rehabilitation after surgically treated hip fracture Settings: ambulatory ‐ predominantly in people's own homes Intervention: supported discharge and multidisciplinary home rehabilitation (duration between three to four weeks) Comparison: usual care
Outcomes	Illustrative comparative risks* (95% CI)		Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Usual care	Intervention
'Poor outcome' (dead or moved to higher level of care or unable to walk) Follow‐up: 12 months	188 per 1000a	172 per 1000 (117 to 254)	RR 0.91 (0.62 to 1.35)	377 participants (3 studies)	⊕⊝⊝⊝ Very lowb	Overall, 78 participants (20.7%) had a 'poor outcome'
Quality of life: SF‐36 Physical Component score (0: worst to 100: best) Follow‐up: 12 months	The mean SF‐36 PC score in the usual care group was 33.3	The mean SF‐36 PC score was 4.70 higher in the intervention group (0.43 lower to 9.83 higher)	‐	56 participants (1 study)	⊕⊝⊝⊝ Very lowc	No evidence of a difference was also reported for the mental component score: MD 1.5 (95% CI ‐2.88 to 5.88)
Mortality (at 4 months) Follow‐up: 3 to 4 months	66 per 1000a	51 per 1000 (23 to 117)	RR 0.77 (0.34 to 1.76)	275 participants (2 studies)	⊕⊝⊝⊝ Very lowb	There were 20 deaths (7.3%) in total
Mortality (end of scheduled follow‐up) Follow‐up: 12 months	125 per 1000a	172 per 1000 (77 to 204)	RR 1.00 (0.61 to 1.63)	377 participants (3 studies)	⊕⊝⊝⊝ Very lowb	There were 53 deaths (14.1%) in total
Independent in personal activities of daily living Follow‐up: 12 months	368 per 1000a	241 per 1000 (152 to 380)	RR 1.06 (0.71 to 1.57)	159 participants (1 study)	⊕⊝⊝⊝ Very lowd	Note, this is a positive outcome. Overall, 60 were independent. Another study (93 participants) reported more favourable results for the intervention groupe
Permanent move to higher level of care (hostel, nursing home, sheltered housing) Follow‐up: 12 months	69 per 1000a	25 per 1000 (5 to 117)	RR 0.35 (0.07 to 1.69)	168 participants (2 studies)	⊕⊝⊝⊝ Very lowb	Overall, 8 survivors (4.8%) had moved to a higher level of care
Unable to walk Follow‐up: 12 months	87 per 1000a	78 per 1000 (34 to 178)	RR 0.89 (0.38 to 2.04)	214 participants (2 studies)	⊕⊝⊝⊝ Very lowb	Overall, 19 survivors (8.9%) were unable to walk
Pain (end of follow‐up)	See comment	‐	‐	‐	‐	None of the three studies reported on hip or lower‐limb pain.
*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; MD: mean difference; RR: risk ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

^aThe estimate was the median control group risk across studies (or control group risk if just one study).
^bEvidence downgraded one level for serious risk of bias and two levels for very serious imprecision (few or very few events, wide confidence interval crossing the line of no effect).
^cEvidence downgraded one level for serious risk of bias and two levels for very serious imprecision (one small study only, wide confidence interval crossing the line of no effect).
^dEvidence downgraded one level for serious risk of bias, two levels for very serious imprecision (one study only, wide confidence interval crossing the line of no effect), and one level for inconsistency (this latter is hypothetical but reflects that data from another study (93 participants) were in favour of the intervention group; see below).
^eThis study reported higher FIM (Functional Independence Measure: 0 to 91; independent) scores in the intervention group (median 85 versus 80, reported P = 0.001).

Summary of findings 3. Supported discharge and multidisciplinary home rehabilitation versus usual care: people living in a nursing home setting (ambulatory setting).

Supported discharge and multidisciplinary home rehabilitation compared with usual care for older people, living in nursing homes, with hip fracture
Patient or population: older people from nursing homes undergoing rehabilitation after surgically treated hip fracture Settings: ambulatory ‐ in nursing homes (in Australia) Intervention: supported discharge and multidisciplinary home rehabilitation (4 week programme) Comparison: usual care
Outcomes	Illustrative comparative risks* (95% CI)		Relative effect (95% CI)	No of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Usual care	Intervention
'Poor outcome' (dead or unable to walk) Follow‐up: 12 months	662 per 1000a	689 per 1000 (576 to 821)	RR 1.04 (0.87 to 1.24)	240 participants (1 study)	⊕⊕⊝⊝ Lowb	Overall, 162 participants (67.5%) had a 'poor outcome'
Quality of life: EQ‐5D (0: dead to 1: best quality)c Follow‐up: 12 months	The mean EQ‐5D score in the usual care group was 0.3	The mean EQ‐5D score was 0.06 lower in the intervention group (0.12 lower to 0.0 lower)	‐	235 participants (1 study)	⊕⊝⊝⊝ Very lowd	The difference is unlikely to be clinically important.
Mortality (at 4 months) Follow‐up: 3 to 4 months	281 per 1000a	295 per 1000 (197 to 439)	RR 1.05 (0.70 to 1.56)	240 participants (1 study)	⊕⊕⊝⊝ Lowb	There were 69 deaths (28.8%) in total
Mortality (end of scheduled follow‐up) Follow‐up: 12 months	430 per 1000a	486 per 1000 (370 to 641)	RR 1.13 (0.86 to 1.49)	240 participants (1 study)	⊕⊕⊝⊝ Lowb	There were 110 deaths (45.8%) in total
Functional dependency: modified Barthel Index (0 to 100: total independence in personal care) Follow‐up: 4 weeks	The mean Barthel score in the usual care group was 23.5	The mean Barthel score was 0.90 higher in the intervention group (4.51 lower to 6.31 higher)	‐	202 participants (1 study)	⊕⊝⊝⊝ Very lowe
Functional dependency: modified Barthel Index (0 to 100: total independence in personal care) Follow‐up: 12 months	The mean Barthel score in the usual care group was 23.5	The mean Barthel score was 4.90 lower in the intervention group (11.69 lower to 1.89 higher)	‐	125 participants (1 study)	⊕⊝⊝⊝ Very lowf
Unable to walk Follow‐up: 12 months	87 per 1000a	78 per 1000 (34 to 178)	RR 0.89 (0.38 to 2.04)	130 participants (1 study)	⊕⊝⊝⊝ Very lowf	Overall, 52 survivors (40%) were unable to walk
Pain: PAINAD (0 to 10; severest pain) Follow‐up: 12 months	The mean pain score in the control group was 0.06	The mean pain score was 0.01 lower in the intervention group (0.44 lower to 0.42 higher)	‐	126 participants (1 study)	⊕⊝⊝⊝ Very lowf
*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; MD: mean difference; PAINAD: Pain Assessment In Advanced Dementia; RR: risk ratio
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

^aThe estimate was the control group risk for the one study testing this comparison.
^bEvidence downgraded one level for serious risk of bias and one level for serious imprecision (single study results, wide confidence interval crossing the line of no effect).
^cEQ‐5D scores are anchored at 1 (full health) and 0 (a state as bad as being dead); negative values are possible and represent health states regarded as worse than being dead. Those who have died are given a zero.
^dEvidence downgraded one level for serious risk of bias and two levels for very serious imprecision (one study only, wide confidence interval and large numbers of deaths that scored 0). The study also reported other quality of life data for survivors at 12 months using the DEMQOL (Dementia Quality of Life) instrument (self‐reported form completed by 70 participants) and the DEMQOL‐Proxy instrument (by 126 proxies). Evidence from these measures was also rated as very low certainty, reflecting downgrading one level for serious risk of bias, two levels for imprecision (reduced sample size and wide confidence interval).
^eEvidence downgraded one level for serious risk of bias and two levels for very serious imprecision (single study results, wide confidence interval crossing the line of no effect). Additionally, we would have downgraded one level for indirectness as the timing of follow‐up is too short and just at the end of the treatment period.
^fEvidence downgraded one level for serious risk of bias and two levels for very serious imprecision (low numbers of participants, wide confidence interval crossing the line of no effect).

Background

This review is an update  of Handoll 2009. The scope of this review has been extended to examine different models of multidisciplinary rehabilitation.

Description of the condition

Fracture of the proximal femur, generally termed a 'hip fracture', occurs most often in frail older people and generally as a result of a simple fall from standing height or less (Norton 1997). In higher‐income countries, the mean age of people sustaining a hip fracture is around 84 years for men and 83 years for women (NICE 2011). Around two‐thirds of hip fractures occur in women (NHFD 2017). People sustaining a hip fracture frequently have many other medical and physical problems; these often hinder their recovery and add to the challenge of managing their fracture. Cognitive impairment and dementia are also major considerations, both in terms of risk of hip fracture and recovery from hip fracture. Despite generally successful surgical treatment, hip fracture poses a major threat to life, mobility and independence (Marottoli 1992). About a third of people sustaining a hip fracture die within the first year (Lyons 1997; NICE 2011), often as a direct consequence of their fracture (Parker 1991). Most of the survivors fail to regain their former levels of mobility and activity, and many become more dependent (Magaziner 2000). About 10% to 20% of the survivors will require a change to a more dependent residential status (Dyer 2016).

The burden on society from hip fracture is immense and is increasing. An estimated 1.6 million hip fractures occurred in 2000 (Johnell 2006), with predictions of numbers rising to 7.3 to 21.3 million by the year 2050; the steepest increases being expected in Asia (Odén 2015). The resources required to provide the institutional and community care for those afflicted are already enormous (Schneider 1990). The direct costs of hip fracture were found to exceed acute myocardial infarction in Italy (Piscitelli 2007). There are data suggesting trends in the incidence of hip fracture vary between countries. In some higher‐income countries, the age‐specific incidence of hip fracture is decreasing, but this is probably not the case in most other countries (Ballane 2014). In higher‐income countries, lifetime prevalence has been estimated to be one in six for white women and somewhat less than half of this for white men (Cummings 2002). Inevitably, the absolute disease burden will also increase with population aging, putting further pressure on already stretched healthcare systems. Additionally, very large costs are generated by the additional requirement for institutional care for many people.

Description of the intervention

Most people with hip fracture have surgery (Handoll 2008), after which there are a wide range of treatments that are used to assist recovery (SIGN 2002). Some of these treatments have specific goals, such as restoring mobility (see Handoll 2011 for a systematic review of mobilisation strategies) and independence in other basic daily living functions, such as bathing, dressing and continence. The focus of this review is the delivery and provision of rehabilitation, specifically using a multidisciplinary approach.

For the purpose of this review, 'rehabilitation' is defined as services provided by a multidisciplinary team with the goal of reducing disability by improving ability to complete activities; for example, walking and dressing.

Care provision after hip fracture varies substantially across the world and may be determined by economic and cultural factors (personal or societal). Many strategies have been employed for the rehabilitation of people with hip fracture. Rehabilitation is generally adapted to an individual's general health, disability status, living circumstances and preferences for types of treatment. For instance, after hip fracture, some people are fit enough to return directly home from the surgical ward – albeit, where possible, with the support of services, including home‐based rehabilitation. The majority have, or require, additional treatment in an institutional setting.

There is an increasing trend to provide treatment early in the hospital stay of people with hip fracture, to encourage mobility and self‐care (NICE 2011). Inpatient rehabilitation can be provided in an orthopaedic ward, but it may take place in a separate rehabilitation ward, to where people are transferred from the acute orthopaedic ward upon recovery from their operation. (Further consideration of the setting for inpatient care is discussed in relation to applicability of the included trials in Overall completeness and applicability of evidence.) The rehabilitation ward is an environment designed to optimise individuals’ function by practicing tasks, including mobility, self‐care and domestic activities. This is enabled by the support of staff, assistive equipment including mobility and bathing aids, and establishing daily routines such as eating meals in a dining room. Multiple health disciplines – for example, physiotherapists, occupational therapists, social workers, nurses and doctors – work with the person with hip fracture to provide a co‐ordinated rehabilitation programme. This is facilitated by formal meetings, usually weekly, and regular discussion about goals of treatment with the patient and their family.

Ideally, people are generally discharged from a hospital‐based rehabilitation programme when able to live with acceptable risk in their usual accommodation or an alternative setting. Following discharge from hospital, several different types of rehabilitation programme have been described. Early Supported Discharge (ESD) is a rehabilitation programme provided in patient's home setting. In the USA, people are often discharged to skilled nursing facilities for rehabilitation. Other types of non‐inpatient rehabilitation programme are provided in some circumstances; for example, in a day hospital or in an outpatient department.

There is no agreed taxonomy for classification of multidisciplinary rehabilitation programmes after hip fracture. These programmes have been conceptualised as 'acute rehabilitation' (meaning rehabilitation in an acute care ward following hip fracture), and 'post‐acute rehabilitation' (Sheehan 2019). The post‐acute rehabilitation programme types are complex. They may be provided in an inpatient rehabilitation ward, as an outpatient or day hospital service, at home or in long‐term care. These components vary between countries; in some countries, the inpatient rehabilitation is provided in nursing care facilities. These two major categories of programmes can be connected by transition services; see Figure 1.

1.

Hip fracture rehabilitation services – programme components (extract from Sheehan 2019)

A different and commonly‐used classification of these programme categories is described in Cameron 2000. This classification established the following models.

• (Geriatric) Hip Fracture Programme (GHFP): meaning an orthogeriatric programme incorporating multidisciplinary rehabilitation as well as other components (such as peri‐operative assessment). These programmes may or may not have an ambulatory component.

• Geriatric Orthopaedic Rehabilitation Unit (GORU): a rehabilitation ward that is exclusively for geriatric orthopaedic patients; thus, people with hip fractures or other fragility fractures. This is more common in the UK.

• Mixed Assessment and Rehabilitation Unit (MARU): a rehabilitation ward where older patients from mixed diagnostic groups are accommodated; thus, people with hip fracture or other fragility fractures in addition to older people requiring rehabilitation for other reasons (for example, stroke). This is the more common model internationally.

• Early Supported Discharge (ESD): as mentioned above, this is an ambulatory programme that aims to discharge people with hip fracture at an earlier stage and support them at home while continuing rehabilitation.

Sheehan 2019 has evolved a working definition of a (Geriatric) Hip Fracture Programme, which includes at least an acute rehabilitation component and a post‐acute rehabilitation component. ESD could be considered as a part of a GHFP or as a stand‐alone transitional service. GORU and MARU are categories within post‐acute inpatient rehabilitation programmes. We also identified an additional category, orthogeriatric programme, provided in the acute care setting that involves orthopaedic and geriatric collaboration and aims to improve functioning. Further details are provided in Table 4; see also Appendix 1.

1. Classification of multidisciplinary rehabilitation programmes.

Descriptor	Label	Description
Hip fracture programme	HFP	Fits NICE definition (NICE 2011, also 2017 version); see Appendix 1. Has components in acute care, can involve inpatient rehabilitation and encourages early supported discharge when feasible.
Orthogeriatric programme	OG	Programme is provided in the acute care setting after hip fracture and involves orthopaedic and geriatric collaboration and aims to improve functioning.
Geriatric orthopaedic rehabilitation unit	GORU	Subacute rehabilitation unit for older people with orthopaedic injuries (see Cameron 2000).
Mixed assessment and rehabilitation unit	MARU	Subacute rehabilitation unit for older people with range of diagnoses. Can be called geriatric evaluation and management unit (GEM) in some countries (see Cameron 2000).
Early supported discharge	ESD	Fits NICE definition and aims to assist selected older people return home after hip fracture. Can be to own home or residential aged care facility (RACF). However, in some cases, there may be more of a focus on enhancing support rather than achieving early or accelerated discharge.
Home‐based rehabilitation	HBR	Home‐based rehabilitation after hip fracture without direct link to hospital services
Outpatient rehabilitation	OPR	Outpatient rehabilitation after hip fracture without direct link to hospital services

The Cameron 2000 classification has also been included in subsequent National Institute for Health and Care Excellence (NICE) Guidelines in the UK (NICE 2011), and also adopted, in part, in the Cochrane Programme Grant on the management of hip fracture. This project, aligned with the NICE Guidelines, specifically focuses on the Hip Fracture Programme, with consideration also of designated Hip Fracture Units, and Early Supported Discharge as part of multidisciplinary care.

In summary, the primary intervention tested in this review is specialised multidisciplinary rehabilitation supervised by a geriatrician, rehabilitation physician or other appropriate physician, with usual care, for older people with hip fracture, in either an inpatient rehabilitation setting, an ambulatory rehabilitation setting, or both. The comparator (control) intervention is usual care. In early studies, this was care in the orthopaedic surgical ward. However, in some more recent studies, it is an alternative rehabilitation programme that usually starts later, and is less intensive and co‐ordinated (Halbert 2007). All of these interventions are complex organisational interventions. Emerging models for multidisciplinary rehabilitation for older people with hip fracture have prompted secondary additions to our approach in terms of the different models, and with direct comparisons of different models of multidisciplinary rehabilitation.

How the intervention might work

The primary goal for the person with hip fracture is return to an optimal level of functioning (WHO 2020). To achieve this, specific goals are set, and the therapeutic input required to achieve these is dependent on the co‐ordinated skills of multiple health professionals. While the process has been described and documented by the World Health Organization (WHO), stroke researchers have most clearly described the components of rehabilitation that are potentially associated with effectiveness (Stroke Collaboration 2013). In this review, we aim to establish the effectiveness and cost‐effectiveness of multidisciplinary rehabilitation overall, rather than attempt to evaluate its components. However, we have now expanded our scope to include comparisons of different models of multidisciplinary rehabilitation.

Why it is important to do this review

This review is required because of the very large number of older people who sustain hip fractures, and the substantial impact – both on the individual and to society – of these fractures. It is, therefore, highly important that the effectiveness of the various strategies employed for the rehabilitation of people with these fractures is assessed. Since the previous review (Handoll 2009), there has been a continuation of the international trend to reduction of acute hospital length of stay, with the continuing development of rehabilitation programmes that operate in other settings. We have again included outpatient or 'ambulatory' rehabilitation, to ensure continuing relevance of the review. Additionally, the evolving models of multidisciplinary rehabilitation for hip fracture merit a reframing of the evidence to inform specific consideration of these models.

Objectives

To assess the effects of multidisciplinary rehabilitation, in either inpatient or ambulatory care settings, for older people with hip fracture.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised trials of post‐surgical care comparing specialised multidisciplinary rehabilitation with conventional care. Trials that used a quasi‐randomisation technique (e.g. allocation by date of birth or days of the week) were also eligible for inclusion.

Types of participants

We included older people with any type of fracture of the proximal femur that had been surgically fixed prior to entry into the care programme. We did not define specific age limits, but as anticipated, most participants were aged 65 years and over. We included younger participants (< 65 years), as the number of these was relatively small, and there appeared to be adequate randomisation, with unbiased distribution of this younger population between the intervention and control groups. We excluded trials of proximal femoral fracture confined to younger populations (e.g. age < 60 years) or to people with multiple trauma, including hip fracture. We also planned to exclude trials focused on the rehabilitation of people with metastatic disease or those focused on people with high‐energy fractures, such as from motor vehicle crashes. We included mixed population trials, provided the number of participants in any of these categories (i.e. people under 60 years of age, with multiple trauma, metastatic disease, or with high‐energy fractures) was relatively small, with an unbiased distribution between the intervention and control groups; or separate data were available for the target population of this review.

Types of interventions

For this review, the type of intervention under scrutiny is treatment in a multidisciplinary rehabilitation programme (where rehabilitation is delivered by a multidisciplinary team, supervised by a geriatrician, rehabilitation physician or other appropriate physician, as opposed to 'usual' care (control group)). Such a programme would aim to improve the functioning of the person with hip fracture. The programme needed to have been provided in an inpatient or ambulatory setting (or both). Here, ambulatory setting covers home (including nursing homes), outpatient department and day hospital locations. 'Usual care' for the control group refers to the usual orthopaedic or medical care, or potentially a rehabilitation programme of lesser intensity, or with different components, to the intervention under study. We further classified the interventions as presented in Table 4.

Revisions to this aspect of the scope of the current version of this review are detailed below.

• We relaxed the criteria for multidisciplinary care to explore the inclusion of trials that did not fulfil our more stringent criterion for supervision by a geriatrician or rehabilitation physician, but otherwise featured a strong component of physician involvement, often at the assessment stage. Where possible, we categorised the interventions according to the models stated in the Description of the intervention. We allowed for combined categories; for instance, a geriatric hip fracture programme set within a hip fracture unit.

• We included comparisons of different models of multidisciplinary rehabilitation. For example, HFP versus GORU. In selecting the control intervention, we used the less specific or intensive intervention.

• We set out a pro forma to examine 'usual care'; this included consideration of whether it and care common to both groups were representative of current management; such as early surgery and early mobilisation.

Types of outcome measures

For this update, we restructured the types of outcome measures into three categories: a) 'critical' or main outcomes for presentation, where appropriate, in the summary of findings tables; b) other 'important' clinical outcomes; and c) economic or resource outcomes. The selection of outcomes was in keeping with those identified via the Cochrane Programme Grant on the management of hip fracture (see Appendix 2). We describe these changes in the Differences between protocol and review section. The previous set of outcomes used in Handoll 2009 is shown in Appendix 3.

Another key shift in our reporting is the greater emphasis on interim outcomes at around four months. This change reflects the finding that quality of life and 'poor outcome' (death or deterioration in residential status) at four months are likely to be consistent with those at 12 months (Griffin 2015). We continued to examine the effects of combining mortality data with measures of greater dependency in activities of daily living and mobility.

The following listing, including the differentiation between 'critical' and 'important' outcomes, is orientated at inpatient rehabilitation trials. We made some adjustments, such as to the timing, for trials conducted in the ambulatory setting. Whilst we did not set out variations in the selection of outcomes for different scenarios (e.g. different patient populations or models of rehabilitation model) beforehand, we discussed the rationale for any changes we made without view of the results and thereby avoided selective reporting bias.

Main or 'critical' outcomes

We selected the following main or 'critical' outcomes for presentation in the summary of findings tables and other summary sections of the review.

• 'Poor outcome', defined as death or deterioration of functional status leading to increased dependency in the community or admission to institutional care. Timing between 4 and 12 months.

• Health‐related quality of life scores (e.g. SF‐36 or EQ‐5D, two widely‐used health questionnaires). Preference was given to trial data at around four months' follow‐up, but the choice depended on data availability, reporting of mortality and whether death‐adjusted estimates were provided (Parsons 2018).

• Mortality (all‐cause); at short‐term (around 4 months, but 'at discharge' data were also accepted) and long‐term (around 12 months).

• Dependency in activities of daily living, primarily based on requiring the assistance of another person, but also based on measures such as the Barthel Index and Functional Independence Measure (FIM); at short‐term (around 4 months) and long‐term (around 12 months).

• Mobility; we selected measures in the following order, according to availability: non‐return to former mobility or indoor/outdoor walking status with emphasis on requiring assistance from another person; mean mobility score; self‐reported walking score. Timing between 4 and 12 months.

• Pain: verbal (numerical) rating or visual analogue score (VAS). Similar considerations on timing as above.

Other important clinical outcomes

The following list covers other important outcomes presented in this review. These include data for the main outcomes that were collected at different time points or where additional to those used above (thus, where multiple outcome measures are used for the same outcome: e.g. SF‐36 and EQ‐5D are presented in the same trial; or multiple measures of mobility are presented).

• 'Poor outcome' at hospital discharge.

• Level of care and extent of support required or provided on discharge (inpatient or from rehabilitation programme).

• Residential status at between 4 and 12 months.

• Medical complications / sequelae: pneumonia; thromboembolism (symptomatic deep vein thrombosis or pulmonary embolism); pressure injury (ulcer, sore); urinary tract infection; delirium.

• Readmission.

• Carer burden and stress.

Subsequent changes on conduct of review update

For the subgroup of trials testing ambulatory rehabilitation only, we also included number of fallers (number of people who fell), number of fall‐related hospital admissions and new fractures.

We decided not to report on 'Level of care and extent of support required or provided on discharge (inpatient or from rehabilitation programme)' as this was too variable in timing, often linked with the interventions themselves and potentially confounded data since it could be influenced by participant, care facility and social care circumstances.

Since delirium was identified as a critical outcome in the protocol for the Cochrane Programme Grant on hip fracture management, we specifically reported on this adverse event, including in the summary of findings table for inpatient rehabilitation.

Economic outcomes

We summarised cost‐analyses reported by the included trials. We reviewed each trial report for costs and resource data that would enable economic evaluation. Data for the following outcomes were routinely collected.

• Length of hospital stay.

• Overall length of stay in secondary care.

• Subsequent admission rates to, and days spent in, institutional care.

Search methods for identification of studies

Electronic searches

For this update, we searched the Cochrane Bone, Joint and Muscle Trauma Group Specialised Register, the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, MEDLINE In‐Process & Other Non‐Indexed Citations and Embase. We did not apply any language restrictions.

The searches up to January 2016 were carried out in two stages. The first search was run from 2009 (the date of the last published search) to August 2013. A second, top‐up search was then run up to 25 January 2016. We also searched ClinicalTrials.gov (January 2016) and the WHO International Clinical Trials Registry Platform (February 2016) for ongoing and recently completed trials. The search was then extended to 25 February 2019. However, the search for CENTRAL was run without a lower date limit to check for trials that might have been excluded on title and abstract alone that were eligible for inclusion under the revised scope. We also extended our search of ClinicalTrials.gov (24 April 2019) and the WHO International Clinical Trials Registry Platform (27 April 2019) for ongoing and recently completed trials. Subsequently, the search was updated to 21 November 2019 (from 1 January 2019) and for CENTRAL, MEDLINE and Embase to 15 October 2020.

The topic‐specific MEDLINE search was combined with the sensitivity‐maximising version of the Cochrane Highly Sensitive Search Strategy for identifying randomised trials (Lefebvre 2011). In Embase, a more precise randomised controlled trial (RCT) filter was used for the second‐stage search. This filter was developed for retrieving RCTs in Embase to add to CENTRAL. Search strategies up to January 2016 for CENTRAL, MEDLINE, Embase and the trial registers are shown in Appendix 4. Those for the extended search up to February 2019 are shown in Appendix 5; the same strategies were used for the search updates on 21 November 2019 and 15 October 2020. Details of previous search strategies are published in the first version of this review (Handoll 2009).

Notably, in the first search of the second stage, we searched our previous set of excluded studies for those that met our revised scope, in terms of a lower intensity or level of organisation of multidisciplinary care and comparisons of different models of multidisciplinary rehabilitation.

Searching other resources

For completeness, other searches from the 2009 version of the review are presented in Appendix 6. When screening the CENTRAL search up to 2000, we checked through a non‐Cochrane Review on rehabilitation following fractures in older people (Cameron 2000). We also screened records of hip fracture rehabilitation trials identified up to November 2018 as part of work on the Cochrane Programme Grant on the management of hip fracture. We checked through the reference lists of all trials and the four systematic reviews we identified in our search (Moyet 2018; Nordstrom 2018; Wang 2015; Wu 2018).

Data collection and analysis

A summary of key changes in this version and the previous version (Handoll 2009), and the published protocol (Cameron 2008), is presented in Differences between protocol and review.

Selection of studies

For the search up to January 2016, initial screening was conducted by pairs of review authors (IC, HH, JM, CP). Subsequently, all five review authors (thus, including TF), independently assessed potentially eligible trials identified via the search for inclusion using a pre‐piloted form. Any disagreement was resolved through discussion. It was not found necessary to seek further information about study methods and interventions from trialists to determine eligibility.

For the search update to February 2019 for the revised review scope, three review authors (IC, HH, JM) conducted initial independent screening. One review author (HH) processed the results of this screening and prepared the documentation for study selection, which was independently performed by the same three review authors. All disagreement was resolved through discussion.

For the search update to October 2020 for the revised review scope, one review author (HH) conducted initial screening. The same review author processed the results of this screening and prepared the documentation for study selection, which was independently performed by HH and IC. All disagreement was resolved through discussion.

Data extraction and management

Using a pre‐piloted data extraction form, pairs of review authors independently extracted trial details and data for the trials not already included in Handoll 2009. Two review authors (always including HH and IC) performed data extraction for all trials, and checked this against data extraction forms provided by the other authors and, where appropriate, the data presented in Cameron 2000, Cameron 2001, Halbert 2007 and Handoll 2009. We sought key additional information from trialists, including the method of randomisation. There was no need to extract results from graphs in trial reports, although this would have been considered where data were not otherwise available. Decisions for data aggregation for calculating 'poor outcome', the designated primary outcome of this review, from mortality and functional data (typically relating to greater dependency) presented in individual trials were made by consensus (IC, HH and JM).

We collected results for the final follow‐up time available, as well as at intermediate assessments, at discharge and at around four months.

We collected the following information and data for each study for presentation in the Characteristics of included studies, for use in risk of bias assessment and assessment of indirectness (applicability), and for summarising elsewhere in the review. We noted where key information was not available.

• Study characteristics: study design; study dates (period of recruitment); setting(s) including hospital type (teaching/university, trauma centre, general), number of centres, country/countries; sample size calculation and primary outcome; length of follow‐up and interim follow‐ups; funding and declaration/conflict of interest.

• Population characteristics: participant inclusion and exclusion criteria; the number of recruited participants, overall and by study arm; baseline demographics (e.g. gender, age, types of fracture, American Society of Anesthesiology (ASA) physical status, cognitive impairment, mixed populations (multi‐trauma, other fractures or injuries, other diagnoses)); surgical treatment.

• Interventions and usual care: full details of the interventions including components, timing, physical location/environments and staff profiles, including for categorisation of rehabilitation model purposes; full details of common care, including major confounders such as time of surgery and mobilisation policy (early mobilisation as opposed to extended bed rest).

• Details of outcomes and outcome measurement.

• Economic and health resource information.

Assessment of risk of bias in included studies

Pairs of review authors independently assessed risk of bias, without masking of the source and authorship of the trial reports, for the new trials included in this update. One review author (HH or IC) checked consistency in assessment at data entry. We used the tool outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008, hereafter referred to as the Cochrane Handbook). This tool incorporates assessment of randomisation (sequence generation and allocation concealment), blinding (of participants, treatment providers and outcome assessors), completeness of outcome data, selection of outcomes reported and other sources of bias. We considered 'hard' outcomes (e.g. death, admission to institutional care, readmission to hospital) and other outcomes (e.g. functional status) separately in our assessment of blinding and completeness of outcome data. We assessed three additional sources of bias: selection bias resulting from imbalances in key baseline characteristics (e.g. cognitive function, prior care); performance bias, such as resulting from lack of comparability in the experience of care providers; and ascertainment bias, such as differences in timing of follow‐up assessment.

Measures of treatment effect

We present quantitative data reported in individual trial reports for outcomes listed in the inclusion criteria in the text and the analyses, using risk ratios (RRs) with 95% confidence intervals (CIs) for dichotomous outcomes, and mean differences (MDs) with 95% CIs for continuous outcomes. We used final scores in preference to change scores. We would have used standard mean differences where pooling final scores for continuous outcomes measured using comparable but different scores.

Unit of analysis issues

Although we would have included cluster‐randomised trials, the unit of randomisation in all the included trials was the individual participant. We were alert to other potential unit of analyses issues, such as the re‐inclusion of participants incurring a second hip fracture in the recruitment period (around 10% of survivors sustain bilateral hip fractures within a year), repeated observation from more than one time‐point, and multiple observations for the same outcome (e.g. multiple medical complications). For the sole multi‐arm trial, we presented the data for the two intervention groups separately in the analyses and split up the numbers in the control group in order to avoid double counting.

Dealing with missing data

We approached trial authors for missing or unclear data. Where appropriate, we have performed intention‐to‐treat analyses to include all people randomised to the intervention groups. As planned, we have investigated the effect of dropouts and exclusions by conducting worst and best scenario analyses. We were alert to the potential mislabelling or non‐identification of standard errors and standard deviations. Unless missing standard deviations could be derived from confidence intervals or standard errors, we did not assume values in order to present these in the analyses.

Assessment of heterogeneity

We assessed heterogeneity by visual inspection of the forest plot (analysis) along with consideration of the Chi² test for heterogeneity and the I² statistic (Higgins 2003). We based our interpretation of the I² statistic results on those suggested by Higgins 2011 (Section 9.5.2):

• 0% to 40%: might not be important;

• 30% to 60%: may represent moderate heterogeneity;

• 50% to 90%: may represent substantial heterogeneity;

• 75% to 100%: considerable (very substantial) heterogeneity.

Assessment of reporting biases

We attempted to reduce the impact of reporting bias by conducting an extensive literature search that included inspection of 'grey literature' and pursuit of trials listed in clinical trial registers. Where there were more than 10 trials in a forest plot, we generated funnel plots to visually examine the potential for publication bias.

Data synthesis

Where appropriate, we pooled results of comparable studies using both fixed‐effect and random‐effects models. We decided on the choice of the model reported by careful consideration of the extent of heterogeneity and whether it could be explained, in addition to other factors, such as the number and size of included studies. We used 95% CIs throughout. We considered not pooling data where there was considerable heterogeneity (I² statistic value of greater than 75%) that could not be explained by the diversity of methodological or clinical features amongst trials. Where it was inappropriate to pool data, we have presented trial data in the analyses or tables for illustrative purposes and report these in the text. In the absence of cluster‐randomised trials, we found it unnecessary to use the generic inverse variance to pool data (Deeks 2005).

The primary analysis is based on the rehabilitation setting: inpatient or ambulatory. We specified a priori that, unless predominantly delivered in an ambulant setting, care spanning over both settings or combinations of inpatient and ambulant settings would be initially included in the inpatient category.

Subgroup analysis and investigation of heterogeneity

Our primary analysis is by setting: inpatient and ambulatory. As an exploratory analysis, we pooled the data from both settings; however, we restricted data from the ambulatory setting to supported discharge and home‐based rehabilitation. In our protocol, we stated that our planned subgroup analyses, which could have been based on the primary setting or the combined setting data, would be by the stage of rehabilitation; participant cognitive function (one measure of this would be the ability versus inability to give individual consent); and pre‐injury residence or dependency status (own home/independent; dependent/nursing home or institutional care). We also stated that presentation in separate subgroups would be considered where there is a fundamental difference in intervention, including types of service provision. Based on our development of the classification for multidisciplinary rehabilitation, we established more transparent categories for subgroup analysis by types of intervention. In the event, we only conducted formal subgroup analyses based on different models of the intervention. We have also presented data stratified by risk of selection bias.

We investigated whether the results of subgroups were significantly different by inspecting the overlap of CIs and performing the test for subgroup differences available in Review Manager 5 (Review Manager 2014).

Sensitivity analysis

Where possible, we planned to do sensitivity analyses examining various aspects of trial and review methodology. As previously, these included checking the effects of missing data, and whether there was selection bias due to major imbalances in participant characteristics in the comparison groups. We considered this second aspect in the context of judgements for selection bias for random sequence generation and allocation concealment. For this update, we also conducted sensitivity analyses to explore the differences between using random‐effects and fixed‐effect models for pooling for the main outcomes with pooled data from five or more trials. We also planned to examine the effects of excluding studies testing interventions that are not currently available worldwide or whose other care is not generally representative of current management, or both.

Summary of findings and assessment of the certainty of the evidence

We used the GRADE approach to assess the certainty of the evidence related to all outcomes listed in the Types of outcome measures (Schünemann 2011). The four levels of evidence certainty are 'high', 'moderate', 'low' or 'very low'. Certainty may be downgraded due to study limitations (risk of bias), imprecision, inconsistency, indirectness or publication bias.

We prepared three summary of findings tables, including for our main comparison of inpatient rehabilitation, and presented the results for each of the critical outcomes listed in Types of outcome measures. The outcomes presented in the summary of findings tables were context‐specific. We added another row for delirium for inpatient rehabilitation, as this was identified as a critical outcome for the Cochrane Programme Grant on the management of hip fracture, in anticipation of informing the update of the NICE guidelines. One review author (HH) produced the summary of findings tables generated in Review Manager 5 (Review Manager 2014). These tables, including the GRADE assessments, were discussed with a second review author (IC), and then shared with the other review authors.

Results

Description of studies

Results of the search

We conducted the search for this updated review (2021) in two main phases. The full search for the first phase, prior to the change in the scope of the review, was from April 2009 to January 2016. Detailed results of this phase are provided in Appendix 7, and the study flow is summarised in Figure 2. Overall, after the first phase, there were 19 included studies, 29 excluded studies, 4 ongoing trials and one study awaiting classification. 

2.

Study flow diagram for the first phase of the search update 2009 to January 2016 (former scope)

We used the same search strategy, updated to February 2019, for the second phase. Although we felt it unlikely that we had missed any studies, we searched the CENTRAL search from database inception. We also added in all 29 studies previously excluded on the basis of the 2016 selection criteria to facilitate a systematic application of the revised selection criteria. We provide detailed results of this second phase in Appendix 7, and the study flow is summarised in Figure 3. This includes the results of search extensions to 21 November 2019 and then 15 October 2020; with one further report found in January 2021 that confirmed exclusion of a study. The second phase resulted in nine newly included studies, four of which had been previously excluded in Handoll 2009. For completeness, the 'Results of the search' section from Handoll 2009 is copied in Appendix 8.

3.

Study flow diagram for the second phase of the search updates up to February/March 2019, November 2019 and October 2020 (revised scope)

Overall, the search update resulted in 15 newly included studies and 13 additional articles pertaining to four studies already included in Handoll 2009. The 15 newly included trials are: Ashe 2019; Baroni 2019; Chong 2013; Crotty 2019; Jette 1987 (previously excluded); Karlsson 2016 (was Gustafson 2009, previously in 'studies awaiting classification' (SAC); Marcantonio 2001 (previously excluded); Prestmo 2015; Sanchez Ferrin 1999 (previously excluded); Shyu 2013a; Singh 2012; Tseng 2019; Uy 2008 (previously in SAC); Watne 2014; and Ziden 2008 (previously in SAC).

Overall, after completing the searches, there are now 28 included studies, 47 excluded studies and 7 studies awaiting classification.

Included studies

Individual trial details of the methods, participants, interventions and outcome measurements for the 28 included trials are presented in the Characteristics of included studies. Appendix 9 summarises the participant characteristics and our assessment of the representativeness of study populations. Appendix 10 summarises early treatment, such as timing and type of surgery, set into the context of the setting and timing of the studies, and our assessment of the representativeness in terms of current practice.

Design

Except for Shyu 2013a, which had three intervention groups, the included trials randomised individual patients into one of two intervention groups. Two trials used quasi‐randomised methods (Chong 2013; Jette 1987).

Sample sizes

The 28 included trials involved a total of 5351 randomised participants, the great majority of whom had undergone surgery for hip fracture. Sample size at randomisation ranged from 11 participants recruited into Uy 2008, to 397 into Prestmo 2015.

Setting

The 28 trials took place in one of 11 countries: Australia (6 trials); Canada (2); Finland (1); Italy (1); Norway (2); Singapore (1); Spain (2); Sweden (4); Taiwan (3); UK (4); and USA (2). The majority were single‐centre trials in terms of recruitment, with the exception of Ashe 2019, Crotty 2003, Crotty 2019 and Singh 2012. Recruitment start dates ranged from 1983 (Jette 1987), to 2012 (Crotty 2019); no information was available for Kennie 1988, and Marcantonio 2001.

Care was provided in an inpatient setting – thus, before hospital discharge – for both intervention groups in 21 trials. Care in the intervention group continued for three to four months, with home visits by a geriatric nurse as part of the intervention, for the three trials conducted in Taiwan (Shyu 2008; Shyu 2013a; Tseng 2019). In three other trials, care was mixed hospital and community care but the community component applied only to the intervention group as follows:  in Crotty 2003, the intervention group participants were discharged from acute care within 48 hours of randomisation; in Karlsson 2016, the intervention involved early discharge and home rehabilitation; and in Ziden 2008, where home rehabilitation was provided after early supported discharge from hospital. Care was provided in the community in four trials: Ashe 2019 at outpatients; Crotty 2019 at nursing homes; Ryan 2006 at home; and Singh 2012 at outpatients and home.

Participants

The mean age of trial participants ranged from 76.5 years in Shyu 2013a to 87 years in Crotty 2019, where all participants were from nursing homes. All but four trials set a lower age limit for trial entry: this ranged from 50 in Cameron 1993 to 70 in five trials (Crotty 2019; Karlsson 2016; Naglie 2002; Prestmo 2015; Stenvall 2007a). Twelve trials set 65 as the lowest age for participation. Four trials included only women (Fordham 1986; Gilchrist 1988; Kennie 1988; Uy 2008). The proportion of men in the remaining trials varied from 17% in Cameron 1993, to 37% in Ashe 2019.

Seven trials explicitly or implicitly excluded people with dementia or cognitive impairment (Ashe 2019; Crotty 2003; Fordham 1986; Galvard 1995; Ryan 2006; Swanson 1998; Ziden 2008). Three trials explicitly excluded people with severe cognitive impairment (Shyu 2008; Shyu 2013a; Tseng 2019); however, a third of the participants of Shyu 2008 had mild cognitive impairment. Jette 1987 did not report on this characteristic. Of the 11 trials documenting dementia prior to the fracture or baseline, the proportion of the trial population with dementia ranged from 5% in Singh 2012, to 77.5% in Crotty 2019. The other six trials documenting cognitive impairment reported this in different ways: 48% of the study population of Cameron 1993 were cognitively impaired; 55% of the study population of Gilchrist 1988 had reduced cognitive scores indicating some cognitive impairment; 47% of Kennie 1988 and 40% of Marcantonio 2001 had mild or greater cognitive impairment; 26% of Naglie 2002 had mild to moderate cognitive impairment but the numbers with more severe impairment were not provided. All 11 participants of Uy 2008 had moderate or severe cognitive impairment.

In Tseng 2019, all 176 participants had diabetes mellitus, the treatment of which was a specific focus of the intervention.

There was no selection based on pre‐fracture residence in 14 trials (Baroni 2019; Cameron 1993; Fordham 1986; Gilchrist 1988; Jette 1987; Karlsson 2016; Kennie 1988; Marcantonio 2001; Naglie 2002; Sanchez Ferrin 1999; Singh 2012; Stenvall 2007a; Vidan 2005; Watne 2014). Two of these studies excluded people whose fracture had occurred in hospital (Kennie 1988; Stenvall 2007a). Twelve trials included only people living in the community (Ashe 2019; Chong 2013; Crotty 2003; Galvard 1995; Huusko 2002; Prestmo 2015; Ryan 2006; Swanson 1998; Shyu 2008; Shyu 2013a; Tseng 2019; Ziden 2008), and except for Prestmo 2015 and Swanson 1998, exclusively in their own home. Crotty 2019 and Uy 2008 included only people who were living in nursing homes before their fracture.

Interventions

All trials, except Chong 2013, compared multidisciplinary rehabilitation (intervention) with a control group ('usual care'). Chong 2013 compared two types of multidisciplinary rehabilitation. The nature of multidisciplinary rehabilitation varied considerably in the included trials. Similarly, there was variation in 'usual care', where described. In 21 trials, multidisciplinary rehabilitation was provided primarily in an inpatient setting. Ambulatory multidisciplinary rehabilitation was tested in seven trials. Table 5 summarises the interventions, split by setting, and usual care for the 28 included trials. Further details of the interventions tested in the individual trials are provided in Characteristics of included studies.

2. Interventions tested in the included studies.

Study ID	Setting	Country; trial start	Intervention	Control	Comments
Inpatient rehabilitation
Baroni 2019	Hospital	Italy; 2012	Orthogeriatric programme	Orthopaedic care; referral to geriatrician as required
Cameron 1993	Hospital	Australia; 1989	Hip fracture programme	Usual care (56% had multidisciplinary rehabilitation)	Intervention emphasised accelerated rehabilitation
Chong 2013	Hospital	Singapore; 2004	Mixed assessment and rehabilitation unit plus structured assessments and checklists	Mixed assessment and rehabilitation unit	Comparison of two types of multidisciplinary rehabilitation
Fordham 1986	Hospital	UK; 1984	Geriatric orthopaedic rehabilitation unit	Orthopaedic management (geriatrician available for advice) including decision to transfer to orthopaedic rehabilitation unit.
Galvard 1995	Hospital	Sweden; 1988	Geriatric orthopaedic rehabilitation unit	Usual orthopaedic care
Gilchrist 1988	Hospital	UK; 1984	Geriatric orthopaedic rehabilitation unit	Usual orthopaedic care in orthopaedic ward. Referral to geriatrician by letter
Huusko 2002	Hospital	Finland; 1994	Orthogeriatric programme	Discharge to local community hospitals, treatment by general practitioners
Jette 1987	Hospital	USA; 1984	Orthogeriatric programme	Usual care: 'standard rehabilitation'
Kennie 1988	Hospital	UK; before 1986	Geriatric orthopaedic rehabilitation unit	Routine orthopaedic care in orthopaedic admission ward
Marcantonio 2001	Hospital	USA; before 1999	Orthogeriatric programme	Orthopaedics team management, including internal medicine or geriatrics consultations on a reactive basis.	The geriatrician input mainly related to reducing delirium
Naglie 2002	Hospital	Canada; 1993	Orthogeriatric programme	Usual care on orthopaedic units which included access to geriatric consultation
Prestmo 2015	Hospital	Norway; 2008	Hip fracture programme	Usual orthopaedic care provided in an orthopaedic ward. Assessment by geriatrician by request only
Sanchez Ferrin 1999	Hospital	Spain; 1996	Mixed assessment and rehabilitation unit (single session of geriatric review)	Usual care under Orthopaedics and Traumatic Surgery Service; consultations to other specialists as required	Low level of geriatrician involvement with emphasis on treating co‐morbidity
Shyu 2008	Hospital	Taiwan; 2001	Hip fracture programme, included home visits	Usual care on trauma or orthopaedic ward. Some consultations of other disciplines	No care provided after discharge at approximately one week in control group
Shyu 2013a	Hospital	Taiwan; 2005	Hip fracture programme (+ home visits): 2 groups 1. HFP + health‐maintenance interventions to manage depressive symptoms and malnutrition and prevent falls 2. HFP	Usual care by orthopaedists.	No care provided after discharge at approximately one week in control group Separate analysis of the two HFP groups not done in this review
Stenvall 2007a	Hospital	Sweden; 2000	Geriatric orthopaedic rehabilitation unit	Usual care on specialist orthopaedic ward	Intervention included an assessment at 4 months by geriatric team
Swanson 1998	Hospital	Australia; 1994	Hip fracture programme	Standard orthopaedic management, geriatrician on referral	Intervention, which included accelerated rehabilitation programme, was based in an orthopaedic ward
Tseng 2019	Hospital and community (intervention group)	Taiwan; 2010	Hip fracture programme, included home visits	Usual care under orthopaedics, with consultations to internal medicine as required	Trial and intervention focused on diabetes‐specific care. No care provided after discharge, around 4 to 5 days after surgery, in control group
Uy 2008	Hospital	Australia; 2001	Hip fracture programme	Usual care: discharge back to the nursing home soon after surgery	All nursing home residents
Vidan 2005	Hospital	Spain; 1997	Orthogeriatric programme	Usual orthopaedic care, counselling from different specialists
Watne 2014	Hospital	Norway; 2009	Hip fracture programme	Usual care on orthopaedic ward, relevant specialists were seen on request
Ambulatory rehabilitation
Ashe 2019	Community (outpatients and home)	Canada; 2011	Outpatient rehabilitation	Usual post‐op rehabilitation	Intervention focused on falls and fracture risk
Crotty 2003	Hospital and community (intervention group)	Australia; 1998	Early supported discharge, included home‐based interdisciplinary rehabilitation	Usual or 'conventional' care, involving routine interdisciplinary hospital care and rehabilitation in hospital.	The extra multidisciplinary rehabilitation component is the home‐based interdisciplinary rehabilitation
Crotty 2019	Community (nursing home / care facility)	Australia; 2012	Early supported discharge to residential aged care facility (nursing home)	Usual care according to usual practice in the nursing home	All nursing home residents. A detailed description of usual care is available.
Karlsson 2016	Hospital and community (intervention group)	Sweden; 2008	Early supported discharge (continuation of rehabilitation in patient's home)	Same inpatient geriatric care and rehabilitation	The extra multidisciplinary rehabilitation component is the home‐based interdisciplinary rehabilitation
Ryan 2006	Community (at home)	UK; 2000	Home‐based rehabilitation: intensive	Home‐based rehabilitation: less intensive	Comparision of two intensities of multidisciplinary rehabilitation
Singh 2012	Community (outpatients and home)	Australia; 2003	Hip fracture programme (included home visits)	Standard care included orthogeriatric care, rehabilitation service, other medical and allied health consultation as required	Intervention extended to multidisciplinary treatment of frailty
Ziden 2008	Hospital and community (intervention group: home)	Sweden; 2004	Supported discharge, included home rehabilitation ward and home visits	Usual care in geriatric ward with no structured rehabilitation after discharge	Geriatrician management applied to both groups

Inpatient rehabilitation

The types of multidisciplinary rehabilitation programmes, where compared to usual care in an inpatient setting, fell into four categories, as defined in Table 4.

