Interventions for improving the appropriate use of imaging in people with musculoskeletal conditions

Abstract

Background

Imaging is commonly performed for musculoskeletal conditions. Identifying interventions to improve the appropriate use of imaging for musculoskeletal conditions could potentially result in improved health outcomes for patients and reduced health care costs.

Objectives

To determine the effects of interventions that aim to improve the appropriate use of imaging for people with musculoskeletal conditions.

Search methods

We searched the Cochrane Effective Practice and Organisation of Care Group Specialised Register (June 2007), The Cochrane Central Register of Controlled Trials (The Cochrane Library 2007, Issue 2), MEDLINE (January 1966 to June 2007), EMBASE (January 1980 to June 2007) and CINAHL (January 1982 to June 2007). We also searched reference lists of included studies and relevant reviews. We undertook citation searches of all included studies, contacted authors of included studies, and contacted other experts in the field of effective professional practice.

Selection criteria

Randomised controlled trials, non‐randomised controlled clinical trials and interrupted time‐series analyses that evaluated interventions designed to improve the use of imaging for musculoskeletal symptoms.

Data collection and analysis

Two review authors independently assessed risk of bias and extracted data. We contacted study authors for additional information.

Main results

Twenty eight studies met our inclusion criteria. The majority of the studies were for the management of osteoporosis or low back pain, and most evaluated interventions aimed at health professionals. To improve the use of imaging in the management of osteoporosis, the effect of any type of intervention compared to no‐intervention controls was modest (absolute improvement in bone mineral density test ordering +10%, IQR 0.0 to +27.7). Patient mediated, reminder, and organisational interventions appeared to have most potential for improving imaging use in osteoporosis. For low back pain studies, the most common intervention evaluated was distribution of educational materials and this showed varying effects. Other interventions in low back pain studies also showed variable effects. For other musculoskeletal conditions, distribution of educational materials, educational meetings and audit and feedback were not shown to be effective for changing imaging ordering behaviour. Across all conditions, increasing the number of intervention components did not increase effect.

Authors' conclusions

For improving the use of imaging in osteoporosis, most professional interventions demonstrated benefit, and patient mediated, reminder, and organisational interventions appeared to have most potential for benefit. For low back pain studies interventions showed varying effects. For other musculoskeletal conditions, no firm conclusions can be drawn.

Plain language summary

Interventions for improving the appropriate use of imaging in people with musculoskeletal conditions

Imaging is the production of a clinical image of the human body using medical techniques such as x‐rays, ultrasound, computed tomography (CT) scan and magnetic resonance (MRI). Imaging is commonly performed for musculoskeletal conditions and is an important aspect of the management of these conditions. In some instances, imaging may not be appropriate, for example x‐rays for acute low back pain, and in other instances, imaging is under utilised, for example bone mineral density testing to diagnose people at risk for osteoporosis. The identification of interventions that improve the appropriate use of imaging (either decreasing inappropriate use or increasing appropriate use) for musculoskeletal conditions would be of great value, potentially resulting in improved health outcomes for patients and reduced healthcare costs. The aim of this systematic review was to identify those interventions that improve appropriate use of imaging and to quantify their effects.

Twenty eight studies met our inclusion criteria. The majority of studies evaluated interventions designed to change health professional behaviour, for example, the distribution of educational materials, reminders to health professionals and patient education. For improving the use of imaging in osteoporosis, most interventions aimed at health professionals demonstrated benefit, and patient mediated, reminder, and organisational interventions appeared to have most potential for benefit. For low back pain studies, the most common intervention evaluated was distribution of educational materials and this showed varying effects. Other interventions in low back pain studies showed variable effects. For other musculoskeletal conditions, distribution of educational materials, educational meetings and audit and feedback were not shown to be effective for changing imaging ordering behaviour. Across all conditions, increasing the number of intervention components did not result in a larger effect of interventions.

Background

Imaging is commonly performed for musculoskeletal conditions and is an important aspect of the management of these conditions. Whether or not imaging is appropriate varies for different musculoskeletal conditions; in some conditions the evidence‐based recommended practice is to undertake imaging, however for other conditions imaging may not be recommended. For example, current evidence concludes that lumbar spine plain x‐rays for uncomplicated acute low back pain are not indicated and this is the recommendation of many back pain clinical guidelines (van Tulder 2004). However, contrary to this, many studies have demonstrated an over utilisation of lumbar spine plain x‐rays for this indication (Freeborn 1997; Carey 1996; Isaacs 2004). In other instances, imaging is under utilised. For example, in people with fracture due to minimal trauma or long‐term use of glucocorticoids, bone mineral density testing is indicated to diagnose osteoporosis, but there is evidence that this test is being under utilised (Feldstein 2003).