• Hip fracture programme: tested in eight trials (Cameron 1993; Prestmo 2015; Shyu 2008; Shyu 2013a; Swanson 1998; Tseng 2019; Watne 2014; Uy 2008).

• Orthogeriatric programme: tested in six trials (Baroni 2019; Huusko 2002; Jette 1987; Marcantonio 2001; Naglie 2002; Vidan 2005).

• Geriatric orthopaedic rehabilitation unit: tested in five trials (Fordham 1986; Galvard 1995; Gilchrist 1988; Kennie 1988; Stenvall 2007a).

• Mixed assessment and rehabilitation unit: tested in one trial (Sanchez Ferrin 1999).

Three trials were distinct from the other 17 trials. In Marcantonio 2001, geriatrician input was mainly related to reducing delirium; in Sanchez Ferrin 1999, geriatrician involvement was low level and focused on treating co‐morbidity; and in Tseng 2019, the intervention was focused on diabetes‐specific care.

Usual care also varied between studies and was not well described in several, especially the older trials. Perusal of Appendix 10 shows the lack of information available for key aspects of early treatment. Of particular note is the minimal nature of usual care provided in three trials conducted in Taiwan, where care ceased after hospital discharge at around one week (Shyu 2008; Shyu 2013a; Tseng 2019).

Two multidisciplinary care interventions were tested versus usual care in Shyu 2013a. Although we present the data for these separately in the analyses, we considered the addition of health‐maintenance interventions to one intervention group did not make the intervention a distinctly different multidisciplinary care programme.

In contrast, we considered the addition of an 'integrated care pathway' to one of the two multidisciplinary rehabilitation interventions tested in Chong 2013 did result in two markedly different types of multidisciplinary rehabilitation. Both interventions were in a mixed assessment and rehabilitation unit setting.

Ambulatory rehabilitation

The multidisciplinary rehabilitation intervention was predominantly in the community in seven trials.

Of four trials testing supported discharge and home rehabilitation, all or the majority of participants in three trials received rehabilitation in their own homes (Crotty 2003; Karlsson 2016; Ziden 2008), whereas the intervention was provided in a nursing home setting for all participants of Crotty 2019. The interventions are summarised below.

• Crotty 2003 compared accelerated discharge from hospital, within 48 hours of randomisation, and home‐based interdisciplinary rehabilitation with usual care involving routine interdisciplinary hospital care and rehabilitation in hospital. The rehabilitation programme at the patient's home focused on early resumption of self‐care and domestic activities. The average duration of home rehabilitation was three weeks (20.3 days).

• Crotty 2019 compared a four‐week ambulatory geriatric multidisciplinary rehabilitation programme delivered in a nursing home setting with usual care, where treatment was provided according to usual practice in the nursing home.

• Karlsson 2016 compared early discharge from hospital with extended multidisciplinary home rehabilitation versus usual care involving routine interdisciplinary hospital care and rehabilitation in hospital. The median duration of the home rehabilitation was three weeks (21 days).

• Ziden 2008 compared home rehabilitation with conventional rehabilitation. Both groups were admitted to rehabilitation wards, but the home rehabilitation group had supported discharge and a three‐week home rehabilitation programme that emphasised self‐efficacy for independent functioning.

Ryan 2006 compared intensive (six visits per week) with less intensive (three or fewer visits per week) multidisciplinary rehabilitation in the participant's own home.

Multidisciplinary rehabilitation was totally or mainly provided at an outpatient clinic in two trials (Ashe 2019; Singh 2012). However, these trials tested otherwise very different interventions. Ashe 2019 compared a post‐discharge geriatrician‐led multidisciplinary clinic, held at 3 to 12 months post fracture, with referral for further rehabilitation as necessary, versus usual care. Singh 2012 compared 12 months of geriatrician‐led multidisciplinary care,  which included high‐intensity weight‐lifting exercise, in an outpatient hospital programme and targeted treatment of frailty at outpatients and home versus usual rehabilitation care.

Outcomes

Only 'poor outcome', our primary outcome, is considered here. We defined this a priori as death or deterioration of functional status leading to increased dependency in the community or admission to institutional care. Where data were available, 'poor outcome' was presented for long‐term follow‐up (between 4 and 12 months) and at hospital discharge. The definitions of deterioration in residential status varied. For some trials, it was based on the requirement for institutional care; in others, the non‐return to home or independent living. Using this definition, deterioration was not measured in the two trials that focused on people from nursing homes who returned to nursing homes (Crotty 2019; Uy 2008). In Crotty 2019, deterioration of functional status was recast in terms of the inability to walk. The same applied to another ambulatory rehabilitation trial, Karlsson 2016, as data on change of residence were unavailable. The definition of poor outcome for individual trials is provided as footnotes in the analyses.

Funding and financial conflicts of interest

Twenty‐one trials acknowledged external funding, and one study acknowledged external support, from public funding bodies or foundations. Six trials did not mention funding (Baroni 2019; Chong 2013; Galvard 1995; Gilchrist 1988; Sanchez Ferrin 1999; Uy 2008). Eleven trials provided explicit statements of no financial conflicts of interest, and 15 trials did not provide statements. Of the two remaining trials, Crotty 2019 stated there was no "relevant conflict", whilst acknowledging funding from industry for other research, and Singh 2012 reported independence from funding organisations and sponsors.

Excluded studies

We list the reasons for exclusion of 47 studies in the Characteristics of excluded studies. The primary reasons for exclusion were: not randomised treatment allocation (15 trials); not multidisciplinary rehabilitation (26 trials); and mixed population without separate data for people with hip fracture (6 trials). Extra consideration of the study design was needed for four of the 15 studies excluded because treatment allocation was not randomised (Boyer 1986; Gonzalez‐Montalvo 2010; Lahtinen 2015; NCT03430193). The first three reported some randomised or quasi‐randomised aspect to their trial but the actual study design, as described in the reasons for exclusion, precluded actual or appropriate random allocation to the treatment groups. The trial registration document for NCT03430193 implied random allocation, but the published protocol and two reports purporting to be about the trial contradicted this implication; they also included major inconsistencies in relation to other details. We excluded Boyer 1986, a mixed population trial, also because of the lack of extractable data on people with hip fracture. Several of the trials ineligible for this review (because they were not testing multidisciplinary rehabilitation) appear in other published Cochrane Reviews on hip fracture rehabilitation: Crotty 2010 includes five excluded trials (Allegrante 2001; Hagsten 2004; Huang 2005; Krichbaum 2007; Tinetti 1999), and Handoll 2011includes Binder 2004. The pending update of Handoll 2011 is likely to include two other trials aimed at improving mobility: Edgren 2015 and Suwanpasu 2013.

Studies awaiting classification

Details of the seven trials in this category are presented in the Characteristics of studies awaiting classification. As detailed in this table, we are unsure of the status of Baroni 2016 (inpatient comparison with a target of 150 participants), or its link with Baroni 2019, despite receiving responses from the contact trialist. We require clarification on the interventions for the four primarily inpatient trials that are insufficiently reported, either in trial registration documentation (NCT01934946, reported sample size of 130 participants; NCT03894709, intended sample size of 304 older people with hip fracture and cognitive impairments, plus their carers), or conference abstracts (Hammond 2017, involved 282 people with hip fracture and cognitive impairment; Wu 2019, involved 157 older people with hip fracture). We also require clarification on the intervention tested in NCT04207788 (intended sample size of 108 participants), which appeared to be mainly delivered in the home setting. Lastly, we require separate data for 335 people with hip fracture for Parsons 2019, which included 403 participants in hospital because of injury.

Risk of bias in included studies

We present a summary of our assessments of the risk of bias associated with various aspects of study design and conduct in Figure 4.

4.

Summary of review authors' assessments (+ = low; ? = unclear; ‐ = high risk of bias) for aspects of study conduct for individual trials

Allocation

We considered 15 trials at low risk of bias for randomisation sequence generation and allocation concealment. Insufficient information on sequence generation prevented judgement of this aspect for four trials (Fordham 1986; Gilchrist 1988; Vidan 2005; Ziden 2008); and on allocation concealment for nine trials (Baroni 2019; Galvard 1995; Gilchrist 1988; Marcantonio 2001; Sanchez Ferrin 1999; Shyu 2008; Shyu 2013a; Swanson 1998; Ziden 2008). Two trials used quasi‐randomised methods and were deemed at high risk of selection bias (Chong 2013; Jette 1987). For the other trials, there were no serious concerns regarding risk of selection bias from inappropriate methods of treatment allocation.

Blinding

Blinded assessors for at least some outcomes were reported for 11 trials (Ashe 2019; Chong 2013; Crotty 2003; Marcantonio 2001; Naglie 2002; Prestmo 2015; Ryan 2006; Stenvall 2007a; Uy 2008; Vidan 2005; Watne 2014). However, the success of this blinding was rarely ascertained, and was found not to be successful in "5‐10% of the cases" in Watne 2014, where the research nurses were unblinded because of information from the trial participants or their relatives. Trial participants were stated to be blinded to treatment allocation in Shyu 2008, Shyu 2013a and Tseng 2019, but we considered this was incompatible with informed consent for Tseng 2019. Despite lack of blinding, many trials clearly applied systematic methods for data collection and often also had independent assessors. We considered the outcomes of death, residence and readmission to be less influenced by lack of blinding than other outcomes, such as function and quality of life measures.

For the first set of outcomes, we considered 12 trials at low risk for these biases, mainly reflecting blinded assessment or systematic assessment of independently‐collected medical record data, and we judged only one trial, Baroni 2019, at high risk. For functional and other related outcomes, all three trials considered at low risk for these biases appeared to have successful blinded assessment (Ryan 2006; Sanchez Ferrin 1999; Vidan 2005). High risk of bias, mainly reflecting lack of, or unsuccessful, blinding, was judged likely for 10 trials (Ashe 2019; Crotty 2019; Fordham 1986; Jette 1987; Karlsson 2016; Kennie 1988; Marcantonio 2001; Tseng 2019; Watne 2014; Ziden 2008). Baroni 2019 did not report on functional outcomes.

Incomplete outcome data

Participant flow information was generally available, although sometimes only after contact with the trial investigators. There were some concerns for attrition bias for 'hard outcomes' such as mortality, residence and readmission data for nine trials (Baroni 2019; Fordham 1986; Huusko 2002; Jette 1987; Prestmo 2015; Sanchez Ferrin 1999; Shyu 2008; Uy 2008; Ziden 2008). We considered two of these trials to be at high risk of attrition bias for 'hard outcomes': Huusko 2002, where there were post‐randomisation exclusions and data excluded from participants without initial mental scores; and Ziden 2008, where over half of the participants were excluded post‐randomisation. We judged the other 19 trials to be at low risk of attrition bias for these outcomes. In contrast, only four trials were at low risk for functional outcomes (Naglie 2002; Stenvall 2007a; Swanson 1998; Vidan 2005), and nine trials at high risk of bias (Chong 2013; Crotty 2019; Fordham 1986; Galvard 1995; Huusko 2002; Marcantonio 2001; Uy 2008; Watne 2014; Ziden 2008). Large losses to follow‐up, imbalanced loss to follow‐up, missing data, data discrepancies indicating missing data or a combination of these were reasons for most of these assessments. The limitations in the functional outcomes data due to the small sample size was the basis for a high risk of bias in Uy 2008. Baroni 2019 did not report on these outcomes.

Selective reporting

The study protocol and trial registration were available for Prestmo 2015 and Watne 2014, both of which we assessed as low risk for selective reporting bias. The two other trials we assessed as low risk are: Crotty 2019, where all outcomes listed at trial registration were reported; and Fordham 1986, where the comprehensiveness of the reporting for this trial made it very unlikely that selective reporting had occurred. We judged 10 trials to be at high risk of selective reporting bias because of missing outcomes and incomplete reporting (Ashe 2019; Baroni 2019; Huusko 2002; Jette 1987; Karlsson 2016; Marcantonio 2001; Shyu 2008; Shyu 2013a; Singh 2012; Tseng 2019). For the remaining trials, which we assessed as having an unclear risk of reporting bias, it seemed likely that the published report included all expected outcomes, including those that were pre‐specified, but no protocol or trial registration document was available to check this.

Other potential sources of bias

Other biases that we considered pertained to selection bias from major imbalances in baseline characteristics, performance bias in terms of important differences in care provision outside of that of the trial interventions, and detection bias resulting from difference in follow‐up procedures. Contamination bias – resulting from the inadvertent application of the intervention being evaluated to people in the control group, potentially leading to reduced differences between the intervention and control groups – was also considered as part of performance bias.

Eight trials had major imbalances at baseline that could have influenced trial findings. In Kennie 1988 and Stenvall 2007a, these imbalances are likely to have favoured the intervention group, whereas the converse was likely in four other trials (Galvard 1995; Huusko 2002; Karlsson 2016; Ryan 2006). It is harder to judge for Ziden 2008, which anyway did not provide the baseline characteristics for over half of the randomised participants. Uy 2008, the other trial at high risk, had too small a sample size for randomisation to be effective, and the imbalances were relatively large in the type of hip fracture and prior use of walking aid.

Insufficient information on care programmes, especially in the control group and on clinicians' experience, prevented judgement on the risks from performance bias. However, we did consider there was a high risk of bias in five trials. In Crotty 2019, there were very likely to be differences in practice between the nursing homes, as well as contamination resulting from shared staff. In Fordham 1986, this was mainly in consequence of the location of the two study sites and arrangements for geriatrician cover. In Galvard 1995, the geriatric hospital had no prior experience with people with hip fracture. In Jette 1987, there was a strong suggestion of contamination between experimental and control groups, with staff coming from the same department. Finally, in Ryan 2006, there was risk of compensatory care provision in the less intensive group. Some contamination bias, resulting from shared care facilities or staff, or both, also seemed likely in 10 trials (Baroni 2019; Chong 2013; Crotty 2019; Fordham 1986, Gilchrist 1988, Jette 1987; Karlsson 2016; Ryan 2006; Sanchez Ferrin 1999; Vidan 2005). Notably, Naglie 2002 purposefully minimised the risk of contamination bias by ensuring "interdisciplinary care and usual care were provided by different staff on different wards", but still provided insufficient information to make a judgement on this item.

Active and systematic methods of follow‐up seemed to preclude detection bias in all but four trials: in Ashe 2019, the start of follow‐up was 25 days later in the control group, a difference that may or may not be important; in Galvard 1995, the imbalance in the follow‐up of functional outcomes may have reflected different research environments at the two study sites; in Jette 1987, there was insufficient information to be certain of comparability; and in Karlsson 2016, data collection differed in the two groups.

Effects of interventions

See: Table 1; Table 2; Table 3

In the following, all results are presented using the fixed‐effect model. We performed sensitivity analysis to check whether there were important changes to the results if the random‐effects model was used instead for pooling. The results from the latter, for key outcomes with data from five or more trials, are presented in Appendix 11; these show that none differed importantly from the results presented below.

Inpatient multidisciplinary rehabilitation versus usual care

Main outcomes

Poor outcome

We combined the outcomes of death or deterioration in residential status (generally, the requirement for institutional care) to give the overall outcome measure of 'poor outcome', which was assessed at the conclusion of follow‐up (6 to 12 months) and at hospital discharge. At final follow‐up, there is moderate‐certainty evidence, downgraded by one level for serious risk of bias, of a lower risk of a poor outcome at long‐term follow‐up in the intervention group (risk ratio (RR) 0.88, 95% confidence interval (CI) 0.80 to 0.98; I² = 0%; 13 studies, 3036 participants; Analysis 1.1; Figure 5). The 13 trials are listed by recruitment start dates, where available. As found in the previous version of our review, this demonstrates that Kennie 1988 remains an outlier: its removal does not substantively change the findings: RR 0.91, 95% CI 0.82 to 1.01; analysis not shown.

1.1. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 1: 'Poor outcome' (long‐term follow‐up at 6 or 12 months)

5.

Multidisciplinary inpatient rehabilitation versus usual care: 'Poor outcome' (long‐term follow‐up at 6 or 12 months)

An exploratory subgroup analysis by type of intervention (GORU, HFP, MARU, orthogeriatric (OG) programme) showed no evidence of subgroup differences (test for interaction: P = 0.25; I² = 26.4%); see Analysis 1.2. Removal of the three trials conducted in Taiwan, where no clinical care was provided after discharge at approximately one week in the 'usual care' group, does not substantively change the findings: RR 0.87, 95% CI 0.79 to 0.97; 2399 participants; analysis not shown. Another sensitivity analysis, based on perceived risk of selection bias (Analysis 1.3), showed no evidence of a difference between the results for trials subgrouped by risk of selection bias (low versus unclear or high risk), which included that linked to probably serious imbalances in participant characteristics (test for interaction: P = 0.316). Evidence is missing from Jette 1987 and Marcantonio 2001. Jette 1987 reported no significant between‐group difference in survival, nor, separately reported, in the "eventual discharge disposition" but did not supply data: at 12 months, 22 had died and 26% of the survivors were not living at home. Marcantonio 2001 reported that 28 (22%) overall had died or were in a new nursing home placement at six months; there was no report of the distribution of this outcome in the two groups.

1.2. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 2: 'Poor outcome' (long‐term follow‐up): subgrouped by intervention type

1.3. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 3: 'Poor outcome' (long‐term follow‐up) by selection bias

Poor outcome at hospital discharge was based on mortality in hospital and discharge location. This outcome is more vulnerable to issues related to indirectness and bias, such as differences in discharge policies and supply of alternative accommodation (as in Galvard 1995). It is notable that there are data for this outcome from three GORU trials, Fordham 1986, Galvard 1995 and Gilchrist 1988, that did not provide data for long‐term 'poor outcome'. There is very low‐certainty evidence, downgraded one level for serious risk of bias, one level for serious indirectness and one level for serious imprecision reflecting the reduced quantity of data available, of a lower risk of a poor outcome in the intervention group (RR 0.87, 95% CI 0.76 to 1.00; ; I² = 24%; 8 studies, 1537 participants; Analysis 1.4). Jette 1987 reported no significant between‐group difference in hospital deaths or in residential status at discharge without reporting separate treatment group data; overall, 40 (53%) had a poor outcome at discharge, either death (4 participants) or discharged to a rehabilitation hospital (28 participants) or newly to a nursing home (8 participants).

1.4. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 4: 'Poor outcome' (at discharge)

Mortality

Mortality data were reported for all 20 trials; however, these were not split by treatment group for Jette 1987 or Marcantonio 2001. Although favouring the intervention, pooled results do not confirm a difference between the two groups at end of follow‐up at 4 to 12 months (RR 0.91, 95% CI 0.80 to 1.05; I² = 0%; 18 studies, 3973 participants; low‐quality evidence downgraded one level for serious risk of bias and one level for serious imprecision as the confidence interval crosses the line of no effect; Analysis 1.5). The results for Cameron 1993 were for the four‐month follow‐up period, the main trial follow‐up period. The inclusion of 12‐month mortality figures for Cameron 1993 did not alter the above finding (RR 0.90, 95% CI 0.79 to 1.03; Analysis 1.6). Jette 1987 reported no significant between‐group difference in mortality at 12 months; overall there were 22 (29%) deaths. Marcantonio 2001 reported only an overall mortality at six months of 15 (12%). Note, the results for Swanson 1998 are 12‐month mortality data provided by the lead trialist, rather than the six months' post‐discharge figures extrapolated from the main trial report. An exploratory subgroup analysis by type of intervention (GORU, HFP, MARU, OG programme) showed no evidence of subgroup differences (test for interaction: P = 0.78; I² = 0%); see Analysis 1.7.

1.5. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 5: Mortality (end of scheduled follow‐up: 4 to 12 months)

1.6. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 6: Mortality (end of scheduled follow‐up) ‐ with 12 month data for Cameron 1993

1.7. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 7: Mortality (end of scheduled follow‐up): subgrouped by intervention type

Hospital mortality also favoured the intervention group, but again the pooled results do not confirm a difference between the two groups (RR 0.77, 95% CI 0.58 to 1.04; I² = 11%; 11 studies, 2455 participants; low‐certainty evidence downgraded one level for serious risk of bias and one level for serious imprecision as the wide confidence interval crosses the line of no effect; Analysis 1.8). Jette 1987 reported no significant between‐group difference in hospital mortality; overall there were 4 (5.3%) deaths.

1.8. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 8: Mortality (at discharge)

Health‐related quality of life

Three trials reported on HRQoL (Prestmo 2015; Shyu 2008; Shyu 2013a). Prestmo 2015 found very low‐certainty evidence, downgraded one level for risk of bias and two levels for imprecision as there are insufficient data from just one study, of potentially better quality of life measured using the EQ‐5D (‐0.59 to 1.00: best quality) in the intervention group at both four months (mean difference (MD) 0.08, 95% CI 0.03 to 0.13; 347 participants) and 12 months (MD 0.07, 95% CI 0.02 to 0.12; 337 participants); see Analysis 1.9). The confidence interval of both results included the minimally clinical important difference (MCID) of 0.8; this value is consistent with that used in hip fracture trial literature (Griffin 2018).

1.9. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 9: Quality of life: EQ‐5D (‐0.594: worse than death, 0: dead to 1: best quality)

The results at 1, 3, 6 and 12 months for the SF‐36 Taiwan version for Shyu 2008 were reported in a later publication (Shyu 2010). As shown in Analysis 1.10, the intervention group had higher scores for all eight domains at 12 months. Pooled data (391 participants) with Shyu 2013a retained this pattern except for bodily pain, where there was no difference between the two groups. In particular, there were marked and clinically important differences favouring multidisciplinary rehabilitation between the two groups in physical domains, especially physical functioning (MD 13.20, 95% CI 6.99 to 19.41) and role limitations due to physical health problems (MD 27.98, 95% CI 19.54 to 36.42; Analysis 1.10). Relative to other populations, the mean scores in the intervention group for several domains, especially the two role limitations categories, are much higher than the norms found in other populations (Walters 2001). We rated this very low‐certainty evidence, downgraded one level for serious risk of bias and two levels for very serious indirectness, reflecting the minimal care provided after hospital discharge in these trials and the unusually high scores that may reflect a difference in the population compared with elsewhere).

1.10. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 10: Quality of life: SF‐36 (Taiwan version) (each domain: 0 to 100; best quality) at 1 year

Kennie 1988, which reported incomplete data at one year for 65 participants, reported no between‐group difference (P = 0.85) in the 'Life Satisfaction Index' (Neugarten 1961), which was slightly modified by the trial investigators in an unrevealed way for use in a Scottish population.

Functional status: dependency in activities of daily living (ADL), primarily based on requiring assistance of another person

Measures of physical functioning and dependency varied considerably between studies, and data pooling was very limited. The numbers of participants for whom measurements were taken were unclear in several trials, and data were often incomplete. Notably, separate group data were not reported for Jette 1987, and Gilchrist 1988 provided no data on function. In Handoll 2009, we presented long‐term (usually one year) results and also framed mobility outcomes in terms of dependency and reported these together with dependency in ADL. We also combined binary data of poor functional result in survivors with final mortality data to check whether this showed any anomalies (i.e. a group may have fewer deaths but more survivors in poorer health). In this update, we reorganised the ADL outcomes into two categories: a) binary data of non‐recovery or greater dependency in ADL; or b) continuous data for measures of ADL such as the Barthel Index. We report these at short‐term follow‐up (1 to 4 months) for survivors and at longer‐term follow‐up (between 6 to 12 months) for survivors and, where practical, for all participants, including those who had died. Mobility outcomes are presented separately. We have not presented data for ADL at hospital discharge or instrumental ADL in the following.

Binary data for non‐recovery of previous ADL or greater dependency for short‐term follow‐up are presented in Analysis 1.11, and for the converse outcome, regain of previous ADL at short‐term follow‐up, in Analysis 1.12. Binary data for non‐recovery of previous ADL or greater dependency for long‐term follow‐up are presented in Analysis 1.13. Where final value means and SD data are available for the Barthel Index and modifications of this measure, these are presented in Analysis 1.14. Non‐parametric data are shown in Analysis 1.15. Change scores for the Katz Index presented in Sanchez Ferrin 1999 are shown in Analysis 1.16. Separate results for participants who were from nursing homes are presented in Analysis 1.17. Other ADL data are presented narratively below.

1.11. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 11: Greater dependency in ADL up to 4 months

1.12. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 12: Regain in former level of ADL independence in the short term (up to 4 months)

1.13. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 13: Greater dependency in ADL in the long term (6 to 12 months)

1.14. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 14: ADL: Barthel scores (higher scores = greater independence)

1.15. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 15: ADL: Barthel Index (0 to 20; best outcome)

ADL: Barthel Index (0 to 20; best outcome)
Study	Scale and timing	MDR (Median (N))	MDR (IQR)	Usual care (Median (N))	Usual care (IQR)	P‐value
Watne 2014	At 4 months	17 (121)	IQR: 10 to 20	16 (121)	IQR: 12 to 20	0.80
	At 12 months	17 (98)	IQR: 9.5 to 19	16 (95)	IQR: 11 to 19	0.44

1.16. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 16: Loss in activities of daily living at 6 months (Katz index; 6 maximum)

1.17. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 17: ADL: Barthel Index (participants who were from nursing homes)

ADL: Barthel Index (participants who were from nursing homes)
Study	Scale and timing	MDR (Median (N))	MDR (Range or IQR)	Usual care (Median (N))	Usual care (Range or IQR)
Uy 2008	Index (0 to 100; best outcome) at 1 month	37 (3)	Range: 10 to 88	14 (7)	Range: 5 to 60
	Index (0 to 100; best outcome) at 4 months	68 (3)	Range: 0 to 88	28 (7)	Range: 0 to 82
Watne 2014	Index (0 to 20; best outcome) at 4 months	8 (35)	IQR: 4 to 13	11 (38)	IQR: 5.5 to 14
	Index (0 to 20; best outcome) at 12 months	8 (26)	IQR: 4 to 11.5	9 (28)	IQR: 4 to 14

Short‐term follow‐up (1 to 4 months)

Pooled data from four trials for greater dependency in ADL favoured the intervention (194/382 versus 218/372; RR 0.87, 95% CI 0.76 to 0.99; I² = 0%; 754 participants; Analysis 1.11). Analysis 1.12 shows the same findings but recast as a positive outcome as presented in these trials in terms of the regain of former ADL (RR 1.18, 95% CI 1.01 to 1.38). Analysis 1.14 shows that Barthel Index (0 to 20; best outcome) scores at four months also favoured the intervention group for Prestmo 2015 (MD 1.01, 95% CI 0.21 to 1.81; 333 participants) as did the pooled results from two studies for the Chinese Barthel Index (0 to 100; best outcome) at three months (MD 7.57, 95% CI 2.87 to 12.27; I² = 52%; 285 participants). The clinical importance of these differences is unknown and may only be slight. Watne 2014 (242 participants) found no evidence of a difference in Barthel scores at four months (reported P = 0.8); Analysis 1.15. Three other studies also reported no evidence of between‐group differences. Huusko 2002 (220 survivors) reported the median change from baseline to three months in the Katz Index (0 to "8"; high score means better functional status) was 0 in both groups (reported P value = 0.5). Jette 1987 (maximum 75 participants) reported there were no significant between‐group differences at three months in activities in daily living either in terms of basic activities, instrumental activities, social/role function or emotional function. Naglie 2002 (257 participants) found no between‐group difference in Barthel scores (0 to 100; best function) at three months: 62.0 versus 62.4.
Taken overall, we are very uncertain whether the intervention improves or makes no difference to short‐term function compared with 'usual care'. This is because we rated the available evidence as very low‐certainty, being downgraded one level for serious risk of bias, one level for imprecision and one level for inconsistency; the last reflecting a general view of the variation in the results from the 11 trials reporting this outcome.

We did not use data collected at hospital discharge by Fordham 1986 (binary data for single item ADLs, such as ability to dress the lower half of the body) and Sanchez Ferrin 1999 (same level based on Katz score as before fracture) given the variation in length of follow‐up. Nor did we use the data from Uy 2008: results for the 10 participants followed up in this trial are reported separately below as they were all from nursing homes.

Long‐term follow‐up (usually 12 months)

Reflecting the differences in the derivation in the reported measures, the results for greater dependency in ADL are presented in separate subgroupings in Analysis 1.13. Two trials (238 participants) found the risk of greater dependency based on the survivors' pre‐fracture Katz index was lower in the intervention group (RR 0.64, 95% CI 0.51 to 0.81). While the risk ratio favoured the intervention group, the 95% CIs crossed the line of no effect for Vidan 2005 (252 participants), which defined this outcome on the incomplete recovery of ADL and mobility (RR 0.88, 95% CI 0.71 to 1.09); and the data for Shyu 2013a (269 participants), which defined this outcome on the non‐recovery of independent self‐care ability (RR 0.85, 95% CI 0.64 to 1.14). For each outcome, the result for survivors, combined with mortality data, did not show an anomalous situation where the direction of effect differed between the result for the function in the survivors and the combined outcome (Analysis 1.13). Thus, there was no evidence that where more had died in one group (here, the usual care group) that this had resulted in the survivors in that group having better function overall in contrast to the survivors in the other group (here, the MDR group), which still included those who were frailer and who may not have survived if they had been in the first group.

Analysis 1.14 shows that Barthel Index (0 to 20; best outcome) scores at 12 months favoured the intervention group for Prestmo 2015 (MD 1.13, 95% CI 0.31 to 1.95; 300 participants); data from this trial also appear in Analysis 1.13. The pooled results from two studies at 12 months for the Chinese Barthel Index (0 to 100; best outcome: MD 4.04, 95% CI ‐1.42 to 9.51; I² = 0%; 293 participants) and the results from Swanson 1998 (60 participants) at six months for a modified Barthel Index (0 to 100; best outcome: MD 6.30, 95% CI ‐0.53 to 13.13) favoured the intervention group but the 95% CIs crossed the line of no effect. The clinical importance of all these differences is unknown and may only be slight. Watne 2014 (242 participants) found no evidence of a difference in Barthel scores at 12 months (reported P = 0.44); Analysis 1.15. Naglie 2002 (241 participants) found no between‐group difference in Barthel scores (0 to 100; best function) at six months: 65.0 versus 65.7.

Sanchez Ferrin 1999 reported no between‐group difference in the losses in ADL as measured by the Katz Index at six months (Analysis 1.16). Huusko 2002 (193 survivors) reported the median change from baseline to one year in the Katz Index (0 to "8"; high score means better functional status) was 0 in both groups (reported P value = 0.5). Jette 1987 (maximum 75 participants) reported there were no significant between‐group differences at 6 or 12 months in activities in daily living either in terms of basic activities, instrumental activities, social/role function or emotional function.

Taken overall, we are very uncertain whether the intervention improves or makes no difference to long‐term function compared with 'usual care'. This is because we rated the available evidence as very low‐certainty, being downgraded one level for serious risk of bias, one level for imprecision and one level for inconsistency; the last reflecting a general view of the variation in the results from the 13 trials reporting this outcome.

Nursing home participants

Analysis 1.17 shows the non‐parametric data for the Barthel Index for participants from nursing homes who were returned back to nursing homes from Uy 2008 (10 participants at 1 and 4 months) and a subgroup from Watne 2014 (73 participants at 4 months, 54 at 12 months). Very little can be made of these data, given the very few and unbalanced number of participants in the two groups of Uy 2008, the different measures used and probable lack of between‐group differences in Watne 2014. The main observation is that these scores are much and expectedly lower than in the overall population of Watne 2014 and those reported in other trials in Analysis 1.14.

Mobility

Data on different aspects of mobility were available in several trials. However, we kept a primary focus on mobility outcomes indicating the need for help from another person at longer‐term follow‐up. Thus, we do not include data on gait speeds or data at hospital discharge, such as those from Fordham 1986, which reported on the numbers able to stand and the numbers able to walk 5 to 10 steps at this stage. We also do not include data from a subgroup of participants (26% of those randomised) in Galvard 1995, which reported all but one participant attending the clinic could walk indoors at one year (38/38 versus 59/60). As well as binary data for greater dependency in walking, we also report on mobility assessed using the Short Physical Performance Battery (SPPB) tool (0 to 12: best mobility).

Pooled data from five trials for greater dependency in mobility measured at 6 or 12 months favoured the intervention (194/382 versus 218/372; RR 0.83, 95% CI 0.71 to 0.98; I² = 0%; 1085 participants; low‐certainty evidence, downgraded one level for serious risk of bias and one level for imprecision as the 95% CI is wide; Analysis 1.18). Naglie 2002 found no difference in function in terms of transfers. For both outcomes, the result for survivors combined with mortality data did not show an anomalous situation where the direction of effect differed between the result for the function in the survivors and the combined outcome.

1.18. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 18: Greater dependency in mobility

SPPB data were available for Prestmo 2015, which found in favour of the intervention at both 4 and 12 months (Analysis 1.19); and for Watne 2014, which reported results for a different population that included nursing home residents. Watne 2014 reported there was no significant between‐group difference at the same follow‐up times (Analysis 1.20). As reported by Prestmo 2015, 1.0 point in the SPPB score is regarded as a substantial meaningful change, and 0.5 points is a small meaningful change. Hence the best estimate and 95% CI reported at 12 months follow‐up for this trial includes a clinically meaningful difference (MD 0.69, 95% CI 0.09 to 1.29; 284 participants). Overall, there is very low‐certainty evidence on whether the intervention improves or makes no difference to mobility assessed using the SPPB compared with 'usual care'. Based on the results from two trials, we downgraded the evidence one level for serious risk of bias, one level for imprecision and one level for inconsistency; although the latter could not be quantified statistically.

1.19. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 19: Mobility: Short Physical Performance Battery (0 to 12: best mobility)

1.20. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 20: Mobility: Short Physical Performance Battery (0 to 12: best mobility)

Mobility: Short Physical Performance Battery (0 to 12: best mobility)
Study	Scale and timing	MDR (Median (N))	MDR (IQR)	Usual care (Median (N))	Usual care (IQR)	P‐value
Watne 2014	At 4 months	4 (121)	IQR: 1 to 8	3 (121)	IQR: 1 to 6	0.13
	At 12 months	3 (98)	IQR: 1 to 7	3 (95)	IQR: 1 to 6	0.14

Pain

Pain data specific to the injury location were available only for Galvard 1995. However, the results for hip pain when walking, sitting and supine were available for a subgroup of participants at one year (38 (21% of 182 in the intervention group) versus 60 (31% of 196 in the control group). We judged these data were not reliable because of the differences in the participant numbers in the two subgroups.

Other outcomes

Poor outcome at hospital discharge

This is reported above, under 'poor outcome'; Analysis 1.4.

Level of care and extent of support required or provided on discharge (inpatient or from rehabilitation programme)

As explained in Types of outcome measures, we decided not to report on this outcome.

Residential status between 4 to 12 months

Fewer survivors of the intervention group were in institutional, mainly nursing home, care between 6 to 12 months (RR 0.90, 95% CI 0.76 to 1.06; I² = 6%; 14 studies, 2497 participants; low‐certainty evidence downgraded one level for serious risk of bias and one level for serious imprecision as the confidence interval crosses the line of no effect; Analysis 1.21).

1.21. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 21: Institutional care at 6 to 12 months (survivors)

Medical complications (morbidity)

In general, there was variation between the trials in the definition, detection and reporting of medical complications, and ensuring these were distinct from underlying morbidity. The available data for in‐hospital medical complications from five studies (Marcantonio 2001; Sanchez Ferrin 1999; Swanson 1998; Vidan 2005; Watne 2014), or complications during 12 months' follow‐up (Huusko 2002), are presented in Analysis 1.22.

1.22. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 22: Complications

Of note is that delirium in hospital was the sole complication reported on by Marcantonio 2001. Cognitive function and delirium were a key focus of Watne 2014. Pooled results from four trials for this complication were in favour of the intervention group (199/483 versus 239/497; RR 0.85, 95% CI 0.74 to 0.98; 980 participants; low‐certainty evidence downgraded one level for serious risk of bias and one level for serious imprecision (wide confidence interval)). Stenvall 2007a (199 participants) reported significantly fewer participants of the intervention group had post‐operative delirium (reported P = 0.003).

Of the more frequent individual complications for which data were available for analysis, there was very low‐certainty evidence of fewer pressure injuries (RR 0.50, 95% CI 0.35 to 0.70; I² = 32%; 3 studies, 854 participants) and urinary tract infections (RR 0.61, 95% CI 0.42 to 0.88; I² = 0%; 2 studies, 535 participants) in the intervention group. The very low‐certainty evidence for participants experiencing any medical complication means that we are uncertain of the finding potentially favouring the intervention (RR 0.91, 95% CI 0.79 to 1.04; I² = 68%; 3 studies, 891 participants; evidence downgraded one level for serious risk of bias, one level for imprecision (CI crossed the line of no effect) and one level for inconsistency (substantial heterogeneity)).

Accounts of these outcomes by other trials are detailed as follows. Cameron 1993 reported that there were no medical complications that could be directly attributed to the intervention. Huusko 2002 reported there was no difference between the two groups in the numbers of people with "some type of" complication during follow‐up. Prestmo 2015 reported that they had "noted no differences in fracture‐related or other complications during the index stay (data not shown)". Identification of new medical disorders was part of the intervention of Gilchrist 1988, who from an inspection of post‐discharge case records with a full set of investigation results, reported significantly fewer cases of untreated medical conditions in the intervention group (5/88 versus 33/69). Kennie 1988 reported there was no difference between the two groups in the "number of overall illnesses post‐discharge". The active prevention, identification and treatment of medical disorders was part of the intervention in Stenvall 2007a, where significantly fewer participants of the intervention were reported with post‐operative delirium (reported P = 0.003), urinary tract infection (P = 0.005) or pressure ulcers (P = 0.01). Vidan 2005 attributed the reduced overall incidence of medical complications in the intervention group (70 versus 100) to the early identification and daily patient care.

We did not use the total complications data for Baroni 2019 because these were internally inconsistent in the report. Jette 1987 (75 participants), which gave overall population data only on medical complications, reported that "half were confused after surgery", 24 developed urinary tract infection, 19 had heart rhythm disturbances, 14 had pneumonia, 11 had depression, and 5 had congestive heart failure.

Re‐operation (unplanned return to operating theatre) or substantive treatment for an adverse effect

Two studies, Vidan 2005 and Watne 2014, reporting the number of participants incurring a surgical complication, found no evidence of a between‐group difference (RR 0.88, 95% CI 0.54 to 1.42; 2 studies, 638 participants). We did not find data on re‐operations.

Hospital readmission

There was low‐certainty evidence, downgraded one level for serious risk of bias and one level for serious indirectness as duration of follow‐up was mixed and inadequately short in several trials and the reasons for readmission were generally not detailed, of little to no between‐group difference in hospital readmission during follow‐up (RR 0.97, 95% CI 0.84 to 1.12; I² = 23%; 11 studies, 2538 participants; Analysis 1.23). Swanson 1998 explicitly reported that no readmission was related to the original admission. One intervention group participant who remained in hospital was included in the numerator for Swanson 1998. A similar pattern in the results applied when mortality data (these were deaths in hospital except for Cameron 1993, where results were for mortality at four months) were added to readmission data (RR 0.91, 95% CI 0.78 to 1.05; 8 studies, 1975 participants; Analysis 1.24).

1.23. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 23: Readmitted to hospital during follow‐up

1.24. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 24: Dead or readmitted to hospital during follow‐up

Carer burden

Two studies reported data relating to carer burden (Cameron 1993; Kennie 1988). A separate report of Cameron 1993 found that carer burden prior to fracture was the strongest predictor of subsequent burden, and that carers of people in nursing homes and people with greater cognitive and physical disability were more burdened. Cameron 1993 found that the intervention did not significantly impact on carer burden. In the 12‐month follow‐up of Kennie 1988, no difference in carer burden was reported between the intervention and control groups. Without supporting data or indication of how outcome was assessed, Huusko 2002 reported no differences between the two groups in home nursing, food service or help from others. No extra data were available for carer burden in relation to the mental health of family carers for Shyu 2013a.

Economic outcomes

Length of stay in hospital and hospital readmission

The reported lengths of stay (the majority of studies considered total length of stay), which usually included initial treatment in the orthopaedic unit and subsequent stay in the rehabilitation setting, varied considerably. For example, the range of mean values in the control groups was from 8.7 days in Baroni 2019 to 56 days in Kennie 1988. Lengths of stay in Huusko 2002 were calculated from the day of surgery to the day of discharge, lasting at least two weeks. Where data were presented showing the distribution of lengths of stay, it was clear that they were not normally distributed but, for completeness, we present the available data for 12 studies, ordered by recruitment start dates, in Analysis 1.25. We did not pool these data, given the clearly considerable heterogeneity. Analysis 1.25 shows a mixed picture, where length of stay in the intervention group is clearly shorter in four trials (Cameron 1993; Kennie 1988; Stenvall 2007a; Swanson 1998), and longer in three trials (Galvard 1995; Naglie 2002; Prestmo 2015). There is no evidence of a between‐group difference for the remaining five trials (Baroni 2019; Gilchrist 1988; Sanchez Ferrin 1999; Shyu 2008; Tseng 2019). Of interest is that lengths of stay tend to be shorter over time. However, the data from two trials conducted in Taiwan, where hospital stays are generally short anyway, may exaggerate this impression.

1.25. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 25: Length of hospital stay (days)

Incomplete data for initial hospital stay are available for six trials. Standard deviations were not available for Fordham 1986 (mean length of stay GORU: 56 days; control: 44 days; no difference reported); and data were presented as medians in Huusko 2002 (median 34 versus 42 days, reported P = 0.05), Vidan 2005 (median 16 versus 18 days, reported P = 0.06); and Watne 2014 (median stay in allocated ward 11 versus 8 days; reported P = 0.001). The length of stay was shorter in both groups in the 102 participants who were returned to their nursing homes (median stay in allocated ward 8 versus 4 days; reported P = 0.001). The distribution of length of stay data in Fordham 1986 also showed more participants of the intervention group (10 versus 4) stayed over 91 days. Jette 1987 reported an overall length of hospital stay of 21 days with no significant difference between the two groups. No between‐group difference in length of stay in the acute hospital setting was reported for Marcantonio 2001 (median 5 days in both groups). Vidan 2005 reported two outliers whose stay was over 100 days.

The total number of days in hospital per participant was 80 in each group of Huusko 2002. There was also no difference in mean stay over six months in institutional care (either acute or rehabilitation hospital or nursing home) in Naglie 2002, which was reported to be 110 days for both groups. Mean hospital stay in Stenvall 2007a over one year was shorter in the intervention group (37.0 versus 51.4 days; reported P = 0.051). Shyu 2013a reported only that the mean hospital stay ranged from 7.93 to 8.47 days in the three groups.

Cost analysis

Five trials reported the results from a cost analysis (Cameron 1993; Fordham 1986; Galvard 1995; Huusko 2002; Prestmo 2015). Cameron 1993, who reported cost outcomes in a separate paper, found that costs (defined as cost per recovered patient) were significantly reduced in the intervention group (10,600 versus 12,800 Australian dollars (AUD)). Costs assessed were direct costs due to treatment and aftercare, up to four months after the fracture. Fordham 1986 concluded that the cost of care per participant (2714 versus 2618 pounds sterling (GBP) at 1985 prices) was slightly greater in the intervention group due to costs generated by travel to the unit. Galvard 1995 reported increased costs for the intervention group (84,537 versus 94,026 Swedish krona at 1989 prices). Though the total direct cost per participant during the first year in the intervention group was estimated at 2000 euros more (1999 prices: 17,900 versus 15,900 euros), Huusko 2002 considered that the costs did not differ remarkably, and furthermore suggested that the costs in the control group were underestimated. Prestmo 2015 noted higher mean index stay (in hospital) costs per participant in the intervention group (2010 prices in euros: 11,868 versus 9537; MD 2331, 95% CI 1483 to 3178; reported P < 0.001). However, after costing for hospital costs after discharge, rehabilitation stay, nursing home stay, and other primary health and care services, Prestmo 2015 reported no significant between‐group difference in the total cost per participant at 12 months (2010 prices in euros: 54,332 versus 59,486; MD ‐5154 favouring the intervention, 95% CI ‐13,311 to 3007; reported P = 0.22). When considered together with the higher QALY (quality‐adjusted life years) results in the intervention group (0.49 versus 0.42; MD 0.07, 95% CI 0.01 to 0.13; reported P = 0.019), Prestmo 2015 reported that the intervention had a 99% probability of being cost‐effective compared with usual care.

Gilchrist 1988 noted that no additional funding had been allocated for the intervention in their trial. An estimate of the cost benefit of the orthopaedic‐geriatrician liaison unit responsible for delivering the intervention in Swanson 1998 was given in the abstract of conference proceedings (Day 1997), but did not appear in the final report.

Subgroup analysis by participant characteristics

Subgroup analyses by participant characteristics were available for Huusko 2002, Marcantonio 2001 and Watne 2014. As argued in Handoll 2009, we considered the data relating to dementia (normal cognition, suspected mild dementia, suspected moderate dementia, suspected severe dementia) presented in a previous report of Huusko 2002 were too compromised to use. Marcantonio 2001 only reported data for delirium split by dementia status. Subgroup data were reported for participants with dementia and those from nursing homes in Watne 2014. As shown in Analysis 1.26, there was no evidence of subgroup differences for mortality at 12 months (test for subgroup differences: Chi² = 0.00, degrees of freedom (df) = 1 (P = 0.99), I² = 0%). Similar findings applied to long‐term mortality when subgrouped by residential status (analysis not shown).

1.26. Analysis.

Comparison 1: Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care, Outcome 26: Subgroup analysis ‐ death at 12 months

Comparison of different types of inpatient multidisciplinary rehabilitation

Multidisciplinary rehabilitation was provided in both groups of Chong 2013, a quasi‐RCT, which tested the addition of an 'integrated care pathway'. There was no evidence of a difference between the two groups at 12 months in 'poor outcome', defined as dead or with reduced mobility (44/76 versus 35/62; RR 1.03, 95% CI 0.77 to 1.37; very low‐certainty evidence downgraded two levels for very serious risk of bias, particularly selection bias, and one level for serious imprecision; Analysis 2.1). The very low‐certainty evidence means we are also uncertain of the findings of little difference in mortality, inability to walk as before or any nursing home stay post‐discharge (Analysis 2.1). Chong 2013 found no differences between the two groups in the physical or mental component summary scores for SF‐12 quality of life measured at 12 months (Analysis 2.2; very low‐certainty evidence). Similarly, Chong 2013 reported no differences in the changes in the modified Barthel Index over time (Analysis 2.3; very low‐certainty evidence). Overall numbers of participants readmitted to hospital by 12 months were not reported; however, the available data for set times did not show evidence of a difference between the two groups (Analysis 2.4; very low‐certainty evidence). The median length of hospital stay was reported to be significantly lower in the intervention group: 35 days versus 48 days; reported P = 0.009).
 

2.1. Analysis.

Comparison 2: Inpatient setting: multidisciplinary rehabilitation (MDR) with an integrated care pathway versus MDR alone, Outcome 1: Poor outcome, dead, couldn't walk as before, or nursing home stay post discharge (12 months)

2.2. Analysis.

Comparison 2: Inpatient setting: multidisciplinary rehabilitation (MDR) with an integrated care pathway versus MDR alone, Outcome 2: SF‐12 scores at 12 months (0: worst to 100: best)

2.3. Analysis.

Comparison 2: Inpatient setting: multidisciplinary rehabilitation (MDR) with an integrated care pathway versus MDR alone, Outcome 3: Modified Barthel Index (0: worst to 100: best): "changes over time"

Modified Barthel Index (0: worst to 100: best): "changes over time"
Study	Care pathway: Mean (SD)	MDR alone: Mean (SD)	P value
At discharge (N = 149)
Chong 2013	22.2 (17.5)	23.9 (19.7)	0.58
At 6 month follow‐up (N = 129)
Chong 2013	32.6 (21.3)	27.7 (20.6)	0.18
At 12 month follow‐up (N = 121)
Chong 2013	33.4 (22.9)	31.8 (19.5)	0.68

2.4. Analysis.

Comparison 2: Inpatient setting: multidisciplinary rehabilitation (MDR) with an integrated care pathway versus MDR alone, Outcome 4: Hospital readmission (at set times)

Ambulatory rehabilitation versus usual care

Seven trials tested rehabilitation in the community setting (Ashe 2019; Crotty 2003; Crotty 2019; Karlsson 2016; Ryan 2006; Singh 2012; Ziden 2008). We pooled data from three of the seven trials in this category as they tested sufficiently similar interventions in comparable settings (Crotty 2003; Karlsson 2016; Ziden 2008); the other four are considered in turn.

Supported discharge and multidisciplinary home rehabilitation versus usual care

While four trials tested this intervention, we pooled data from only three trials (Crotty 2003; Karlsson 2016; Ziden 2008). This is mainly because all participants from Crotty 2019 were from nursing homes, and the multidisciplinary intervention and usual care in the nursing home setting was substantively different from those of the other three trials. Typically, the home rehabilitation lasted between three to four weeks.

Intervention solely or mainly aimed at people living in their own homes

Main outcomes

The available data for poor outcome, mortality, transfer to a higher level of care and inability to walk at three to four months follow‐up are shown in Analysis 3.1, and at 12 months' follow‐up in Analysis 3.2. The very low‐certainty evidence, downgraded one level for serious risk of bias and two levels for very serious imprecision (few events, wide confidence intervals crossing the line of no effect) means we are uncertain of the findings of little or no between‐group difference in poor outcome at three months (RR 1.22, 95% CI 0.63 to 2.39; 1 study, 209 participants) or at one year (38/190 versus 40/187; RR 0.91, 95% CI 0.62 to 1.35; I² = 0%; 3 studies, 377 participants). The same conclusion of uncertainty of the findings of little to no between‐group difference applies to mortality at three or four months (RR 0.77, 95% CI 0.34 to 1.76; I² = 0%; 2 studies, 275 participants), and mortality at one year (27/190 versus 26/187; RR 1.00, 95% CI 0.61 to 1.63; I² = 0%; 3 studies, 377 participants). Based on very few events, there is very low‐certainty evidence of little or no between‐group differences in the inability to walk at three months (RR 2.47, 95% CI 0.68 to 9.02; 1 study, 183 participants) or at 12 months (RR 0.89, 95% CI 0.38 to 2.04; I² = 14%; 2 studies, 214 participants); or in the move to a higher level of care by 12 months (RR 0.35, 95% CI 0.07 to 1.69; I² = 0%; 2 studies, 168 participants).

3.1. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 1: 'Poor outcome', mortality, and unable to walk (3 or 4 months)

3.2. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 2: 'Poor outcome', mortality, institutional care and unable to walk (12 months)

While Crotty 2003 and Ziden 2008 excluded people with dementia, 50% of the participants had dementia in Karlsson 2016. For illustrative purposes only, subgroup analysis split according to whether participants had dementia or not in Karlsson 2016 are shown for poor outcome and mortality at 12 months in Analysis 3.3 and Analysis 3.4. There was no statistically significant differences between the results for participants with or without dementia for either poor outcome (test for subgroup differences (interaction): P = 0.72) or mortality (test for interaction: P = 0.16).

3.3. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 3: Subgroup analysis: poor outcome (dead or non‐recovery of indoor walking ability) at 1 year, subgrouped by dementia status

3.4. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 4: Subgroup analysis: mortality at 1 year, subgrouped by dementia status

Only Crotty 2003, reporting results for 56 of 66 randomised participants, provided usable data on quality of life: see Analysis 3.5. There is very low‐certainty evidence, downgraded one level for serious risk of bias and two levels for very serious imprecision, of little between‐group difference in quality of life at 12 months assessed using the SF‐36 scores (0 to 100; best quality of life) split by the physical (MD 4.70, 95% CI ‐0.43 to 9.83) and mental (MD 1.50, 95% CI ‐2.88 to 5.88) components. Quality of life was incompletely reported in Ziden 2008, which selectively reported on two domains, physical function and bodily pain, of the eight domains of the SF‐36: these were found to favour the intervention group at 6 and 12 months. We considered these results unreliable when taken out of context in this way. Quality of life assessed via the EQ‐5D data are listed in the trial registration document but not yet reported for Karlsson 2016.

3.5. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 5: SF‐36 scores at 12 months (0: worst to 100: best)

We present further data, from Karlsson 2016 and Ziden 2008, linked with assessment of dependency, in Analysis 3.6 ('Independence in personal activities of daily living (PADL) and outdoor walking', at 3 and 12 months), Analysis 3.7 ('Function (ADL): Barthel index'; non‐parametric data at 3 and 12 months), and Analysis 3.8 ('Daily activities scores at 1 year'; non‐parametric data for FIM (Functional Independence Measure) and FAI (Frenchay Activity Index)). Not presented are findings at four months' follow‐up for Crotty 2003, which found statistically significant differences favouring the intervention group for the modified Barthel Index (median 97.0 versus 94.0; clinical importance not established). The evidence for all these outcomes is very low‐certainty evidence, downgraded one level for serious risk of bias and two levels for very serious imprecision. Thus, we are uncertain of the findings of minimal between‐group difference in independent functioning and mobility, and the Barthel Index findings by Karlsson 2016; the findings in favour of the intervention group for both daily activities scores found by Ziden 2008; and the findings for the modified Barthel Index reported by Crotty 2003.

3.6. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 6: Independence in personal activities of daily living (PADL) and outdoor walking

3.7. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 7: Function (ADL): Barthel Index (0 to 20: total independence in personal care)

Function (ADL): Barthel Index (0 to 20: total independence in personal care)
Study	Scale and timing	MDR (Median (N))	MDR ( IQR)	Usual care (Median (N))	Usual care (IQR)
Karlsson 2016	At 3 months	15 (95)	IQR: 8 to 20	16 (89)	IQR: 11 to 19
	At 12 months	17 (80)	IQR: 8 to 20	17 (79)	IQR: 9 to 19

3.8. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 8: Daily activities scores at 1 year

Daily activities scores at 1 year
Study	Home rehabilitation (N = 45)	Usual care (N = 48)	Reported P value
FIM (Functional Independence Measure: 0 to 91; independent)
Ziden 2008	Median = 85 Range: 46 to 91	Median = 80 Range: 29 to 91	P = 0.001
FAI (Frenchay Activity Index: 0 to 45; best activity)
Ziden 2008	Median = 27 Range: 0 to 40	Median = 20 Range: 0 to 42	P = 0.028

None of the trials reported on hip or lower‐limb pain.

Other outcomes

All the evidence for the other outcomes is rated at very low certainty, downgraded one or two levels for serious or very serious risk of bias and two levels for very serious imprecision, reflecting few events, small sample sizes and wide confidence intervals.

Ziden 2008 reported all participants were alive at discharge. Six participants in the intervention group did not participate in the home rehabilitation programme: three were discharged to a short‐term nursing home and three to another acute clinic. In the control group, deviation from 'usual care' comprised nine participants discharged to short‐term nursing home care, eight who received community home rehabilitation and four who visited outpatient physiotherapists.

Karlsson 2016 found no evidence of a between‐group difference in the numbers of people incurring a medical complication or who had delirium recorded after hospital discharge (Analysis 3.9).

3.9. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 9: Complications (discharge to 12 months)

There was no evidence of a difference in the numbers of participants who were readmitted to hospital within 4 or 12 months (RR 1.20, 95% CI 0.83 to 1.74; I² = 0%; 2 studies, 265 participants; Analysis 3.10). Thirteen participants, roughly spread between the two groups, had a reoperation in Karlsson 2016; see Analysis 3.10.

3.10. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 10: Readmission to hospital and reoperation

Fall and further fracture data from Crotty 2003 (up to 4 months) and Karlsson 2016 (from discharge to 12 months) are presented in Analysis 3.11. There is very low‐certainty evidence of little to no difference between the two groups for these outcomes.

3.11. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 11: Falls outcomes

Crotty 2003, reporting on care burden, as rated by the Caregiver Strain Index (CSI: 0 to 12: worst stress), found a one‐point difference in medians at four months (1.0 versus 2.0; maximum number of carers: 21 versus 18) and a three‐point difference at one year (median CSI 1.0 versus 4.0; reported P = 0.02) in favour of the home rehabilitation group.

Economic outcomes

Consistent with a policy of accelerated discharge, participants in the home‐based rehabilitation groups of Crotty 2003 and Karlsson 2016 had shorter stays in hospital. This was by six days in both Crotty 2003 (MD ‐6.50 days, 95% CI ‐11.30 to ‐1.70 days; Analysis 3.12) and Karlsson 2016 (median 17 days (interquartile range (IQR) 12 to 26) versus 23 days (IQR 17 to 32)). Although the mean length of hospital stay was over a day shorter in the home‐based rehabilitation group of Ziden 2008, the data do not confirm a between‐group difference (MD ‐1.60 days, 95% ‐4.59 to 1.39; Analysis 3.12). In all three trials, overall duration of rehabilitation in the intervention group would have been longer, as shown by data from Crotty 2003 (MD 14.00 days, 95% CI 7.84 to 20.16 days). Therapists visited participants of the home‐based group an average of 13.6 times in Crotty 2003, 14.2 times in Karlsson 2016, and 4.9 times in Ziden 2008.

3.12. Analysis.

Comparison 3: Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation, Outcome 12: Lengths of hospital or rehabilitation stays (days)

In a separate publication reporting results at one month follow‐up, Ziden 2008 observed that the home rehabilitation group spent 376 fewer days within institutional care (short‐term nursing homes and geriatric clinics) than the usual care group.

Crotty 2003 found there were no statistically significant differences in carer time, GP visits or use of community services at four months.

Intervention for people living in a nursing home setting

Crotty 2019 compared a four‐week ambulatory geriatric multidisciplinary rehabilitation programme delivered in a nursing home with usual care, where treatment was provided according to usual practice in the nursing home. The 240 previously mobile people from 76 nursing homes (nursing care facilities) with hip fracture were discharged from acute hospital at about a week after their injury; most were unable to transfer or were confined to bed at the time of transfer. Follow‐up was reported at four weeks, at the end of the intervention, and one year. The lead trial investigator provided additional mortality data at four months.

Main outcomes

There is low‐certainty evidence, downgraded one level for serious risk of performance bias and one for serious imprecision, of no or minimal between‐group differences at 12 months in 'poor outcome', defined as dead or unable to walk (82/119 versus 80/121; RR 1.04, 95% CI 0.87 to 1.24), mortality (58/119 versus 52/121; RR 1.13, 95% CI 0.86 to 1.49), or survivors who were unable to walk (24/61 versus 28/69; RR 0.97, 95% CI 0.64 to 1.48); see Analysis 4.1. At four weeks, there was very low‐certainty evidence of fewer deaths in the intervention (10 versus 22; RR 0.46, 95% CI 0.23 to 0.93). However, this difference was gradually lost, and there was low‐certainty evidence of no or very little between‐group difference by four months (35 versus 34 deaths; RR 1.05, 95% CI 0.70 to 1.56); see Analysis 4.2.

4.1. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 1: 'Poor outcome', mortality, and unable to walk (12 months)

4.2. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 2: All cause mortality

The large loss to follow‐up, predominantly from death at one year, means that the certainty of evidence is very low for quality of life, dependency, mobility and pain data for survivors at one year. Thus, we are very uncertain of the findings of a very slightly better self‐reported quality of life in the intervention group when measured via the DEMQOL (12 to 112; best QoL) (MD 7.40, 95% CI 2.38 to 12.42; 70 participants) and, conversely, of very little between‐group difference in favour of the control group when measured via the DEMQOL‐Proxy (31 to 124; best QoL) (MD ‐3.20, 95% CI ‐6.94 to 0.54; 126 participants); see Analysis 4.3. The data for the EQ‐5D are more complete, as these include deaths, which were set at zero. While the data show a small quantitative difference in favour of the control group (MD ‐0.06, 95% CI ‐0.12 to ‐0.00; 235 participants; Analysis 4.4), the difference is unlikely to be clinically important (Walters 2005).

4.3. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 3: Quality of life at 12 months: DEMQOL & DEMQOL‐Proxy

4.4. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 4: EQ‐5D quality of life index at 12 months (0 dead to 1 best quality)

We are uncertain of the findings of minimal between‐group difference in functional dependency assessed using the modified Barthel Index (0 to 100: total independence in personal care) at four weeks (MD 0.90, 95% CI ‐4.51 to 6.31; 202 participants) or at one year (MD ‐4.90, 95% CI ‐11.69 to 1.89; 125 participants); see Analysis 4.5. Mobility, measured using the Nursing Home Life‐Space Diameter (0 bed/chair‐bound to 50 leaving the facility daily), was slightly higher at one month in the intervention group, but it is unknown whether this small difference is clinically important (MD 1.90, 95% CI 0.56 to 3.24; 203 participants; Analysis 4.6). There was no evidence of a between‐group difference in this outcome at one year (MD 0.30, 95% CI ‐1.40 to 2.00; 126 participants; Analysis 4.6).

4.5. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 5: Modified Barthel Index (0 to 100: total independence in personal care)

4.6. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 6: Nursing Home Life‐Space Diameter (0 to 50; leaves facility daily)

There was no evidence of a between‐group difference in pain, assessed using the Pain Assessment In Advanced Dementia scale (0 to 10; severest pain) at four weeks or at one year; see Analysis 4.7.

4.7. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 7: Pain: PAINAD (0 to 10; severe pain)

Other outcomes

There is very low‐certainty evidence of a greater number of people in the intervention group who fell in the first four weeks (Analysis 4.8). The numbers of fall‐related hospital admissions (27 in all) and of hip fractures (4 in all) were too small to draw conclusions (Analysis 4.8).

4.8. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 8: Fall outcomes (adverse events)

Economic analysis

The mean per participant 12‐month Australian Medicare costs at 2015/16 unit prices were AUD 2076 higher in the intervention group than in the control arm, but these differences were reported not to be statistically significant (reported 95% CI AUD ‐220 to 4360); see Analysis 4.9. Crotty 2019 based their cost‐effectiveness analysis on the DEMQOL data at one year; notably these were statistically in favour of the intervention group, unlike those for DEMQOL‐Proxy or EQ‐5D data. The incremental cost‐effectiveness ratio (ICER) was reported at AUD 328,685 per QALY gained (95% CI 82,654 to 75,007,056). Crotty 2019 noted that, given that this ICER is substantially greater than the implicit cost‐effectiveness threshold of AUD 50,000 per QALY gained currently applied by regulatory bodies in Australia, the intervention would not be considered cost‐effective.

4.9. Analysis.

Comparison 4: Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes, Outcome 9: Mean costs per patients (in AU $): base case analysis

Mean costs per patients (in AU $): base case analysis
Study	Costs	MDR: mean (SD)	Usual care: mean (SD)	Mean difference	Bootstrapped 95% CI
Crotty 2019	Medical Benefits Schedule fees	1570 (114)	1743 (142)	‐173	‐534 to 166
	Pharmaceutical Benefits Schedule	1164 (210)	983 (111)	180	‐214 to 787
	Inpatient costs (AR‐DRGs)	2945 (762)	3174 (829)	‐229	‐2479 to 1683
	Intervention costs	2298 (76)	‐	2298	‐
	Total	5900 (855)	7977 (825)	2076	‐220 to 4360

Intensive compared with less intensive multidisciplinary home rehabilitation

Ryan 2006 compared intensive with less intensive home‐based multidisciplinary rehabilitation, lasting a maximum of 12 weeks, in a subgroup of 71 people who had been recently discharged from hospital after hip fracture surgery. (The other subgroup included in Ryan 2006 comprised 89 people recovering from stroke.) The evidence for all reported outcomes for the hip fracture subgroup is of very low certainty, downgraded one level for risk of bias and two levels for serious imprecision. There is no evidence of differences between the two groups of hip fracture participants in any of the measured outcomes. The data for 'poor outcome' (mortality or institutional care at 12 months), mortality and institutional care are presented in Analysis 5.1. The findings for quality of life measured using the EQ‐5D, dependence assessed via the Barthel Index, and activity assessed via the Frenchay Activities Index at 3 and 12 months are shown in Analysis 5.2. Of note in terms of the population, is that the median values of the Barthel Index were at the top end of the range at three months, having been 16 and 17 at baseline in the two groups. Mobility and pain were not reported. Intensification of intervention was based on at least doubling the number of contacts made by the multidisciplinary team per week. This aim was not achieved, although statistically significantly more visits were made to the intervention group (see Analysis 5.3: mean difference 6.50 visits; 95% CI 3.01 to 9.99).

5.1. Analysis.

Comparison 5: Ambulatory setting: intensive versus less intensive community rehabilitation, Outcome 1: 'Poor outcome', mortality and institutional care (12 months)

5.2. Analysis.

Comparison 5: Ambulatory setting: intensive versus less intensive community rehabilitation, Outcome 2: Quality of life, dependence and activities scores at 3 and 12 months

Quality of life, dependence and activities scores at 3 and 12 months
Study	Timing	Intensive (Median (N))	Intensive (IQR)	Less intensive (Median (N))	Less intensive (IQR)	Reported P value
Quality of life: EQ‐5D (0: dead to 1: best quality)
Ryan 2006	3 months	0.62 (30)	0.52 to 0.77	0.67 (28)	0.59 to 0.79	0.3
	12 months	0.7 (30)	0.59 to 0.8	0.7 (28)	0.62 to 0.74	0.67
Barthel Index (0 to 20: independent)
Ryan 2006	3 months	20 (30)	19 to 20	20 (28)	19 to 20	0.83
	12 months	20 (30)	19 to 20	20 (28)	19 to 20	0.18 (probably 0.81)
Frenchay Activities Index (0 to 45: best activity)
Ryan 2006	3 months	19 (30)	14 to 23	19 (28)	14 to 24	0.82
	12 months	22 (30)	16.5 to 29.5	21 (28)	13 to 26	0.27

5.3. Analysis.

Comparison 5: Ambulatory setting: intensive versus less intensive community rehabilitation, Outcome 3: Number of contacts over 12 weeks (for participants with hip fracture or stroke)

Multidisciplinary care, including progressive resistance training for one year, versus usual care

Singh 2012 randomised 124 participants to 12 months of geriatrician‐led multidisciplinary care, including high‐intensity weight‐lifting exercise, in an outpatients hospital programme and targeted treatment of frailty at outpatients and home, or usual rehabilitation care. Follow‐up was 12 months after the fracture – thus, at the end of the intervention. All results were very low‐certainty evidence, downgraded one level for risk of bias and two levels for imprecision, reflecting low numbers of events and wide confidence intervals.

The available data for poor outcome, mortality and nursing home admission at 12 months are shown in Analysis 3.1. At 12 months, fewer intervention group participants had a poor outcome (dead or in institutional care) than in the control group (7/62 versus 16/62; RR 0.44, 95% CI 0.19 to 0.99); had died (4/62 versus 8/62); or were in institutional care (5/62 versus 12/62).

Basic activities of daily living (ADLs) measured using the Katz ADL Index (0 to 12; higher scores = greater dependency) were marginally better in the intervention group at 12 months (mean 0.5 versus 1.0); the reported adjusted mean difference also favoured the intervention group (MD ‐0.09, 95% CI ‐1.9 to 0.2; reported P = 0.06).

Singh 2012 (99 participants) reported assistive device use was "significantly lower at 12 months (P < 0.02) in the intervention group" but did not provide details of the continuous measure used.

Singh 2012 found no evidence of a difference between the two groups in ALSAR: Assessment of Living Skills And Resources scores; either relating to skills or to resource availability (scores 0 to 22; lower scores = better outcome); data for 99 participants; Analysis 6.2.

6.2. Analysis.

Comparison 6: Ambulatory setting: extended multidisciplinary ambulatory rehabilitation versus usual care, Outcome 2: ALSAR: Assessment of Living Skills And Resources

Post‐discharge multidisciplinary clinic, with referral for further rehabilitation as necessary, versus usual care

Ashe 2019 compared a post‐discharge geriatrician‐led multidisciplinary clinic, with referral for further rehabilitation as necessary, with usual care, in 53 community‐dwelling older adults who were on average 32.5 weeks (range 3 to 12 months) post‐fracture. Currently, only very limited outcome data on physical activity and sedentary behaviour are available for this study (Zusman 2019); none of the reported outcomes feature in those presented in this review. We received confirmation that no participants died or moved to residential care during the study, although data were missing for one participant in the intervention group (Analysis 7.1).

7.1. Analysis.

Comparison 7: Outpatient multidisciplinary clinic between 3 to 12 months post fracture versus usual care, Outcome 1: Mortality and institutional care

Care in both settings

As per our protocol, we pooled the data from both settings as an exploratory analysis. However, we restricted the data from the ambulatory setting to supported discharge and own‐home‐based rehabilitation. Inclusion of data from Crotty 2003, Karlsson 2016 and Ziden 2008, where the care in the intervention group was predominantly at the participant's own home, made very little difference (RR 0.89, 95% CI 0.80 to 0.98; I² = 0%; 16 studies, 3413 participants; Analysis 8.1) to the results for 'poor outcome' at long‐term follow‐up. (Those for inpatient rehabilitation only are: RR 0.88, 95% CI 0.80 to 0.98; 13 studies, 3036 participants.) Of note is that the exploratory subgroup analysis by type of intervention (GORU, HFP, MARU, OG programme; and supported discharge and home‐based rehabilitation) showed no evidence of subgroup differences (test for interaction: P = 0.39; I² = 2.5%). Similar observations apply to pooled data for mortality at the end of scheduled follow‐up (RR 0.92, 95% CI 0.81 to 1.05; I² = 0%; 21 studies, 4350 participants; Analysis 8.2) compared with inpatient data only (RR 0.91, 95% CI 0.80 to 1.05; I² = 0%; 18 studies, 3973 participants; Analysis 1.5).

8.1. Analysis.

Comparison 8: Exploratory analysis: inpatient and supported discharge (home‐based) settings, Outcome 1: 'Poor outcome' (long‐term follow‐up): subgrouped by intervention type

8.2. Analysis.

Comparison 8: Exploratory analysis: inpatient and supported discharge (home‐based) settings, Outcome 2: Mortality (end of scheduled follow‐up): subgrouped by intervention type

Discussion

Summary of main results

This updated review includes 28 trials involving a total of 5351 randomised participants, almost all of whom had undergone surgery for hip fracture. Four trials included women only, and women were in the majority for the other 24 trials. The mean age of trial participants ranged from 76.5 years to 87 years. Care was provided in an inpatient setting – thus, before hospital discharge – for both intervention groups in 21 trials. Care was provided predominantly or exclusively in the community (ambulatory rehabilitation) for the other seven trials. One of the 21 inpatient trials compared two different types of multidisciplinary rehabilitation.

Inpatient multidisciplinary rehabilitation versus usual care

The evidence for this comparison in this setting, which was tested by 20 trials, is summarised in Table 1.

• There is moderate‐certainty evidence (data from 13 studies) that multidisciplinary rehabilitation in an inpatient setting results in fewer cases of 'poor outcome' (death or deterioration in residential status, generally requiring institutional care) at 6 to 12 months' follow‐up. Based on an illustrative risk of 347 people with hip fracture with poor outcome in 1000 people followed up between 6 and 12 months (data based on the median control group risk data from the 13 studies), this equates to 41 (95% CI 7 to 69) fewer people with poor outcome after multidisciplinary rehabilitation. Expressed in terms of numbers needed to treat for an additional harmful outcome (NNTH), 25 people (95% CI 15 to 100) would need to be treated to avoid one 'poor outcome'. Subgroup analysis by type of intervention showed no evidence of subgroup differences. The very low‐certainty evidence means we have very little confidence in findings from one trial of a marginally better quality of life in the multidisciplinary rehabilitation group. Of note, however, is that the Hip Fracture Programme was tested in eight of these studies, including three of the four recently completed studies.

• There is low‐certainty evidence (data from 11 studies) that multidisciplinary rehabilitation in an inpatient setting may result in fewer deaths in hospital but could also result in slightly more. Based on an illustrative risk of 75 people with hip fracture dying in hospital in 1000 people followed up until hospital discharge (data based on the median control group risk data from the 11 studies), this equates to 17 fewer (95% CI 31 fewer to 3 more) people dying in the multidisciplinary rehabilitation group.

• There is low‐certainty evidence (data from 18 studies) that multidisciplinary rehabilitation in an inpatient setting may result in fewer deaths at 4 to 12 months' follow‐up but could also result in slightly more. Based on an illustrative risk of 182 people with hip fracture dying in hospital in 1000 people followed up (data based on the median control group risk data from the 18 studies), this equates to 16 fewer (95% CI 36 fewer to 10 more) people dying in the multidisciplinary rehabilitation group. Subgroup analysis by type of intervention showed no evidence of subgroup differences.

• The evidence based on a variety of measures in 11 trials for assessing greater dependence in personal activities of daily living at one to four months' follow‐up was of very low certainty. This means we have very little confidence in the mixed findings of some or no difference from multidisciplinary rehabilitation on dependence in activities of daily living in survivors.

• A similar conclusion applied for evidence from 13 studies on greater dependence in personal activities of daily living at 6 to 12 months' follow‐up.

• There is low‐certainty evidence (data from 5 studies) that multidisciplinary rehabilitation in an inpatient setting may result in less decline in mobility in survivors at 6 to 12 months' follow‐up. Based on an illustrative risk of 403 people with hip fracture with greater dependency in mobility in 1000 people followed up between 6 and 12 months (data based on the median control group risk data from the 5 studies), this equates to 68 (95% CI 8 to 116) fewer people with poorer mobility after multidisciplinary rehabilitation.

• There was no evidence on long‐term hip‐related pain.

• We included delirium in hospital, when delirium or confusion was documented as a complication, as an extra critical outcome given the growing awareness of the importance of delirium in this setting. There is low‐certainty evidence (data from 4 studies) that multidisciplinary rehabilitation in an inpatient setting may result in fewer cases of delirium in hospital.

Comparison of different types of inpatient multidisciplinary rehabilitation

The only study comparing different types of inpatient multidisciplinary rehabilitation tested the effects of adding an 'integrated care pathway' to multidisciplinary rehabilitation. This quasi‐RCT of 162 participants provided very low‐certainty evidence, which means we have very little confidence in findings of little to no between‐group differences in 'poor outcome', mortality, inability to walk as before or any nursing home stay post‐discharge, quality of life, and dependency.

Ambulatory rehabilitation

We reported on four comparisons – one of which was split according to participants' former residence – tested in seven trials. Five of these trials were newly included in this update. The comparisons are considered below.

Supported discharge and multidisciplinary home rehabilitation versus usual care

Intervention solely or mainly aimed at people living in their own homes

The evidence for this comparison in this setting, which was tested by three trials, is summarised in Table 2. The very low‐certainty evidence, where the wide confidence intervals crossed the line of no effect, means that we have very little confidence in the findings of little to no between‐group difference in 'poor outcome' (mortality or move to a higher level of care or inability to walk) at one year (3 studies, 377 participants); in quality of life at one year (1 study, 56 participants); in mortality at four months (2 studies, 275 participants) or at 12 months (2 studies, 214 participants); in being independent in personal activities of daily living (data presented for 1 study, 159 participants); in a permanent move to a higher level of care (2 studies, 168 participants); or being unable to walk (2 studies, 214 participants). None of the trials reported on hip or lower‐limb pain.

Intervention for people living in a nursing home setting

The evidence for this comparison in this setting, which was tested by one trial of 240 participants, is summarised in Table 2. There is low‐certainty evidence that there may be no or minimal between‐group differences at 12 months in 'poor outcome' defined as dead or unable to walk; or in mortality at 4 months or at 12 months. The very low‐certainty evidence for the remaining outcomes reflected the large loss to follow‐up, predominantly from death. This means we have very little confidence in the findings of no between‐group differences in quality of life at 12 months (125 participants); dependency at 4 weeks (202 participants) or at 12 months (125 participants); inability to walk at 12 months (130 participants); or pain at 12 months (126 participants).

Intensive with less intensive multidisciplinary home rehabilitation

An intervention that aimed to double the number of weekly contacts at the patient's home from a multidisciplinary rehabilitation team was tested by one trial that included 71 participants recently discharged from hospital after hip fracture surgery. The very low‐certainty evidence available for this comparison means that we have very little confidence in the lack of between‐group differences in 'poor outcome', mortality, quality of life, and dependency in activities of daily living at one year follow‐up. Mobility and pain were not reported. Of note is that the aimed doubling of the number of contacts in the intervention group was not achieved.

Multidisciplinary care, including progressive resistance training for one year, versus usual care

This was tested in one trial that included 124 participants. The very low‐certainty evidence, where the wide confidence intervals typically crossed the line of no effect, means that we have very little confidence in the findings that marginally favoured the intervention for 'poor outcome', death, or survivors in institutional care, basic activities of daily living or use of assistive devices. Quality of life and pain were not reported.

Post‐discharge multidisciplinary clinic, with referral for further rehabilitation as necessary, versus usual care

This intervention, which started between 3 and 12 months post‐fracture, was tested in one trial of 53 community‐dwelling older adults. Aside from receiving confirmation that no participants died or moved to residential care during the study, there are as yet no data available on outcomes relevant to this review.

Overall completeness and applicability of evidence

Completeness of the evidence

The review now includes 28 trials involving a total of 5351 randomised participants. Focusing on the three main comparisons presented in the summary of findings tables:

• of the 20 trials that compared multidisciplinary rehabilitation versus usual care in the inpatient setting, data for 'poor outcome' at final follow‐up were available for only 13 trials and 72% of participants (3036/4214). Data were most complete for long‐term mortality (18 trials, 3973 participants); only available for three trials for quality of life measures; and not available for long‐term hip pain. Dependence in activities of daily living was reported for 11 trials up to 4 months, and 13 trials at long‐term follow‐up. However, the variety and incompatibility of measures, as well as incomplete data for dependence in activities of daily living, substantially reduced the potential for pooling.

• of the three ambulatory rehabilitation trials, involving 377 participants, that compared supported discharge and multidisciplinary rehabilitation versus usual care for home dwellers, complete data were available for 'poor outcome' (although a different definition of this outcome was accepted for one trial), and long‐term mortality. Only one of these trials, reporting data for 56 participants, reported on quality of life. Again, no data were available for hip pain.

• data were available for all main outcomes for the other ambulatory rehabilitation trial that tested a similar intervention, but in a nursing home setting. However, there was a large loss to follow‐up, predominantly from death, in this trial.

Applicability of the evidence

The characteristics of multidisciplinary rehabilitation and 'usual care' varied considerably in the 28 included trials, and all findings need to be viewed in the context of the clinical heterogeneity of the trial interventions, trial populations and outcome measurement. Our observation in Handoll 2009, that the included trials are "a disparate group", still applies. Nonetheless, all study populations fit with the general characteristics, in terms of age and gender, of the typical population of older people with hip fracture, and the interventions are consistent with the broad conceptual basis of multidisciplinary rehabilitation, where rehabilitation is delivered by a multidisciplinary team, supervised by a geriatrician, rehabilitation physician or other appropriate physician. In the following, we consider applicability in terms of participants, interventions and outcomes.

COVID‐19 pandemic

All included trials were conducted before the onset of the COVID‐19 pandemic. In our view, the basic characteristics in terms of population, interventions and outcomes remain as before. However, the typical population for hip fracture is the one most vulnerable to the serious consequences of COVID‐19, as illustrated by Hall and colleagues’ finding that COVID‐19 was independently associated with a threefold increased 30‐day mortality rate for this population in Scotland (Hall 2021). Hall 2021 noted that length of stay was the only modifiable risk factor for infection, highlighting "the importance of high‐quality and timely care in this patient group". In some countries, where COVID‐19 has largely been contained, care of this vulnerable population is essentially as before but in others, the impact of COVID‐19 on health and social care systems has been immense, and inevitably has impacted and will continue to impact the access to appropriate care for these individuals. It has been estimated that existing rehabilitation services in 60% to 70% of countries have been disrupted due to the COVID‐19 pandemic (WHO 2020).

Populations

To aid consideration of the representativeness of the study populations, we prepared a table featuring key characteristics; see Appendix 9. In our judgement of the representativeness (see 'Target population'), we considered this applied to all studies except Jette 1987, where there was not enough information to be certain. However, for many studies, the population represented a key subgroup. For example, all participants of Tseng 2019 had diabetes, and all participants in Crotty 2019 and Uy 2008 were nursing home residents. Being able to walk, return to living at home, having adequate cognition (absence of dementia), or one or more of these, also defined key subpopulations, particularly for some ambulatory rehabilitation trials. The typical living situation for older people also varies between countries, and we noted in Appendix 9 the possibility of more home dwellers in Italy (Baroni 2019) than in some other countries (e.g. the UK). The same observation applies to the studies based in Taiwan (Shyu 2008; Shyu 2013a; Tseng 2019), where we also emphasised that "Taiwan differs substantially in case mix, culture and social organisation from 'Western' societies".

Interventions

When revising our protocol for this update, we set out a plan to collect data on the key components of the multidisciplinary rehabilitation tested in the included trials, in order to examine and summarise these in terms of common components. To this end, we developed a pro forma, with a primary focus on team members and treatment components, and piloted the pro forma on six studies. Our pilot showed that this was a substantial project, which was not essential for the review and, moreover, could seriously impede the review's completion. Additionally, the pilot and consideration of other trials indicated that under‐reporting of the intervention and control groups, especially in older trials, was a serious problem, as well as the clinical heterogeneity in the trial settings. The under‐reporting is also illustrated in Appendix 10, which summarises the information on key items of early treatment of hip fractures, with a particular focus on assessing how representative these are in current practice.

More recent trials have provided more detailed accounts of the interventions and standard care. Notable are Prestmo 2015, where the development and delivery of patient treatment for the 'Trondheim Hip Fracture Trial' are reported in Saltvedt 2012; and Crotty 2019, where the usual care provided in nursing homes was described in Killington 2020.

Whilst we have provided a classification of different types of multidisciplinary rehabilitation programmes, the subgroup analyses do not show differences in outcomes by type of programme, and the overall analyses do not show evidence of important heterogeneity.

Inpatient care is provided in the hospital setting in the included trials. However, models of treatment have been developed whereby the intervention can be provided in a variety of settings. These can range from a ward providing acute hospital treatment and early rehabilitation (for example, the Trondheim Hip Fracture Trial) to an area in a care home where specialised rehabilitation is provided (for example, this is common in the Netherlands). The pressure to reduce length of stay in the acute setting is likely to influence this trend. Future iterations of this review will need to address these and other developments to the intervention.

Outcomes

The issues relating to applicability of outcomes revolve mainly round availability and data incompleteness. This particularly applied to quality of life and functional outcomes. However, poor definitions of outcome measurement, presentation of data for components – rather than overall results – of tools, such as the SF‐36, and use of modified or non‐validated scales also hamper assessment of applicability. Of note is that non‐return to pre‐fracture function or mobility can also be problematic as it is susceptible to measurement error at baseline from recall or at time of assessment. There is scope for standardisation for activities of daily living; some have proposed use of the Barthel Index rather than the Katz Index on account of the former's better score distribution (Liem 2013). Overall, the ability to investigate specific outcomes (such as mobility) that are relevant to older people is very limited, due to the lack of inclusion of suitable measures in the trials.

Certainty of the evidence

We have reported our judgements on the GRADE certainty of evidence in our accounts for all comparisons in Effects of interventions and restated these in the summary of findings tables for three comparisons. The gradings ranged from 'moderate certainty', where we are moderately confident in the effect estimate, for one outcome alone ('poor outcome'; Table 1), through to 'very low certainty', where we have very little confidence in the effect estimate, for the majority of outcomes.

We downgraded all evidence one level for study limitations that could or would result in serious risk of bias. Although most trials appeared well designed – an impression sometimes enhanced upon gaining additional information from trial investigators – all were at unclear or at high risk of bias for some domains, notably blinding‐related and incomplete data domains, and possible confounding related to potential or known imbalances in key baseline characteristics (e.g. cases with dementia in Huusko 2002; a younger treatment group with higher cognitive function in Kennie 1988). Some questions remained unanswered on conduct of some trials, including those relating to performance and contamination biases.

We downgraded the evidence for most outcomes either one or two levels for serious or very serious imprecision. This was typically because of insufficient data, such as few events, wide confidence intervals, or both.

We rarely downgraded for serious inconsistency. Notably, where pooling was performed, the I² values were very low and indicated unimportant heterogeneity at most. In assessing the mixed outcome measurement evidence for dependency in activities in daily living for the first comparison, we downgraded the evidence one level for serious inconsistency based on a general view of the variation of effect in the results from the group of trials reporting this outcome (Table 1).

We downgraded for indirectness in a very few cases. For example, we judged 'poor outcome' at hospital discharge was vulnerable to issues related to indirectness and bias, such as differences in discharge policies and supply of alternative accommodation, that hindered applicability of the results.

We did not downgrade for publication bias. Inspection of funnel plots for 'poor outcome' (13 trials) and mortality at 4 to 12 months for our first comparison did not indicate publication bias (plots not shown). Data for other comparisons were not available to assess this.

Potential biases in the review process

Given the sustained effort over the years in searching for trials in this area, and the continued involvement of one author (IC) and his colleagues in primary research, we think it is unlikely that we have overlooked fully published trials in this relatively slow‐moving research area. It is hard to ascertain whether we have missed unpublished trials, including any in non‐English reports. The need to obtain clarification on the trial status of Baroni 2019 (included), Baroni 2016 (in Studies awaiting classification), and multiple publications, including abstracts, over several years for several trials illustrates how problematic the literature in this area can be.

Whilst we consider that we have included and excluded trials appropriately, it was sometimes hard to decide whether (included) trials evaluated multidisciplinary rehabilitation or not. To some extent, consideration of what we would accept as a legitimate intervention when adjusting the scope of our review for this update (as described in Differences between protocol and review) made this easier. This was mainly because we relaxed our criteria regarding the role of the geriatrician or rehabilitation physician, and accepted that whilst there should be a strong component of physician involvement, often relating to assessment, it did not have to continue subsequently. This resulted in the inclusion of Sanchez Ferrin 1999, which was excluded in the previous version of our review (Handoll 2009). We protected against bias related to inappropriate study selection decisions through independent study selection by two review authors and arbitration where required.

Despite the clear heterogeneity of trial comparisons, populations, outcome assessment and some aspects of trial quality, we pooled data for several 'hard' outcomes provided by the trials comparing multidisciplinary rehabilitation with usual care in the inpatient setting. This activity in itself is not a potential bias in the review process, but we acknowledge that it does hamper interpretation and considerations of external validity. Sensitivity analyses did not show important differences in findings. Whilst performed in a limited and exploratory way for different types of multidisciplinary rehabilitation, we acknowledge there were insufficient data for this and for the examination of the effects of clinical heterogeneity through subgroup analysis.

In this version of the review, we have extended our focus to various measures of function and independence, that we had acknowledged had received less attention than merited (Handoll 2009). Nonetheless, as anticipated, the data for the various validated measures of function in use are generally missing or incomplete and, where available, only limited pooling was possible and appropriate. Whilst unlikely to be a source of bias, we took care in our selection of which measures to summarise for outcomes, such as extent of dependence in activities of daily, measured in a variety of ways, and took into account the evidence and claims from all trials reporting an outcome.

A strength of the review is the often successful acquisition of extra data and details from trial investigators. This can add to the complexity of the data extraction and checking processes, and lead to disparities between the published and subsequently provided data. In this review, this process has reinforced the perception of the inherent dangers of reporting percentages without the data from which they are derived.

Agreements and disagreements with other studies or reviews

This review stems from Handoll 2009, which in turn stemmed from Cameron 2001, of which the fourth update appeared in Issue 2, 2003 of the Cochrane Library. We noted in Handoll 2009 that, despite the accumulation of evidence, our conclusions had remained consistent with those of Cameron 2001; namely: "While there is no conclusive evidence of the effectiveness of multidisciplinary inpatient rehabilitation following hip fracture surgery in older people, there is a trend towards effectiveness in all main outcomes." Our current review has been expanded in scope, with some re‐evaluation of the inclusion criteria and updating of review methods, including the incorporation of GRADE for assessing the certainty of the evidence. The addition of data for 'poor outcome' from five, and for long‐term mortality from seven, of the nine newly included studies with 'usual care' controls conducted in the inpatient setting, has strengthened the evidence, such that we are now 'moderately confident' that multidisciplinary rehabilitation reduces the risk of 'poor outcome'. Also of note is the relative increase in the number of trials of ambulatory rehabilitation from two to seven.

In Handoll 2009, we referred to an earlier systematic review by Halbert 2007, which had concluded that people who received multidisciplinary rehabilitation were at lower risk of a poor outcome, defined as death or admission to a nursing home at discharge from rehabilitation. Our observation at the time, in terms of the timing of outcome measurement, still holds: "Our review focused on long‐term outcome, which ranged between four months and one year, but crucially was less susceptible to differences in discharge policies or practicalities between intervention groups in individual trials." We noted also that while Halbert 2007 had pooled data from Crotty 2003 with those from the inpatient rehabilitation trials, we had received feedback at the protocol stage for our review that had persuaded us that we should only pool from different settings in an exploratory way. Our exploratory analysis for this review update showed that adding data from the three ambulatory rehabilitation trials, including Crotty 2003, which tested supported discharge and own‐home‐based rehabilitation, to those of the inpatient rehabilitation trials made almost no difference for either poor outcome or long‐term mortality.

We note here two recent reviews: Moyet 2018 and Nordstrom 2018. Moyet 2018, which investigated orthogeriatric models of care, included 18 studies of mixed designs; the four included RCTs are also included in our review (Gilchrist 1988; Huusko 2002; Vidan 2005; Watne 2014). Nordstrom 2018 conducted a meta‐analysis of the effects of rehabilitation by geriatric interdisciplinary teams for people with hip fracture, and reported that the intervention "increases physical function and mobility significantly compared with conventional care" but that discharge "to one’s own home and survival are not influenced". Nordstrom 2018 included seven studies, of which six are included in our review (Huusko 2002; Naglie 2002; Prestmo 2015; Stenvall 2007a; Vidan 2005; Ziden 2008). The other study, a quasi‐RCT, was excluded from our review because of compromised methods (Gonzalez‐Montalvo 2010). Ziden 2008 was considered to be conducted in an ambulatory setting in our review.

We found insufficient data to perform viable subgroup analysis of 'poor outcome' or mortality, according to participant cognitive function. A related Cochrane Review on enhanced rehabilitation and care models for adults with dementia following hip fracture surgery was updated in 2020 (Smith 2020). This reported evidence from a total of 555 participants with dementia or cognitive impairment included in seven trials investigating models of care for all older people following hip fracture. Six of these trials (where different, study IDs in Smith 2020 are in brackets) are included in our review: Huusko 2002 (Huusko 2000); Marcantonio 2001; Shyu 2008 (Shyu 2012); Stenvall 2007a (Stenvall 2012); Uy 2008; Watne 2014 (Wyller 2012). Smith 2020 found low‐certainty evidence that "some of the models of enhanced rehabilitation and care used in the included trials may show benefits over usual care for preventing delirium and reducing length of stay for people with dementia who have been treated for hip fracture". The certainty for all other results was very low. Smith 2020 proposed that determining "the optimal strategies to improve outcomes for this growing population of patients should be a research priority".

Authors' conclusions

Implications for practice.

In a hospital inpatient setting, there is moderate‐certainty evidence that rehabilitation after hip fracture surgery, when delivered by a multidisciplinary team, supervised by a geriatrician, rehabilitation physician or other appropriate physician, results in fewer cases of 'poor outcome' (death or deterioration in residential status). There is low‐certainty evidence that multidisciplinary rehabilitation may result in fewer deaths in hospital and at 4 to 12 months; however, it may also result in slightly more. There is low‐certainty evidence that multidisciplinary rehabilitation may reduce the numbers of people with poorer mobility at 12 months. No conclusions can be drawn on other outcomes for which the evidence is of very low certainty.

The generally very low‐certainty evidence available for supported discharge and multidisciplinary home rehabilitation means we are very uncertain whether the findings of little or no difference for all outcomes between the intervention and usual care is true.

Implications for research.

Future research evaluating the effectiveness of specialised inpatient rehabilitation needs to pay attention to several factors. A modest effect size, such as a 10% reduction in 'poor outcome', would be an important result for this population. To obtain definitive evidence for such a reduction, larger trials are needed. Measures of outcome should be standardised (particularly with regard to functional status) and assessed by a blinded observer. Trials should include careful monitoring of direct and indirect costs, as well as assessing cost‐effectiveness and carer burden. Components of the interventions used in trials, and details of the treatment provided to the control group, should be carefully specified. Some account should be taken of other strategies for rehabilitation, including those of early discharge and a greater emphasis on community‐based rehabilitation.

Participant characteristics, including cognitive status, should be clearly presented, and if subgroup analyses are considered, these should be a priori, sufficiently powered, and an analysis of the outcome of participants with or without evident dementia considered. Although this basic dichotomy would be instructive, further categorisation by participant characteristics would be valuable, as within the hip fracture population, there are very different groups of older people, ranging from independent older people to older people with severe frailty and cognitive impairment. Achieving an international consensus on a suitable classification, such as based on pre‐fracture dependence in personal activities of daily living or extent of frailty, that could be used in future studies would be likely to reduce the heterogeneity in study outcomes.

As the effectiveness of multidisciplinary rehabilitation overall has now been established, evaluation of its components should occur, with an emphasis on optimising cost‐effectiveness. New service models for rehabilitating people with hip fracture should be rigorously evaluated before they are implemented and become established.

As well as trial registration and making the trial protocol publicly available, it is important that trial reports conform to the requirements of the nonpharmacologic CONSORT statement (Boutron 2017). Complying with this version of CONSORT should mean that adequate descriptions of the interventions under test will be available.

For the sake of continuity, we have retained the protocol scope and title of this Cochrane Review with its emphasis on 'rehabilitation'. However, when assessing the scope of the next update for this topic area, consideration should be given to whether the scope needs refining to fit with the broader concept of 'multidisciplinary management of hip fracture'. This would be consistent with recommendations and terminology in other publications (Cameron 2000; Ranhoff 2019; Riemen 2016; Swift 2016).

What's new

Date	Event	Description
25 May 2021	New search has been performed	The key changes for this version of the review, published in 2021, are summarised below. We revised the protocol beforehand to incorporate: changes to the scope. In particular, the relaxation in the criteria for geriatrician or rehabilitation physician involvement; the inclusion of comparisons of different models of multidisciplinary rehabilitation; and a restructuring of the outcomes, with a shift to and extra collection of interim outcome at around four months; an enhanced classification and description of the interventions; a separate examination and assessment of the 'usual' care provided in trials in relation to current management, such as early mobilisation; updated methodology such as GRADE assessment in terms of certainty and the production of Summary of findings tables; additional sensitivity analyses. The Background and related sections were updated. The search was updated to October 2020. Fifteen more trials were included (Ashe 2019; Baroni 2019; Chong 2013; Crotty 2019; Jette 1987; Karlsson 2016; Marcantonio 2001; Prestmo 2015; Sanchez Ferrin 1999; Shyu 2013a; Singh 2012; Tseng 2019; Uy 2008; Watne 2014; Ziden 2008). Four extra comparisons were added in. An additional author (CP) was added to the byline.
25 May 2021	New citation required and conclusions have changed	All conclusions were framed in relation to GRADE assessment of certainty of the available evidence. Additional evidence resulted in stronger conclusions on the effects of multidisciplinary rehabilitation in an inpatient setting. Additional evidence was available for six other comparisons, five of which applied to the ambulatory setting.

History

Protocol first published: Issue 2, 2008
Review first published: Issue 4, 2009

Date	Event	Description
7 July 2008	Amended	Converted to new review format
19 February 2008	Amended	This protocol, published in Issue 2, 2008, precedes the update and scope expansion of a former review ("Co‐ordinated multidisciplinary approaches for inpatient rehabilitation of older patients with proximal femoral fractures"; last updated Issue 2, 2003), which included inpatient rehabilitation only.

Appendices

Appendix 1. Hip Fracture Programme (extract from NICE guidelines 2011)

4.2.8 Multidisciplinary management

➢ From admission, offer patients a formal, acute orthogeriatric or orthopaedic ward based Hip Fracture Programme that includes all of the following:

• orthogeriatric assessment
• rapid optimisation of fitness for surgery
• early identification of individual goals for multidisciplinary rehabilitation to recover mobility and independence, and to facilitate return to prefracture residence and long‐term wellbeing.
• continued, coordinated, orthogeriatric and multidisciplinary review
• liaison or integration with related services, particularly mental health, falls prevention, bone health, primary care and social services.
• clinical and service governance responsibility for all stages of the pathway of care and rehabilitation, including those delivered in the community.

Appendix 2. Outcomes relating to the Cochrane Programme Grant on hip fracture management

The following set of outcomes resulted from consultation with stakeholders, including a formal consultation workshop in January 2018, consideration of the UK core outcome set for hip fracture trials (Haywood 2014), and the NICE clinical guideline Hip fracture: management (CG124). There is an emphasis on outcome at around four months.

[GRADE] 'Critical'

• Health‐related quality of life (HRQoL) (e.g. SF‐36, EQ‐5D)

• Delirium

• Mortality (early (< 4 months) and late (> 4 months)

• Activities of daily living (e.g. Barthel Index (BI), Functional Independence Measure (FIM))

• Functional status (region specific) (e.g. hip rating questionnaire, Harris Hip Score, Oxford Hip Score)

• Mobility (e.g. indoor/outdoor walking status, mean mobility score, self‐reported walking scores)

• Adverse events, including unplanned return to theatre

[GRADE] 'Important'

• Pain (verbal rating or visual analogue scale)

• Length of stay

• Residence

Appendix 3. Previous types of outcome measures

In the 2009 version of this review, we presented the types of outcome measures as shown below (Handoll 2009).

The primary outcome is 'poor outcome', defined as death or deterioration of functional status leading to increased dependency in the community or admission to institutional care.

Information was sought on this primary outcome, and its constituent outcomes, and other outcomes as listed below. These were treated as secondary outcomes.

• All‐cause mortality;

• Morbidity, including postoperative complications and treatment of newly recognised but pre‐existing disease or illness;

• Patients' postoperative functional status including cognitive functioning, mobility and ability to perform activities of daily living;

• Resource use, including length of hospital stay and subsequent admission rates to, and days spent in, institutional care;

• The level of care and extent of support required or provided on discharge;

• The patients' perceived quality of life, and well‐being, after discharge;

• Carer burden and stress;

• Direct, indirect, hidden and opportunity costs (the cost of treatments forgone);

• Adherence to rehabilitation under examination (including changes to other types of rehabilitation).

Timing of outcome assessment

Results were collected for the final follow‐up time for which these were available. We, however, took note where trial results may not have been representative of final functional status. This was based on a minimum follow‐up of six months from time of injury.

Appendix 4. Search strategies up to January 2016

CENTRAL

Stage 1 (Wiley Online Library; 2013 Issue 9)

#1 MeSH descriptor: [Hip Fractures] explode all trees (1019)
#2 ((hip* or ((femur* or femoral*) near/3 (neck or proximal))) near/4 fracture*):ti,ab,kw (2083)
#3 #1 or #2 (2083)
#4 MeSH descriptor: [Physical Therapy Modalities] this term only (2368)
#5 MeSH descriptor: [Rehabilitation] explode all trees (13009)
#6 MeSH descriptor: [Activities of Daily Living] this term only (3443)
#7 MeSH descriptor: [Early Ambulation] this term only (273)
#8 MeSH descriptor: [Occupational Therapy] explode all trees (483)
#9 MeSH descriptor: [Critical Pathways] this term only (235)
#10 MeSH descriptor: [Physical Therapy Specialty] this term only (101)
#11 MeSH descriptor: [Rehabilitation Nursing] explode all trees (37)
#12 (rehab* or (early near/1 (mobil* or discharg*or ambulat*)) or occupational therap* or physiotherap* or physical therap* or multidisciplin*):ti,ab,kw (22798)
#13 geriatr* or geriatr*‐orthop*or ortho*‐geriatr* or orthogeriatr*:ti,ab,kw (3265)
#14 "hospital at home":ti,ab,kw (70)
#15 MeSH descriptor: [Hospitalization] explode all trees (10991)
#16 Any MeSH descriptor with qualifier(s): [Rehabilitation ‐ RH] (12052)
#17 #4 or #5 or #6 or #7 or #8 or #9 or #10 or #11 or #12 or #13 or #14 or #15 or #16 (47478)
#18 #3 and #17 in Trials (355)

Stage 2 (CRS Online; 2013 Issue 10 to 2016 Issue 1)

#1 MESH DESCRIPTOR Hip Fractures EXPLODE ALL TREES (962)
#2 (((hip* or ((femur* or femoral*) adj3 (neck or proximal))) adj4 fracture*)):TI,AB,KY (2208)
#3 #1 OR #2 (2208)
#4 MESH DESCRIPTOR Physical Therapy Modalities (2522)
#5 MESH DESCRIPTOR Rehabilitation EXPLODE ALL TREES (15943)
#6 MESH DESCRIPTOR Activities of Daily Living (3823)
#7 MESH DESCRIPTOR Early Ambulation (284)
#8 MESH DESCRIPTOR Occupational Therapy EXPLODE ALL TREES (525)
#9 MESH DESCRIPTOR Critical Pathways (143)
#10 MESH DESCRIPTOR Physical Therapy Specialty (105)
#11 MESH DESCRIPTOR Rehabilitation Nursing EXPLODE ALL TREES (47)
#12 ((rehab* or (early adj1 (mobil* or discharg*or ambulat*)) or occupational therap* or physiotherap* or physical therap* or multidisciplin*)):TI,AB,KY31537
#13 (geriatr* or geriatr*‐orthop*or ortho*‐geriatr* or orthogeriatr*):TI,AB,KY (4498)
#14 ("hospital at home"):TI,AB,KY (61)
#15 MESH DESCRIPTOR Hospitalization EXPLODE ALL TREES (10197)
#16 #4 OR #5 OR #6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13 OR #14 OR #15 (52887)
#17 #3 AND #16 (477
#18 31/10/2013 TO 29/02/2016:DL (215492)
#19 #17 AND #18 (164)

MEDLINE

Stage 1 (Ovid Online; 2009 to August Week 4 2013)

1 exp Hip Fractures/ (17601)
2 ((hip$ or ((femur$ or femoral$) adj3 (neck or proximal))) adj4 fracture$).tw. (18566)
3 or/1‐2 (24233)
4 exp Physical Therapy Modalities/ or Rehabilitation/ or "Activities of Daily Living"/ or Early Ambulation/ or Occupational Therapy/ or Critical Pathways/ or Physical Therapy Specialty/ (199567)
5 Rehabilitation Nursing/ (1085)
6 (rehab* or (early adj1 (mobil* or discharg*or ambulat*)) or occupational therap* or physiotherap* or physical therap* or multidisciplin*).tw. (177454)
7 (geriatr* or geriatr*orthop* or orthop?edic‐geriatr* or ortho*‐geriatr* or orthogeriatr*).tw. (32903)
8 hospital at home.tw. (306)
9 exp Hospitalization/ (157721)
10 rh.fs. (163789)
11 or/4‐10 (587583)
12 and/3,11 (4087)
13 Randomized controlled trial.pt. (384981)
14 Controlled clinical trial.pt. (89120)
15 randomized.ab. (300266)
16 placebo.ab. (161737)
17 Drug therapy.fs. (1748887)
18 randomly.ab. (212565)
19 trial.ab. (316255)
20 groups.ab. (1352336)
21 or/13‐20 (3382653)
22 exp Animals/ not Humans/ (4027008)
23 21 not 22 (2899484)
24 and/12,23 (1050)
25 (2009* or 2010* or 2011* or 2012* or 2013*).ed. (4936930)
26 24 and 25 (404)

Stage 2 (Ovid Online; August 2013 to January Week 2 2016)

1 exp Hip Fractures/ (19241)
2 ((hip* or ((femur* or femoral*) adj3 (neck or proximal))) adj4 fracture*).tw. (20767)
3 or/1‐2 (27097)
4 exp Physical Therapy Modalities/ or Rehabilitation/ or "Activities of Daily Living"/ or Early Ambulation/ or Occupational Therapy/ or Critical Pathways/ or Physical Therapy Specialty/ (200587)
5 Rehabilitation Nursing/ (1219)
6 (rehab* or (early adj1 (mobil* or discharg*or ambulat*)) or occupational therap* or physiotherap* or physical therap* or multidisciplin*).tw. (203458)
7 (geriatr* or geriatr*orthop* or orthop?edic‐geriatr* or ortho*‐geriatr* or orthogeriatr*).tw. (35225)
8 hospital at home.tw. (299)
9 exp Hospitalization/ (176044)
10 rh.fs. (173328)
11 or/4‐10 (633914)
12 and/3,11 (4474)
13 randomized controlled trial.pt. (404220)
14 controlled clinical trial.pt. (89993)
15 randomized.ab. (333041)
16 placebo.ab. (165383)
17 drug therapy.fs. (1810104)
18 randomly.ab. (240533)
19 trial.ab. (343915)
20 groups.ab. (1506012)
21 or/13‐20 (3635421)
22 exp animals/ not humans.sh. (4173052)
23 21 not 22 (3125877)
24 and/12,23 (1124)
25 (201308* or 201309* or 201310* or 201311* or 201312* or 2014* or 2015* or 2016*).ed,dc. (3084896)
26 24 and 25 (225)

Embase

Stage 1 (Ovid Online; 2009 to 2013 Week 36)

1 exp Hip Fracture/ (27944)
2 ((hip* or ((femur* or femoral*) adj3 (neck or proximal))) adj4 fracture*).tw. (22269)
3 or/1‐2 (33448)
4 exp Rehabilitation/ (208510)
5 Daily Life Activity/ (54030)
6 Mobilization/ (16612)
7 Occupational Therapy/ or exp Physiotherapy/ (62735)
8 Clinical Pathway/ (5495)
9 Hospitalization/ (198978)
10 Patient care/ or Rehabilitation Care/ or exp Elderly Care/ (251561)
11 exp Health Care Delivery/ (1805141)
12 Geriatric Assessment/ (9091)
13 rh.fs. (132559)
14 rehab*.tw. (139085)
15 (early adj1 (mobil* or discharg*or ambulat*)).tw. (3302)
16 (occupational therap* or physiotherap* or physical therap*).tw. (48166)
17 (multidisciplin* or interdisciplinar* or multidisciplinar* or multiprofessional* or multimodal* or integrated).tw. (251860)
18 (geriatr* or geriatr*orthop* or ortho*‐geriatr* or orthogeriatr* or GORU).tw. (45429)
19 hospital at home.tw. (360)
20 or/4‐19 (2647029)
21 3 and 20 (11135)
22 Randomized controlled trial/ (355507)
23 Clinical trial/ (888404)
24 Controlled clinical trial/ (405007)
25 Randomization/ (63366)
26 Single blind procedure/ (18204)
27 Double blind procedure/ (117458)
28 Crossover procedure/ (38360)
29 Placebo/ (224986)
30 Prospective study/ (248839)
31 ((clinical or controlled or comparative or placebo or prospective* or randomi#ed) adj3 (trial or study)).tw. (719048)
32 (random* adj7 (allocat* or allot* or assign* or basis* or divid* or order*)).tw. (176106)
33 ((singl* or doubl* or trebl* or tripl*) adj7 (blind* or mask*)).tw. (158193)
34 (cross?over* or (cross adj1 over*)).tw. (67900)
35 ((allocat* or allot* or assign* or divid*) adj3 (condition* or experiment* or intervention* or treatment* or therap* or control* or group*)).tw. (224545)
36 RCT.tw. (12493)
37 or/22‐36 (1861488)
38 Case Study/ or Abstract Report/ or Letter/ (905479)
39 37 not 38 (1824021)
40 21 and 39 (2843)
41 (2009* or 2010* or 2011* or 2012* or 2013*).em. (5886015)
42 40 and 41 (1223)

Stage 2 (Ovid Online; 2013 to 2016 Week 04)

1 exp Hip fracture/ (33569)
2 ((hip* or ((femur* or femoral*) adj3 (neck or proximal))) adj4 fracture*).tw. (27586)
3 or/1‐2 (40406)
4 exp Rehabilitation/ (277089)
5 Daily Life Activity/ (63316)
6 Mobilization/ (23521)
7 Occupational Therapy/ or exp Physiotherapy/ (79599)
8 Clinical Pathway/ (6829)
9 Hospitalization/ (243707)
10 Patient care/ or Rehabilitation Care/ or exp Elderly Care/ (295018)
11 exp Health Care Delivery/ (2255341)
12 Geriatric Assessment/ (10769)
13 rh.fs. (134894)
14 rehab*.tw. (173306)
15 (early adj1 (mobil* or discharg*or ambulat*)).tw. (4250)
16 (occupational therap* or physiotherap* or physical therap*).tw. (63254)
17 (multidisciplin* or interdisciplinar* or multidisciplinar* or multiprofessional* or multimodal* or integrated).tw. (327637)
18 (geriatr* or geriatr*orthop* or ortho*‐geriatr* or orthogeriatr* or GORU).tw. (56472)
19 hospital at home.tw. (421)
20 or/4‐19 (3278022)
21 3 and 20 (14245)
22 exp Randomized Controlled Trial/ or exp Single Blind Procedure/ or exp Double Blind Procedure/ or Crossover Procedure/ (445496)
23 (random* or RCT or placebo or allocat* or crossover* or 'cross over' or trial or (doubl* adj1 blind*) or (singl* adj1 blind*)).ti,ab. (1492139)
24 22 or 23 (1571547)
25 (exp Animal/ or Animal.hw. or Nonhuman/) not (exp Human/ or Human cell/ or (human or humans).ti.) (5752721)
26 24 not 25 (1387184)
27 21 and 26 (1798)
28 (2013* or 2014* or 2015* or 2016*).em,dd. (5045464)
29 27 and 28 (610)

WHO ICTRP (February 2016)

1. Hip AND fracture AND rehab* OR Hip AND fracture AND mobil* OR Hip AND fracture AND discharg* OR Hip AND fracture AND ambulat* OR Hip AND fracture AND occupational OR Hip AND fracture AND multidisciplin* OR Hip AND fracture AND interdiscip* OR Hip AND fracture AND activit* OR Hip AND fracture AND pathway Hip AND fracture AND geriatr* OR Hip AND fracture AND orthogeriatr* OR Hip AND fracture AND physical OR Hip AND fracture AND discharge (86 records for 81 trials)

2. Fem* AND fracture AND rehab* OR Fem* AND fracture AND mobil* OR Fem* AND fracture AND discharg* OR Fem* AND fracture AND ambulat* OR Fem* AND fracture AND occupational OR Fem* AND fracture AND multidisciplin* OR Fem* AND fracture AND interdiscip* OR Fem* AND fracture AND activit* OR Fem* AND fracture AND pathway OR Fem* AND fracture AND geriatr* OR Fem* AND fracture AND orthogeriatr* OR Fem* AND fracture AND physical OR Fem* AND fracture AND discharge (35 records for 34 trials)

ClinicalTrials.gov (January 2016)

fracture AND (femur OR hip) AND (rehabilitation OR mobilization OR discharge OR ambulation OR occupational OR multidisciplinary OR activity OR pathway OR geriatric OR orthogeriatric OR physical); received 01/01/2009 (371 records)

Appendix 5. Search strategies: second phase, mainly up to February 2019; with extensions to 21 November 2019 and 15 October 2020

CENTRAL

CRS web; to 2019 Issue 2

#1 MESH DESCRIPTOR Hip Fractures EXPLODE ALL AND CENTRAL:TARGET (1113)
#2 (((hip* or ((femur* or femoral*) adj3 (neck or proximal))) adj4 fracture*)): AB,EH,KW,KY,MC,MH,TI,TO AND CENTRAL:TARGET (3187)
#3 #1 OR #2 (3343)
#4 MESH DESCRIPTOR Physical Therapy Modalities AND CENTRAL:TARGET (3068)
#5 MESH DESCRIPTOR Rehabilitation EXPLODE ALL AND CENTRAL:TARGET (28229)
#6 MESH DESCRIPTOR Activities of Daily Living AND CENTRAL:TARGET (4401)
#7 MESH DESCRIPTOR Early Ambulation AND CENTRAL:TARGET (319)
#8 MESH DESCRIPTOR Occupational Therapy EXPLODE ALL AND CENTRAL:TARGET (644)
#9 MESH DESCRIPTOR Critical Pathways AND CENTRAL:TARGET (174)
#10 MESH DESCRIPTOR Physical Therapy Specialty AND CENTRAL:TARGET (113)
#11 MESH DESCRIPTOR Rehabilitation Nursing EXPLODE ALL AND CENTRAL:TARGET (52)
#12 ((rehab* or (early adj1 (mobil* or discharg*or ambulat*)) or occupational therap* or physiotherap* or physical therap* or multidisciplin*)): AB,EH,KW,KY,MC,MH,TI,TO AND CENTRAL:TARGET (53652)
#13 (geriatr* or geriatr*‐orthop*or ortho*‐geriatr* or orthogeriatr*): AB,EH,KW,KY,MC,MH,TI,TO AND CENTRAL:TARGET (7393)
#14 ("hospital at home"): AB,EH,KW,KY,MC,MH,TI,TO AND CENTRAL:TARGET (82)
#15 MESH DESCRIPTOR Hospitalization EXPLODE ALL AND CENTRAL:TARGET (12260)
#16 #4 OR #5 OR #6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13 OR #14 OR #15 (85803)
#17 #3 AND #16 (846)

Search update 01 Jan 2019 to 21 Nov 2019: 192 hits

Search update 21 Nov 2019 to 16 Oct 2020: 142 hits

MEDLINE

Ovid Online; January 2016 to Week 3 2019

1 exp Hip Fractures/ (22217)
2 ((hip* or ((femur* or femoral*) adj3 (neck or proximal))) adj4 fracture*).tw. (22325)
3 1 or 2 (29179)
4 exp Physical Therapy Modalities/ or Rehabilitation/ or "Activities of Daily Living"/ or Early Ambulation/ or Occupational Therapy/ or Critical Pathways/ or Physical Therapy Specialty/ (232851)
5 Rehabilitation Nursing/ (1382)
6 (rehab* or (early adj1 (mobil* or discharg*or ambulat*)) or occupational therap* or physiotherap* or physical therap* or multidisciplin*).tw. (222898)
7 (geriatr* or geriatr*orthop* or orthop?edic‐geriatr* or ortho*‐geriatr* or orthogeriatr*).tw. (38774)
8 hospital at home.tw. (333)
9 exp Hospitalization/ (216583)
10 rh.fs. (188359)
11 4 or 5 or 6 or 7 or 8 or 9 or 10 (723640)
12 3 and 11 (5290)
13 randomized controlled trial.pt. (476026)
14 controlled clinical trial.pt. (92894)
15 randomized.ab. (377174)
16 placebo.ab. (177577)
17 drug therapy.fs. (2084259)
18 randomly.ab. (261703)
19 trial.ab. (391416)
20 groups.ab. (1628635)
21 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 (4036225)
22 exp animals/ not humans.sh. (4548295)
23 21 not 22 (3443925)
24 12 and 23 (1405)
25 (2016* or 2017* or 2018* or 2019*).ed,dt. (2727589)
26 24 and 25 (359)

Search update 1 Jan 2019 to 21 Nov 2019: 180 hits

Search update 1 Nov 2019 to 15 Oct 2020: 250 hits

Embase

Ovid Online; 2016 to 2019 Week 08

1 exp Hip fracture/ (33665)
2 ((hip* or ((femur* or femoral*) adj3 (neck or proximal))) adj4 fracture*).tw. (33659)
3 1 or 2 (46311)
4 exp Rehabilitation/ (353719)
5 Daily Life Activity/ (80155)
6 Mobilization/ (29512)
7 Occupational Therapy/ or exp Physiotherapy/ (91896)
8 Clinical Pathway/ (7939)
9 Hospitalization/ (320978)
10 Patient care/ or Rehabilitation Care/ or exp Elderly Care/ (351590)
11 exp Health Care Delivery/ (2882545)
12 Geriatric Assessment/ (14466)
13 rh.fs. (143184)
14 rehab*.tw. (210407)
15 (early adj1 (mobil* or discharg*or ambulat*)).tw. (5508)
16 (occupational therap* or physiotherap* or physical therap*).tw. (80123)
17 (multidisciplin* or interdisciplinar* or multidisciplinar* or multiprofessional* or multimodal* or integrated).tw. (452863)
18 (geriatr* or geriatr*orthop* or ortho*‐geriatr* or orthogeriatr* or GORU).tw. (67854)
19 hospital at home.tw. (524)
20 4 or 5 or 6 or 7 or 8 or 9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 (4147352)
21 3 and 20 (18380)
22 exp Randomized Controlled Trial/ or exp Single Blind Procedure/ or exp Double Blind Procedure/ or Crossover Procedure/ (598569)
23 (random* or RCT or placebo or allocat* or crossover* or 'cross over' or trial or (doubl* adj1 blind*) or (singl* adj1 blind*)).ti,ab. (1928182)
24 22 or 23 (2017299)
25 (exp Animal/ or Animal.hw. or Nonhuman/) not (exp Human/ or Human cell/ or (human or humans).ti.) (6048098)
26 24 not 25 (1784074)
27 21 and 26 (2387)
28 (2016* or 2017* or 2018* or 2019*).dc,yr. (5195299)
29 27 and 28 (648)

Search update 01 Jan 2019 to 21 Nov 2019: 219 hits

Search update 1 Jan 2019 to 16 Oct 2020: 466 hits

WHO ICTRP (April 27 2019)

1. Hip AND fracture AND rehab* OR Hip AND fracture AND mobil* OR Hip AND fracture AND discharg* OR Hip AND fracture AND ambulat* OR Hip AND fracture AND occupational OR Hip AND fracture AND multidisciplin* OR Hip AND fracture AND interdiscip* OR Hip AND fracture AND activit* OR Hip AND fracture AND pathway Hip AND fracture AND geriatr* OR Hip AND fracture AND orthogeriatr* OR Hip AND fracture AND physical OR Hip AND fracture AND discharge (168 records for 158 trials; searched first 100 records; 21/03/2013 onwards)

2. Fem* AND fracture AND rehab* OR Fem* AND fracture AND discharg* OR Fem* AND fracture AND multidisciplin* OR Fem* AND fracture AND interdiscip* OR Fem* AND fracture AND activit* OR Fem* AND fracture AND pathway OR Fem* AND fracture AND geriatr* OR Fem* AND fracture AND orthogeriatr* OR Fem* AND fracture AND discharge (53 records for 52 trials; searched all records)

Search update to 21 Nov 2019: 222 hits

ClinicalTrials.gov (April 24 2019)

fracture AND (femur OR hip) AND (rehabilitation OR mobilization OR discharge OR ambulation OR occupational OR multidisciplinary OR activity OR pathway OR geriatric OR orthogeriatric OR physical); posted from 01/01/2016 (88 records)

Search update 01 Jan 2019 to 21 Nov 2019: 136 hits

Appendix 6. Other searches as reported in the 2009 version of the review

As well as following up any reports of relevant trials presented at conferences attended in person (IC), we included the findings from handsearches of Orthopaedic Transactions, supplements of Acta Orthopaedica Scandinavica and the British Volume of the Journal of Bone and Joint Surgery (1996 to 2007). In addition, we handsearched the final programmes of SICOT (1996, 1999 and 2007), the British Orthopaedic Association Congress (2000, 2001, 2002, 2003, 2005 and 2006), and abstracts of the American Orthopaedic Trauma Association annual meetings (1996 to 2006). We scrutinised weekly downloads of "Fracture" articles in new issues of 15 journals (Acta Orthop Scand; Am J Orthop; Arch Orthop Trauma Surg; Clin J Sport Med; Clin Orthop; Foot Ankle Int; Injury; J Am Acad Orthop Surg; J Arthroplasty; J Bone Joint Surg Am; J Bone Joint Surg Br; J Foot Ankle Surg; J Orthop Trauma; J Trauma; Orthopedics) and of "Rehabilitation Medicine" in a further 10 journals (Am J Phys Med Rehabil; Arch Phys Med Rehabil; BMJ; Clin Rehabil; J Am Geriatr Soc; J Rehabil Med; JAMA; Lancet; Phys Ther; Scand J Rehabil Med) from AMEDEO .

Results from a comprehensive search for trials (up to August 1998) for a non‐Cochrane review on rehabilitation following fractures in older people (Cameron 2000) were screened, as well as those from a more recent non‐Cochrane review on multidisciplinary rehabilitation (Halbert 2007).

Appendix 7. Detailed report of the search results of the two‐phased search for this review update (2019)

First phase (up to January 2016)

The search was updated from April 2009 to January 2016 in two stages, and both stages are reported together here. Overall, we screened a total of 3539 records from the following databases: Cochrane Bone, Joint and Muscle Trauma Group Specialised Register (20 records); CENTRAL (519), MEDLINE (629), Embase (1833), the WHO International Clinical Trials Registry Platform (167) and ClinicalTrials.gov (371). We also identified six references to potentially eligible studies from other sources: these were three studies awaiting classification and two ongoing studies from our review (Handoll 2009), and a cost‐effectiveness article of an excluded trial published in March 2017.

The search update resulted in the identification of 21 new studies. Of these, 6 trials (22 references) were selected for inclusion (Prestmo 2015; Shyu 2013a; Singh 2012; Uy 2008; Watne 2014; Ziden 2008); 10 studies (20 references) were excluded; 4 trials (5 references) were classified as 'ongoing trials'; and one trial was classified as 'awaiting classification'. Additional articles were found for two already included trials: Shyu 2008 (5 articles) and Stenvall 2007a (2 conference abstracts).

Five of the 21 studies had been listed in Handoll 2009: Prestmo 2015 was listed as ongoing under Sletwold 2009; Uy 2008 and Ziden 2008 were both listed as awaiting classification under their current study IDs; Lahtinen 2015, now excluded, was listed as awaiting classification under Jalovaara 2009; and one ongoing trial, ISRCTN15738119, now included trial Karlsson 2016 but formerly Gustafson 2009, remained in the same category until the second phase.

Overall, after this first phase, there were 19 included studies, 29 excluded studies, 4 ongoing trials and one study awaiting classification.

Second phase (up to February/March 2019); revised scope implemented

Overall, 1853 records were obtained from the following databases: CENTRAL (846), MEDLINE (359) and Embase (648). After removal of 329 duplicates, we screened 1524 records. Additionally, we screened 88 records from ClinicalTrials.gov, 153 records from the WHO trials registry, 114 records via the Cochrane hip fracture programme, and reassessed the 29 previously excluded trials (51 reports). Spot searches by review authors and contact with trial investigators identified nine extra records in May and June 2019. No new trials were identified from bibliography checks of recent systematic reviews.

The retrospective search up to the previous date of the last search identified six new reports of four already included trials (Cameron 1993, 1 report; Huusko 2002, 1 report; Prestmo 2015, 2 reports; Shyu 2008, 2 reports); four new reports of two excluded trials (Allegrante 2001; Edgren 2015; 2 each), and 10 new trials (11 reports) all of which were excluded (Bai 2009; Barone 2006; Binder 2004; Boyer 1986; Gonzalez‐Montalvo 2010; Hagsten 2004; Lin 2009; Martin‐Martin 2014; NCT02058329; Tinetti 1999).

Our prospective database searches (CENTRAL, MEDLINE, Embase) identified 10 new reports of three already included trials (Prestmo 2015, 5 reports; Shyu 2008, 1 report; Shyu 2013a, 4 reports); five new reports of two already excluded trials (Edgren 2015, 2 reports; ISRCTN22464643; 3 reports); and three new reports for three trials (Crotty 2019 eventually included; ISRCTN15738119 eventually included as Karlsson 2016; NCT01254942 eventually included as Ashe 2019; 1 each); and five new trials (7 reports). Our search through the trial registries identified 10 new records not already found; one of which is a retrospective registration of an excluded trial eventually included (Chong 2013), and three were of newly identified trials (1 for Asplin 2017; 2 for NCT03430193, eventually excluded).

Overall, we put forward 56 reports for full‐text screening: 29 records pertained to 12 already identified trials and 27 records (21 trials) were new to the review.

Extra searches for reports of an ongoing trial (ISRCTN15738119; now Karlsson 2016) identified an additional report for an included trial, Stenvall 2007a; and one for the ongoing trial (ISRCTN15738119; now included trial Karlsson 2016). Contact with authors for the trial status of a newly identified trial reported in a conference abstract (Baroni 2016) resulted in notification of a full report of Baroni 2019 published in April 2019 by the same investigators that linked with a study previously awaiting classification (Serra 2013). The status of Baroni 2016, now in studies awaiting classification, remains unclear despite requests to trial investigators. A search (12 June) on FIT‐HIP trial (NTR5695; now FIT‐HIP 2019) found a protocol and full publication (2 references); thus, this study was moved from 'ongoing' to 'studies awaiting classification' (allocation changed after November 2019 update). Extra searches or contact with study investigators resulted in four reports for three trials: Crotty 2019 (2 reports for this now included, previously ongoing trial); NCT01254942 (1 report for this trial in 'studies awaiting classification'; now Ashe 2019, and allocation changed after November 2019 update); and NCT01051830 (1 secondary analysis report for ongoing trial; allocation changed with the identity of the main report after November 2019 update: Tseng 2019).

From 21 new trials, we excluded 16 (20 references): Asplin 2017; Bai 2009; Barone 2006; Binder 2004; Boyer 1986; Gonzalez‐Montalvo 2010; Hagsten 2004; HIPFRAC 2017; IRCT2016072119141N2; ISRCTN28376407 (Femur‐3 trial); Joeris 2017; Lin 2009; Martin‐Martin 2014; NCT02058329; NCT03822247; Tinetti 1999. We added two to 'studies awaiting classification' as their interventions, particularly in relation to medical clinician involvement, are not clear: NCT01934946 and NCT03894709. We also moved Baroni 2016 to 'studies awaiting classification', given the uncertainties revolving round this trial, as well as FIT‐HIP 2019, upon finding the two extra references, published protocol and final report, for NTR5695 on 12 June 2019. Lastly, we moved NCT03430193 to 'ongoing studies'.

Reassessment of the 29 excluded trials in Handoll 2009 resulted in the inclusion of four trials (6 reports): Chong 2013 because it compared two types of multidisciplinary rehabilitation; Jette 1987 because it was considered inappropriate to exclude this because of a lack of comparative data; and Marcantonio 2001 and Sanchez Ferrin 1999 because these now fitted our revised criteria for geriatrician involvement and oversight.

Overall, the second phase up to February 2019 resulted in the inclusion of seven new included studies, four of which had been previously excluded in Handoll 2009.

Second phase: search update on 21 November 2019 (from 1 January 2019)

Overall, 949 records were obtained from the following databases: CENTRAL (192), MEDLINE (180), Embase (219), ClinicalTrials.gov (136) and WHO ICTRP (222). After deduplication, 763 records were sent for screening by Maria Clarke, the Cochrane Bone, Joint and Muscle Trauma (BJMT) Group Information Specialist. After screening, 14 reports were put forward for full‐text screening; this included one other article, a protocol, found separately. Three articles had been found previously. Study selection of these 14 articles resulted in the inclusion of two trials: Tseng 2019; one new report, one report already included as well as another report appearing under NCT01051830 as an ongoing trial; and Ashe 2019; one report in the November search had been found previously, all four references for this trial appeared under NCT01254942 in 'studies awaiting classification'. Six articles belonged to six excluded trials: four were newly identified (ACTRN12619000296134 2019; Beaupre 2019; Dautel 2019; NCT01537523); one was the trial registration document for an already excluded trial (Edgren 2015), and one already identified article prompted the revaluation of FIT‐HIP 2019 (3 articles), which was moved from 'studies awaiting classification'. The remaining five articles reported three studies that were placed in 'studies awaiting classification': Hammond 2017 (1 protocol and 2 abstracts); Parsons 2019 (1 full report); Wu 2019 (1 abstract).

Second phase: search update on 15 October 2020 (CENTRAL (CRS Web) from 21 November 2019; MEDLINE (Ovid) from 1 November 2019; Embase (Ovid) from 1 January 2019)

Overall, 858 records were obtained from the following databases: CENTRAL (142), MEDLINE (250) and Embase (466). After deduplication, 696 records were sent by Maria Clare to the authors for screening. Of 18 relevant references, we put forward four for full‐text screening. One was another article for Karlsson 2016, an included trial; one (Aftab 2020) was an article linked with NCT03430193 (then an ongoing trial); one new trial registration was added to 'studies awaiting classification' (NCT04207788); and one was a reference update for Parsons 2019, previously in press.

Overall, the second phase resulted in the inclusion of nine new studies, four of which had been previously excluded in Handoll 2009.

January 2021. Additional article found for NCT03430193, moved from ongoing to excluded.

Final review study population

Overall, after completing the searches, there are 28 included studies, 47 excluded studies and 7 studies awaiting classification. Thirteen additional reports were identified for four studies already included in Handoll 2009.

Appendix 8. Report of search results in previous version of the review

First version: Issue 4, 2009 (Date of last search: April 2009)

Systematic searches, carried out for this review and for previous reviews (Cameron 2000; Cameron 2001; Halbert 2007), yielded a total of 37 eligible studies, of which 13 are included, 19 are excluded, two are ongoing, and three are studies awaiting classification. All 13 included trials are reported in full and in English; exceptionally, Fordham 1986 is available only as a report. Multiple publications, often conference abstracts but also reports reporting longer‐term follow‐up results, are available and included for several trials. Personal communications resulting in additional information and data have been recorded in the Included studies section of the references: some pertain to correspondence belonging to a previous review (Cameron 2001). Four other reports of Stenvall 2007a that focused on different aspects of the intervention were also available for this trial.

Appendix 9. Trial participant characteristics and representativeness in terms of the target population

Study ID	Age in years (mean); Sex (% female)	Fracture described?	Other fracture (%; description) / conditions (e.g. stroke)	Comordity restriction	Cognitive impairment (mental status)	Pre‐fracture / randomisation status (mobilitya; independence)	Prior residence	Target population?	Comment; subgroup restriction?
Inpatient rehabilitation
Baroni 2019	83; 77%	Yes ‐ all hip	Excluded	None	20% dementia	NR	94% from own home	Yes	Italy; more home dwellers?
Cameron 1993	84; 83%	Yes ‐ all hip 51% intracapsular	Excluded	None	48% cognitively impaired	46% limited disability; 17% moderate to severe disability; rest in nursing home	18% living alone 37% from care home	Yes	Note: participants stratified by prior living arrangements and disability
Chong 2013	78; 68.5%	Yes ‐ all hip 50.6% intertrochanteric 45.7% neck of femur 3.7% subtrochanteric	NR	NR	17% had dementia	NR	Excluded nursing home residents	NR	‐
Fordham 1986	86% over 75 years; 100%	Yes ‐ all hip	NR	Not terminal illness; not confusion	Long standing dementia excluded	NR	From own home: 77%; residential care: 21%	Yes but subgroup	All females and no long‐term dementia
Galvard 1995	79; 74%	Yes ‐ all hip 48% intracapsular, 8% subtrochanteric	NR	Probably none	Not recorded but none implied as participants were independently living at home	Independently living	All lived in own home	Yes but subgroup	All independently living at home
Gilchrist 1988	81; 100%	Yes – all hip	3% pathological	None	55% some cognitive impairment	89% needed support (from family or community services)	82% from community	Yes	All females
Huusko 2002	80; 72%	Yes – all hip	Excluded	Not terminal illness or serious early complication. Only 6% had no chronic diseases	20% had dementia	All independent ambulation 44% were functionally independent	All community dwelling. 54% lived alone	Yes but subgroup	All independent ambulators and living at home
Jette 1987	78; 67%	Yes ‐ all hip	NR	None	NR	ADL scores indicated some dependence, mainly instrumental and social	84% at home; 16% nursing home	Unclear (written consent required; just from patients?)	May not have included people with dementia
Kennie 1988	medians: 79 and 84; 100%	Yes ‐ all hip. 49% intracapsular	Excluded	No mention	47% had mild, moderate or severe impairment; (32% had moderate and severe impairment)	45% independent in basic ADL; 61% socially independent or only slightly dependent	89% at home; 11% in residential home	Yes	All females
Marcantonio 2001	79; 79%	Yes ‐ all hip	Excluded	None ‐ unless life expectancy < 6 months	40% mild or greater impairment	23% impairments in > 1 full ADL	93% at home; 7% nursing home	Yes	‐
Naglie 2002	84; 80%	Yes ‐ all hip Subcapital 43%	Excluded	None	26% had mild or moderate cognitive impairment (selected subgroup). Numbers with severe dementia not known.	Use of a walking aid 52%	At home 55% At retirement home 17% Nursing home 28%	Yes	Extreme end of spectrum: dependency in ambulation excluded.
Prestmo 2015	83; 74%	All hip Femoral neck 62% Trochanteric 31% Subtrochanteric 7%	Excluded	Excluded short life expectancy (3 months)	13% had previous diagnosis of dementia	All could walk 10 m before injury 239 (60%) lived alone Mean Barthel Index (0‐20) 18.2	All home dwelling. No nursing home. Sheltered housing 12%	Yes but subgroup	All home dwellers and able to walk 10 m.
Sanchez Ferrin 1999	82; 76%	Yes ‐ all hip	Concurrent fractures 6%	None	30% had dementia	68% went out by themselves	82% at home; 18% residential accommodation	Yes	‐
Shyu 2008	78; 69%	Yes ‐ all hip	NR	Excluded terminal illness	Excluded severe cognitive impairment. 32% had mild cognitive impairment	85% had independent walking ability	Presume all from own or relative's home	Yes but subgroup and based in Taiwan	Not severe cognitive impairment; predominantly independent. Taiwan differs substantially in case mix, culture and social organisation from 'Western' societies
Shyu 2013a	76.5; 64%	Yes ‐ all Femoral neck: 58%; intertrochanteric: 40%; subtrochanteric 2%	NR	Excluded terminal illness	Excluded severe cognitive impairment	95% had independent walking ability 67% independent in ADLs	All from home setting (excluded nursing homes)	Yes but subgroup and based in Taiwan	Not severe cognitive impairment; predominantly independent. Taiwan differs substantially in case mix, culture and social organisation from 'Western' societies
Stenvall 2007a	82; 74%	Yes ‐ all hip; but just intracapsular	Excluded pathological	Not severe renal failure, rheumatoid arthritis	32% had dementia	85% independent walking indoors; 51% without walking aid 58% independent in personal ADLs	63% lived independently	Yes but subgroup	All intracapsular fractures (not explained why)
Swanson 1998	78; 78%	Yes ‐ all hip	Excluded pathological	None	None had dementia	All independently mobile Social support: none: 27 (38%); family or community 42 (59%)	Resided in home or hostel	Yes but subgroup	All mobile, no dementia; no nursing home
Tseng 2019	78; 77%	Yes ‐ all hip Femoral neck: 53%; intertrochanteric: 45%; subtrochanteric 2%	Excluded cancer	Not cancer. All had diabetes. Mean number of comorbidities 2.84	Excluded those who couldn't communicate with researchers; e.g. with severe impairment	Prefracture independence in walking: 116 (66%) Prefracture independence in ADLs: 88 (50%)	All in the community (at home?); exclusion of 2 living in institutional care before their fracture	Yes but subgroup	All had diabetes (mean 12.7 years since diagnosis). Also based in Taiwan (see comment for Shyu 2008)
Uy 2008	medians: 80 and 83; 100%	Yes ‐ all hip	NR	Median Charlson Index = 1	100%. All had moderate to severe cognitive impairment	Ambulant prior to hip fracture; 55% used walking aid	All lived in a nursing home	Yes, but subgroup	All female, with cognitive impairment and living in nursing homes
Vidan 2005	82; 81.5%	Yes ‐ all hip. Subcapital: 44%	Excluded	None ‐ unless life expectancy < 12 months	24.5% had dementia	All could walk. 74% independent ambulation "fewer than 40% were independent in an all ADLs"	84% at home	Yes	Extreme end of spectrum; dependency in all ADL excluded
Watne 2014;	median 85; 75%	Yes ‐ all hip Femoral neck: 59%; Intertrochanteric: 40%; subtrochanteric: 1%	Excluded high‐energy trauma	Not terminal illness	49% had dementia	Modified Barthel Index (0 to 20; fully independent in personal care) median 18	31% from institutional care (nursing home)	Yes	High % dementia (opinion‐based assessment)
Ambulatory rehabilitation
Ashe 2019	79.5; 63%	All hip	NR	NR	None; excluded	Able to walk at least 10 m	Community dwelling; 42% lived alone. Excluded residential care	Yes but subgroup	All community dwellers and able to walk 10 m. No dementia.
Crotty 2003	82.5; 67%	Yes ‐ all hip	Excluded	Medically stable	None; all needed to be able to participate in rehabilitation programme	52% used mobility aids; 38% lived alone	All home living and expected to return home post discharge	Yes but subgroup	All participants expected to return home and able to participate in rehabilitation
Crotty 2019	87; 74%	Yes ‐ all hip	Excluded	None	77.5% recorded dementia diagnosis	All were indoor ambulators; most used a walking frame	All nursing home residents	Yes but subgroup	All from nursing homes
Karlsson 2016	83; 72%	Yes ‐ all hip. 72% intracapsular	Excluded	None	50% had dementia	69% walked independently outdoors; 72% lived alone	69% at home; 31% residential care facilities	Yes, but on the severe end of spectrum	Larger proportion of people with dementia and from residential homes; and intracapsular fractures
Ryan 2006	81; 76%	All hip	Stroke also	Not Parkinson's nor dementia	None had dementia	Median Barthel score 16 versus 17 indicates some disability (score 0 to 20: independence in personal care). 37% had co‐residing carer	All from home and returning home	Yes but subgroup	No dementia, all returned to home
Singh 2012	79; 69%	Yes ‐ all hip surgery	Excluded pathological	Excluded terminal illness	5% had dementia	Mobility NR Most were independent in basic ADLs: Katz 1970	86% house or unit; 7% hostel; 2% nursing home; 4% other	Yes but subgroup	Fairly independent. Had sufficient understanding to participate in home rehabilitation
Ziden 2008	82; 69%	Yes ‐ all hip Cervical: 53%; trochanteric: 37%; subtrochanteric 10%	Excluded	Not serious disease; cognitive impairment	None had cognitive impairment	All community dwellers 33% no help	All community dwellers 64% living alone	Yes but subgroup	Community dwelling, no cognitive impairment

Footnotes

^aAnd/or weight‐bearing status; depending on trial setting
ADL = activity/activities in daily living
NR = Not reported

Appendix 10. Early treatment ‐ representativeness in current practice

Study ID	Setting	Country; trial start	Early surgery (< 48 hours)	Anaesthesia	Type of surgery	Early mobilisation?	Early supported discharge?	Current treatment practice?	Notes
Inpatient rehabilitation
Baroni 2019	Hospital	Italy; 2012	No (31%)	NR	NR, 7 (3%) had conservative treatment	NR	No	No	Italy ‐ generally has longer time to surgery
Cameron 1993	Hospital	Australia; 1989	Mean 1.8 days	NR	NR	No (part of intervention)	No (part of intervention)	No	Usual care not well described
Chong 2013	Hospital	Singapore; 2004	NR	NR	Inconsistent data from 93% population. bipolar hemiarthroplasty 20.5%; Moore's hemiarthroplasty 17%; dynamic hip screw 50%; others.	NR ‐ unlikely 49% were not weight‐bearing on admission to rehabilitation	No	No (except for Singapore)	Not enough information but high rate of non‐weight‐bearing
Fordham 1986	Hospital	UK; 1984	No, 48% within 2 days, but 30% 5 days or over	NR	Mixed but some dated. Included 59 (55%) screws; 20 (18.5%) pin and plates; 16 (15%) cemented Hastings	NR, unlikely	No	No	Dated trial
Galvard 1995	Hospital	Sweden; 1988	NR	NR	Cervical: 2 nails; trochanteric: sliding screw	NR	No, long hospital stays	No	Dated trial
Gilchrist 1988	Hospital	UK; 1984	Yes (90%)	96% general	NR	NR	NR	Unlikely	Dated trial but early surgery
Huusko 2002	Hospital	Finland; 1994	Likely. Median 1 day	NR	NR	No (part of intervention)	Not really. Moved to rehabilitation or local hospital ward	Unclear	Dated trial but early surgery
Jette 1987	Hospital	USA; 1984	NR	NR	In part but some old implants	No ‐ 6 weeks for unstable fractures	No	No	Dated trial
Kennie 1988	Hospital	UK; before 1986	NR (but recommended in a later paper 1990)	spinal 24% general 76%	42% hemiarthroplasty 43% dynamic hip screw	No (part of intervention)	No	No	Dated trial
Marcantonio 2001	Hospital	USA; before 1999	NR	NR	NR in full	No (part of intervention)	NR	Unclear	Focus on delirium ‐ hard to judge
Naglie 2002	Hospital	Canada; 1993	Mean 1.35 days	NR	Pin + plate 60% hemiarthroplasty 30% other 10%	No (part of intervention)	No	No	Dated trial (surgery)
Prestmo 2015	Hospital	Norway; 2008	Mean 29 hours	Mostly spinal anaesthesia (after pre‐op femoral nerve block)	41% hemiarthroplasty; 18% screws; 33% bone plates and screws; others	No (part of intervention)	Yes for intervention group	Unclear	Intervention includes several aspects of current practice
Sanchez Ferrin 1999	Hospital	Spain; 1996	NR	NR	Vague	NR	NR	No	Spain ‐ generally has longer time to surgery
Shyu 2008	Hospital	Taiwan; 2001	NR	NR	37% hemiarthroplasty; 63% internal fixation	No (part of intervention)	NR but policy to send home quickly	No (except for Taiwan)	Taiwan’s healthcare system differs from those in most Western countries, including underdevelopment of community‐based, long‐term‐care services.
Shyu 2013a	Hospital	Taiwan; 2005	No, but mean 2.2 days	NR	38% arthroplasty; 62% internal fixation	No (part of intervention)	NR but policy to send home quickly	No (except for Taiwan)	Comment in Shyu 2008 is likely to apply
Stenvall 2007a	Hospital	Sweden; 2000	Mean 25 hours	NR	2 hook‐pins 35%; bipolar hemiarthroplasty: 56%; dynamic hip screw: 8.5% for basocervical fractures	Yes policy within 24 hours	No	Unclear	Types of surgery more common in Sweden
Swanson 1998	Hospital	Australia; 1994	60 (84.5%) had surgery within 48 hours (11 exceptions; reasons given) (part of intervention)	Regional anaesthesia used (63.2% versus 42.4%); (part of intervention)	internal fixation (65%) or hemiarthroplasty (35%)	No (part of intervention)	No (part of intervention)	Unclear	Intervention includes several aspects of current practice
Tseng 2019	Hospital and community (intervention group)	Taiwan; 2010	NR	NR	45.5% arthroplasty; 54.5% internal fixation	No (part of intervention)	NR but policy to send home quickly	No (except for Taiwan)	Comment in Shyu 2008 is likely to apply
Uy 2008	Hospital	Australia; 2001	NR	NR	hemiarthroplasty for intracapsular fractures; compression screw and plate for trochanteric fractures	Likely (also part of intervention)	Special group (part of intervention)	Unclear	All came from nursing care facilities and control group of early return reflects usual practice
Vidan 2005	Hospital	Spain; 1997	No, mean 77.2 hours	96% regional anaesthesia	Non‐specific internal fixation (60%), replacement (35%), other	No ‐ rehabilitation started on average 9.3 days	No	No	Spain ‐ generally has longer time to surgery
Watne 2014	Hospital	Norway; 2009	Yes, median 26 and 24 hours	92% spinal anaesthesia	44% hemiarthroplasty 54% osteosynthesis 1 total hip replacement not operated 3 (1%)	Yes (policy)	Partly (part of intervention too). Length of stay data indicated this applied generally	Yes (see next box)	Note that no mention of specific early discharge for nursing home patients
Ambulatory rehabilitation
Ashe 2019	Community (outpatients and home)	Canada; 2011	NR	NR	NR	NR	NR	Unsure	No details to judge
Crotty 2003	Hospital and community (intervention group)	Australia; 1998	NR	NR	Internal fixation 41%; replacement 59%; 1 total hip replacementa	Yes, median 1 day to mobilise	No (part of intervention)	Unsure	Intervention includes several aspects of current practice
Crotty 2019	Community (nursing home / care facility)	Australia; 2012	Yes (88% within 24 hours)	NR	More intramedullary nail for extracapsular fractures (79%); very few total hip replacements for intracapsular fractures (2.3%)a	NR but unlikely in context	Early discharge as back to nursing home	Yes (for these patients)	Country‐ and circumstances‐dependent. All came from nursing care facilities
Karlsson 2016	Hospital and community (intervention group)	Sweden; 2008	NR	NR	NR	Yes (policy)	Yes for intervention group	Unclear	‐
Ryan 2006	Community (at home)	UK; 2000	NR	NR	NR	NR	No. Length of hospital stay was probably over a month on average	Unclear	Unlikely in terms of discharge timing
Singh 2012	Community (outpatients and home)	Australia; 2003	NR	NR	NR	NR	NR	Unclear	No information on these
Ziden 2008	Hospital and community (intervention group: home)	Sweden; 2004	NR	NR	NR	Yes, policy: within 48 hours	No: supported discharge was an aim of one intervention but no indication of early	Unclear	Insufficient evidence to judge

Footnotes

^aLow numbers of total hip replacements are usual in this population.
NR: not reported

Appendix 11. Sensitivity analyses: random‐effects results

Analysis	Outcome	Results
Analysis 1.1	Poor outcome (long‐term)	RR 0.88, 95% CI 0.80 to 0.97
Analysis 1.4	Poor outcome at discharge	RR 0.88, 95% CI 0.75 to 1.03
Analysis 1.5	Mortality (long‐term)	RR 0.92, 95% CI 0.80 to 1.05
Analysis 1.8	Mortality at discharge	RR 0.81, 95% CI 0.58 to 1.13
Analysis 1.11	Greater dependency in ADL up to 4 months	RR 0.88, 95% CI 0.78 to 1.00
Analysis 1.18	Non‐recovery in walking (long‐term)	RR 0.85, 95% CI 0.72 to 1.00
Analysis 1.21	Instititional care (long‐term)	RR 0.90, 95% CI 0.75 to 1.07

Data and analyses

Comparison 1. Inpatient setting: multidisciplinary rehabilitation (MDR) versus usual care.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 'Poor outcome' (long‐term follow‐up at 6 or 12 months)	13	3036	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.80, 0.98]
1.2 'Poor outcome' (long‐term follow‐up): subgrouped by intervention type	13	3036	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.80, 0.98]
1.2.1 Geriatric orthopaedic rehabilitation unit	2	307	Risk Ratio (M‐H, Fixed, 95% CI)	0.68 [0.51, 0.91]
1.2.2 Hip fracture programme	7	1685	Risk Ratio (M‐H, Fixed, 95% CI)	0.89 [0.77, 1.02]
1.2.3 Mixed assessment and rehabilitation unit	1	203	Risk Ratio (M‐H, Fixed, 95% CI)	0.94 [0.67, 1.31]
1.2.4 Orthogeriatric programme	3	841	Risk Ratio (M‐H, Fixed, 95% CI)	0.97 [0.79, 1.17]
1.3 'Poor outcome' (long‐term follow‐up) by selection bias	13	3036	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.80, 0.98]
1.3.1 Low risk of selection bias	6	1751	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.78, 1.00]
1.3.2 Unclear or high risk of selection bias	7	1285	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.74, 1.05]
1.4 'Poor outcome' (at discharge)	8	1537	Risk Ratio (M‐H, Fixed, 95% CI)	0.87 [0.76, 1.00]
1.5 Mortality (end of scheduled follow‐up: 4 to 12 months)	18	3973	Risk Ratio (M‐H, Fixed, 95% CI)	0.91 [0.80, 1.05]
1.6 Mortality (end of scheduled follow‐up) ‐ with 12 month data for Cameron 1993	18	3973	Risk Ratio (M‐H, Fixed, 95% CI)	0.90 [0.79, 1.03]
1.7 Mortality (end of scheduled follow‐up): subgrouped by intervention type	18	3973	Risk Ratio (M‐H, Fixed, 95% CI)	0.91 [0.80, 1.05]
1.7.1 Geriatric Orthopaedic Rehabilitation Unit	5	1008	Risk Ratio (M‐H, Fixed, 95% CI)	0.95 [0.75, 1.22]
1.7.2 Hip fracture programme	8	1698	Risk Ratio (M‐H, Fixed, 95% CI)	0.95 [0.77, 1.19]
1.7.3 MARU	1	206	Risk Ratio (M‐H, Fixed, 95% CI)	0.74 [0.44, 1.23]
1.7.4 Orthogeriatric programme	4	1061	Risk Ratio (M‐H, Fixed, 95% CI)	0.86 [0.66, 1.14]
1.8 Mortality (at discharge)	11	2455	Risk Ratio (M‐H, Fixed, 95% CI)	0.77 [0.58, 1.04]
1.9 Quality of life: EQ‐5D (‐0.594: worse than death, 0: dead to 1: best quality)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
1.9.1 At 4 months	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
1.9.2 At 12 months	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
1.10 Quality of life: SF‐36 (Taiwan version) (each domain: 0 to 100; best quality) at 1 year	2		Mean Difference (IV, Fixed, 95% CI)	Subtotals only
1.10.1 Bodily pain	2	391	Mean Difference (IV, Fixed, 95% CI)	0.11 [‐4.49, 4.70]
1.10.2 General health perceptions	2	391	Mean Difference (IV, Fixed, 95% CI)	5.03 [‐0.44, 10.50]
1.10.3 Vitality	2	391	Mean Difference (IV, Fixed, 95% CI)	6.41 [2.13, 10.69]
1.10.4 Social functioning	2	391	Mean Difference (IV, Fixed, 95% CI)	6.52 [1.10, 11.94]
1.10.5 Role limitations because of emotional problems	2	391	Mean Difference (IV, Fixed, 95% CI)	3.93 [‐1.44, 9.30]
1.10.6 General mental health	2	391	Mean Difference (IV, Fixed, 95% CI)	6.80 [2.45, 11.15]
1.10.7 Physical functioning	2	391	Mean Difference (IV, Fixed, 95% CI)	13.20 [6.99, 19.41]
1.10.8 Role limitations because of physical health problems	2	391	Mean Difference (IV, Fixed, 95% CI)	27.98 [19.54, 36.42]
1.11 Greater dependency in ADL up to 4 months	4	754	Risk Ratio (M‐H, Fixed, 95% CI)	0.87 [0.76, 0.99]
1.12 Regain in former level of ADL independence in the short term (up to 4 months)	4	754	Risk Ratio (M‐H, Fixed, 95% CI)	1.18 [1.01, 1.38]
1.13 Greater dependency in ADL in the long term (6 to 12 months)	4		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
1.13.1 More dependent at 1 year	2	238	Risk Ratio (M‐H, Fixed, 95% CI)	0.64 [0.51, 0.81]
1.13.2 More dependent or dead at 1 year	2	300	Risk Ratio (M‐H, Fixed, 95% CI)	0.71 [0.60, 0.84]
1.13.3 Incomplete recovery of ADL and mobility at 1 year	1	252	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.71, 1.09]
1.13.4 Incomplete recovery of ADL and mobility, or dead at 1 year	1	319	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.75, 1.04]
1.13.5 Non‐recovery of independent self‐care ability at 1 year	1	269	Risk Ratio (M‐H, Fixed, 95% CI)	0.85 [0.64, 1.14]
1.13.6 Non‐recovery of independent self‐care ability or dead at 1 year	1	282	Risk Ratio (M‐H, Fixed, 95% CI)	0.93 [0.65, 1.34]
1.14 ADL: Barthel scores (higher scores = greater independence)	4		Mean Difference (IV, Fixed, 95% CI)	Subtotals only
1.14.1 Barthel Index (0 to 20) at 4 months	1	333	Mean Difference (IV, Fixed, 95% CI)	1.01 [0.21, 1.81]
1.14.2 Chinese Barthel Index (0 to 100) at 3 months	2	285	Mean Difference (IV, Fixed, 95% CI)	7.57 [2.87, 12.27]
1.14.3 Barthel Index (0 to 20) at 12 months	1	300	Mean Difference (IV, Fixed, 95% CI)	1.13 [0.31, 1.95]
1.14.4 Chinese Barthel Index (0 to 100) at 12 months	2	293	Mean Difference (IV, Fixed, 95% CI)	4.04 [‐1.42, 9.51]
1.14.5 Modified Barthel Index (0 to 100) at 6 months	1	60	Mean Difference (IV, Fixed, 95% CI)	6.30 [‐0.53, 13.13]
1.15 ADL: Barthel Index (0 to 20; best outcome)	1		Other data	No numeric data
1.16 Loss in activities of daily living at 6 months (Katz index; 6 maximum)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
1.17 ADL: Barthel Index (participants who were from nursing homes)	2		Other data	No numeric data
1.18 Greater dependency in mobility	5		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
1.18.1 Non‐recovery/decline in walking at long‐term follow‐up	5	1085	Risk Ratio (M‐H, Fixed, 95% CI)	0.83 [0.71, 0.98]
1.18.2 Non‐recovery in walking or dead at long‐term follow‐up	5	1276	Risk Ratio (M‐H, Fixed, 95% CI)	0.82 [0.72, 0.92]
1.18.3 Decline in transfers (bed to chair etc) at long‐term follow‐up	1	241	Risk Ratio (M‐H, Fixed, 95% CI)	0.96 [0.69, 1.34]
1.18.4 Decline in transfers or dead at long‐term follow‐up	1	279	Risk Ratio (M‐H, Fixed, 95% CI)	0.93 [0.72, 1.21]
1.19 Mobility: Short Physical Performance Battery (0 to 12: best mobility)	1		Mean Difference (IV, Fixed, 95% CI)	Subtotals only
1.19.1 At 4 months	1	325	Mean Difference (IV, Fixed, 95% CI)	0.74 [0.19, 1.29]
1.19.2 At 12 months	1	284	Mean Difference (IV, Fixed, 95% CI)	0.69 [0.09, 1.29]
1.20 Mobility: Short Physical Performance Battery (0 to 12: best mobility)	1		Other data	No numeric data
1.21 Institutional care at 6 to 12 months (survivors)	13	2497	Risk Ratio (M‐H, Fixed, 95% CI)	0.90 [0.76, 1.06]
1.22 Complications	6		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
1.22.1 confusion, delirium in hospital	4	980	Risk Ratio (M‐H, Fixed, 95% CI)	0.85 [0.74, 0.98]
1.22.2 heart failure, cardiac complications	2	648	Risk Ratio (M‐H, Fixed, 95% CI)	1.46 [0.89, 2.39]
1.22.3 cardiac insufficiency	1	206	Risk Ratio (M‐H, Fixed, 95% CI)	0.25 [0.05, 1.15]
1.22.4 chest infection, cardiac problem, bedsore	1	71	Risk Ratio (M‐H, Fixed, 95% CI)	0.40 [0.17, 0.94]
1.22.5 pressure injuries (sores, ulcers)	3	854	Risk Ratio (M‐H, Fixed, 95% CI)	0.50 [0.35, 0.70]
1.22.6 pneumonia, pulmonary complications	2	648	Risk Ratio (M‐H, Fixed, 95% CI)	1.46 [0.83, 2.57]
1.22.7 respiratory infection	1	206	Risk Ratio (M‐H, Fixed, 95% CI)	1.33 [0.48, 3.71]
1.22.8 urinary tract infection	2	535	Risk Ratio (M‐H, Fixed, 95% CI)	0.61 [0.42, 0.88]
1.22.9 stroke, emboli, cerebral complications	2	400	Risk Ratio (M‐H, Fixed, 95% CI)	3.99 [0.70, 22.79]
1.22.10 thrombo‐embolic complications	1	329	Risk Ratio (M‐H, Fixed, 95% CI)	5.09 [0.25, 105.24]
1.22.11 renal failure	1	329	Risk Ratio (M‐H, Fixed, 95% CI)	3.06 [0.63, 14.92]
1.22.12 ionic anomalities	1	206	Risk Ratio (M‐H, Fixed, 95% CI)	0.17 [0.02, 1.36]
1.22.13 gastrointestinal complications	1	329	Risk Ratio (M‐H, Fixed, 95% CI)	1.27 [0.35, 4.66]
1.22.14 dehydration	1	206	Risk Ratio (M‐H, Fixed, 95% CI)	0.57 [0.17, 1.89]
1.22.15 any medical complication	3	891	Risk Ratio (M‐H, Fixed, 95% CI)	0.91 [0.79, 1.04]
1.22.16 any surgical complications	2	638	Risk Ratio (M‐H, Fixed, 95% CI)	0.88 [0.54, 1.42]
1.23 Readmitted to hospital during follow‐up	11	2538	Risk Ratio (M‐H, Fixed, 95% CI)	0.97 [0.84, 1.12]
1.24 Dead or readmitted to hospital during follow‐up	8	1975	Risk Ratio (M‐H, Fixed, 95% CI)	0.91 [0.78, 1.05]
1.25 Length of hospital stay (days)	12		Mean Difference (IV, Random, 95% CI)	Totals not selected
1.26 Subgroup analysis ‐ death at 12 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
1.26.1 With dementia	1	162	Risk Ratio (M‐H, Fixed, 95% CI)	1.09 [0.75, 1.60]
1.26.2 Without dementia	1	167	Risk Ratio (M‐H, Fixed, 95% CI)	1.10 [0.53, 2.26]

Comparison 2. Inpatient setting: multidisciplinary rehabilitation (MDR) with an integrated care pathway versus MDR alone.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Poor outcome, dead, couldn't walk as before, or nursing home stay post discharge (12 months)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.1.1 Poor outcome (death or reduced mobility)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.1.2 Death	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.1.3 Unable to walk as before	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.1.4 Any nursing home stay post discharge	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.2 SF‐12 scores at 12 months (0: worst to 100: best)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
2.2.1 Physical component summary scores	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
2.2.2 Mental component summary scores	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
2.3 Modified Barthel Index (0: worst to 100: best): "changes over time"	1		Other data	No numeric data
2.3.1 At discharge (N = 149)	1		Other data	No numeric data
2.3.2 At 6 month follow‐up (N = 129)	1		Other data	No numeric data
2.3.3 At 12 month follow‐up (N = 121)	1		Other data	No numeric data
2.4 Hospital readmission (at set times)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.4.1 At discharge	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.4.2 Within 3 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
2.4.3 Within 1 year	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

Comparison 3. Ambulatory setting: supported discharge and multidisciplinary home‐based rehabilitation versus usual inpatient rehabilitation.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 'Poor outcome', mortality, and unable to walk (3 or 4 months)	2		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
3.1.1 Poor outcome (dead or unable to walk)	1	209	Risk Ratio (M‐H, Fixed, 95% CI)	1.22 [0.63, 2.39]
3.1.2 Mortality	2	275	Risk Ratio (M‐H, Fixed, 95% CI)	0.77 [0.34, 1.76]
3.1.3 Unable to walk	1	183	Risk Ratio (M‐H, Fixed, 95% CI)	2.47 [0.68, 9.02]
3.2 'Poor outcome', mortality, institutional care and unable to walk (12 months)	3		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
3.2.1 Poor outcome (dead or higher level of care / unable to walk)	3	377	Risk Ratio (M‐H, Fixed, 95% CI)	0.91 [0.62, 1.35]
3.2.2 Mortality	3	377	Risk Ratio (M‐H, Fixed, 95% CI)	1.00 [0.61, 1.63]
3.2.3 Moved to higher level of care	2	168	Risk Ratio (M‐H, Fixed, 95% CI)	0.35 [0.07, 1.69]
3.2.4 Unable to walk	2	214	Risk Ratio (M‐H, Fixed, 95% CI)	0.89 [0.38, 2.04]
3.3 Subgroup analysis: poor outcome (dead or non‐recovery of indoor walking ability) at 1 year, subgrouped by dementia status	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
3.3.1 Had dementia	1	103	Risk Ratio (M‐H, Fixed, 95% CI)	1.02 [0.81, 1.28]
3.3.2 Didn't have dementia	1	102	Risk Ratio (M‐H, Fixed, 95% CI)	0.90 [0.48, 1.70]
3.4 Subgroup analysis: mortality at 1 year, subgrouped by dementia status	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
3.4.1 Had dementia	1	103	Risk Ratio (M‐H, Fixed, 95% CI)	0.87 [0.45, 1.66]
3.4.2 Didn't have dementia	1	102	Risk Ratio (M‐H, Fixed, 95% CI)	2.43 [0.66, 8.87]
3.5 SF‐36 scores at 12 months (0: worst to 100: best)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
3.5.1 Physical component summary scores	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
3.5.2 Mental component summary scores	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
3.6 Independence in personal activities of daily living (PADL) and outdoor walking	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
3.6.1 Independent PADL at 3 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
3.6.2 Independent PADL at 12 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
3.6.3 Independent outdoor walking at 3 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
3.6.4 Independent outdoor walking at 12 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
3.7 Function (ADL): Barthel Index (0 to 20: total independence in personal care)	1		Other data	No numeric data
3.8 Daily activities scores at 1 year	1		Other data	No numeric data
3.8.1 FIM (Functional Independence Measure: 0 to 91; independent)	1		Other data	No numeric data
3.8.2 FAI (Frenchay Activity Index: 0 to 45; best activity)	1		Other data	No numeric data
3.9 Complications (discharge to 12 months)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
3.9.1 Had complication post discharge	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
3.9.2 Had delirium recorded post discharge	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
3.10 Readmission to hospital and reoperation	2		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
3.10.1 Readmission to hospital	2	265	Risk Ratio (M‐H, Fixed, 95% CI)	1.20 [0.83, 1.74]
3.10.2 Reoperation	1	199	Risk Ratio (M‐H, Fixed, 95% CI)	1.40 [0.48, 4.14]
3.11 Falls outcomes	2		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
3.11.1 Number of fallers	2	263	Risk Ratio (M‐H, Fixed, 95% CI)	1.09 [0.80, 1.50]
3.11.2 Falls requiring hospitalisation	1	64	Risk Ratio (M‐H, Fixed, 95% CI)	0.94 [0.06, 14.38]
3.11.3 Additional fracture	1	199	Risk Ratio (M‐H, Fixed, 95% CI)	1.90 [0.75, 4.80]
3.12 Lengths of hospital or rehabilitation stays (days)	2		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
3.12.1 Length of hospital stay	2		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
3.12.2 Length of rehabilitation (hospital + home)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

Comparison 4. Ambulatory setting: outreach multidisciplinary rehabilitation versus usual care in nursing homes.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 'Poor outcome', mortality, and unable to walk (12 months)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.1.1 Poor outcome (dead or unable to walk)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.1.2 Mortality	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.1.3 Unable to walk	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.2 All cause mortality	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.2.1 At 4 weeks	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.2.2 At 4 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.2.3 At 12 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.3 Quality of life at 12 months: DEMQOL & DEMQOL‐Proxy	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.3.1 DEMQOL score (12 to 112; best QOL)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.3.2 DEMQOL‐Proxy score (31 to 124; best QOL)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.4 EQ‐5D quality of life index at 12 months (0 dead to 1 best quality)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.5 Modified Barthel Index (0 to 100: total independence in personal care)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.5.1 At 4 weeks	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.5.2 At 12 months	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.6 Nursing Home Life‐Space Diameter (0 to 50; leaves facility daily)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.6.1 At 4 weeks	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.6.2 At 12 months	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.7 Pain: PAINAD (0 to 10; severe pain)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.7.1 At 4 weeks	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.7.2 At 12 months	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4.8 Fall outcomes (adverse events)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.8.1 Number of fallers at 4 weeks	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.8.2 Number of hospital admissions for falls	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.8.3 Number of hip fractures	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
4.9 Mean costs per patients (in AU $): base case analysis	1		Other data	No numeric data

Comparison 5. Ambulatory setting: intensive versus less intensive community rehabilitation.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
5.1 'Poor outcome', mortality and institutional care (12 months)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
5.1.1 Poor outcome (dead or in institutional care)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
5.1.2 Mortality	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
5.1.3 Institutional care	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
5.2 Quality of life, dependence and activities scores at 3 and 12 months	1		Other data	No numeric data
5.2.1 Quality of life: EQ‐5D (0: dead to 1: best quality)	1		Other data	No numeric data
5.2.2 Barthel Index (0 to 20: independent)	1		Other data	No numeric data
5.2.3 Frenchay Activities Index (0 to 45: best activity)	1		Other data	No numeric data
5.3 Number of contacts over 12 weeks (for participants with hip fracture or stroke)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

Comparison 6. Ambulatory setting: extended multidisciplinary ambulatory rehabilitation versus usual care.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
6.1 'Poor outcome', mortality and institutional care (12 months)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
6.1.1 Poor outcome (dead or in institutional care)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
6.1.2 Mortality	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
6.1.3 Institutional care	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
6.2 ALSAR: Assessment of Living Skills And Resources	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
6.2.1 Skills score (0 to 22; lower score = better ADL skills)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
6.2.2 Resources score (0 to 22; lower score = more resources available)	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

6.1. Analysis.

Comparison 6: Ambulatory setting: extended multidisciplinary ambulatory rehabilitation versus usual care, Outcome 1: 'Poor outcome', mortality and institutional care (12 months)

Comparison 7. Outpatient multidisciplinary clinic between 3 to 12 months post fracture versus usual care.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
7.1 Mortality and institutional care	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
7.1.1 Mortality at 12 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected
7.1.2 Institutional care at 12 months	1		Risk Ratio (M‐H, Fixed, 95% CI)	Totals not selected

Comparison 8. Exploratory analysis: inpatient and supported discharge (home‐based) settings.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
8.1 'Poor outcome' (long‐term follow‐up): subgrouped by intervention type	16	3413	Risk Ratio (M‐H, Fixed, 95% CI)	0.89 [0.80, 0.98]
8.1.1 Geriatric orthopaedic rehabilitation unit	2	307	Risk Ratio (M‐H, Fixed, 95% CI)	0.68 [0.51, 0.91]
8.1.2 Hip fracture programme	7	1685	Risk Ratio (M‐H, Fixed, 95% CI)	0.89 [0.77, 1.02]
8.1.3 Mixed assessment and rehabilitation unit	1	203	Risk Ratio (M‐H, Fixed, 95% CI)	0.94 [0.67, 1.31]
8.1.4 Orthogeriatric programme	3	841	Risk Ratio (M‐H, Fixed, 95% CI)	0.97 [0.79, 1.17]
8.1.5 Supported discharge and home rehabilitation	3	377	Risk Ratio (M‐H, Fixed, 95% CI)	0.91 [0.62, 1.35]
8.2 Mortality (end of scheduled follow‐up): subgrouped by intervention type	21	4350	Risk Ratio (M‐H, Fixed, 95% CI)	0.92 [0.81, 1.05]
8.2.1 Geriatric Orthopaedic Rehabilitation Unit	5	1008	Risk Ratio (M‐H, Fixed, 95% CI)	0.95 [0.75, 1.22]
8.2.2 Hip fracture programme	8	1698	Risk Ratio (M‐H, Fixed, 95% CI)	0.95 [0.77, 1.19]
8.2.3 MARU	1	206	Risk Ratio (M‐H, Fixed, 95% CI)	0.74 [0.44, 1.23]
8.2.4 Orthogeriatric programme	4	1061	Risk Ratio (M‐H, Fixed, 95% CI)	0.86 [0.66, 1.14]
8.2.5 Supported discharge and home rehabilitation	3	377	Risk Ratio (M‐H, Fixed, 95% CI)	1.00 [0.61, 1.63]

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Ashe 2019.

Study characteristics
Methods	Randomised trial using a computerised programme independently administered by a statistical company Assessor blinding: yes, for research staff but not participants Length of follow‐up: 12 months, also 6 months
Participants	Three hospitals, Vancouver, British Columbia, Canada Conducted: May 2011 to April 2013 53 people with hip fracture; 37% male. Mean age 79.5 years. Cognitive status: none, people with dementia were excluded. Inclusion criteria: aged 65 years or older community‐dwelling adults with a recent history (within 3 months in protocol; 3 to 12 months post‐fracture in Zusman 2019 report; see Notes) of hip fracture who were admitted from home and expected to be discharged to home (protocol). Written informed consent. Exclusion criteria: unable to ambulate 10 metres prior to the femoral fracture (with/without a walking aid), discharged to a residential care facility, or diagnosed with any type of dementia (Alzheimer's, vascular, etc.) Assigned: 26/27 [Geriatrician‐led follow‐up clinic; Usual care] Assessed (12 months): 23/23 (missing accelerometry data) Loss to follow‐up: 1 (intervention group) but 1 in control group did not wear accelerometer at baseline
Interventions	Randomisation took place "prior to discharge from hospital, rehabilitation unit and/or from a discharge list of eligible older adults". Recruitment was between 3 and 12 months from hip fracture. (1) Attendance at an enhanced post‐fracture follow‐up clinic for outpatient management of falls and fracture risk. The clinic had been designed based on the need to assess both falls risk factors and bone health. A geriatrician led the clinic and provided post‐fracture assessment and management. The clinical assessment focused on four key areas to address secondary prevention of future injury by addressing key domains that contribute to fall and fracture risk, and included assessment of: balance, bone health, brain function and bladder function = B4 Clinic. Based on identified need during the geriatric assessment, participants were also referred to additional health professionals, such as occupational therapist, registered dietician, social worker, etc. Participants were also assessed by a physiotherapist at the clinic, who recommend a standardised home exercise programme and assessed need for subsequent onsite outpatient physiotherapy visits. Following discharge from outpatient physiotherapy, participants received up to four additional telephone calls based on motivational interviewing. If, during these sessions, the participant identified a mobility concern, they were encouraged to follow‐up with the B4 Clinic. Intervention lasted 12 months. (2) Usual orthopaedic and rehabilitative post‐operative treatment for the hip fracture. Fracture follow‐up clinic offered at 12 months.
Outcomes	Data reported for physical activity and sedentary behaviour outcomes in Zusman 2019. The following 'generic' outcomes were recorded in the protocol but most are not available at this time. Primary outcome 1. Mobility, measured by Short Physical Performance Battery at 12 months Secondary outcomes 2. Frailty 3. Rehospitalisations 4. Falls risk factors 5. Quality of life 6. Physical activity and sedentary behaviour 7. Acceptability of programme by participants and clinic staff Economic evaluation also planned in the first year
Funding and conflict of interest statements	Study was funded by the Canadian Institutes of Health Research (CIHR) grant (FRN 99051) and career award support for Dr Ashe from CIHR, the Michael Smith Foundation for Health Research, and the Canada Research Chairs Program. Explicit statement of no conflicts of interest.
Notes	Published protocol available. The inclusion criteria differed slightly between registration (e.g. recent history (12 months) of femoral fracture at trial registration). Study planned to recruit 130 adults from 2 academic teaching hospitals. A notable change between protocol and report in inclusion criteria was that time since hip fracture was "within past 3 months" in the protocol and "3 to 12 months after hip fracture" in Zusman 2019. Actual intervention: "For the participants in the intervention group, the median (p10, p90) number of health visits to the geriatrician was 2 (1, 4); PT was 2 (1, 11.6); and occupational therapist was 1 (1, 4.3). The geriatrician made referrals to other health professionals (within the clinic), such as a continence nurse (n = 3 participants); dietician (n = 2 participants); and social worker (n = 1 participant)." A qualitative study on hip fracture recovery is available (Stott‐Eveneshen 2017). A report, limited to activity and sedentary behaviour outcomes, was located in June 2019 (Zusman 2019); and also identified in November 2019 search update: study renamed to Ashe 2019. Request sent to Maureen Ashe on 2 January 2020 asking if 1) there are plans to publish a full report of the trial; 2) what happened to the additional 9 participants described in the trial registration document ("actual enrollment: 62") not accounted for in Zusman 2019; 3) results for death and in residential care at 6 and 12 months; and 4) an explanation for change in 'timing since hip fracture' between protocol and trial report (see above). Response received from Maureen Ashe on 2 January 2020. "1. We are in the process of trying to get a manuscript published on the primary outcome (SPPB). 2. There were only 53 participants in the main trial, however, in advance of starting the study, we conducted a pilot study with 9 participants who all received the intervention. They were not part of the main study. 3. As far as I understand, no participants died or moved to residential care during the study, but 1 participant was loss to follow up. 4. Recruitment was a challenge, thus we changed our inclusion criteria to older adults who fractured within the previous 12 months."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	“An independent statistical company randomized participants using a computerized program to remotely allocate participants into either intervention (n = 26) or control (n = 27) groups. We stratified randomization by sex and recruitment site.”
Allocation concealment (selection bias)	Low risk	“An independent statistical company randomized participants using a computerized program to remotely allocate participants into either intervention (n = 26) or control (n = 27) groups. We stratified randomization by sex and recruitment site.”
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	"All research study staff were blinded to group allocation, except the project manager (who did not collect quantitative data)." Although data collection is less unlikely to be affected for these outcomes, "Participants were aware whether they were randomized to receive the intervention."
Blinding (performance bias and detection bias) Function, QOL	High risk	"All research study staff were blinded to group allocation, except the project manager (who did not collect quantitative data)." "Participants were aware whether they were randomized to receive the intervention."
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Clarification on numbers randomised received from principal investigator and confirmation of no deaths or placement in residential care by 12 months.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Most outcomes not reported. Losses in terms of missing data comparable between the two groups. However, incompletely reported activity outcome results in Zusman 2019.
Selective reporting (reporting bias)	High risk	Prospective trial registration and published protocol but discrepancies between these and, so far, only partly reported despite the trial being completed in September 2014. Currently only sedentary behaviour, light physical activity, moderate or vigorous physical activity and step count reported.
Free of bias resulting major imbalances at baseline	Unclear risk	Baseline characteristics not reported in full.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Usual care was not well defined. “As part of our protocol, we collected data monthly via phone calls to all of our participants. We recognize that this may have served as a form of cointervention that prompted participants in the control group to be more physically active than they would have been otherwise; in our qualitative interviews, some participants in the control group expressed appreciation of having regular phone calls.”
Free of detection bias (from e.g. differences in follow‐up procedures)?	Unclear risk	Appeared similar and systematic but time since fracture and thus start of follow‐up was 25 days later in the control group (215 versus 240 days).

Baroni 2019.

Study characteristics
Methods	Randomised trial using a coin flip Assessor blinding: no Length of follow‐up: until discharge from acute hospital (mean 8.5 days); also 12 months for mortality
Participants	Teaching hospital, Perugia, Italy Conducted: March to August 2012 229 people with hip fracture; 23% male. Mean age 83 years (data from conference abstract: Serra 2013) 220 people with hip fracture; 23% male. Mean age 83 years (data from main report: Baroni 2019). Cognitive status: 43 (20%) had dementia Inclusion criteria: aged 65 years or older admitted to hospital because of a "proximal native or low‐impact femur fracture", informed consent. Exclusion criteria: peri‐prosthetic, cancer‐related, multiple trauma and inherited bone disorder fractures were excluded. Assigned: 119/110 (Serra 2013); 112/108 (main report) [Orthogeriatic co‐management / Orthopaedic with geriatric consultation] Assessed (up to discharge; 12 months): 112/108 (at discharge) Loss to follow‐up: 9 reported in Serra 2013 are not accounted for
Interventions	(1) Orthogeriatric co‐management (OGC) An orthopaedic surgeon and a geriatrician who, with a nurse, performed daily rounds from Monday to Saturday. The appointed geriatricians usually were rapidly notified of participants assigned to OGC; they took care of patients at the earliest from admission and during the entire hospital stay. Orthopaedic surgeons changed weekly due to their high turnover rate. Geriatric and orthopaedic care providers shared responsibility for patients’ care throughout their entire in‐hospital stay. They evaluated and defined patients’ priorities, i.e. surgical and/or medical, and treated them according to a standardised comprehensive geriatric assessment and multidisciplinary management. The traumatologist decided on the suitability of the surgical treatment, technique to use, and when progressive weight bearing could begin. The geriatrician managed comorbidities and polypharmacy to make patients clinically stable and ready for surgery, to reduce peri‐operative complications and promote early functional recovery. The geriatrician wrote daily notes about medically indicated orders in patients’ charts, and coordinated additional medical or surgical consultations as required. Both traumatologist and geriatrician shared surgical and clinical information at the daily briefing with anaesthesiologists in the pre‐operative phase for the trauma list. While in the early post‐operative phase, they started rehabilitation and discharge planning with physical therapists and nurses, respectively. The geriatrician was not available at weekends or public holidays, when the traumatologists were on‐call. The OGC was completely dedicated to the management of persons with hip fractures in their acute phase. Such a model was developed based on the reorganization of existing local staff and resources. (2) Consultant geriatric service provided to orthopaedic care (GCS) Patients were mainly under the management of the traumatologists who decided upon the need for any geriatric consultation during the hospital stay according to their perception of patients’ clinical conditions. There were neither standard clinical pathways nor a multidisciplinary treatment plan. Referrals for medical problems were made to the GCS through a formal request of a consultation. The GCS consisted of two geriatricians skilled in the care of hip fracture patients. They worked in the geriatric acute ward of the hospital, the same one of the geriatricians involved in OGC. The geriatric consultant team assured patients’ evaluation within 8 hours from the request and provided assessment, interventions, and orders for patients’ care by written notes in the charts. Traumatologists took responsibility for the management of patients’ medical and surgical problems throughout the entire hospital stay.
Outcomes	Mortality (in hospital and at 1 year), length of acute hospital stay Not used: time to surgery, in‐hospital medical complication (internally inconsistent data), prescription of anti‐fracture medication
Funding and conflict of interest statements	No mention of funding. Explicit statement of no conflicts of interest.
Notes	First identified as a conference abstract (Serra 2013). Notice of full report provided by Professor Carmelinda Ruggiero. While we received a detailed response to our requests for clarification of the link with the study reported in an abstract (Baroni 2016), currently awaiting classification, it was confusing. Also reported was a retrospective comparison with traditional orthopaedic care (230 participants in Serra 2013; 210 in Baroni 2019) in March to August 2011.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	“As informed consent was obtained, participants were randomly assigned to OGC or GCS by orthopedic resident on call, in collaboration with the orthopedic surgeon in charge, using the coin‐flipping procedure.”
Allocation concealment (selection bias)	Unclear risk	No information of safeguards to allocation concealment.
Blinding (performance bias and detection bias) Death, residence, readmission	High risk	Not blinded. Aside from death (very few in hospital), reported outcomes were susceptible to biases resulting from lack of blinding.
Incomplete outcome data (attrition bias) Death, residence, readmission	Unclear risk	Although follow‐up was just to hospital discharge, there was no participant flow and there were differences between numbers reported in a conference abstract (119 versus 110 compared with those in the full report (112 versus 108). It is unknown whether this slight between‐group difference in the missing participants made a difference to the results.
Selective reporting (reporting bias)	High risk	No trial registration or published protocol. Very short active follow‐up (until acute hospital discharge). Missing key outcomes; some of which (e.g. ADL) were assessed at baseline.
Free of bias resulting major imbalances at baseline	Unclear risk	Clear baseline imbalances. Particularly notable imbalance in fracture type. 38.4% versus 55.1% neck ('medial') fractures (reported P = 0.01).
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Geriatricians were in common to the two interventions and gave continuity of care. The traumatologists (orthopaedic surgeons) may also have been in common over the period but were changed over every week. This increased the risk of ‘contamination’.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	No indication of a difference between the two groups in follow‐up to hospital discharge

Cameron 1993.

Study characteristics
Methods	Randomised trial: stratification (using sealed envelopes) Assessor blinding: no Length of follow‐up: 4 months (12 months for some outcomes)
Participants	General hospital in Sydney, Australia. Conducted: January 1989 to October 1990. 252 people with hip fracture; 17% male. Mean age 84 years. Cognitive status: 122 were cognitively impaired. Inclusion criteria: aged > 50 years, had undergone surgery within 7 days of injury for an uncomplicated hip fracture, resident in district, informed consent. Exclusion criteria: pathological fracture, other fractures, fracture sustained while a hospital inpatient, transferred to another hospital for surgery. Assigned: 127/125 [Accelerated care / Usual care] Assessed (4 and 12 months): 127/125 Loss to follow‐up: none lost to follow‐up.
Interventions	Randomisation took place at time of surgery. (1) Accelerated rehabilitation and early discharge. Early assessment by physician experienced in rehabilitation and geriatric medicine, to identify and treat concurrent illness, and establish rehabilitation goals. Early commencement of rehabilitation (mobilisation and self‐care), greater emphasis on retraining physical independence, closer family care giver support and more detailed discharge planning. Nursing home patients: contact by rehabilitation physician and arrangements made for mobilisation; supervised by nursing staff, visiting physiotherapist. Review of progress by rehabilitation physician; and orthopaedic review according to need. Other patients, mobilisation in hospital, supervised by physiotherapist, continued by nursing staff. Joint 3 or 4 times weekly review by orthopaedic surgeon and rehabilitation physician. Patients with limited disability direct discharge to home: seen by occupational therapist, input too from social worker and nutritionist. Additional / more intensive treatment for those with moderate to severe disability prior to fracture: transfer to rehabilitation ward with interdisciplinary rehabilitation programme and regular contact with nursing, medical, physiotherapy, occupational therapy and social workers. Additional support upon discharge, e.g. delivered meals, nursing help, physiotherapy, day hospital attendance. (2) Conventional care and rehabilitation (of these 56% had multidisciplinary rehabilitation) as per standard treatment provided at study hospital at time of trial. Patients living in nursing homes and those with limited disability were discharged when deemed orthopaedically appropriate. Patients likely to need additional assistance to return home were referred to Rehabilitation and Geriatric service several days after surgery. Many were transferred to a rehabilitation ward. The following summary is given: "In summary, accelerated rehabilitation differs from conventional care in its early assessment of rehabilitation goals, early commencement (usually within 24 hours of surgery), greater emphasis on retraining for physical independence, closer family care‐giver contact and more detailed discharge planning." Common care Timing of surgery Anaesthesia and analgesia Type(s) of surgery Mobilisation policy Discharge policy
Outcomes	'Poor outcome': mortality at discharge or discharge to institutional care; mortality at 12 months or institutional care. Other outcomes: mortality (4 and 12 months), ADL/functional status (Barthel Index), length of stay, place of residence, readmissions, carer burden, cost effectiveness.
Funding and conflict of interest statements	The study was supported by a grant from the Australian Department of Health, Housing and Communty Services. Care burden study also acknowledges a grant from University of Sydney. There were no conflict of interest statements.
Notes	Data and clarifications received from Ian Cameron (including emails: 26/02/1997, 19/06/2000) on mortality at 4 and 12 months, institutional care, dependency, SDs for hospital stay.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"A stratified randomization procedure was used." Stratification by prior living arrangements and disability.
Allocation concealment (selection bias)	Low risk	"The surgeon was blind to the outcome of randomization, which was performed at the time of surgery." Use of sealed envelopes.
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Not blinded but data collection unlikely to be affected for these outcomes.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	Clinical assessor was not blinded to group assignment. However, there was high correlation between the findings and that of an independent blind assessor who evaluated Barthel scores for 10 participants.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	"All 252 patients were successfully followed up until death or 4 months after injury." However, there were discrepancies in the mortality data between the reports and those provided by the author from his thesis.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Despite reporting that all patients were followed up, discrepancies in mortality data (see above) meant that this was unclear.
Selective reporting (reporting bias)	Unclear risk	Incompletely reported but data provided subsequently. The study protocol is not available but it seems very likely that the published reports included all expected outcomes, including those that were pre‐specified.
Free of bias resulting major imbalances at baseline	Unclear risk	Intervention group was slightly younger (82.4 versus 85.4); this was reportedly due to "differences in the subgroup from nursing homes".
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Insufficient details for the control group to judge this.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Follow‐up procedures were the same in both groups and all participants were accounted for.

Chong 2013.

Study characteristics
Methods	Quasi‐randomised trial based on odd and even ID numbers. Assessor blinding: yes. Some outcome assessment was by trained research assistants who were blinded with respect to the patient’s allocation. Length of follow‐up: 12 months
Participants	Subacute rehabilitation hospital, Singapore Conducted: September 2004 to June 2006 162 people admitted for step‐down care after hip fracture. 31.5% male. Mean age 78.4 years. Cognitive status: 28 had dementia Inclusion criteria: admitted for rehabilitation after hip fracture (surgical treatment data only for 151 (93%)). Exclusion criteria: pre‐morbid non‐ambulatory status, nursing home residents, palliative care patients, or patients previously enlisted for the trial. Assigned: 92/70 [care pathway / usual multidisciplinary care] Assessed (12 months): 66/56 Loss to follow‐up = 26 versus 14 (withdrew (follow‐up refused), deceased)
Interventions	Comparison of different models of multidisciplinary rehabilitation (1) Multidisciplinary team care plus structured assessments and checklists. The intervention group had the following as part of the integrated care pathway (ICP): 1. Medical assessment on admission for risk factors for falls, using a template. 2. A protocol was developed for the early detection and management of complications consisting of a weekly assessment of complications, including pain, deep venous thrombosis, anaemia, wounds and pressure ulcers, depression, delirium, constipation, urinary retention and malnutrition, and interventions for osteoporosis. Patients reporting pain were started on pain assessment and monitoring charts. 3. The therapists coordinated their work through the use of a combined physiotherapy and occupational therapy assessment form, which facilitated goal setting and helped to avoid duplication of work. Five‐week physiotherapy and occupational therapy guidelines with recommended milestones were developed and applied by the therapists. Different milestones were set for the full, partial and non‐weight bearing groups. 4. Physiotherapists in the intervention group used the Physiotherapy Clinical Outcome Variables Scale to assess baseline mobility, define outcome goals and direct treatment plans. 5. A postoperative hip precaution handout providing information on avoiding hip prosthesis dislocation in patients with total hip replacement or hemiarthroplasty was given to patients and their caregivers. (2) Usual multidisciplinary care
Outcomes	'Poor outcome': mortality at 12 months or non‐return to previous ambulatory status Other outcomes: mortality (12 months), Montebello Rehab Factor score, ADL/functional status (Modified Barthel Index), quality of life (SF‐12), length of stay, any nursing home stay after discharge, place of residence, readmissions, achieving premorbid ambulatory status, weight bearing at discharge
Funding and conflict of interest statements	No mention of funding No conflict of interest statements
Notes	Previously excluded: randomised evaluation of a clinical care pathway. Both groups received multidisciplinary care. IC: After reading the study data in more detail, I now feel that this is eligible because, as Helen has said, it is a comparison of different models of multidisciplinary rehabilitation. Therefore INCLUDE
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	"Administrative staff allocated patients to either ICP or usual care according to the last digit of their National Registration Identity Card (NRIC) numbers, odd numbers to the intervention group and even numbers to the control group. Patients were then admitted to 1 of 2 intervention wards or 1 of 3 control wards. Patients were enrolled by the principal investigators only after moving into their respective wards because of workflow limitations." Quasi‐RCT, thus non‐random sequence generation
Allocation concealment (selection bias)	High risk	"Our study is the only pseudo‐randomised control trial of ICPs for hip fracture in community rehabilitation hospitals" Quasi‐RCT, allocation not concealed
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Outcomes less susceptible to risk of bias.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	"Clinical assessments were performed by trained nurses and therapists, and research baseline and outcome assessments were performed by trained research assistants, the latter being blinded with respect to the patient’s allocation to either intervention or control group." Not blinded in terms of performance but this attempt at blinding of outcome assessors is likely to reduce the risk of bias for these outcomes.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Some disparity between the two groups in those lost to follow‐up at 12 months: 16/92 (17%) versus 8/70 (11%) but probably no loss to follow‐up for mortality and does not affect these outcomes.
Incomplete outcome data (attrition bias) Function, QOL	High risk	Plus mortality, high and a bit imbalanced loss to follow‐up between the two groups at 1 year (26 (28%) versus 14 (20%)). Exact denominators not provided by separate groups for data at 6 months and for scores such as the SF‐12.
Selective reporting (reporting bias)	Unclear risk	The study protocol is not available but a detailed list of measurements and outcomes were provided in the methods section. It is notable that data collected at 12 months for some measures (the Mini Mental State Examination (MMSE) and the Geriatric Depression Scale (GPS: 0 to 15) were not reported. Even though these were not listed as the trial outcomes, it is not clear why these would not merit reporting. For instance, 55% were depressed on admission (GPS ≥ 5), so knowing how depressed participants were by 12 months would appear to be insightful.
Free of bias resulting major imbalances at baseline	Unclear risk	Groups were balanced in most characteristics. However, more had visual impairment in the intervention group (43 (46.7%) versus 20 (28.6%); reported P = 0.02) and this group also had a lower MMSE score (0 to 30) 16.7 versus 18.7 (reported P = 0.057).
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	"It was difficult to avoid cross‐over contamination between the trial wards. There was some control of this problem in that there were separate medical, nursing and rehabilitation teams for the intervention and control wards, although inevitably some cross‐over coverage and experience might have occurred. The long recruitment period for the trial also allowed for more contamination between wards."
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Follow‐up procedures were the same in both groups

Crotty 2003.

Study characteristics
Methods	Randomised trial: using computer‐generated randomisation and sealed, opaque envelopes; independent pharmacist with no other involvement with study. Assessor blinding: yes Length of follow‐up: 12 months
Participants	3 hospitals in Adelaide, Australia. Conducted: July 1998 to March 1999. 66 people with hip fracture, 33% male. Mean age 82.5 years. Cognitive status: 14 had confusion; otherwise needed to be able to participate in rehabilitation programme. Inclusion criteria: aged 65+ years, had undergone surgery for fall‐related hip fracture; medically stable; adequate medical and mental capacity to participate in a rehabilitation programme; expected to return home after hospital discharge; home environment suitable for rehabilitation; agreed to hospital admission should complications occur; written consent. Exclusion criteria: inadequate support in community; no telephone at home; or not resident in Adelaide's Southern Metropolitan Region. Assigned: 34/32 [Accelerated care / Usual care] Assessed (12 months): 28/28 Loss to follow‐up: 3 lost to follow‐up; 7 died
Interventions	Randomisation took place in acute care, after surgery (1) Accelerated discharge and home‐based interdisciplinary rehabilitation. Initial assessment by study co‐ordinator who visited patient's home and organised modifications, equipment and assistive aids before patient's discharge. GPs were contacted. Participants were discharged from acute care within 48 hours of randomisation and were visited by physiotherapists, occupational therapists, speech pathologists, social workers and therapy aides, who negotiated a set of realistic, short‐term, and measurable treatment goals with both participants and their carers. Standard therapy services (e.g. podiatry, nursing care, assistance with light domestic tasks) provided as required. Weekly multidisciplinary case conference also attended by specialist in rehabilitation medicine or geriatrician. Decision to discharge from rehabilitation programme made in consultation with rehabilitation consultant. When necessary, participants were referred to community agencies for ongoing care. (2) Usual or 'conventional' care, involving routine interdisciplinary hospital care and rehabilitation in hospital. Inpatient services, care pathways and discharge planning.
Outcomes	'Poor outcome': mortality at 12 months or living in more dependent residence (hostel or nursing home). Other outcomes: mortality (4 and 12 months), quality of life (SF‐36: physical component summary and mental component summary); ADL (Modified Barthel Index), Balance Confidence Scale, mobility (timed up and go test and inability to walk), length of stay in hospital/of rehabilitation (including at home), change in residence (indicating higher level of care), readmissions, home visits by therapists (as part of intervention), use of community services, GP visits, Caregiver Strain Index, fallers and hospitalisation for falls.
Funding and conflict of interest statements	Study funded by grant from the South Australian Department of Human Services, Australia. There are no conflict of interest statements.
Notes	Data and clarifications received from Michelle Miller (17/12/2008): separate data by group for deaths, admission to higher level of care, inability to walk, readmission to hospital, SDs for SF‐36 data and hospital stay.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"computer‐generated allocation sequence"
Allocation concealment (selection bias)	Low risk	"Randomization was undertaken by the hospital pharmacy department, which maintained a computer‐generated allocation sequence in sealed opaque envelopes." "Randomization ... was performed by a hospital pharmacist who was unaware of the medical status of the patient and who had no other involvement in the study."
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	"Follow‐up data were collected at 4 and 12 months ... by an assessor who was blinded to treatment allocation."
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	"Follow‐up data were collected at 4 and 12 months ... by an assessor who was blinded to treatment allocation." However, trial participants were not blinded.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Loss to follow‐up and death fully reported and data split by treatment group provided.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Although denominators stated the 3 lost to follow‐up were all in the accelerated care group.
Selective reporting (reporting bias)	Unclear risk	Incompletely reported but data provided subsequently. The study protocol is not available but it seems very likely that the published reports included all expected outcomes, including those that were prespecified.
Free of bias resulting major imbalances at baseline	Low risk	Baseline characteristics, as well as prior care and timing, seemed comparable.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Not enough detail to confirm this.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Comparable follow‐up, and set times.

Crotty 2019.

Study characteristics
Methods	Randomised controlled trial Randomised trial: using computer‐generated randomisation and sealed, opaque envelopes; independent pharmacist with no other involvement with study. Assessor blinding: yes Length of follow‐up: 12 months
Participants	Randomised at 3 hospitals, Adelaide, Australia but intervention carried out in nursing homes / care facilities Conducted: June 2012 to December 2014 240 people from 76 nursing homes (nursing care facilities) with hip fracture; 26% male. Mean age 86.6 years. Cognitive status: in 97% of cases, informed consent was obtained from family members because of cognitive impairment. 186 (77.5%) recorded diagnosis of dementia. Inclusion criteria: a recent hip fracture (proximal femoral fracture) treated surgically, aged 70 years or older, living in an aged care facility (nursing home) within the catchments of the local hospital prior to injury. Ambulant prior to fracture either without assistance, with aids or with the assistance of one other person and medically stable and ready for discharge. Exclusion criteria: unable to provide informed consent or gain this from a suitable proxy, pathological and peri‐prosthetic fractures, terminal illness and receiving palliative care, hip fracture treated non‐surgically, or severe cognitive impairment, unable to follow a one‐step command at recruitment. Assigned: 119/121 [Multidisciplinary rehabilitation / Usual care] Assessed (12 months): 60/66 Loss to follow‐up: 110 deaths, 2 lost to follow‐up and 2 withdrawals
Interventions	Intervention started within 24 hours of return to nursing home; all residents received usual medical care from their general practitioner. 201 (84%) were discharged from hospital within 48 hours of randomisation; most were unable to transfer or were confined to bed. The mean length of stay in the acute hospital was 7.3 days (range 3 to 33 days). (1) Multidisciplinary rehabilitation. Four‐week ambulatory geriatric rehabilitation programme. Aimed to resemble that routinely delivered to older adults with hip fracture who return to community living, geriatric hip fracture programmes and early supported discharge programmes. The programme included: medical assessment by a geriatrician prior to discharge if possible and post‐discharge at residential facility; and review of medications and comorbidities by a geriatrician prior to discharge if possible and post‐discharge at residential facility. Following discharge, participants received 4 weeks of physiotherapy focused on restoration of transfers and limited mobility with a physiotherapist and allied health assistant. Other allied health assessment and support, including dietetics assessment, were provided as required. In addition, the care providers worked with aged care staff to encourage a rehabilitation approach between formal therapy sessions, and a formal meeting was set up between geriatrician and families within a fortnight to discuss progress. Intervention was 'low intensity' and involved 13 hours of input spread over 4 weeks. (2) Usual care. Participants continued treatments (which may have included sessions of physiotherapy) according to usual practice in the nursing home. The control group on all sites received orthogeriatric care in hospital and medical care from a general practitioner after discharge.
Outcomes	Follow‐up: 1 and 12 months post‐randomisation 'Poor outcome': mortality at 12 months or unable to walk ("unable to walk at all, uses a wheelchair to get around, or stays in bed") Mortality, quality of life (DEMQOL, DEMQOL‐proxy), EQ‐5D‐5L (for cost‐effectiveness analysis), inability to ambulate, mobility autonomy measured using Nursing Home Life‐Space Diameter (0, signifying bed‐ or chair‐bound to 50, signifying leaving the facility daily); modified Barthel Index (0 to 100: total independence); Cornelll Scale for Depression in Dementia; cognition (Mini Mental State Examination); delirium (Confusion Assessment Method); nutrition (Mini Nutritional Assessment); pain (Pain Assessment IN Advanced Dementia); medications (appropriateness based on 10 criteria); recovery (Functional Recovery Score; Zuckerman 2000); medical and adverse events (falls and injuries). Incremental cost‐effectiveness analysis. For DEMQOL and DEMQOL‐proxy, a zero was assigned for people who had died. Missing values for these were imputed for those who had answered at least half the questionnaire by the mean value of complete questions as recommended by the instrument developers.
Funding and conflict of interest statements	Funder: National Health and Medical Research Council (NHMRC), Australia Conflict of interest declared for M Crotty relating to two previous clinical drug trials on community dwelling hip fracture patients; and for M Chehade relating to institutional grants from Stryker to support hip fracture research. Neither posed "a relevant conflict".
Notes	A review author (IC) is an investigator on this trial. Prospectively registered: 24/01/2012; updated 28/07/2014 A qualitative investigation conducted in parallel with this trial was reported in 2016 (Killington 2016). Title: SACRED ‐ Southern Adelaide co‐ordinated regional hip and debility rehabilitation programme to improve quality of life Paper describing intervention submitted (notified by IC 19/07/19) HH sent email on 04/09/19 to Maria Crotty requesting data on number of nursing homes, hospital length of stay, mortality at 4 months, number of bed‐bound/non‐ambulating survivors at 12 months, and clarification on delirium data. Response with answers on all these points received 30/09/19.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"A computer generated random sequence with random block sizes was used by a pharmacist external to the project to allocate people with a hip fracture"
Allocation concealment (selection bias)	Low risk	"A computer generated random sequence with random block sizes was used by a pharmacist external to the project to allocate people with a hip fracture" Independent pharmacist ensured allocation concealment
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Not blinded but outcomes are less susceptible to blinding
Blinding (performance bias and detection bias) Function, QOL	High risk	Not blinded. Outcomes at risk.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Participant flow diagram provided. Very few lost to follow‐up.
Incomplete outcome data (attrition bias) Function, QOL	High risk	High loss to follow‐up because of mortality; large difference between groups at 4 weeks.
Selective reporting (reporting bias)	Low risk	Corrections and changes in trial registration document were all accounted for and plausible
Free of bias resulting major imbalances at baseline	Unclear risk	Imbalances in some reported characteristics are small but it is not known if there are imbalances in time to randomisation from injury; type of anaesthesia; co‐morbidities etc.
Free of performance bias (from non‐trial differences in care provision)?	High risk	This is unavoidable with probable contamination and differences between nursing homes.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	No problem detected.

Fordham 1986.

Study characteristics
Methods	Randomised trial: method unclear although independent of clinicians providing care Assessor blinding: no mention Length of follow‐up: to discharge
Participants	General hospital in Huddersfield, UK (intervention patients transferred to community hospital). Conducted: March 1985 to March 1985. 108 women with hip fracture (surgically treated). Mean age not stated. Cognitive status: none retained in trial with long‐standing dementia. Inclusion criteria: women aged over 65 years with hip fracture. Exclusion criteria: (post‐randomisation) patients transferred from other hospitals or health authorities. Non consent. Confusion and terminal illness. Assigned: 50/58 [Joint geriatric and orthopaedic care / Orthopaedic care] Assessed (mortality): 50/58 Loss to follow‐up: none, but no data on 25 omitted (14 versus 11) post‐randomisation.
Interventions	Randomisation took place immediately after the operation. (1) Joint geriatric and orthopaedic management involving early post‐admission assessment by geriatrician on the orthopaedic ward, joint decision for transfer to geriatric‐orthopaedic ward (6 beds in orthopaedic rehabilitation ward converted to geriatric‐orthopaedic beds), joint clinical rounds once a week and liaison away from bed side, joint responsibility for rehabilitation programme, joint decisions about and facilitation of discharge or transfer, joint junior medical staff. (2) Orthopaedic management (geriatrician available for advice) including decision to transfer to orthopaedic rehabilitation unit.
Outcomes	'Poor outcome': mortality at discharge + discharge to nursing home (or remaining in hospital at 2 months) Other outcomes: mortality (discharge), patient function / activities of daily living (standing, getting in/out of bed, getting in/out of a chair; walking; dressing) assessed using 9 categories (1 = independent), length of stay, place of residence (at discharge or 2 months), transfer to community hospital, costs (staff and hospital stay)
Funding and conflict of interest statements	Funding received from the Kings Fund, London and Yorkshire Regional Health Authority. No conflicts of interest statements
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Randomly assigned" by research team within first few hours of hospital admission.
Allocation concealment (selection bias)	Low risk	Randomisation by research team ‐ then appropriate clinicians were informed.
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Unlikely that mortality data affected by lack of blinding. However, discharge location might have been.
Blinding (performance bias and detection bias) Function, QOL	High risk	The lack of blinding may have influenced these outcomes.
Incomplete outcome data (attrition bias) Death, residence, readmission	Unclear risk	Post‐randomisation exclusions means that there is some risk of bias. Full data for death and residence at discharge provided for all those retained in trial.
Incomplete outcome data (attrition bias) Function, QOL	High risk	Functional outcome data (standing, getting in/out of bed, getting in/out of a chair; walking; dressing) were incomplete, with data missing for some participants; with some imbalances between groups in those not assessed.
Selective reporting (reporting bias)	Low risk	Very comprehensive and planned analyses.
Free of bias resulting major imbalances at baseline	Unclear risk	No significant differences reported in population after post‐randomisation exclusions (mainly not in catchment area) but larger numbers of subtrochanteric (12/50 versus 5/58) fractures and less hip replacement in the intervention group.
Free of performance bias (from non‐trial differences in care provision)?	High risk	The three‐mile distance between the hospitals, with no clinics held in the acute hospital by the geriatricians, may have affected performance. Potential contamination as some clinicians were involved in patient management in both groups.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Follow‐up appeared thorough.

Galvard 1995.

Study characteristics
Methods	Randomised trial: use of a random number generator Assessor blinding: no mention Length of follow‐up: 12 months
Participants	General hospital in Malmo, Sweden. Conducted: November 1988 to December 1989. 378 people with hip fracture. Of 371: 26% male. Mean age 79 years, range 52 to 102 years. Cognitive status: no information but all independent. Inclusion criteria: people living in own home, perhaps receiving meals on wheels or home help service, with hip fractures. Patient consent. Exclusion criteria: people resident in nursing home or waiting for a nursing home bed, or already in hospital. Second hip fracture. Assigned: 182/196 [Geriatric hospital / Orthopaedic hospital] Assessed (mortality): 179/192 Loss to follow‐up varied according to outcome assessed: it was markedly incomplete for hip pain (38/60), walking ability and gait speed.
Interventions	Randomisation took place immediately after the operation. (1) Geriatric rehabilitation: transfer to geriatric hospital (usually on second post‐operative day) and weekly visit by orthopaedic surgeon (2) Usual orthopaedic care
Outcomes	'Poor outcome': mortality at discharge + discharge to nursing home (or other) Other outcomes: mortality (discharge and 1 year), patient function (subset at 1 year: hip pain and mobility), length of stay, place of residence (at discharge), readmissions (during 1 year), costs
Funding and conflict of interest statements	No information on funding No conflict of interest statements
Notes	Although 378 patients were entered into the study, 7 were excluded (protocol violations): 4 with a second hip fracture wrongly included; 2 did not give consent, 1 did not live locally.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomisation took place at the orthopaedic department immediately after the operation, using a random number generator."
Allocation concealment (selection bias)	Unclear risk	No details.
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	All patient records, from a central data registration facility, reviewed by the trial investigators.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	No reference made to blinding but it was considered unlikely that the trial allocation influenced the physiotherapist assessing mobility at one year follow‐up.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Full data for death, residence at discharge and readmission provided.
Incomplete outcome data (attrition bias) Function, QOL	High risk	Functional outcome data (mobility and pain) only provided for a subgroup of trial participants (38 versus 60) attending follow‐up session at 1 year. Report noted that a potential contributing factor to the difference in numbers of attendees was "that patients treated at the orthopaedic department traditionally participate in trials and studies, while patients treated at the geriatric hospital had no such tradition at the time."
Selective reporting (reporting bias)	Unclear risk	The study protocol is not available but it is very likely that the published report includes all expected outcomes, including those that were prespecified.
Free of bias resulting major imbalances at baseline	High risk	As reported by the trial authors, there is significant bias due to: a) the men in the geriatric hospital group being older (79.1 versus 73.6 years); and b) the higher proportion of patients with subtrochanteric fractures, which often require longer rehabilitation, in the same group (21/179 (12%) versus 8/192 (4%)).
Free of performance bias (from non‐trial differences in care provision)?	High risk	In contrast to the orthopaedic hospital, the geriatric hospital had no prior experience with hip fracture patients. It was noted that "the orthopaedic department had first priority to home help services, alterations in patients' flats, and convalescent homes." This and the fact that there was a lack of nursing home beds in Malmo was mooted as a factor contributing to the longer hospital stay in geriatric hospital.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Unclear risk	While detection bias was unlikely for outcomes such as death, the imbalance in the follow‐up of functional outcomes could have been related to the different research environments at the two trial sites (see above).

Gilchrist 1988.

Study characteristics
Methods	Method of randomisation: unclear, stratified by fracture type Assessor blinding: no mention Length of follow‐up: 6 months after discharge (for mortality)
Participants	General hospital in Glasgow, UK. Conducted: October 1984 to July 1986 222 women with hip fracture. Mean age 81 years. Cognitive status: initial mental function assessed using mini object test (101 had scores > 19/30, would have had better mental function). Inclusion criteria: women aged over 65 years with hip fractures, transferred to the hospital for rehabilitation. Exclusion criteria: none (6 patients had pathological fractures). Assigned: 97/125 [Orthopaedic‐geriatric ward / Orthopaedic ward] Assessed (mortality): 97/125 Loss to follow‐up: none.
Interventions	Randomisation took place at transfer, which occurred at a mean of 10 days after admission for hip fracture, from another hospital. (1) Combined geriatric‐orthopaedic care in special designated unit. Weekly ward round with a geriatrician, an orthopaedic senior registrar and a senior ward nurse; followed by a case conference with, in addition, a physiotherapist, occupational therapist and a social worker. (2) Usual orthopaedic care in orthopaedic ward. Referral to geriatrician (different person from the one attending the geriatric‐orthopaedic ward) by letter. All patients had similar nursing cover, access to similar paramedical services and remained under the care of the orthopaedic surgical staff.
Outcomes	'Poor outcome': mortality at discharge and discharge to nursing home (or institution) Other outcomes: mortality (discharge, 3 and 6 months post‐discharge), length of stay, place of residence (at discharge), medical conditions (recognised; untreated), length of time patient seen by geriatrician
Funding and conflict of interest statements	No information on funding No conflict of interest statements
Notes	Although both groups were reported to have had home visits at 3 and 6 months after discharge, there were no data pertaining to follow‐up at these times, aside from mortality.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Randomised". Method not stated. However, "stratified on the basis of the site of the fracture (intracapsular;extracapsular)" and a 5 to 4 bias introduced in favour of the orthopaedic wards due to a larger number of beds.
Allocation concealment (selection bias)	Unclear risk	No details.
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Not blinded, nor was blinding possible, but prospective and active follow‐up for death. Unclear if discharge placement was influenced by knowledge of group allocation.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	No blinding. No function or QOL data but, while some standardisation was evident, assessment of morbidity (untreated conditions) may have been influenced by knowledge of group allocation.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	No loss to follow‐up or post‐randomisation exclusions; no transfer between wards. However, there was an incomplete account of placement of participants who had not been living in the community before their fracture.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Not clear if function assessed.
Selective reporting (reporting bias)	Unclear risk	The study protocol is not available but it is very likely that the choice of main outcomes was prespecified.
Free of bias resulting major imbalances at baseline	Low risk	No important differences in age, gender (all female), type of fracture, dementia (mini object score) and time to transfer.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	None detected. However, possible contamination with geriatric intervention 'spilling over' to the control group. A greater interest was taken by the orthopaedic surgeons in the medical and social aspects of care in the control group such that patient management (e.g. more comprehensive discharge notes and fewer drugs) had "considerably changed". This may have been influenced by the rotation of junior staff during training.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Follow‐up appeared active and systematic.

Huusko 2002.

Study characteristics
Methods	Randomised using opaque sealed envelopes; independently computer generated allocation sequence. Assessor blinding: no Length of follow‐up: 12 months.
Participants	Specialist district hospital in Jyvaskyla, Finland. Conducted: October 1994 to December 1998. 260 people with hip fracture. Of 243: 28% male. Mean age 80 years, range 66 to 97 years. Cognitive status: 52 had dementia. Inclusion criteria: community dwelling, > 64 years with surgically treated hip fracture, informed consent, independent ambulation. Exclusion criteria: pathological fracture, multiple fractures, terminally ill, serious early complication, receiving calcitonin, unable to communicate. Assigned: 130/130 [Intensive care/ Usual care]; 120/123 (after post‐randomisation exclusions) Assessed (12 months): 120/123 (mortality); 119/119 (residence); 95/98 (ADL)
Interventions	Patient consent obtained on first post‐operative day; assume randomisation followed this. All patients mobilised on first post‐operative day. (1) Intensive geriatric rehabilitation within hospital: multidisciplinary geriatric team (geriatrician, specially trained GP and nurses, occupational therapist, physiotherapists, social worker, neuropsychologist) working with other specialists (in physical medicine, neurology, psychiatry) providing intensive rehabilitation programme (assessment; twice daily physiotherapy; ADL practice; daily schedule; counselling; information; discharge plan; home visits, treatment at home after discharge) based in geriatric ward in same hospital as surgery. During first two months after discharge, patients discharged to independent living were visited 10 times by physiotherapist. (2) Discharge to local community hospitals, treatment by general practitioners; physiotherapist usually available. Transfer 2 to 5 days after surgery. Around half of the patients on the long‐term wards are long‐term patients. (No details given of actual non‐specialist care.)
Outcomes	'Poor outcome': mortality at discharge or non‐return to independent living; transfer to more dependent residence; mortality or institutional care or hospital at 12 months. Other outcomes: mortality (in hospital and 12 months), functional status (ADL and instrumental ADL), morbidity (any complication during 12 months), length of stay, discharge location (incomplete data); place of residence (12 months), any complication during follow‐up, costs, physiotherapy sessions, home services.
Funding and conflict of interest statements	Support by grants from the Central Finland Health Care District, Kuopio University Hospital, Emil Aaltonen Foundation, Uulo Arhio Foundation and Novartis Finland Ltd. Authors declared no conflicts of interests
Notes	Patients were also randomised to nasal calcitonin or placebo for 3 months (full report in 2002). The distribution of patients taking calcitonin was claimed to be similar in the two groups considered here. A pre‐planned subgroup analysis by cognitive impairment (based on the mini mental state examination) was performed and reported before (Huusko 2000) the report of the full population (2002). Data received from Huusko on return to independent living at discharge; on independence in ADL; and overall length of hospital stay over 1‐year follow‐up.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Allocation sequence was computer‐generated"
Allocation concealment (selection bias)	Low risk	"Allocation sequence was computer‐generated and sealed in numbered, opaque envelopes in Helsinki, Finland, by the information technology department of Novartis before the study was started. The envelopes were stored on the orthopaedic ward by the head nurse until patients were randomised."
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	Not blinded, but unlikely to affect results
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	"We could not blind the staff doing the interventions or assessments". Geriatrician in charge of study didn't take part in geriatric rounds or team meetings.
Incomplete outcome data (attrition bias) Death, residence, readmission	High risk	Baseline data were not presented for 17 participants (10 versus 7): 11 'violation of randomization criteria'; 3 'withdrawal of consent'; 3 'protocol violation'. Residence data also not available for 5 participants without initial mental state scores.
Incomplete outcome data (attrition bias) Function, QOL	High risk	As well as post‐randomisation exclusions, additional losses for functional outcomes (ADL and instrumental ADL) because results provided only for those with data measured at baseline.
Selective reporting (reporting bias)	High risk	No protocol available and incomplete reporting.
Free of bias resulting major imbalances at baseline	High risk	Imbalance in patient characteristics. For example, intervention group had a greater number with dementia (32/120 versus 20/123); fewer were functionally independent in ADL before hip fracture (41 versus 66).
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	No details of staff training or motivation. Hard to determine for complex packages.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Appeared systematic and thorough.

Jette 1987.

Study characteristics
Methods	Quasi‐randomised based on physician’s emergency on‐call schedule. Assessor blinding: no Length of follow‐up: 12 months (also 6 weeks, 3 and 6 months).
Participants	Teaching hospital in Boston, Massachusetts, USA. Conducted: 1 June 1983 to 30 July 1984 75 people with hip fracture, 33% male. Mean age 78 years. Cognitive status: mention of use of the Short Portable Mental Status Questionnaire to assess mental competence but no information otherwise. Inclusion criteria: > 54 years with surgically treated hip fracture, written consent. Exclusion criteria: not stated. Assigned: 35/40 [Intensive rehabilitation / Usual rehabilitation] Assessed (12 months): not known
Interventions	Patient consent obtained soon after surgery. (1) Intensive rehabilitation: standard care plus education sessions, a geriatric team evaluation, weekly team meetings, home visit post‐discharge, monthly follow‐up by phone). Six components: Standard rehabilitation (details under Usual care). One to two hours individual patient and family rehabilitation by physical therapist (+ booklet on hip fracture) to establish realistic expectations, goal setting, and enhanced commitment to the rehabilitation programme. Geriatric team evaluation focused on a comprehensive assessment and follow‐up of the medical needs of the patient. Weekly team meetings attended by geriatrician, team physical therapist, social worker focused on co‐ordinating rehabilitation, discharge planning, follow‐up on the geriatric team assessment. Home visit by hospital physical therapist and/or social worker within 2 days of discharge to assess patient’s home environment and enhance continuity of the home rehabilitation programme with patient and family; or a visit with staff for patients discharged to a rehabilitation hospital or nursing home to enhance continuity of the rehabilitation provided in that facility. Telephone calls by hospital physical therapist at least once a month post‐discharge to monitor / facilitate progress. (2) Usual care: "standard rehabilitation". Breathing exercises and quad sets on first day post‐surgery, progressing to daily active exercises. Stable fractures: out of bed on second day post‐surgery and began ambulating daily with weight‐bearing as tolerated. Unstable fractures: non‐weight‐bearing for 6 weeks. Note on personnel: although situated on the same hospital floor, each orthopaedic service had its own physicians, physical therapists and social workers. However, some nursing staff, particularly for night shifts, were shared.
Outcomes	No comparative data available. 'Poor outcome': mortality and not living at home at 12 months; mortality and not discharged back to former residence (discharged to rehabilitation hospital or, newly, to a nursing home) Mortality, discharge and final residence (home, nursing home), shortened version of the Functional Status Index (basic ADL, instrumental ADL and social / role ADL), emotional function (well being), medical complications including confusion, length of hospital stay.
Funding and conflict of interest statements	Research was supported by a grant from the Robert Wood Johnson Foundation. No conflict of interest statements provided.
Notes	Formerly excluded because of a lack of comparative data available (Handoll 2009). Requests for information and comparative data were unsuccessful. Email from Alan Jette to Ian Cameron on 9 August 1997 said that "I am sorry but I do not have time to dig into my archives for further information on a study done in the mid 1980s."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	“Patients were assigned randomly to one of two orthopaedic services based on the physician’s emergency on‐call schedule.” Appears to be quasi‐randomised.
Allocation concealment (selection bias)	High risk	“Patients were assigned randomly to one of two orthopaedic services based on the physician’s emergency on‐call schedule.” Appears to be quasi‐randomised; no allocation concealment.
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	No mention of blinding. Outcomes are less susceptible to lack of blinding.
Blinding (performance bias and detection bias) Function, QOL	High risk	No mention of blinding and outcomes susceptible to bias because of this.
Incomplete outcome data (attrition bias) Death, residence, readmission	Unclear risk	No participant flow and possible data disparities when calculating whole numbers from percentages. Scale of key plot differs from claimed proportions for mortality and living situation. No report of separate data for the two groups.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	No participant flow or assurances of full follow‐up of survivors for these outcomes. No report of separate data for the two groups.
Selective reporting (reporting bias)	High risk	No trial registration and the study protocol is not available. Separate data for the two groups are not reported.
Free of bias resulting major imbalances at baseline	Unclear risk	Unable to judge. Baseline characteristics were not split by treatment group; claimed “no statistically significant differences in background characteristics”
Free of performance bias (from non‐trial differences in care provision)?	High risk	Follow‐up in attending orthopaedic clinics differed: at 12 months, > 80% of surviving participants of the intensive rehabilitation group attended compared with 43% in the usual care group. Authors suggest that contamination between experimental and control groups may have occurred: staff were from the same department.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Unclear risk	Insufficient information to allow judgement although set follow‐up times were indicated

Karlsson 2016.

Study characteristics
Methods	Randomised based on sequentially numbered opaque, sealed envelopes Assessor blinding: no Length of follow‐up: 12 months (also 3 months) after surgery.
Participants	Teaching Hospital, Umeå, Sweden Conducted: May 2008 to June 2011 209 people who had acute hip fracture surgery. Of 205: 28% male. Mean age 82.9 years. Cognitive status: 103 had dementia. Inclusion criteria: aged 70 years or older, had undergone surgery for acute hip fracture; living in the municipality of Umeå in ordinary housing or in residential care facilities. People with cognitive impairment or dementia were included. Patients needed to be referred to the Geriatric Department for rehabilitation: by virtue of the usual clinical pathway in the hospital, all cervical (intracapsular) fracture patients were referred to the Geriatric Department but only some of those with trochanteric fractures, who were otherwise returned postoperatively to the Orthopaedic Department. Exclusion criteria: people with pathological fractures and those who fractured their hips in the hospital were excluded. Assigned: 108/101 [Multidisciplinary care / Usual rehabilitation] Baseline: 107/98 Assessed (12 months): 80/79; 95/89 (data available on complications) Loss to follow‐up: 11 (7 withdrew; 1 moved; 3 lost to follow‐up); 39 died
Interventions	Patient consent obtained after surgery. (1) Multidisciplinary home rehabilitation. “Participants in the intervention group received the same geriatric care and rehabilitation as did those in the control group, but with the aim of early discharge from the hospital and continuation of rehabilitation in their homes for a maximum of 10 weeks. Participants were discharged from hospital when no serious medical obstacle existed and when they could manage basic transfers (such as getting in and out of bed and using the bathroom), and/or when they had the help they needed at home from next of kin or social home services. The intervention began directly after discharge …. In short, the geriatric interdisciplinary home rehabilitation team, trained in comprehensive geriatric assessment, implemented the intervention with regular meetings and evaluation of participants’ individualized goals. The team focused on prevention, detection, and treatment of complications after discharge. All team members reported symptoms, such as delirium, pain, and sleeping disturbances, to the nurse and geriatrician, who assessed and treated the participants to minimize the risk of further complications. The nurse and geriatrician also evaluated pain medication use and participants’ ability to handle their medicines safely.” (2) Standard home rehabilitation. "Geriatric care and rehabilitation consisted of a multidisciplinary and multifactorial intervention programme implemented at the ward beginning in 2000 [based on intervention in Stenvall 2007a]. The staff worked in teams to conduct comprehensive geriatric assessment, with regular meetings and individual care planning." The programme included early mobilisation, participation of the whole staff in the patient’s everyday activities, specific training with the occupational therapist and physiotherapist, and thorough discharge planning. These patients were in hospital as long as necessary. Participants living in ordinary housing and in need of further rehabilitation after discharge were referred to primary health care, and 3 months after the fracture, they could also receive rehabilitation at a geriatric outpatient rehabilitation unit connected to the Geriatric Department. For participants living in residential care facilities, the physiotherapists and occupational therapists in the facilities were contacted before discharge. Both groups had comprehensive geriatric assessments, management and rehabilitation with active prevention, detection and treatment of postoperative complications such as falls, delirium, pain and infections, for example. The staff worked in teams, with regular meetings and individual care planning. Those randomised to home rehabilitation received some of these assessments and rehabilitation at home.
Outcomes	'Poor outcome': mortality and unable to walk (3 and 12 months); other data on change of residence are not available. Other outcomes: mortality (at 3 and 12 months), mobility (walking, gait speed), length of stay in acute hospital, re‐operation, readmission, number with complications (discharge to 12 months; see Notes) and delirium (discharge to 12 months); number of fallers and new fractures; information on actual treatment provided / health care utilisation.
Funding and conflict of interest statements	Funding support from the University of Umeå and County Council of Västerbotten, the Foundation of the Medical Faculty, the Swedish Dementia Association, the Swedish Research Council 2014 (grant K2014‐99X‐22610‐01‐6), and the Strategic Research Program in Care Sciences, Sweden. The authors declared no potential conflicts of interest.
Notes	This was listed as an ongoing trial in Handoll 2009. The start of the intervention and control groups are in the hospital; decisions on hospital discharge in the light of managing complications was group‐dependent. This affects the interpretation of complications data during hospitalisation. Additionally, in hospital data are likely to be a mixture of pre‐randomisation / baseline data and outcome data. Hence, we did not use these in this review.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The randomization was stratified into 2 categories according to type of housing (ordinary housing or residential care facilities) and type of fracture (cervical or trochanteric). Before arriving in the geriatric ward, the participants were randomized to either the intervention group (i.e., conventional geriatric care and rehabilitation with GIHR after discharge) or to the control group (i.e., conventional geriatric care and rehabilitation). The randomization was carried out using sequentially numbered lots in opaque, sealed envelopes drawn by a nurse at the ward, not involved in the study." Random sequence generation not described but seems very likely.
Allocation concealment (selection bias)	Low risk	"The randomization was carried out using sequentially numbered lots in opaque, sealed envelopes drawn by a nurse at the ward, not involved in the study."
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Not blinded but outcomes are less susceptible to blinding.
Blinding (performance bias and detection bias) Function, QOL	High risk	Not blinded. Outcomes at risk.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Clear participant flow diagram. Few and balanced losses.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Due mainly to mortality, 27/107 (25%) versus 19/98 (19%), there is some potential for attrition bias at 12 months.
Selective reporting (reporting bias)	High risk	Prospective trial registration document lists several outcomes that have yet to be reported, including primary outcomes; those that are not reported (yet) are listed below Primary outcomes: 3. Depression, assessed by the Geriatric Depression Scale (GDS) 5. Functional performance regarding activities of daily living (ADL), assessed by the Katz ADL index and Barthel ADL index Secondary outcomes 1. Balance, assessed by the Bergs Balance Scale and Fear of Falling 2. Quality of life, assessed by EQ‐5D 3. Nutritional status, assessed by the mini nutritional assessment (MNA) 4. Costs 5. Self‐perceived health, assessed by the 36‐item Short Form (SF‐36) health survey 6. Pain, assessed by a visual analogue scale (VAS) 7. Subjective well‐being, assessed by the Philadelphia Geriatric Morale Scale (PGCM)
Free of bias resulting major imbalances at baseline	High risk	Reasonably balanced, except the use of antidepressants (49 versus 26) and diagnosis of depression (47 versus 30) was greater in the intervention group. Notably, of the 11 on Parkinson medications at discharge, 10 were in the intervention group; important for falls results.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Potential contamination. During hospital stay, both groups were treated at the same hospital ward. The geriatricians working at the ward were occasionally responsible for both groups, including responsibility for discharge decisions.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Unclear risk	Although the follow‐up was systematic, there were potential issues as indicated: "We do not know to which extent the participants in the control group, who were living in residential care facilities, received rehabilitation after discharge. The residential care facilities have their own rehabilitation staff, and there was no access to their documentation."

Kennie 1988.

Study characteristics
Methods	Randomised using random numbers and sealed envelopes Assessor blinding: not done Length of follow‐up: 12 months
Participants	District hospital in Stirling, UK. Conducted: over an 18 month period 108 women with hip fracture. Median ages of the 2 groups: 79 and 84 years, range 65 to 94. Cognitive status: initial mental function assessed using short portable mental status questionnaire (56 with at least mild cognitive impairment; 35 had moderate or severe impairment). Inclusion criteria: women aged over 65, fractured proximal femur requiring operative fixation Exclusion criteria: not fit enough for enrolment into trial; pathological fracture; likely to be discharged within 7 days of trial entry; remaining unfit for transfer, e.g. prolonged traction Assigned: 54/54 [Multidisciplinary care / Orthopaedic care] Assessed (12 months): 53/54 Loss to follow‐up: one patient, who was known to be alive.
Interventions	Randomisation when orthopaedic surgeon judged the patient was fit enough to go to a rehabilitation ward. (1) Multidisciplinary care (general practitioner (GP), geriatrician and advice when needed from orthopaedic specialist) in orthopaedic beds at peripheral hospital. Transfer 0 to 7 days after trial entry. Day to day care by GP. Each week, geriatrician attended 2 ward rounds and one multidisciplinary team conference. Orthopaedic surgeon gave advice on demand. (2) Routine orthopaedic care in orthopaedic admission ward (a few moved to other short stay wards). Allied health care (physiotherapy, occupational therapy; orthotic and other services) received by both groups. Attempt to standardise these overall.
Outcomes	'Poor outcome': mortality or in nursing care at 1 year (or mortality or physically dependent at 1 year); mortality in hospital or discharge to institutional care Other outcomes: mortality (discharge, 1 year), functional status (mobility; physical independence), length of stay, place of residence, life satisfaction index (modified for Scottish population), carer burden (at 12 months); medical conditions (no difference post‐discharge).
Funding and conflict of interest statements	Project funded by Forth Valley Health Board and "a donation from BP Chemicals" No conflict of interest statements
Notes	A few patients from the control group (number not stated) were moved into other short stay wards at the discretion of the consultant orthopaedic surgeon, and were 'encouraged early discharge'. Data for mortality and residence at one year are taken from the Reid 1989 paper. There were still discrepancies with those provided in Burns 1992, which was also internally inconsistent and thus not included here. The data trends were the same.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Randomised: "random sequence determined before the start of the study after the method of Tukey".
Allocation concealment (selection bias)	Low risk	"The allocation was in sealed envelopes held by a departmental secretary."
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	"No attempt was made to blind either staff or patients". Assessments of independence in the activities of daily living were made jointly by an occupational therapist and a physiotherapist, and all other assessments were made by a doctor. Blinding at 1‐year assessment was not stated, but seems unlikely.
Blinding (performance bias and detection bias) Function, QOL	High risk	"No attempt was made to blind either staff or patients". Assessments of independence in the activities of daily living were made jointly by an occupational therapist and a physiotherapist. Meetings were held to ensure consistent assessment throughout the study. Blinding at 1 year assessment was not stated, but seems unlikely.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	"No patient was excluded on grounds of non‐compliance." One patient lost to follow‐up was confirmed to be alive.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	As above, but some discrepancies between the Reid and the Burns papers in the numbers followed up.
Selective reporting (reporting bias)	Unclear risk	The study protocol is not available but it is very likely that the choice of main outcomes was prespecified.
Free of bias resulting major imbalances at baseline	High risk	The treatment group were younger (median ages 79 versus 84) and had higher cognitive function (P = 0.06).
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	"the consultant geriatrician ... may have ..used community resources more fully, and been more willing than the orthopaedic surgeon to take calculated risks when discharging frail elderly patients back to the community." In the control group: "the demand for beds encouraged their early discharge". Both treatment and control groups received physiotherapy, occupational therapy, and orthotic and other services. However, the intervention group may have received more physiotherapy and nursing care. It is unclear whether there "was a greater ability of the nurses to conduct functionally orientated care when working outside the acute orthopaedic ward with its emphasis on caring for highly dependent patients".
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Follow‐up appeared active and systematic.

Marcantonio 2001.

Study characteristics
Methods	Randomised using random numbers and sealed envelopes Assessor blinding: blinded assessment by independent interviewer for delirium Length of follow‐up: 1 month
Participants	Teaching hospital in Boston, USA Conducted: not stated; first reported in May 1999 126 people with hip fracture, 21% male. Mean age 79 years. Cognitive status: 50 had mild or greater mental impairment Inclusion criteria: aged 65+ years, admitted for hip fracture surgery; resident in community or nursing home. Exclusion criteria: metastatic cancer or other comorbid illnesses likely to reduce life expectancy to less than 6 months, inability to obtain informed consent within 24 hours of surgery or 48 hours of admission, no consent by patient or designated healthcare proxy. Assigned: 62/64 [Multidisciplinary care; geriatrician consultation / Usual care] Assessed (1 month): 62/64 Loss to follow‐up: 0
Interventions	Randomisation after assessment on admission to hospital. (1) Geriatrics consultation, started preoperatively (38 participants) or within 24 hours postoperatively (24 participants), and performed daily for duration of hospitalisation. Geriatrician made targeted recommendations based on a structured protocol. This included 10 modules: adequate central nervous system oxygen delivery; fluid/electrolyte balance; treatment of severe pain; elimination of unnecessary medications; regulation of bowel/bladder function; adequate nutritional intake; early mobilisation and rehabilitation; prevention, early detection, and treatment of major postoperative complications; appropriate environmental stimuli; treatment of agitated delirium. Each module contained 2 to 5 specific recommendations. These were prioritised and limited to no more than 5 after the initial visit and 3 at follow‐up visits. (Consultants did not recommend things that the orthopaedists or nurses were already doing; recommendations were made only when something was not being done that they felt should be.) (2) Management by the orthopaedics team, including internal medicine or geriatrics consultations on a reactive rather than a proactive basis.
Outcomes	Delirium during acute hospitalisation (any, severe, days per episode, at discharge); acute hospital length of stay; discharge to institutional setting (nursing home, rehab hospital), ADL decline (reported in conference abstract in 1999).
Funding and conflict of interest statements	Partly funded by a pilot project grant from the Older Americans Independence Center P60‐AG08812‐06, and a grant from the Charles Farnsworth Trust. Dr Marcantonio was funded in part by an Academic Award from the National Institute on Aging KO8‐AG00648 and a postdoctoral fellowship from the Medical Foundation: Charles A. King Trust. No conflict of interest statements provided.
Notes	The main report indicated that the impact on postdischarge outcomes would be reported separately. In August 2019, we identified a related publication that reported outcomes at 6 months for the whole cohort but not split by treatment group (Marcantonio 2000). This reported on mortality (3 at 1 month; 15 at 6 months), ADL decline; ambulation decline; and 'poor outcome' (death or new nursing home placement: 33 at 1 month; 28 at 6 months) for the whole cohort. This trial was previously excluded in Handoll 2009. However, upon further consideration, reflecting the adjustments to scope of this update, we considered that as well as a geriatric consultation aimed at reducing delirium, the intervention contains components that can be considered a rehabilitation programme. Subgroup analysis data by pre‐fracture dementia and pre‐fracture functional impairment were reported for delirium only.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	“... subjects were randomized to proactive geriatrics consultation or usual care by opening a sealed envelope containing the randomization assignment derived from a random number table.”
Allocation concealment (selection bias)	Unclear risk	"... subjects were randomized to proactive geriatrics consultation or usual care by opening a sealed envelope containing the randomization assignment derived from a random number table." Inadequate mention of safeguards
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	“The research interviewer was trained in the use of these instruments [for assessing delirium] and conducted the assessments blinded to the intervention status of the subjects. Blinding was successfully maintained on all enrolled subjects by staggering the timing of the geriatrician and interviewer patient encounters.” Medical record, review for other outcomes “was not blinded to the intervention status of the patient but was conducted by an independent nurse reviewer who did not communicate with the research interviewer performing delirium assessments.”
Blinding (performance bias and detection bias) Function, QOL	High risk	No blinding of ADL.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	No loss to follow‐up (although short follow‐up); intention‐to‐treat analysis performed.
Incomplete outcome data (attrition bias) Function, QOL	High risk	No blinding of ADL
Selective reporting (reporting bias)	High risk	No trial registration or protocol. However, the very short follow‐up (1 month) and focus on delirium means that the study fails to include results for key outcomes that would be expected to have been reported for such a study. Small data inconsistency for people with delirium in results table (20 versus 32) and table showing subgroup analysis by pre‐fracture dementia (20 versus 30). This 'selective reporting' included omitting mortality data.
Free of bias resulting major imbalances at baseline	Unclear risk	There were slightly higher levels of pre‐fracture dementia (37% versus 51%) and ADL impairment (19% versus 31%) in the usual care group (both P values > 0.1).
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Insufficient information provided for usual care group.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Although limited, follow‐up was active and systematic.

Naglie 2002.

Study characteristics
Methods	Randomised trial: using computer‐generated randomisation (block size of 4), stratified by age and place of residence, sequentially numbered, sealed, opaque envelopes. Assessor blinding: yes Length of follow‐up: 6 months
Participants	Teaching hospital in Toronto, Canada. Conducted: June 1993 to March 1997. 280 people with hip fracture. Of 279: 20% male. Mean age 84 years. Cognitive status: 74 had mild to moderate mental impairment; unknown numbers with no impairment or severe impairment. Inclusion criteria: aged 70+ years, had undergone surgery for hip fracture; resident in community or nursing home. Exclusion criteria: fracture occurred in acute care setting or hospital, pathological fracture, multiple trauma, previous surgery of fractured hip, expected survival < 6 months, nursing home and dependent for ambulation before fracture, technical failure of surgery, no bed available on interdisciplinary unit. Assigned: 141/139 [Multidisciplinary care / Usual care] Assessed (6 months): 141/138 Loss to follow‐up: one withdrawal from usual care group
Interventions	Randomisation took place in post‐anaesthetic care unit when the patient was considered stable enough for transfer to a ward. (1) Interdisciplinary care consisting of routine postoperative surgical care plus daily medical care by internist‐geriatrician, and regular care by physiotherapist, occupational therapist, social worker and clinical nurse specialist. Twice weekly interdisciplinary rounds to set goals and monitor patients' progress. (Emphasis on preventing medical problems, such as pressure sores; early mobilisation; early participation in self‐care; individualised discharge planning; priority assessment & specialised training.) (2) Usual care on orthopaedic units (same hospital but different care teams) which included access to geriatric consultation and allied health care professionals if requested; access to occupational therapists and clinical nurse specialists was limited. There was no specialist follow‐up care.
Outcomes	'Poor outcome': mortality at discharge or transfer to more dependent residence (e.g. nursing home or rehabilitation hospital); mortality at 6 months or living in more dependent residence (nursing home etc.); mortality or decline in ambulation; or mortality or decline in ambulation or chair or bed transfers and place of residence. Other outcomes: mortality (at discharge, 3 and 6 months), ADL (Barthel Index), mobility (transfers and walking), length of stay, days in hospitals over 6 months, change in residence (indicating deterioration), health care utilisation.
Funding and conflict of interest statements	Funded by grants from the Ontario Ministry of Health, Canada and, towards data analysis, from the Research Institute of Queen Elizabeth Hospital, Toronto, Canada Explicit statement of no conflicts of interest
Notes	The discrepancies between the full report of the trial, published in 2002, and an abstract (1999) in the numbers randomised resulted from the exclusion of 18 participants in the abstract report. These comprised one patient who withdrew and 17 who were deemed ineligible after randomisation. The latter 17 were included in an intention‐to‐treat analysis in the full report.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Stratified, computer‐generated randomization scheme with a block size of 4." Stratification by age (70 to 79; 80+ years) and place of residence at admission (private residence; retirement home; nursing home).
Allocation concealment (selection bias)	Low risk	"Orthopaedic residents, who were blinded to block size, assigned the patients to treatment group according to sequentially numbered, sealed opaque envelopes that were colour coded by stratum."
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	"All follow‐up assessments were conducted by research assistants blinded to group assignments."
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	"All follow‐up assessments were conducted by research assistants blinded to group assignments." Providers and participants were not blinded.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Intention‐to‐treat analysis. Data missing only for one participant who withdrew from the trial. Flow diagram presented. (At review, 17 patients were deemed ineligible (8 versus 9) but they were included in the full report.)
Incomplete outcome data (attrition bias) Function, QOL	Low risk	As above.
Selective reporting (reporting bias)	Unclear risk	The study protocol is not available but it is very likely that the choice of main outcomes was prespecified.
Free of bias resulting major imbalances at baseline	Low risk	None of the differences between treatment groups was statistically significant.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	"To minimise the risk of contamination bias, interdisciplinary care and usual care were provided by different staff on different wards." Insufficient information ‐ there was no specialist follow‐up care.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Systematic and active follow‐up.

Prestmo 2015.

Study characteristics
Methods	Randomised trial: using a web‐based computer‐generated randomisation service Assessor blinding: limited to data collection of medical record information Length of follow‐up: 12 months (after surgery)
Participants	Teaching hospital in Norway Conducted: April 2008 to December 2010 397 people with hip fracture. 26% male. Mean age 83 years. Cognitive status: 53 had dementia (previous diagnosis). Inclusion criteria: home‐dwelling (home, sheltered housing and related institution), aged 70 years or over, admitted with acute hip fracture, able to walk 10 m before fracture surgery, written consent by patient or next of kin. Lived in nearby municipalities. Exclusion criteria: pathological fractures, multiple trauma, short life expectancy (expected to die within 3 months), living permanently in nursing home. Assigned: 198/199 [Comprehensive geriatric care / Usual care] Assessed (12 months): 162/149 (remaining in trial) Loss to follow‐up: 19 withdrew, 67 deceased (2 had completed follow‐up)
Interventions	Randomisation at the emergency department and participants transferred to the allocated wards, with pre‐ and post‐operative care provided by separate teams, directly after treatment allocation. (1) Comprehensive geriatric care in a dedicated geriatric ward. Assessment by orthopaedic surgeon by request only. Pre‐ and postoperative treatment in an orthogeriatric unit (geriatric evaluation and management). Structured, systematic interdisciplinary comprehensive geriatric assessment and care focusing on: somatic health (comorbidity management, review of drug regimens, pain, nutrition, elimination, hydration, osteoporosis, and prevention of falls); mental health (depression, delirium); function (mobility, p‐ADL (personal activities) and i‐ADL (instrumental)) and social situation. Early discharge planning. Early mobilisation and initiation of rehabilitation. (2) Usual orthopaedic care provided in an orthopaedic ward. Assessment by geriatrician by request only. Treatment on ordinary orthopaedic unit (usual orthopaedic care). Participants in both groups "received care and physiotherapy in accordance with national and international guidelines". After hospital discharge, responsibility passed to primary health‐care services. No routine hospital‐based follow‐up after discharge; when needed, the orthopaedic surgeon arranged follow‐up at the orthopaedic outpatient clinic for participants in both groups.
Outcomes	'Poor outcome': mortality or transferred to permanent nursing home but alive at 12 months (data provided by authors) Other outcomes: mortality (at 1, 4 and 12 months), ADL (Barthel Index and Nottingham Extended ADL Scale), mobility (Short Physical Performance Battery (SPPB), Timed Up and Go (TUG)), cognition (Clinical Dementia Rating Scale, Mini Mental Status Examination), quality of life (EQ‐5D‐3L), fear of falling (Falls Efficacy Scale International‐short form), mood (Geriatric Depression Scale), length of hospital stay, discharged directly home, living at home, rehabilitation, readmission, short stay in nursing home, permanent stay in nursing home, health care use (costs per patient)
Funding and conflict of interest statements	Funded by the Norwegian Research Council, the Central Norway Health Authority, the St Olav Hospital Trust, the Department of Neuroscience at the Norwegian University of Science and Technology, the Foundation for Scientific and Industrial Research at the Norwegian Institute of Technology (SINTEF) and St Olav Hospital Fund for Research and Innovation, and the Municipality of Trondheim, Norway. Authors declared they had no competing interests.
Notes	Request for numbers of participants admitted permanently to a nursing home by 1 year follow‐up and numbers either dead or in a nursing home at 1 year (primary outcome of review) sent to Ingvild Saltvedt on 8 August 2017. Response 8 August 2017 ‐ passed on to colleague Vidar Halsteinli ‐ with data supplied 16 August 2017. Not used in this review are data for continuous outcomes (mobility, ADL, and cognition) from subgroup analysis by gender, age (cut‐off 80 years), and pre‐fracture instrumental‐ADL (cut‐off 45 on the Nottingham scale); reported in Prestmo 2016. Also, 'free‐living physical behaviour' (time spent in upright activities) data, reported in Taraldsen 2014 and Taraldsen 2015; drug prescriptions (at admission and discharge) as reported in Heltne 2017; and gait characteristics as reported in Thingstad 2016. However, independence in mobility data were used from the lattermost article.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The randomisation sequence was computer‐generated in blocks of a size unknown to the investigators. We used a web‐based, computer‐generated service prepared by the Unit of Applied Clinical Research, Norwegian University of Science and Technology (NTNU)."
Allocation concealment (selection bias)	Low risk	"Provided eligibility criteria were confirmed, patients were enrolled and randomly assigned in a 1:1 ratio by a nurse to either the orthopaedic ward for orthopaedic care or the geriatric ward for comprehensive geriatric care. Patients were transferred to the allocated wards directly from the emergency department after treatment allocation."
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	"Medical information, including complications, admissions to hospital, and visits to outpatient clinics was obtained from hospital records. Data collection for discharge destination, place of residence, and use of health‐care services was undertaken with the group allocation concealed."
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	Not blinded for these outcomes: “Masking of the patients and staff delivering the treatment was not possible, and we were only partly able to accomplish masking of assessors during follow‐up” "Whenever possible during data collection, patients were the primary informant."
Incomplete outcome data (attrition bias) Death, residence, readmission	Unclear risk	Participant flow diagram provided. Although mortality fully accounted for, the presentation of nursing home admission and permanent stay and hospital readmission results are split by follow‐up. This means that the longer‐term data cannot be used because of double counting. The consequences are unclear in terms of bias.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Participant flow diagram provided. Difference in loss to follow‐up between multidisciplinary care and orthopaedic care groups: e.g. at 12 months for Barthel Index: 30 (15%) versus 56 (28%) could affect outcome for these.
Selective reporting (reporting bias)	Low risk	Trial registration and protocol available. All outcome measures described in protocol reported on across multiple publications. (Limitation of cost‐effectiveness analysis is absence of baseline EQ‐5D‐3L measurements, making it impossible to control for potential imbalances in baseline values.)
Free of bias resulting major imbalances at baseline	Low risk	Baseline characteristics did not differ between the groups
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Elements of individualised rehabilitation programmes and primary healthcare follow‐up not standardised
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Systematic and active follow‐up as per protocol

Ryan 2006.

Study characteristics
Methods	Randomised trial: random numbers table, sealed opaque envelopes Assessor blinding: yes Length of follow‐up: 12 months
Participants	In the community, Sheffield, UK Conducted: July 2000 to June 2002 71 people with hip fracture. 24% male. Mean age 81 years. Cognitive status: none had dementia Inclusion criteria: aged 65 years or over recovering from (stroke or) hip fracture, recently discharged from hospital to their own home. Contact within 5 days of notification of patient eligibility. Exclusion criteria: concomitant disease (e.g. Parkinson’s disease or dementia). Assigned: 37/34 [intensive / less intensive] Assessed (12 months): 30/28 Loss to follow‐up = 7 versus 6 (withdrew, deceased, unavailable, in hospital)
Interventions	Randomisation within 5 working days of referral to community rehabilitation team. Maximum duration of treatment period was 12 weeks. All treatment (and assessment) undertaken in patient's own home. (1) Intensive: 6 or more face‐to‐face contacts per week from members of a multidisciplinary rehabilitation team (2) Less intensive: 3 or less face‐to‐face contacts per week from members of a multidisciplinary rehabilitation team The service (both augmented and routine) was provided to both groups of patients by a local multidisciplinary team (physiotherapist, occupational therapist, speech and language therapist or therapy assistant). There were 4 teams, divided by geographical area, but belonging to the same city‐wide service managed by the same Primary Care Trust.
Outcomes	'Poor outcome': mortality at 12 months and admission to institutional care Other outcomes: mortality (3 and 12 months), function (Barthel Index; Frenchay activities index), admission to institutional care, days spent in hospital (*stroke and hip fracture), use of home care services readmissions (*), number of contacts during treatment period (*), EQ‐5D, therapy outcome measure (no data).
Funding and conflict of interest statements	Funded by NHS Executive Trent, UK There were no conflict of interest statements
Notes	Some data for hip fracture patients received from Dr Ryan on 5 February 2009: gender and co‐residing carer; death and institutional care at 3 and 12 months follow‐up.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomization was carried out using a random number table (blocks of 10) and sealed opaque envelopes. Randomization was stratified on the basis of the diagnosis of hip fracture or stroke and locality team."
Allocation concealment (selection bias)	Low risk	"sealed opaque envelopes. .... An administrator based within each of the teams carried out the allocation procedure. The result was recorded on each patient file .. and was immediately communicated to the research team."
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	"Blinded follow‐up assessments were carried out"
Blinding (performance bias and detection bias) Function, QOL	Low risk	"Blinded follow‐up assessments were carried out". "Blinding was achieved by follow‐up being undertaken by trained therapy assistants who were not involved in the provision of service to participating patients and who worked in another part of the city to that of the patient."
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Participant flow diagram provided.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	"A method of imputation, using the median for each group, was used for missing data."
Selective reporting (reporting bias)	Unclear risk	The study protocol is not available but it is very likely that the published report includes all expected outcomes, including those that were prespecified.
Free of bias resulting major imbalances at baseline	High risk	There were more males (12/37 versus 5/34) and significantly more patients with a residing carer in the intensive group.
Free of performance bias (from non‐trial differences in care provision)?	High risk	The same multidisciplinary team provided care to both groups. In this pragmatic trial, therapists were requested to provide a more intensive service to those allocated to the intervention group. However, what occurred within those additional face‐to‐face contacts was not prescribed, neither was it known.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Active and systematic follow‐up

Sanchez Ferrin 1999.

Study characteristics
Methods	Randomised using random number generator Assessor blinding: yes, blinded nurse at admission and discharge Length of follow‐up: 6 months
Participants	Teaching hospital in Barcelona, Spain. Conducted: April 1996 to June 1997 208 people with hip fracture. Of 206: 24% male. Mean age 82 years. Cognitive status: 61 had dementia (previous diagnosis). Inclusion criteria: patients with hip fracture admitted to hospital aged over 64 years; all but 2 had surgery. Study report clarified that patients with previous physical or psychiatric disabilities were not excluded. Exclusion criteria: patients in an unstable condition, usually from severe trauma, who were admitted to the intensive care unit Assigned: 104/104 [Geriatric unit intervention / Usual care] (2 of these, 1 in each group, who were transferred to another hospital immediately after surgery were excluded.) Assessed (6 months): 81/75 (derived from percentages for admission to institutional care) Loss to follow‐up = 23 versus 29 (death 47, exclusion 2, probably missing 3)
Interventions	Interventions started soon after surgery: mean 3.9 days (SD 10 days) after fracture admission. (1) Evaluation by the Functional Geriatric Unit (FGU) during hospitalisation. Initial basic geriatric assessment by nurse from the FGU within 48 hours of admission. Results of evaluation discussed during geriatric team meeting, where decisions were made about treatment and need for consultation from other professions in the team, e.g. geriatrician, social worker, physiotherapist, psychologist. It was the responsibility of the nurse and doctor to ensure the treatment and management recommendations were implemented, and to review results of investigations and specialist consultations sought at time of first visit. Note: medical geriatric review was not continuous – it occurred on a single session. (2) Usual care: under care of the Orthopaedics and Traumatic Surgery Service (OTSS), with consultations to other specialists as required. The OTSS team doctor decided which patients from each group needed referral to the Rehabilitation Service during their hospital stay. The geriatrician was also able to refer patients to this service. At discharge, appointments for outpatient physiotherapy were organised and made by the Rehabilitation service. All patients requiring post‐hospital care were discharged to the area Community Health Centre, which required assessment by the Functional Geriatric Unit prior to admission. Intervention and control group patients shared rooms and ward staff.
Outcomes	'Poor outcome': mortality at discharge or discharge to institutional care; mortality and institutional care admission at 6 months Other outcomes: mortality, ADL, mobility (goes out into street), complications (delirium; respiratory infection; urinary infection; dehydration; pressure sores; cardiac insufficiency; local complications), length of stay, discharge location, readmission, institutional care; specialist consultations
Funding and conflict of interest statements	No mention of funding. No conflict of interest statements provided.
Notes	This trial was excluded in Handoll 2009 as it was deemed as geriatric assessment, not multidisciplinary rehabilitation. However, the reduced level of geriatrician involvement was considered appropriate for inclusion in line with our revised inclusion criteria.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Translation: "Patient allocation to either group was carried out with the aid of a computer programme that generated random numbers."
Allocation concealment (selection bias)	Unclear risk	Translation: "Patient allocation to either group was carried out with the aid of a computer programme that generated random numbers." No details of safeguards.
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Translation: “A nurse from outside the Unit and blinded to the patient’s group allocation, interviewed all patients at the time of admission and at discharge.” This included collection of residential status. Other outcomes also less susceptible.
Blinding (performance bias and detection bias) Function, QOL	Low risk	Translation: “A nurse from outside the Unit and blinded to the patient’s group allocation, interviewed all patients at the time of admission and at discharge.”
Incomplete outcome data (attrition bias) Death, residence, readmission	Unclear risk	Two patients who were transferred to another hospital immediately after surgery were excluded; this is unlikely to be a source of bias. Participant flow and losses to follow‐up are not reported. There are small numbers missing upon calculating denominators from percentages.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Two patients who were transferred to another hospital immediately after surgery were excluded; this is unlikely to be a source of bias. Participant flow and losses to follow‐up are not reported. There are small numbers missing upon calculating denominators from percentages.
Selective reporting (reporting bias)	Unclear risk	Neither a published protocol nor trial registration document were available. The outcomes mentioned in the methods are presented in the results but selective reporting cannot be ruled out.
Free of bias resulting major imbalances at baseline	Unclear risk	Groups were comparable in most ways except proportionately more participants in the intervention group had dementia (37 (35.9%) versus 24 (23.3%)); probably also reflected in slightly higher numbers in residential accommodation (22 versus 15).
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	No information available. Staff appear to have been in common, which may reduce performance bias but also may increase the risk of ‘contamination’. Translation: “it was not possible to isolate the control group patients, given they shared rooms with the intervention group patients and were cared for by the same ward staff as the intervention group patients.” "11 control group patients were, in fact, assessed by the Unit nurse in order to effect an early discharge".
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	None apparent

Shyu 2008.

Study characteristics
Methods	Randomised trial: coin tossed by a third party Assessor blinding: partly: independent data collectors. (Also, participants only approved data collection rather than trial inclusion.) Length of follow‐up: 12 months.
Participants	Teaching hospital in Taiwan Conducted: September 2001 to November 2003 162 people with hip fracture. 31.5% male. Mean age 78 years. Cognitive status: About a third of participants (51/160: 32%) had mild cognitive impairment; severe impairment not included. Inclusion criteria: aged 60 years or over, admitted for acute hip fracture surgery, unilateral hip fracture, resided in Northern Taiwan, pre‐fracture score Chinese Barthel index > 70; able to perform full range of motion against gravity and resistance. Approval sought for data collection. Exclusion criteria: severe cognitive impairment; completely unable to follow instructions (< 10 on Chinese mini‐mental state examination); or terminally ill. Assigned: 80/82 [Multidisciplinary care / Usual care] Assessed (12 months): 60/62 Loss to follow‐up (patient withdrew) = 16 versus 14
Interventions	Randomisation took place soon after admission to emergency department. (1) Interdisciplinary programme of geriatric consultation, continuous rehabilitation and discharge planning. Geriatrician and geriatric nurses provided geriatric assessment/consultation; physiotherapist, geriatric nurses, and rehabilitation physician were responsible for rehabilitation programme; geriatric nurse co‐ordinated discharge planning service. Early mobilisation, home visit and follow‐up services provided. In addition to routine care, 1 physical therapy session a day from geriatric nurse (total 4 times, ~ 30 minutes each), 2 assessments by a physical therapist (each 20 minutes), one visit from rehabilitation physician (20 minutes). After hospital discharge: 4 home visits during first month (once a week for ~ 30 minutes) and 4 home visits during second and third month (~30 minutes each) from a geriatric nurse. (2) Usual care on trauma or orthopaedic ward. Some consultations with other disciplines (e.g. internal medicine) occasionally made depending on patient's condition. Exercises taught by nurses in first 2 to 3 days. Number of physical therapy sessions varied because of insurance policy: 18 received 3 sessions, the rest only 1 session. Discharged approximately 7 days from surgery with no care provision subsequently.
Outcomes	'Poor outcome': mortality at 12 months and admission to institutional care; mortality + functional deterioration at 12 months (non‐recovery of walking function). Other outcomes: mortality (1, 3, 6 and 12 months), function (mobility), function (modified Barthel Index), quality of life (SF‐36 Taiwan version), depressive symptoms, Chinese Barthel Index, admission to institutional care, length of hospital stay, readmissions (emergency room visit), falls.
Funding and conflict of interest statements	Financial support from the National Health Research Insitute, Taiwan, and Ghang Gung Memorial Hospital, Taiwan Authors declared they had no competing interests in a secondary report.
Notes	Confirmation that the study populations of the 2005 and 2008 reports were the same was received from Dr Shyu. Additional data for 12 months follow‐up also received on 9 December 2008. Retrospective trial registration and another report (link in trial registration document) of trial identified in search update January‐February 2016. The two previous reports (2005 and 2008) are referred to with regard to the development of the rehabilitation programme in the 2010 report but not in relation to trial results (Shyu 2010). Shyu 2010 reported on quality of life results. Also, 3 additional references that consider cognitive impairment: Shyu 2012a; Shyu 2012b; Shyu 2013b, providing follow‐up to 24 months. However, these post‐hoc investigations have data inconsistencies. For example, in the numbers of deaths at 12 months (4 versus 6, instead of 7 versus 7) differ in comparison with the main report (2008); and the numbers with cognitive impairment (severe cognitive impairment was an excluded criterion), which include those with delirium, differ between Shyu 2012a (27 versus 24) and the other two articles (30 versus 29). We therefore decided that using these data would not be appropriate for subgrouping by 'dementia' (such data were not collected by Shyu et al, who confirmed that "dementia and delirium were not differentiated" (Shyu 2013b). Likewise, we did not include two‐year follow‐up data, reported in two reports (2010 and 2011), and note that retrospective trial registration was for one‐year follow‐up. As pointed out in the trial report: "Taiwan differs substantially from Western developed countries in its healthcare system, clinical practice, case mix, culture, and social organisation. For example, anthropometric measures differ between ethnic groups. Unlike elderly patients in Western countries, the majority of older people on Taiwan continue to reside with their adult married children. Taiwan’s healthcare system also differs from those in most Western countries because of Taiwan’s national health insurance, the ability of its citizens to choose and visits in medical centers without a referral, and the underdevelopment of community‐based, long‐term‐care services.”
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"randomly assigned to an experimental or control group by the flip of a coin"
Allocation concealment (selection bias)	Unclear risk	"a neutral third party not involved in delivering the intervention or assessing outcomes flipped the coin". Adherence probable but not confirmed.
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	Insufficient information but data collection was probably independent.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	"participants were blinded but evaluators were not". However, different nurses (data collectors) from those who delivered the intervention conducted the follow‐up assessments at participants' homes.
Incomplete outcome data (attrition bias) Death, residence, readmission	Unclear risk	Participant flow obtained. Some discrepancies in data (e.g. mortality) (between papers and data from author) although not an imbalance in numbers of participants who withdrew.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	As above.
Selective reporting (reporting bias)	High risk	The study protocol is not available. The appearance of retrospective trial registration and a report of the SF‐36 results undermined our previous judgement that all outcomes reported in the methods were collected. We thus changed this to a high risk of selective reporting bias in this update.
Free of bias resulting major imbalances at baseline	Low risk	No serious imbalance.
Free of performance bias (from non‐trial differences in care provision)?	Low risk	Insurance policy determined number of physiotherapy sessions in control group. However, the difference was between one or three sessions. Notably, the active follow‐up of the control group can be considered additional to usual care.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Sufficiently similar and independent follow‐up procedures

Shyu 2013a.

Study characteristics
Methods	Randomised trial: dice throw by a third party Assessor blinding: partial, as participants were unaware of the interventions being tested Length of follow‐up: 12 months
Participants	Teaching hospital in Taiwan Conducted: September 2005 to July 2010 299 people with hip fracture. 36.1% male. Mean age 76.5 years. Cognitive status: severe cognitive impairment not included. Inclusion criteria: aged 60 years or over, admitted for accidental first‐time unilateral hip fracture, received hip arthroplasty or internal fixation, able to perform full range of motion against gravity and against at least some resistance of the unaffected limb at admission, living in Northern Taiwan, pre‐fracture score Chinese Barthel index > 70. Exclusion criteria: severe cognitive impairment and completely unable to follow instructions (< 10 on Chinese mini mental state examination); terminally ill; unable to communicate; admitted from a nursing home. Assigned: 99/101/99 [Comprehensive care / Interdisciplinary care / Usual care] Assessed (12 months): 92/92/85 Loss to follow‐up (death or patient withdrew: "refused to participate") = 7 versus 9 versus 14
Interventions	Randomisation took place soon after admission to emergency department. (1) Comprehensive care. This included interdisciplinary care (described below) supplemented with health‐maintenance interventions to manage depressive symptoms and malnutrition and prevent falls. Additionally, participants were scheduled to receive one nursing visit per month from 4 to 6 months following discharge and once every 2 months from 6 to 12 months following discharge, and one physical therapist visit at 6 months following discharge. (2) Interdisciplinary programme of geriatric consultation, continuous rehabilitation and discharge planning with post‐hospital services. Geriatrician and geriatric nurses provided geriatric assessment/consultation; physiotherapist, geriatric nurses, and rehabilitation physician were responsible for rehabilitation programme; geriatric nurse co‐ordinated discharge planning service. Early mobilisation, home visit and follow‐up services provided. In addition to routine care, 1 physical therapy session a day from geriatric nurse (total 4 times, ~ 30 minutes each), 2 assessments by a physical therapist (each 20 minutes), one visit from rehabilitation physician (20 minutes). After hospital discharge: 4 home visits during first month (once a week for ~ 30 minutes) and 4 home visits during second and third month (~30 minutes each) from a geriatric nurse. Additionally, one visit from a physical therapist within the first week, third week, and third month after discharge. (3) Usual care by orthopaedists. Some consultations for internal medicine care occasionally made depending on patient's condition. Exercises taught by nurses in first 2 to 3 days. In‐hospital physiotherapy on 3rd day with number of sessions depending on patient's condition (actual mean 1.82 sessions). Discharge between 5 to 7 days from surgery with health education provided by nurses. No home rehabilitation. Clinical follow‐up at 1, 3 and 6 months after discharge.
Outcomes	'Poor outcome': mortality at 12 months and admission to institutional care. Other outcomes: mortality (1, 3, 6 and 12 months), recovery to self‐care ability (incomplete data), cognitive impairment during hospitalisation, quality of life (SF‐36 Taiwan version), risk for depression, risk for malnutrition, admission to institutional care, length of hospital stay (incomplete data), readmissions (emergency room visits; cumulative data over 12 months not available; just used those occurring within one month), falls.
Funding and conflict of interest statements	Financial support from the National Health Research Insitute, Taiwan (grant number: NHRI‐EX98‐9404PI) and Ghang Gung Memorial Hospital, Taiwan In a secondary report, authors declare they have no competing interests.
Notes	Orthopaedic care. Consultations for internal medicine care occasionally made, depending on the patient’s condition. Physical therapy usually started on the second or third day without any in‐home rehabilitation provided or telephone follow‐ups. Patients encouraged to ambulate with protected weight‐bearing for 3 months. Sent request for nursing home data and 'poor outcome' to Shyu on 26/04/16. Response with data received on 04/05/16. No additional outcome data were collected from publications reporting secondary analyses: a) relating to nutritional status (Liu 2014a) or the mental health of family carers (Liu 2015). As emphasised in Shyu 2008: "Taiwan differs substantially from Western developed countries in its healthcare system, clinical practice, case mix, culture, and social organisation." Expanded in Notes of Shyu 2008.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The randomisation sequence was generated by a throw of dice by a neutral person not involved in delivering interventions and assessing outcomes."
Allocation concealment (selection bias)	Unclear risk	No details of safeguards
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	"The participants only knew that they were in a study but were blinded to which intervention group they were in." However, as acknowledged, assessors were not blinded.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	"The participants only knew that they were in a study but were blinded to which intervention group they were in." However, as acknowledged, assessors were not blinded.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Twice as many participants in the usual care group refused follow‐up at 12 months (4 versus 4 versus 9). However, these were still low numbers and the authors declared that: "No significant differences were found between those who completed all follow ups (n = 269) and those who did not (n = 30)." The latter included 13 deaths.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Although the above applies, the actual denominators for quality of life were not explicit and there were discrepancies in the table for recovery to independent self‐care ability, with the percentages not corresponding exactly to the expected numbers of participants.
Selective reporting (reporting bias)	High risk	The retrospective trial registration (in 2011) made no mention of recording SF‐36 data; these were reported in a separate paper. Thus, we think there is a risk of selective reporting bias.
Free of bias resulting major imbalances at baseline	Unclear risk	The usual care group had more participants (P < 0.05) who were illiterate, more who had intertrochanteric fracture. The effect on outcome is not clear.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Insufficient information to judge this.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Sufficiently similar and independent follow‐up procedures

Singh 2012.

Study characteristics
Methods	Randomised trial: using computer‐generated randomisation and sequentially numbered, sealed, opaque envelopes; stratified by gender. Assessor blinding: yes for disability outcomes Length of follow‐up: 12 months (after fracture)
Participants	Recruitment from one teaching hospital and surrounding geriatric and rehabilitation hospitals in New South Wales, Australia. Conducted: February 2003 to April 2007. 124 people with hip fracture, 31% male. Mean age 79.3 years. Cognitive status: all had sufficient cognitive ability to understand informed consent process; 5% (6) had dementia. Inclusion criteria: aged 55+ years, had surgical repair of hip fracture, had sufficient cognitive ability and English‐language skills to understand informed consent process. Exclusion criteria: terminal illness, pathological fracture, no surgical repair, or geographical distance precluding participation. Assigned: 62/62 [Extended care / Usual care] Assessed (12 months): 62/62 for death & nursing care; 49/50 for 'other' outcomes (however, data were imputed) Loss to follow‐up: 13 (9 versus 4) lost to follow‐up for other outcomes and 12 (4 versus 8) died
Interventions	Randomisation took place in acute care, after surgery. (1) Extended multidisciplinary treatment of frailty. Twelve months of geriatrician‐supervised high‐intensity weight‐lifting exercise and targeted treatment of balance, osteoporosis, nutrition, vitamin D/calcium, depression, cognition, vision, home safety, polypharmacy, hip protectors, self‐efficacy, and social support. Participants were prescribed high‐intensity progressive resistance training (80% of peak upper and lower body muscle strength) supervised by research staff in the outpatient clinic of the hospital 2 days per week for 12 months as the core treatment. Weight lifting began after standard physiotherapy ceased, approximately 6 to 8 weeks after fracture. Other intervention arms began after assessment in hospital or at home. All interventions were coordinated by the research staff via weekly interdisciplinary team meetings. All received a monthly phone call and a monthly residential visit by their trainer. Thus, total contacts prescribed in addition to usual care averaged 80 supervised exercise training sessions, 10 home visits, and 10 phone calls over 12 months. (2) Standard care (provided to all trial participants) as offered for hip fracture in the area health service, including orthogeriatric care, rehabilitation service, other medical and allied health consultation as required, and physiotherapy.
Outcomes	'Poor outcome': mortality at 12 months or nursing home admissions Other outcomes: mortality (4 and 12 months), nursing home admission (at any time point: hospital discharge; rehabilitation discharge; 4 or 12 months after hip fracture). Others reported for 12 months: ADL (Katz ADL, total and subcategories); disability (National Health and Nutrition Examination Survey (NHANES) score; Modified Barthel Index); dependency (Functional Independence Measure (FIM); ALSAR (Assessment of Living Skills and Resources); use of assistive devices
Funding and conflict of interest statements	Funded by the Australian National Health and Medical Research Council (project grant 211226) No conflict of interest statements except declared independence from funders and sponsors
Notes	The target sample size was 284 participants. However: "The final sample size (n = 126) was smaller than planned because of significant funding reductions and fewer numbers of hip fractures than projected by the area health service report during the years of enrollment."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Participants were individually randomized to 1 of 2 parallel groups in a 1:1 ratio, stratified by gender via computer‐generated randomly permuted blocks (available at www.randomization.com)."
Allocation concealment (selection bias)	Low risk	"Masked randomization assignments were generated by an offsite investigator and sent electronically to research assistants, who then distributed the written treatment assignments to participants in sequentially numbered opaque sealed envelopes within 1 to 2 days of consent."
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Aside from function: "All other outcomes (....) were collected by unblinded research assistants." However, data collection unlikely to be affected for these outcomes.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	"The postfracture disability outcomes (FIM and ALSAR scores) were collected by an otherwise uninvolved blinded assessor after discharge." However, participants were not blinded.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	"Mortality and nursing home status were obtained in 100% of the cohort."
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	Although similar losses to follow‐up for function and related outcomes in both groups (21% versus 19%), the effect on the results of these losses is unclear and could have differed between the two groups.
Selective reporting (reporting bias)	High risk	Prospective title registration. However, 4‐month results (while measured) not reported and other outcomes on nutritional status, muscle strength, gait and balance, and depression not reported.
Free of bias resulting major imbalances at baseline	Low risk	Baseline characteristics were adequately balanced between the two groups.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Insufficient information to judge this.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	The same follow‐ups applied to the two groups.

Stenvall 2007a.

Study characteristics
Methods	Randomised trial: sequentially numbered, opaque sealed envelopes Assessor blinding: no, but independent data collectors Length of follow‐up: 12 months.
Participants	Teaching hospital in Sweden Conducted: May 2000 to December 2002 199 people with hip fracture. 26% male. Mean age 82 years. Cognitive status: 64 had dementia. Inclusion criteria: aged 70 years or over, admitted for acute hip fracture surgery, oral and written consent by patient or carer. (All had intracapsular fractures, clarified in Karlsson 2016.) Exclusion criteria: severe rheumatoid arthritis or hip osteoarthritis; pathological fracture; severe renal failure; bedridden; fracture sustained in hospital. Assigned: 102/97 [Multidisciplinary care / Usual care] Assessed (12 months): 84/76 (survivors) Loss to follow‐up (patient withdrew) = 2 versus 2
Interventions	Randomisation envelope received in the emergency room but envelope not opened until immediately before surgery. (1) Geriatric unit specialising in geriatric orthopaedic care. Multidisciplinary team providing comprehensive geriatric assessments and rehabilitation. Active prevention, detection and treatment of post‐operative complications, such as falls, delirium, pain, pressure sores and malnutrition. Early mobilisation (within 24 hours), with daily training, provided by physiotherapists (2 full time), occupational therapists (2 full time) and care staff, including dietician (0.2) during hospital stay. Staffing ratio 1.07 nurses or aides per bed. A geriatric team, including a physician, assessed patients at 4 months, including for further rehabilitation needs. (2) Usual care on specialist orthopaedic ward. Care included that by physiotherapists (2 full time), occupational therapist (0.5 full time) and care staff, not including dietician, during hospital stay. Staffing ratio 1.01 nurses or aides per bed in orthopaedic ward; 1.07 in geriatric control ward used for longer stays. Included early mobilisation (within 24 hours). No 4‐month assessment.
Outcomes	'Poor outcome': mortality at discharge and not discharged to previous residence; mortality + not in same residence at 12 months; or mortality and reduced activity of daily living. Other outcomes: mortality (in hospital; 4 and 12 months), function (mobility), function (Katz ADL), independent living, complications including falls, length of hospital stay (initial episode; over 1 year), return to previous residence, readmissions, falls.
Funding and conflict of interest statements	The study was supported by the “Vårdal Foundation”, the Joint Committee of the Northern Health Region of Sweden (Visare Norr), the JC Kempe Memorial Foundation, the Dementia Fund, the Foundation of the Medical Faculty, the Borgerskapet of Umeå Research Foundation, the Erik and Anne‐Marie Detlof’s Foundation, University of Umeå and the County Council of Västerbotten (“Dagmar”, “FoU”, and “Äldre Centrum Västerbotten”) and the Swedish Research Council, grants K2002‐27VP‐14165‐02B, K2002‐27VX‐14172‐02B, K2005‐27VX‐15357‐01A. The authors declared no financial conflicts of interest in one paper.
Notes	Separate publications produced by trial group reporting outcomes (delirium, nutritional status, and falls; Berggren 2008; Lundstrom 2007; Olofsson 2007; Stenvall 2007b) in association with selected components of the intervention but set in the context of the whole programme, are not considered in this review. Other reports published after 2009 include a cohort analysis of the comorbities and causes of deaths at three‐year follow‐up and a subgroup analysis of the 64 participants (32%) with dementia (Stenvall 2012).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Mention of 'lots' in the envelopes; stratification by methods of surgery.
Allocation concealment (selection bias)	Low risk	Randomised "using opaque sealed envelopes". "The lots in the envelopes were numbered sequentially." "The selection procedures were carried out by people not involved in the study."
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	Systematic data collection. "A geriatrician, who was unaware of the study‐group allocation, analysed all assessments and documentations after the study was finished."
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	Not blinded but independent assessors.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Participant flow diagram provided.
Incomplete outcome data (attrition bias) Function, QOL	Low risk	Data provided.
Selective reporting (reporting bias)	Unclear risk	The main report of the trial that covers the whole intervention is probably protocol driven. This doesn't apply to the subsidiary publications; see Notes.
Free of bias resulting major imbalances at baseline	High risk	Significant differences between the two groups in the numbers of patients with diagnosed depression (33 versus 45) and on antidepressants (29 versus 45).
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Outpatient rehabilitation was not standardised.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	No apparent problems

Swanson 1998.

Study characteristics
Methods	Randomised trial: use of a computer‐generated randomisation list Assessor blinding: no Length of follow‐up: 12 months post hospital discharge.
Participants	Teaching hospital in Brisbane, Australia. Conducted: October 1994 to July 1995 71 people with hip fracture. 22% male. Mean age 78 years. Cognitive status: none had dementia. Inclusion criteria: aged 55 years or over, admitted for acute hip fracture surgery, resided at home or in a hostel, independently mobile, accessible for follow‐up, public‐health patient, informed consent. Exclusion criteria: dementia, pathological fracture, inadequate English; residing in a nursing home. Assigned: 38/33 [Multidisciplinary care / Usual care] Assessed: 38/33 No loss to follow‐up.
Interventions	Randomisation took place before surgery. Eligible patients admitted to Accident and Emergency or to the Orthopaedic department were identified to the trial co‐ordinator who then sought patient consent. (1) Accelerated rehabilitation programme involving multidisciplinary team (orthopaedic surgeon, geriatrician, nurse‐coordinator, physiotherapist, occupational therapist, social worker) based in orthopaedic ward. Early surgery, minimal narcotic analgesia, early mobilisation, intense twice‐daily physiotherapy, daily visits by occupational therapist and social worker, early review by geriatrician, two additional ward rounds attended by all staff, weekly case conference, home assessment visit before discharge, community services referrals arranged, follow‐up at 1 and 6 months after discharge. (2) Standard orthopaedic management including daily visits from physiotherapist. Social worker or occupational therapist by request, geriatrician on referral, weekly discharge planning meeting. Other care including home visit as requested / needed.
Outcomes	'Poor outcome': mortality at discharge + discharge to nursing home or hostel; mortality + institutional care at 12 months; or mortality and functional deterioration at 12 months. Other outcomes: mortality (discharge, 6 and 12 months after discharge), functional decline (mobility), function (modified Barthel Index), ADL, mobility, complications, dynamic balance, length of stay, place of residence (institutional care at discharge and 12 months), social support (dependency), readmissions (all for "unrelated causes"), analgesia, costs.
Funding and conflict of interest statements	Study was funded by Medicare Incentives Hospital Access Programme (1993‐1998), initiated by the Australian Government. There were no conflict of interest statements.
Notes	Six months post‐discharge results were presented in the main trial report (Swanson 1998). A further report (Day 2001) reported phone and postal follow‐up at 12 months post‐discharge. Clarification of method of randomisation, mortality and length of data results and data for readmissions, functional decline and institutional care received from Cheryl Swanson on 5 December 2001.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"computer‐generated randomisation list"
Allocation concealment (selection bias)	Unclear risk	".. trial co‐ordinator, who, having sole possession of the treatment codes, obtained consent then randomised each patient to one of the treatment arms." However, trial co‐ordinator also co‐ordinated care in the intervention group.
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	Specific discharge criteria were established to eliminate bias. Data obtained from medical records.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	Intervention group patients followed‐up at six months while attending an outpatient clinic. Usual care patients were followed up at outpatient clinic or phoned up by trial co‐ordinator.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	No loss to follow‐up and intention‐to‐treat analysis implied.
Incomplete outcome data (attrition bias) Function, QOL	Low risk	No loss to follow‐up and intention‐to‐treat analysis implied.
Selective reporting (reporting bias)	Unclear risk	The study protocol is not available but it is very likely that the published report includes all expected outcomes, including those that were prespecified.
Free of bias resulting major imbalances at baseline	Unclear risk	Slight imbalance in male/female composition of groups (males: 11/38 versus 5/33).
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Given the comprehensive nature of the intervention group, hard to tell. No hint of contamination of the control group.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Sufficiently similar follow‐up procedures.

Tseng 2019.

Study characteristics
Methods	Randomised trial: computer‐generated random table Assessor blinding: participant blinding claimed but seems incompatible with informed consent Length of follow‐up: 24 months (also 1, 3, 6, 12 and 18 months)
Participants	Teaching hospital in Taiwan Conducted: January 2010 to August 2014 176 people with hip fracture and diabetes mellitus. 32.7% male. Mean age 78 years. Cognitive status: severe cognitive impairment not included. Inclusion criteria: aged 60 years or older, diagnosed with diabetes; admitted to Chang Gung Memorial hospital from its emergency department due to hip fracture; able to communicate; living in northern Taiwan; informed consent. Exclusion criteria (trial registration): clinical diagnosis of cancer; unable to communicate with researchers (examples in trial report: severe cognitive impairment; hearing problems). Reasons given for non‐eligibility in report: did not undergo surgery; other health issues during hospitalisation (e.g. too weak to walk, severe comorbidities); lived in institution before hip fracture. Assigned: 88/88 [Interdisciplinary + diabetes‐specific care / Usual care] Assessed (12 and 24 months): 62/68; 53/58 Loss to follow‐up (death, participant refused to participate or could not be contacted) = 16 versus 20 (at 12 months); 35 versus 30 (at 24 months)
Interventions	Randomisation took place soon after admission to emergency department. (1) Interdisciplinary care (including geriatric consultation, discharge planning, and in‐home rehabilitation) as developed in Shyu 2008 with diabetes‐specific care (including dietary and diabetes education, blood pressure control, dyslipidemia management, a glycemic treatment regimen, and rehabilitation exercises). Geriatric consultations were delivered by a geriatrician and a geriatric nurse to detect potential problems that would delay surgery. Rehabilitation started on the first day after surgery and included in‐hospital sessions and in‐home rehabilitation for 4 months after discharge. Both in‐patient and in‐home rehabilitation programmes contain a hip fracture‐oriented intervention and a general intervention for deteriorated physical fitness. The rehabilitation emphasised pain relief, while enhancing range of motion, muscle strength, endurance, proprioception, and balance, and aerobic capacity. After hospitalisation, a geriatric nurse visited the patient weekly during the first month, and biweekly during months 2 to 4. At the same time, a physical therapist visited the patient during the first, fourth, and twelfth weeks after discharge. Discharge planning was delivered by geriatric nurses and comprised a structured assessment of home environment, referrals, and reminders for clinical follow‐up. A diabetes‐specific care component was also involved. A diabetes nurse trained in diabetes management (dietary, education, blood pressure control, dyslipidemia management, a glycemic treatment regimen, and exercises) delivered these in a home setting with referral and consultation as required with dieticians and geriatricians specialising in internal medicine. Individual education with written information was provided during hospitalisation and reinforced during home visits at 1, 3, 6 and 12 months after discharge. Clinic visits were arranged where the patient did not have a regular physician. (2) Usual care (control group). This entailed one or two in‐hospital rehabilitation sessions. Cared for by orthopaedists, but consultations for internal medicine care occasionally made, depending on the patient's condition. In the first 2 to 3 days after hip fracture surgery, nurses taught patients bed exercises. Physical therapy usually started on the second day following surgery. Around 4–5 days after surgery, patients were usually discharged from the hospital. After hospital discharge, patients received no in‐home programmes for rehabilitation or nursing care. Additionally, diabetes‐related health education and consultation were not routinely provided. Note: follow‐up involved assessment at 1, 3, 6, 12 and 24 months after discharge through home visits, and families were also asked to keep weekly diaries. Both of these were additional to usual care and thus could have acted as a co‐intervention.
Outcomes	'Poor outcome': mortality and admission to institutional care (nursing home) at 6 and 12 months Other outcomes (1, 3, 6, 12 and 24 months): mortality, admission to nursing home, readmission & emergency department visits (cumulative data over 24 months not available; just used those occurring within one month), recovery to self‐care ability (Chinese Barthel Index), recovery of instrumental activities of daily living (incomplete data), length of acute hospital stay (incomplete data), falls. Data are also available for range of motion, quadriceps strength, heart rate variables. No data yet for outcomes mentioned in relation to the trial: Quality of life (SF‐36 Taiwan version), listed as the primary outcome; and depressive symptoms (Geriatric Depression Scale, Chinese version)
Funding and conflict of interest statements	Funding support from the National Health Research Institutes, Taiwan (NHRI‐EX103‐9905PI); and Chang Gung Medical Foundation (BMRP297) The authors declared no financial conflicts of interest.
Notes	Refers to their previous "NHRI‐funded study: Intervention program for elderly patients with hip fracture, 2001‐2003"; Shyu 2008 (same principal investigator) The stated purpose of the study is to "1) develop a well‐conceived and feasible protocol for hospital discharge and subacute care for hip‐fractured elderly persons with DM, and 2) compare the costs and effectiveness of this DM‐specific model with those of an effective subacute care model previously developed by our research team." Search in June 2019 identified a secondary data publication that reported on the numbers recruited and that the main report was "under review". Main report now published and name changed. Sent request for nursing home and 'poor outcome' data at 6 and 12 months to Shyu on 07/01/20. Response with data received 14/01/20. The presentation of the results for this trial are confusing; usually the results are reported between different time periods and thus not relative to baseline. The continuous outcomes may be change scores but notably anyway, the tabular data do not match well with the graphically presented trajectories.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	“Elderly patients and their caregivers who agreed to participate were then randomly assigned to the experimental or the control group using a computer‐generated random table by a neutral person who did not conduct assessment or intervention.”
Allocation concealment (selection bias)	Low risk	"Elderly patients and their caregivers who agreed to participate were then randomly assigned to the experimental or the control group using a computer‐generated random table by a neutral person who did not conduct assessment or intervention.”
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	“Patients and their families were blinded to whether they were in the experimental or control group.” This seems incompatible with informed consent. Care providers and assessors were not blinded. However, these outcomes are less susceptive to detection bias.
Blinding (performance bias and detection bias) Function, QOL	High risk	“Patients and their families were blinded to whether they were in the experimental or control group.” This seems incompatible with informed consent. Care providers and assessors were not blinded.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	There was a high loss to follow‐up (33.5% including deaths and 19.3% excluding deaths at 1 year follow‐up); this was acceptably balanced between the two groups. Aside from deaths (27 in all), most of the losses to follow‐up were refusals (5 were not contactable) and so for the three outcomes (death, readmission and admitted to nursing home) reported in this category, the risk of bias is likely to be low.
Incomplete outcome data (attrition bias) Function, QOL	Unclear risk	There was a high loss to follow‐up (33.5% including deaths and 19.3% excluding deaths at 1 year follow‐up). However, this was reasonably balanced between the two groups (35% versus 32%). The authors acknowledged that the "results of this study were limited by around 20% of participants dropping out due to refusal or loss of contact”, but pointed to their analysis which had controlled for attrition and the similar results between “on‐protocol” and “whole sample” analyses. It remains difficult to judge this.
Selective reporting (reporting bias)	High risk	Trial was prospectively registered. However, quality of life data were not reported (this was the primary outcome in the trial registration document), and an emphasis was given instead to those outcome measures, such as hip flexion range of motion, that showed differences.
Free of bias resulting major imbalances at baseline	Low risk	There was no indication of imbalance in the reported baseline characteristics.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Insufficient information to judge this; this includes whether the staff were in common for the two trial arms. The extra home visits for follow‐up could be considered a co‐intervention in the usual care group, however.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Appeared comparable.

Uy 2008.

Study characteristics
Methods	Randomised trial: use of numbered opaque envelopes, with the randomised sequence generated from a random number table Assessor blinding: yes for functional outcomes, collected by a research nurse who was masked to the allocation of the study participant Length of follow‐up: 4 months
Participants	General hospital in Sydney, Australia. Conducted: 2001 11 people with hip fracture; all female. Of 10: median age 80 years (intervention group); 83 years (control group). Cognitive status: "all had moderate to severe cognitive impairment" Inclusion criteria: female, living in a nursing home in the hospital catchment area, ambulant prior to hip fracture, able to follow commands at time of seeking informed consent. Informed consent by patient or legally authorised proxy. Exclusion criteria: non‐consent or not fulfilling the inclusion criteria. Assigned: 4/7 [Accelerated care / Usual care] Assessed (1 and 4 months): 3/7 Loss to follow‐up: 1 death.
Interventions	Randomisation took place after surgery. (1) Inpatient multidisciplinary rehabilitation programme involving accelerated rehabilitation (as in Cameron 1993). This is summarised as following: "In summary, accelerated rehabilitation differs from conventional care in its early assessment of rehabilitation goals, early commencement (usually within 24 hours of surgery), greater emphasis on retraining for physical independence, closer family care‐giver contact and more detailed discharge planning." (2) Usual care: discharge back to the nursing home soon after hip fracture surgery.
Outcomes	'Poor outcome': not available Other outcomes: mortality ("early", up to 4 months), function (Barthel Index at 1 and 4 months), and mobility (gait velocity at 1 and 4 months)
Funding and conflict of interest statements	There was no statement on funding. There were no conflicts of interest statements.
Notes	The trial was ended prematurely after recruiting only 11 participants because of a change in Australian Government regulations that created financial incentives to have only immobile residents in nursing homes ‐ thus, severely restricting the number of potentially eligible participants. The trial was included in this update, given that another trial (Watne 2014) had included (and provided a subgroup analysis) patients from nursing homes; a previously overlooked patient group.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"..randomisation sequence generated from a random number table"
Allocation concealment (selection bias)	Low risk	"Concealed randomisation using numbered opaque envelopes, with the randomisation sequence generated from a random number table"
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	No blinding for mortality data but unlikely to have been affected by lack of blinding.
Blinding (performance bias and detection bias) Function, QOL	Unclear risk	"Follow‐up data were collected by a research nurse who was masked to the allocation of the study participant" However, there is no confirmation of the success of this.
Incomplete outcome data (attrition bias) Death, residence, readmission	Unclear risk	Only mortality data provided. Sample size too small.
Incomplete outcome data (attrition bias) Function, QOL	High risk	Small sample size means that the outcome data for these (Barthel index and gait velocity) are very limited.
Selective reporting (reporting bias)	Unclear risk	No trial registration or protocol available. Trial stopped prematurely, although reasons given. Baseline data not given for the early death.
Free of bias resulting major imbalances at baseline	High risk	The sample size is too small for randomisation to be effective and the imbalances, including in the number in each group, are relatively large in the type of hip fracture and prior use of walking aid.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Not confirmed.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Systematic follow‐up at one and four months.

Vidan 2005.

Study characteristics
Methods	Randomised trial: closed envelopes in a secure box, administered by one person; stratified by pre‐fracture ADL level Assessor blinding: yes, for patient interviews Length of follow‐up: 12 months
Participants	General hospital in Madrid, Spain. Conducted: February 1997 to December 1997. 321 people with hip fracture. Of 319: 18.5% male. Mean age 82 years, (SD 7.7). Cognitive status: 24.5% had dementia. Inclusion criteria: previously ambulant, aged 65 years or over, admitted for acute hip fracture surgery. Informed consent for patient or proxy. Exclusion criteria: unable to walk before fracture; dependency in all ADL; pathological hip fracture; known terminal illness; life expectancy < 12 months. Assigned: 157/196 [Multidisciplinary care / Usual care] Assessed (mortality): 148/164 Loss to follow‐up included 2 in Multidisciplinary care group that did not have surgery.
Interventions	Randomisation took place after baseline assessment, within 48 hours of admission. All participants had an orthopaedic surgeon and a nurse assigned on hospital admission. All shared same orthopaedic wards and used same hospital support services, including physical therapy and social work. Orthopaedic surgeon determined discharge in both groups. (1) Multidisciplinary care: Geriatric team (geriatrician, rehabilitation specialist, specific social worker); geriatric assessment; daily visits from geriatrician who was responsible for medical care; rehabilitation specialist planned schedule, duration and intensity of physical therapy; social worker assessed social environment and advised where necessary. Within 72 hours of admission, an interdisciplinary meeting, including orthopaedic and geriatric teams, to formulate a comprehensive treatment plan. Meeting repeated weekly to monitor compliance with plan. (2) Usual orthopaedic care: Management by orthopaedic surgeon and nurses. Counselling from different specialists as required.
Outcomes	'Poor outcome': mortality at 12 months + admission to long‐term care institution; mortality in hospital + discharge to rehabilitation facility (incomplete data) Other outcomes: mortality (in hospital and 1 year), patient function (mobility; recovery in ADL at 3, 6 and 12 months), length of stay, discharge to rehabilitation facility; admission to long‐term care institution, readmissions (during 1 year); morbidity, medical and surgical complications; receipt and duration of physical therapy.
Funding and conflict of interest statements	Funding support via a grant from the Findo de Investigaciones Sanitarias, Ministerio de Sanidad, Spain. There was no sponsor. There were no conflict of interest statements.
Notes	Details of method of randomisation and data received from trial investigator: Maria Vidan.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No details: stratified by pre‐fracture ADL level (independent in 4 or more ADLs or < 4 ADLs)
Allocation concealment (selection bias)	Low risk	"We used a system of closed envelopes in a secure place (box), administrated by only one person." (personal communication from trialist)
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	Data obtained from medical records.
Blinding (performance bias and detection bias) Function, QOL	Low risk	"A research assistant who was not involved in patient care and was blinded to the allocation group conducted all interviews."
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	Full data for death, residence at discharge and readmission provided.
Incomplete outcome data (attrition bias) Function, QOL	Low risk	Data on ADL and mobility provided.
Selective reporting (reporting bias)	Unclear risk	The study protocol is not available but it is very likely that the published report includes all expected outcomes, including those that were prespecified.
Free of bias resulting major imbalances at baseline	Low risk	No major differences in baseline characteristics.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Trial authors noted that there was a shorter length of stay in the control group than in previous years that they thought may be explained by the Hawthorne effect (that the improvement resulted because the physicians were being observed). There were, however, some delays in the discharge of patients with dementia and ADL dependency "because home care services or rehabilitation centre facilities could not be arranged in a timely manner." Potential for contamination bias given both groups had same location and staff.
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Identical follow‐up procedures.

Watne 2014.

Study characteristics
Methods	Randomised trial: sealed, opaque, numbered envelopes; stratified by whether or not the patients were admitted from nursing homes Assessor blinding: yes, for study nurses assessing participants at follow‐up visits and two specialists assessing dementia Length of follow‐up: 12 months
Participants	University hospital in Oslo, Norway. Conducted: September 2009 to January 2012 (recruitment). 332 people with hip fracture. 25.1% male. Median age 85 years, range 46 to 101. Cognitive status: of 329, 49.2% had dementia. Inclusion criteria: acute admission for hip fracture (femoral neck fracture, a trochanteric or a sub‐trochanteric fracture). Informed consent by patient or relative. Exclusion criteria: hip fracture was a part of a high‐energy trauma (defined as a fall from higher than one metre), terminal illness on admission. Not resident in catchment area of hospital. Assigned (3 'moribund patients' excluded): 163/166 [Geriatric ward / Orthopaedic ward] Assessed (mortality): 163/166 Loss to follow‐up (at 12 months): 65 versus 71 (deceased, withdrew, hospitalised, unreachable).
Interventions	Randomisation took place in the emergency department, overseen by the duty orthopaedic surgeon (thus before surgery). Transfer to the allocated wards was as soon as possible, with return to the same ward after surgery and a few hours in the postoperative care unit. (1) Acute geriatric ward: Preoperative and postoperative orthogeriatric care integrated in an acute geriatric ward. Comprehensive Geriatric Assessment (CGA) was used as a basis for treatment planning. All team members (geriatrician, nurse, physiotherapist and occupational therapist) were expected to assess patients during their first day on the ward. The team had daily meetings to co‐ordinate treatment and to plan discharge. Piloted clinical routines, with checklists, included medication reviews, early and intensive mobilisation, optimising pre‐ and postoperative nutrition (including supplementation) and early discharge planning. (2) Orthopaedic ward. There were no multidisciplinary meetings and no geriatric assessments. Early mobilisation was emphasised, and hip fracture patients were seen by a physiotherapist soon after surgery. Relevant specialists were seen on request. The postoperative care unit was within the orthopaedic ward, where all trial participants were observed after surgery. All participants were offered a follow‐up appointment in the orthopaedic outpatient clinic four months after surgery. There was no additional intervention after discharge from hospital.
Outcomes	'Poor outcome': mortality at 12 months + new nursing home admissions + lost‐to‐follow‐up because of being hospitalised or too ill to approach (also available at 4 months) Other outcomes: mortality (in hospital, 4 and 12 months); patient function (mobility; ADL at 4 and 12 months); length of stay; discharge to rehabilitation facility; new nursing home admissions (4 and 12 months); readmissions (at 4 and 12 months); morbidity, medical and surgical complications; delirium and dementia
Funding and conflict of interest statements	Funding was mainly by the Research Council of Norway (grant 187980/H10). Also Oslo University Hospital, the Sophies Minde Foundation, the Norwegian Association for Public Health and Civitan's Research Foundation. The sponsors had no role in the design, conduct and reporting of the trial. The authors declared they had no competing interests.
Notes	The acute geriatric ward (20 beds) mainly admitted older patients suffering from acute medical disorders; the only surgical patients treated in the ward were the hip fracture patients included in the trial. Since the acute geriatric ward was regularly full, the ward was instructed to admit included hip fracture patients even if these had to be treated in the corridor until a room was available, usually within the first 24 hours. The orthopaedic ward (52 beds) admitted a range of elective and non‐elective orthopedic patients. The staff‐patient ratio was similar to that of the acute geriatric ward. Subgroup data available for participants with dementia and those from nursing homes.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Randomization was based on computer‐generated random numbers (blocks of variable and unknown size) and was carried out by a statistician (ES) not involved in the clinical service. Randomization was stratified according to whether or not the patients were admitted from nursing homes."
Allocation concealment (selection bias)	Low risk	"Allocation was by sealed, opaque, numbered envelopes."
Blinding (performance bias and detection bias) Death, residence, readmission	Unclear risk	Participants and care personnel were not blinded. However, these outcomes less susceptible to detection bias.
Blinding (performance bias and detection bias) Function, QOL	High risk	Participants and care personnel were not blinded. Data collected at the 4 and 12 months' follow‐up were collected by study nurses blind to allocation. However, as reported in the protocol, in "5‐10% of the cases" the research nurses were unblinded because of information from the trial participants or their relatives.
Incomplete outcome data (attrition bias) Death, residence, readmission	Low risk	This assessment after consideration of two minor intention‐to‐treat deviations: the post‐randomisation exclusion of 3 moribund patients (incorrectly included) and the analysis of the results of two patients according to the ward they received treatment (one in each group) instead of their allocated ward.
Incomplete outcome data (attrition bias) Function, QOL	High risk	In part reflecting that 27% had died by 12 months, but data for function were only available for 59% of the population at that time.
Selective reporting (reporting bias)	Low risk	Although trial was registered after recruitment had started, and the protocol reported on the numbers randomised, there did not appear to be problems with selective outcome reporting. All recorded outcomes were reported.
Free of bias resulting major imbalances at baseline	Low risk	Participants in the two groups were "well matched" with no major differences in baseline characteristics.
Free of performance bias (from non‐trial differences in care provision)?	Low risk	This was equivalent as evidenced: "The staff‐patient ratio was similar [in the orthopaedic ward] to that of the acute geriatric ward." "All patients included in the trial were offered a control in the orthopedic outpatient clinic four months after surgery. There was no additional intervention after discharge from hospital."
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Identical follow‐up procedures.

Ziden 2008.

Study characteristics
Methods	Randomised trial: sealed envelopes opened by "geriatric nurse" Assessor blinding: none reported Length of follow‐up: 12 months
Participants	Sahlgrenska University hospital in Gothenburg, Sweden Conducted: November 2004 to March 2006 (recruitment). 212 people with surgery for hip fracture of whom 102 were included in the trial. 30.4% male. Mean age 82 years, range 65 to 99. Cognitive status: patients with cognitive impairments were excluded. Inclusion criteria: acute hip fracture surgery, community dwelling, medically approved by the responsible geriatric doctor as being in need of geriatric care and rehabilitation, aged 65 or above, and able to speak and understand the Swedish language. Exclusion (post randomisation list): cognitive impairments, lived in wrong part of city, serious disease, abuse or dead, wrong diagnosis, inappropriate to seek consent, non‐consent Assigned: 105/107 [Home rehabilitation / Conventional care] Included in study: 48/54 Loss to follow‐up of 102 (at 12 months): 3 dead versus 6 (4 dead, 1 drop‐out, 1 unreachable)
Interventions	Randomisation occurred before the patients were transferred to the respective geriatric ward (one with home rehabilitation; one with usual care), hence before they were informed and asked about participation in the study. Subsequently those remaining suitable for inclusion were asked to participate and informed consent obtained. (1) A geriatric, multi‐professional home rehabilitation programme focused on supported discharge, independence in daily activities, and enhancing physical activity and confidence in performing daily activities. As well as usual care, patients at the home rehabilitation ward followed a specially designed programme, consisting of two parts; one started upon ward admission and one after discharge. The programme included early goal‐setting together with the patient and supported discharge, including close cooperation with the patient’s relatives and social home services. This aimed to reduce the patients’, and their relatives’, possible uneasiness about going home instead of staying longer in hospital or being transferred to a short‐term nursing home. Each intervention was individually designed. The programme included multi‐professional actions aimed at mobilising the person’s motivation and self‐efficacy. The second part of the programme consisted of a brief intervention period of a maximum of 3 weeks that included home visits made mainly by physiotherapists and occupational therapists. The same therapists were involved in both parts of the programme. When necessary, nurses or assistant nurses made visits. The hospital geriatrician was also medically responsible for the patients’ care during the second part of the programme. (2) Conventional care with no structured rehabilitation after discharge. Usual care had involvement of all personnel in everyday rehabilitation in the geriatric ward. Early mobilisation and standard rehabilitation programme: daily individual training and physiotherapy group training session. All rehabilitation measures were adapted to the patients’ individual medical and functional status and personal goals. The physiotherapist and occupational therapist made a home visit to the patient in connection with discharge when it was assessed as necessary.
Outcomes	'Poor outcome': mortality at 12 months or move to permanent sheltered housing (by 12 months) Other outcomes: mortality (1, 6 and 12 months), move to permanent sheltered housing (during 12 months). At one month follow‐up: length of hospital stay, discharge to short‐term nursing home care (more a compliance issue), walking, Functional Independence Measure (FIM), Falls Efficacy Scale, degree of independence in and frequency of activities, physical tests ('Timed Up and Go' and 'Sit to Stand'). Outcomes reported for 6 and 12 months: dependency in ADL (FIM) and in i‐ADL (IAM: Instrumental Activity Measure), frequency of activities (FAI: Frenchay's Activity Scale); Timed Up and Go'; muscle strength; balance confidence; HRQoL (SF‐36); mood and perceived recovery
Funding and conflict of interest statements	Support but not an explicit mention of funding was acknowledged from the Vårdal Institute, the Hjalmar Svensson's Foundation and the Geriatric Section of the Swedish Association of Registered Physiotherapists There were no conflict of interest statements.
Notes	Randomisation and allocation occurred at emergency unit without patient consent. Subsequent exclusion of 99 patients (cognitive impairments (16), lived in wrong part of city (10), serious disease, abuse or dead (16), wrong diagnoses (2), no bed on allocated ward (30), moved from clinic or too ill to be asked to participate (25)) before patients invited to participate. A further 11 declined consent, resulting in 102 study participants. Six participants in the intervention group did not participate in the home rehabilitation programme; 3 discharged to a short‐term nursing home and 3 to another acute clinic. Different care and rehabilitation than usual in the control group involved 20 participants: 9 went to short‐term nursing home care; 8 received community home rehabilitation and four visited outpatient physiotherapists. All were included in their allocated treatment group in the analyses.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Patients with hip fracture ... were referred directly from the emergency unit to either of two different geriatric wards, one with HR and one with CC. The randomization therefore occurred before the patients were transferred to the respective geriatric ward, and thus before they were informed and asked about participation in the study." No information on sequence generation
Allocation concealment (selection bias)	Unclear risk	"The nurses at the emergency unit reported all patients who fulfilled the basic inclusion criteria (registration diagnosis, age, and living situation) to a geriatric nurse who performed the randomization using sealed envelopes." Insufficient details on allocation concealment safeguards.
Blinding (performance bias and detection bias) Death, residence, readmission	Low risk	Participants and care personnel were not blinded. But these data, which were collected from medical records, are unlikely to be affected.
Blinding (performance bias and detection bias) Function, QOL	High risk	Not blinded. "The follow‐up visits in the study were carried out in the participants’ homes by the same researcher (LZ), who was blind to previous assessments but not to the group to which patients had been randomized."
Incomplete outcome data (attrition bias) Death, residence, readmission	High risk	Post‐randomisation exclusions (52%), some of whom were because they had died before consent being given, were not included in the analyses.
Incomplete outcome data (attrition bias) Function, QOL	High risk	Post‐randomisation exclusions (52%) were not included in the analyses.
Selective reporting (reporting bias)	Unclear risk	No trial registration or published protocol. Although all listed outcomes reported, the 6‐ and 12‐month data for functional outcomes are unhelpfully presented; since these are reported differently from the 1‐month data, it is possible that these had been a change in protocol / analysis plan.
Free of bias resulting major imbalances at baseline	High risk	The baseline characteristics were only provided for the 102 included in the trial rather than those randomised. There are some differences between the two groups that may be important: there were more men in the home rehabilitation group (39.6% versus 22.2%) and fewer participants were living alone (54.2% versus 72.2%). Of the 110 randomised but not included, the stated reasons (such as declined to participate) were balanced between the two groups. Nonetheless, there is no information to make a judgement on the originally randomised population.
Free of performance bias (from non‐trial differences in care provision)?	Unclear risk	Such trials are hard to assess for this aspect. However, greater attention was given to the monitoring of the home rehabilitation programme. "The researcher (LZ) was not involved in the HR programme. However, one way to control the intervention was through regular meetings, when the researcher and the HR staff."
Free of detection bias (from e.g. differences in follow‐up procedures)?	Low risk	Identical follow‐up procedures.

ADL: activities of daily living
GP: general practitioner
PFF: proximal femoral fracture
QOL/QoL: quality of life
vs: versus

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Abe 2001	Trial of a four‐week (short) versus eight‐week (long) rehabilitation training programme in surgically treated hip fracture patients. A translation from Japanese revealed that the two intervention groups were not concurrent (historical control).
ACTRN12619000296134 2019	Not randomised: prospective study evaluating a "multidisciplinary care bundle" after hip fracture surgery
Allegrante 2001	Multicomponent as opposed to multidisciplinary rehabilitation. Trial included in Crotty 2010.
Asplin 2017	Not randomised: study compared a coordinated rehabilitation programme with early onset of patient participation and intensified occupational therapy and physiotherapy versus orthopaedic care after hip fracture surgery.
Bai 2003	Complete care package rather than multidisciplinary care. Control group received minimal rehabilitation.
Bai 2009	Not hip fracture patients. Participants had hip arthroplasty for osteoarthritis.
Barone 2006	Not randomised: report of a cohort study of a comprehensive geriatric intervention.
Beaupre 2019	Not randomised allocation. The intervention was a mobility intervention and not multidisciplinary.
Binder 2004	Randomised evaluation of a physical therapy intervention comprising six months of supervised exercise carried out three times weekly. The intervention did not meet our definition of multidisciplinary rehabilitation. Trial is included in Handoll 2011, which covers mobilisation interventions.
Boyer 1986	Study, which tested an acute care geriatric unit, was described as an "experimental randomized pretest‐posttest design controlled study". This implies it was randomised but probably not in a comparative way. The mixed population study gave no specific detail or data on participants. There were no extractable data on hip fracture patients.
Choong 2000	Randomised evaluation of a clinical care pathway. The intervention did not meet our definition of multidisciplinary rehabilitation.
Cuncliffe 2004	Only 28% of participants had fractures: additionally, the percentage with hip fracture is not reported.
Dautel 2019	The multifactorial intervention focusing on tailored exercise and counselling did not meet our definition of multidisciplinary care; there also did not appear to be geriatrician assessment or involvement.
Deneckere 2012	Cluster‐randomised trial carried out in four countries testing a care pathway (active components: a formative evaluation of quality and organisation of the care setting, a set of evidence‐based key interventions, and support of the development and implementation of the care pathway) versus usual care for patients with hip fracture or chronic obstructive pulmonary disease. This is not a trial evaluating multidisciplinary rehabilitation. This appears to be an extension of the trial, including people with hip fracture only, tested in Vanhaecht 2012.
Deschodt 2011	Not randomised: controlled trial based on "assignment by convenience" comparing a multidisciplinary geriatric intervention versus usual care during hospitalisation after hip fracture.
Edgren 2015	The multicomponent home‐based rehabilitation programme started six weeks after discharge from hospital involved a physiotherapist; this does not meet our definition of multidisciplinary rehabilitation. This trial (ProMo) is likely to be included in the update of Handoll 2011.
FIT‐HIP 2019	Not fitting our definition of multidisciplinary rehabilitation. Intervention was "targeted treatment of fear of falling in geriatric rehabilitation (GR) after hip fracture using a multi‐component cognitive behavioural intervention (FIT‐HIP)".
Fordham 1995	Not randomised: observational study of a geriatric orthopaedic programme at one hospital and conventional orthopaedic care at another. Study produced equivalent groups at both hospitals.
Gonzalez‐Montalvo 2010	Not randomised: although alternation was used it was based on the two wards involved being on call on alternate days; with patients being admitted to the usual care ward when smaller acute orthogeriatric ward was full. This capacity issue already compromised, at best, a quasi‐randomised allocation.
Hagsten 2004	Randomised evaluation of an occupational therapy programme, delivered by an occupational therapist. The intervention did not meet our definition of multidisciplinary rehabilitation. This trial is included in Crotty 2010.
Hempsall 1990	Not randomised: allocation to Geriatric Orthopaedic Rehabilitation Unit (GORU) by area of residence.
HIPFRAC 2017	Randomised evaluation of a functional training programme, initiated by nurses, in additional to usual care occupational therapy programme, delivered by an occupational therapist. The intervention did not meet our definition of multidisciplinary rehabilitation.
Houldin 1995	Pilot study evaluating a psychological intervention in eight hip fracture patients. Not in scope of review.
Huang 2005	Not multidisciplinary rehabilitation: discharge planning led by nurse only (and then extra care as requested). This trial is included in Crotty 2010.
IRCT2016072119141N2	Trial registration for a randomised evaluation of a training booklet. The intervention did not meet our definition of multidisciplinary rehabilitation.
ISRCTN22464643	Randomised trial comparing six extra sessions of supervised physiotherapy or occupational therapy rehabilitation, tailored to their individual needs on top of usual multidisciplinary care versus usual multidisciplinary care. Excluded as intervention is not multidisciplinary care. This feasibility study (Fracture in the Elderly Multidisciplinary Rehabilitation (FEMuR)) is likely to be included in the update of Handoll 2011.
ISRCTN28376407	Ongoing trial (Femur3), linked with ISRCTN22464643, testing up to six extra rehabilitation sessions plus providing participants with a workbook and a goal‐setting diary. Study dates: August 2018 to January 2022. Excluded as intervention is not multidisciplinary care.
Joeris 2017	Not randomised: multicentre observational cohort study.
Krichbaum 2007	While nurse co‐ordinator, who could not provide direct care, arranged multidisciplinary care, this is not multidisciplinary rehabilitation. This trial is included in Crotty 2010.
Kuisma 2002	Unidisciplinary intervention (physiotherapy) only. This trial compared discharge from an acute ward to home with visits by a physiotherapist versus usual care in a rehabilitation centre in 81 hip fracture patients. It was also excluded by Crotty 2010 because it is "primarily a home versus hospital comparison".
Lahtinen 2015	The unusual method of allocation into three separate groups (physiatric‐oriented rehabilitation (POR) in Oulu, Finland; geriatric‐oriented rehabilitation (GOR) in Oulu, Finland; routine rehabilitation (control) in health centre hospitals in surrounding counties (province)) comprised two randomisation lines: one for patients based in city of Oulu (two thirds randomised to GOR and one third to POR) and the other for patients living in the province (one third randomised to POR and two thirds to control). In consequence, GOR was not randomised with control (based in geographically distinct locations) and only half the participants in the POR group were randomised in each of the two comparisons. This unusual design means that we cannot consider the trial data to be randomised, with protection against selection bias. This was acknowledged in the Discussion section of Lahtinen 2017, who acknowledged that "One potential weakness of our study was that patients' residences influenced their randomization to rehabilitation groups". Thus, this study is excluded because of its inappropriate study design.
Lin 2009	No details of how allocation of participants was "randomly divided". Evaluation of discharge‐planning programme applied by nursing staff. Excluded as not multidisciplinary rehabilitation and thus not in scope of review.
Lofgren 2015	Not a randomised trial. Comparison of care in two sites.
Martin‐Martin 2014	Randomised trial evaluating additional occupational therapy carried out by an occupational therapist. Excluded as intervention is not multidisciplinary care.
NCT00951691	Mixed population of 26 participants, only 3 of whom had hip fracture (all in the intervention group).
NCT01537523	The planned intervention, an early mobilisation programme to improve physical and psychological aspects in the community, does not meet our definition of multidisciplinary care; there also did not appear to be geriatrician involvement.
NCT02058329	Registration document available for a trial evaluating home visits by a geriatric fellow. Excluded as intervention is not multidisciplinary care.
NCT03430193	The trial registration document (NCT03430193) for this study implied random allocation to a no‐rehabilitation group (discharged patients not transferred to rehabilitation unit after surgery for hip fracture). However, the protocol indicates that this would be an age‐ and sex‐matched "historically controlled group". We found two publications that referred to the trial registration document. One, Aftab 2020, reported a 2‐week follow‐up of 39 participants who were recruited August 2017 to January 2018, which was before the start date of the trial (12 February 2018). The other, Choi 2020, was a non‐randomised (historic) comparison testing home rehabilitation; all participants of which had the intervention (Fragility fracture integrated rehabilitation management (FIRM)) that was to be tested according to NCT03430193. The historical control arm had been in hospital from March 2017 to February 2018; thus again before the start date of NCT03430193. Excluded as this confirms that it is not a randomised trial.
NCT03822247	Registration document available for a mixed population trial (hip and knee fractures; hip and knee arthritis) testing a multidisciplinary recovery programme. Excluded as population is mixed and intervention is not multidisciplinary care.
Pearson 1988	Specialised nursing unit rather than Geriatric Orthopaedic Rehabilitation Unit (GORU). Mixed population of stroke, amputation and hip fracture patients (40%). Only limited outcome data for hip fracture patients.
Richards 1998	Location (evaluating early supported discharge and hospital at home) rather than multidisciplinary care. Mixed population: 31% hip fracture.
Roder 2003	Not a randomised trial, although treatment was left to chance.
Rubenstein 1984	No extractable data for fracture patients (23% of trial population): unclear how many had hip fracture.
Siu 1996	Mixed population: < 5% (17) had hip fracture.
Suwanpasu 2013	The physical activity‐enhancing programme tested in this trial did not meet our definition of multidisciplinary rehabilitation. This trial is likely to be included in the update of Handoll 2011.
Tinetti 1999	Trial assesses multicomponent rehabilitation provided by physiotherapist and nursing staff. Not multidisciplinary rehabilitation. Trial included in Crotty 2010.
Vanhaecht 2012	Cluster‐randomised trial carried out in four countries testing a care pathway (active components: a formative evaluation of quality and organization of the care setting, a set of evidence‐based key interventions, and support of the development and implementation of the care pathway) versus usual care for hip fracture patients. This is not a trial evaluating multidisciplinary rehabilitation. An extension of the trial including people with chronic obstructive pulmonary disease tested in Deneckere 2012.

Characteristics of studies awaiting classification [ordered by study ID]

Baroni 2016.

Methods	RCT. "Randomisation will be achieved via a web based, computer generated, block method with unknown block size."
Participants	150 patients with hip fractures aged 65 years or older, who were able to walk outdoors before their fracture. Conducted in Santa Maria della Misericordia Hospital, Perugia, Italy
Interventions	1. Acute hip fracture comprehensive geriatric care in a dedicated orthogeriatric ward 2. Usual orthopaedic care
Outcomes	Follow‐up: discharge, 6 and 12 months Primary outcome (reflecting stated primary aim) Prescription of appropriate antifracture therapy at hospital discharge. Secondary outcomes (reflecting secondary aim) Mention also of expectations of a lower risk of falls and fragility refracture; hence these are probably measured too.
Notes	We have provided the following record of our, eventually, successful attempt in getting further clarification about the status of this trial. It shows that sometimes persistence pays where there is good will! We thank the authors for their patience. Giulio Pioli in response to Ian Cameron on 15 May 2019 proposed that he contacted Carmelinda Ruggiero for information on this trial. In Carmelinda Ruggiero's response on 16 May 2019, she explained that "We recently published the results of the study in Osteoporosis International." (This was Baroni 2019.) Prof Ruggiero clarified that "Participants were randomly assigned to OGC or GCS by orthopedic resident on call, in collaboration with the orthopedic surgeon in charge, using the coin‐flipping procedure." This is distinct from the method described in this abstract (see Methods). In response to a request for clarification by Ian Cameron "we are trying to understand whether the following are two studies or two reports from one study", Prof Ruggiero's reply on 23 May 2019 included " You are perfectly right: they are two manuscripts from the same study protocol but they do not belong to the same study sample." We were unsure of the correct interpretation of this but decided that pressing further would not be productive. HH sent an email to Marta Baroni on 23 May 2019. Reply received 10 July 2019 indicated there were there were two different study protocols (Baroni 2016 and Boroni 2019) but was vague regarding the status of this Baroni 2016. HH sent an email to Marta Baroni on 30 July 2019. Reply received on 29 August 2019 included the following information of the trial status: "The trial is ended, we formally enrolled patients from March 2016 to April 2017. We are preparing the manuscript ..."

Hammond 2017.

Methods	Cluster‐randomised controlled trial in 11 hospitals in England and Scotland
Participants	282 people with hip fracture and cognitive impairment 21 November 2016 and closed on 31 January 2018. Inclusion criteria: 1. Patient must have had confirmed proximal hip fracture requiring an operation and be aged ≥ 60 years at the time of the operation 2. Patient has a preoperative Abbreviated Mental Test Score ≤ 8 in England (including those with zero because of an inability to answer questions) or a 4AT score ≥1 in Scotland. (4AT: Assessment tool for delirium and cognitive impairment (Alertness, Abbreviated mental test 4, Attention, Acute change or fluctuating course)) 3. Patient must have a “suitable informant” (e.g. relative, unpaid or paid carer, care home manager) who has a minimum of once a month face‐to‐face or telephone contact with the patient and is able, and consents to, provide information on proxy measures 4. Patient and a suitable informant must be recruited into the trial within 7 days of the hip fracture operation 5. Patient must spend a minimum of 5 days on the study ward Exclusion criteria: 1. Decision taken not to undergo hip surgery 2. Patient not expected to survive beyond 4 weeks 3. Patient already enrolled in a clinical trial of an investigational medicinal product (CTIMP)
Interventions	1. Complex intervention called PERFECT‐ER (featuring 68 patient level care and 15 organisational items) made up of: a best practice checklist, a staff training manual, staff time to implement the items on the checklist, staff time to train colleagues and a process to help continuous improvement in care. 2. Usual care
Outcomes	Follow‐up at 1, 3 and 6 months (patient and suitable informants) Cognition (mini mental state examination MMSE‐2 SV; Clinical Dementia Rating; Informant Questionnaire on Cognitive Decline in the Elderly); Quality of life (DEMQOL; EQ‐5D‐5L); mobility (Timed Up & Go); Physical function (Bristol Activities of Daily Living Scale (BADLS)); Patient and carer perspectives on care (howRwe; howRthey); place of residence; discharge location; service use (various: length of hospital stay; days in instructional care); delirium; comorbidity (Charlson Comorbidity Index).
Notes	There is insufficient information on the PERFECT‐ER intervention to make a judgement. Trial results are reported only in two conference abstracts.

NCT01934946.

Methods	RCT: "The subjects from wards are randomly divided into group A, B and Group C."
Participants	150 people with hip fracture from hospitals in Taipei, Taiwan; recruitment 1 January 2014 to 31 December 2014. Inclusion criteria: more than 60 years old, hip fracture "s/p" within 3 months, potential for rehabilitation. Exclusion criteria: active infection, such as pneumonia, urinary tract infection, osteomyelitis, wound infection; other uncontrolled medical condition, such as dyspnoea on exertion, acute renal failure, acute coronary syndrome, acute heart failure, uncontrolled diabetes, active malignancy, deep vein thrombosis, possibility of pulmonary embolism, acute psychosis; no caregiver; no potential for rehabilitation, such as severe dementia, poor cognition, poor cooperation and motivation, poor family support; end stage renal diagnosis on haemodialysis
Interventions	1. (Group C) 14 days hospitalisation for comprehensive rehabilitation and 6 times home rehabilitation services within 3 months 2. (Group B) received 14 days hospitalisation for comprehensive rehabilitation 3. (Group A) control group; probably home rehabilitation 6 times of PT home visit within 3 months after discharge The intervention is described as follows: "Comprehensive rehabilitation PT OT once per day for 10 times within 14 days hospitalization"
Outcomes	Follow‐up at 2 weeks and 3 months after intervention Pain scale Barthel Activity of Daily Living Harris Hip Score
Notes	s/p: usually means status post. We have left it in the description of inclusion criteria just in case. Status was 'recruiting' at trial registration on 4 September 2013. No response to request sent 17 May 2019 from HH to Koa M and Huang K on the status of their trial and more details of their intervention. Clarification is needed on the intervention in order to classify this trial.

NCT03894709.

Methods	Randomised, double‐blind trial
Participants	304 older hip fracture patients with cognitive impairment and their carers, based in Taiwan; study dates: 1 January 2015 to 31 October 2019 (estimated). Inclusion criteria: aged 60 years or older, admitted to CGMH with one‐side hip fracture, diagnosed as needing surgery, assessed as having cognitive impairment by the Chinese Mini‐Mental State Examination (CMMSE) (CMMSE score < 21 with < 6 years education, or CMMSE < 25 with ≥ 6 years education; has a primary family caregiver (> 20 years of age who provides direct or supervises care); living in northern Taiwan. Exclusion criteria: cognitively intact by CMMSE, without a primary family caregiver, terminally ill, severe cognitive impairment such that they are completely unable to follow orders (CMMSE < 10).
Interventions	1. Family‐centered care model "Interventions include a family‐centered approach to interdisciplinary care and a family caregiving‐training component to enhance family caregivers' competence in providing post‐operative care and handling behavioral problems of adults with cognitive impairment. The interdisciplinary care model consists of geriatric consultation, continuous rehabilitation, and discharge planning. The family‐centered approach involves family caregivers using a structured guide to assess the condition of the hip‐fractured patient with cognitive impairment. Habits, daily routines, preferences, behavioral problems and environmental safety and stimuli are explored. The strengths, weakness, and resources of the family are assessed. The behavioral problems and symptoms to target are identified. Both the research nurse and the caregiver will then collaborate on a tentative plan to minimize the behavioral problems." 2. Usual care "During hospitalization, patients receive health teaching for exercise while still in bed. Physical therapy usually starts only for those who received arthroplasty of hip replacement. Physical therapists train patients to use a walker and get in/out of bed through consultation. Usually, patients are discharged from the hospital without home assessment, nor are in‐home programs provided for rehabilitation or nursing care. The usual care does not involve interdisciplinary care protocols, continuity of care, or specific care for hip‐fractured patients with cognitive impairment."
Outcomes	Follow‐up (from date of discharge) at 1, 3, 6 and 12 months Primary outcomes All change scores from baseline Total range of motion of hip (sum of hip flexion and hip extension in degrees) Muscle strength of lower limbs Flexibility (in cm) Physical function to one year; performance of activities of daily livings (ADLs) assessed by the Chinese Barthel Index (CBI) Cognitive function to one year measured using the MMSE Taiwan version Behavioral problems to one year, measured using the Chinese version Cohen‐Mansfield Agitation Inventory (CMAI), community form Caregiver competence to one year measured using the 17‐item Chinese version Caregiver Competence Scale (Huang & Shyu, 2003) Delirium to one year measured using the Delirium Rating Scale, revision 98 (DRS‐R‐98) Secondary outcomes: Change from one‐month service utilization to one year Health‐related quality of life (HRQoL) to one year measured by the SF‐36 Taiwan version Change from baseline cost of care to one year Other outcomes: All change scores from baseline Nutritional status to one year; assessed using the Mini Nutritional Assessment (MNA) Depressive symptoms to one year; assessed for severity using the 15‐item Geriatric Depression Scale (GDS) short form. Social support to one year; by the Medical Outcomes Study (MOS) social support survey
Notes	Clarification is needed on the intervention in order to classify this trial.

NCT04207788.

Methods	Multicentre randomised controlled trial
Participants	108 (target) people with hip fracture Inclusion criteria: recent hip fracture, aged 65 years or older, living at home prior to fracture in Herlev, Gentofte, Furesoe, Rudersdal or Lyngby‐Taarbæk municipalities (Denmark), ability to give informed consent. Exclusion criteria: not expected to be discharged to home or rehabilitation centres in the municipality; unable to speak or understand Danish, or both; prior severe physical or mental disabilities, or both
Interventions	1. 'HIP‐REP programme' (add‐on activity‐focused interventions). HIP‐REP programme; an add‐on to usual care which includes 5 add‐on individual activity‐focused interventions that are tailored to the needs of elderly with hip fracture in liaison with the health care professionals. Home visits and follow‐up phone call are part of the programme. 2. Usual care: with a focus on mobilisation within 24 hours postoperative as an inpatient. "In the municipality physiotherapy is offered twice a week for approximately 12 weeks postoperative focusing on neuromotor control over hip, activation and strengthening of muscles as well as increase mobility. When relevant, the elderly receives a home visit by an occupational therapist who assess ADL behaviours, the use of assistive devices and the environmental hazards, preparing and adapting the home environment to prevent falls."
Outcomes	Follow‐up: at 3 and 6 months Primary outcome Assessment of Motor and Process Skills: change from baseline at motor and process skills 3 and 6 months after baseline testing Secondary outcomes European Quality of Life Questionnaire Verbal Rating Scale Functional Recovery Score
Notes	A full description of the interdisciplinary care will be needed to check whether the trial meets the inclusion criteria.

Parsons 2019.

Methods	Randomised trial: computer‐generated randomisation sequence
Participants	335 hip fracture patients (out 403 older people in hospital because of injury that required hospital admission and subsequent rehabilitation), based in Waikato, New Zealand; study recruitment dates: December 2013 to July 2015. Inclusion criteria: 65 years of age; in hospital at time of referral and did not require ongoing acute hospital‐based treatment (in the judgment of the consultant geriatrician); consented to being treated at home; and agreed with the objectives set by the referring inter‐disciplinary team. Further, following assessment by the referring team, the participant was considered to have potential for partial or complete recovery with suitable home rehabilitation within 6 weeks; was able to stand and transfer with one person (with or without the help of a resident carer); had a recent injury and was at a borderline level of function with an associated reduction in activities of daily living (ADL) and/or instrumental ADL (IADL); and who without input from the team was considered likely to fail to recuperate full potential of functional recovery or was likely to fail to manage satisfactorily at home despite conventional community support and, therefore, would be at risk of hospital re‐admission or institutionalisation. Exclusion criteria: implicit by contradicting the above
Interventions	1. Referral to a supported discharge team (SDT). Post‐acute home‐based rehabilitation programme delivered by trained healthcare assistants, up to 4 times a day, 7 days a week, under the guidance of registered nurses, allied health, and geriatricians for up to 6 weeks. 2. Usual care
Outcomes	Follow‐up at 4 and 12 months Primary outcome: number of days in hospital during the index admission Secondary outcomes: readmission measured by days in hospital over 12 months; functional capability ascertained via the interRAI Contact Assessment (interRAI‐CA) which assesses multiple domains of function; ADL self‐performance scale (domains of bathing, personal hygiene, dressing lower body, locomotion and toilet use); and IADL capacity scale (meal preparation, ordinary housework, managing medications, and stairs). Healthcare costs.
Notes	Trial report was in press in 2019 and published in 2020; we did not update study ID. Mixed diagnosis trial; separate data not published for hip fracture patients, whose numbers differed in the two groups: 126 (62.4%) versus 109 (54.0%).

Wu 2019.

Methods	Randomised trial: "randomly divided"
Participants	157 "elderly" patients with hip fractures (femoral neck and trochanteric) admitted into their department [hospital in China]; study dates: 1 September 2015 to 31 December 2017
Interventions	1. multidisciplinary team collaboration 2. traditional treatment
Outcomes	Follow‐up at 3 and 6 Harris hip function score at 3 and 6 months Treatment times: preoperative length of stay, total hospitalisation time and the time of antibiotic use Off‐bed activity time Postoperative complications Total costs of treatment and costs for bed, laboratory examinations, and nursing
Notes	Limited information available from abstract at the 8th Chinese Congress on Gerontology and Health Industry, CCGI 2019. China

OT: occupational therapist
PT: physical therapist 
 

Differences between protocol and review

Differences in version published 2021

The changes to protocol for this update (2021) resulted from updated methodology; developments in the classification of different types of multidisciplinary rehabilitation and related emerging themes and concepts; and in response to guidance resulting from the prioritisation process underpinning the Cochrane Programme Grant on the management of hip fracture.

Participants

We clarified that we would have excluded trials focused on the rehabilitation of people with metastatic disease or those on people with high‐energy fractures, such as from motor vehicle crashes.

Interventions

We developed the classification of different types of multidisciplinary rehabilitation further in Description of the intervention, and enhanced our description of the interventions in the Characteristics of included studies table. We relaxed the criteria for multidisciplinary care to explore the inclusion of trials that nonetheless featured a strong component of geriatrician or rehabilitation physician involvement. We included comparisons of different models of multidisciplinary rehabilitation. We also examined 'usual care' and the common care provided, especially in terms of their representativeness of current management, such as early surgery and early mobilisation.

Outcomes

We restructured the types of outcome measures into main or 'critical' outcomes for presentation in summary of findings tables, other important clinical outcomes and economic outcomes. With the exception of treatment adherence, all of the outcomes collected in previous versions still feature (see Appendix 3). Previously, our primary focus was on one‐year follow‐up, which reflected that recovery from hip fracture is a protracted process. However, there is increasing loss to follow‐up over the year and some evidence of consistency between quality of life and 'poor outcome' (death or deterioration in residential status) at 4 and 12 months (Griffin 2015). Hence, a key shift in our reporting is the greater onus on interim outcome at around four months for quality of life and function outcomes, but particularly for trials conducted in an inpatient setting.

For the subgroup of trials testing ambulatory rehabilitation only, we also included number of fallers, number of fall‐related hospital admissions and new fractures. This was prompted by considerations of a potential change of emphasis in rehabilitation conducted outside the inpatient setting.

The selection of outcomes was informed by those listed in the commissioning brief generated from the Cochrane Programme Grant on the management of hip fracture (see Appendix 2).

Data collection and analysis

We updated data collection and analysis to be consistent with methodology presented in Higgins 2011. However, we framed GRADE assessment and production of the summary of findings tables in terms of 'certainty', rather than 'quality', of the evidence. A key change related to Subgroup analysis and investigation of heterogeneity, with the clear adoption of better‐defined intervention categories. In the Sensitivity analysis section, we added exploration of pooling using random‐effects compared with fixed‐effect models, and the effects of excluding trials unrepresentative of current interventions and care programmes.

When pooling the data in a prespecified exploratory analysis from both settings, we restricted data from the ambulatory setting to supported discharge and home‐based rehabilitation.

Differences in version published 2009

We renamed 'outpatient' rehabilitation as 'ambulatory' rehabilitation because the latter term better covers the various settings: home, outpatient department or day hospital.

We clarified that supervision of multidisciplinary care, particularly in the ambulatory setting, could be by a specialist clinician other than a physician.

We assessed risk of bias using the tool outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2008).

Contributions of authors

The protracted nature of producing this update means that authors contributed to different extents to the key tasks at different stages. All five authors participated in study screening, selection, assessed risk of bias and extracted data from trial reports, and provided critical feedback on drafts. Ian Cameron (IC) and Helen Handoll (HH) co‐ordinated the production of the review, contacted trialists and, together with Jenson Mak (JM), revised the protocol. With regular discussion with IC on decisions taken, HH completed the first draft, including data entry and data analyses in Review Manager 5 (Review Manager 2014). Ian Cameron and Helen Handoll are the guarantors of the review.

Contributions of the authors to the protocol and original review are provided in Handoll 2009.

Sources of support

Internal sources

• University of Teesside, Middlesbrough, UK

• John Walsh Centre for Rehabilitation Research, Kolling Institute, The University of Sydney, Australia, Australia

• University of Edinburgh, Edinburgh, UK

External sources

• No sources of support provided

Declarations of interest

None of the authors have a conflict of interest. As Ian Cameron was an investigator in three of the included trials, these trials were assessed independently by other review authors.

New search for studies and content updated (conclusions changed)