Inappropriate overuse and under use of imaging in musculoskeletal disorders has a number of potential adverse outcomes. There may be poorer health outcomes due to inefficient diagnosis and subsequent inappropriate management and also a significant impact on health service resource utilisation. In addition, inappropriate use of imaging may result in unnecessary radiation exposure of patients. For example, lumbar spine radiographs result in one of the highest cumulative doses of radiation to the reproductive organs of any radiological study (Rosenstein 1988). In the case of not performing imaging when it is recommended, for example, a woman may sustain a hip fracture because decreased bone density was not detected and subsequent preventive management was not instituted.

Many different interventions can be used to target improved imaging ordering behaviour of healthcare professionals. The Cochrane Effective Practice and Organisation of Care (EPOC) Group (EPOC 2007) has classified these intervention types with a taxonomy of professional, organisational, financial and regulatory interventions. Table 1 provides details of the relevant interventions for this review from the EPOC taxonomy. Two previous systematic reviews have examined the effect of various interventions directed at health professionals to improve the use of bone mineral density test ordering in osteoporosis (Elliot‐Gibson 2004; Morris 2004). However, the authors of these reviews were unable to determine which interventions were most effective. We are unaware of any other systematic reviews that examine the use of interventions to improve the appropriate use of imaging in musculoskeletal conditions.

1. Classification of relevant interventions from EPOC taxonomy.

Intervention	Description
Professional:
Audit and feedback	Any summary of clinical performance of health care over a specified period of time.  The summary may also have included recommendations for clinical action.  The information may have been obtained from medical records, computerised databases, or observations from patients.
Distribution of educational materials	Distribution of published or printed recommendations for clinical care, including CPGs, audiovisual materials and electronic publications. The materials may have been delivered personally or through mass mailings.
Educational meetings	Health care providers participating in conferences, lectures, workshops or traineeships.
Educational outreach visits	Use of a trained person who met with providers in their practice settings to give information with the intent of changing the provider’s practice.  The information given may have included feedback on the performance of the provider(s).
Reminders	Patient or encounter specific information, provided verbally, on paper or on a computer screen, which is designed or intended to prompt a health professional to recall information.  This would usually be encountered through their general education, in the medical records or through interactions with peers, and so remind them to perform or avoid some action to aid individual patient care.  Computer aided decision support and drugs dosage are included.
Other	Other professional intervention categories to be agreed in consultation with the EPOC editorial team, e.g. patient mediated interventions.
Financial:
Provider incentives	Provider received direct or indirect financial reward or benefit for doing a specific action.
Organisational:
Case management	Including co‐ordination of assessment, treatment and arrangement for referrals.
Clinical multidisciplinary team	Creation of a new team of health professionals of different disciplines or addition of new team members who work together to care for patients.

The identification of interventions that improve the appropriate use of imaging for musculoskeletal conditions would be of great value, potentially resulting in improved health outcomes for patients and reduced health care costs. The aim of this systematic review was to identify those interventions that improve appropriate use of imaging and to quantify their effects.

Objectives

To determine the effects of interventions that aim to improve the appropriate use of imaging for musculoskeletal conditions. The following questions were addressed.

1. What are the effects of interventions that aim to improve the appropriate use of imaging for people with musculoskeletal conditions (on all outcomes including use of imaging, cost, and clinical outcomes)?

2. Do modifications in the characteristics of the interventions (for example, timing, content, etc) modify the effects?

3. Do the effects of interventions vary depending on the setting, type of professionals targeted, type of behaviour (decreasing use, increasing use or modifying use) or whether or not the intervention is based on theory?

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs), non‐randomised controlled clinical trials (CCTs) and interrupted time‐series (ITS) studies.

Types of participants

We included studies of interventions to improve the use of imaging for musculoskeletal symptoms, such as neck pain, back pain and other regional pain, possible or known arthritis (including osteoarthritis, rheumatoid arthritis and spondylo‐arthropathies) or osteoporosis. We excluded studies where the people included had musculoskeletal injuries and trauma.

Interventions to improve the appropriate use of imaging may have been directed toward health professionals, policy‐makers and the general public, or a combination of these groups of people.

Types of interventions

We included studies that evaluated any intervention designed to improve the appropriate use of imaging for diagnostic, screening or monitoring purposes, including assessing the response to treatment. We considered any type of educational, organisational, financial or regulatory intervention. Studies were only included where the primary objective, or one of the primary objectives, was to determine the efficacy of an intervention to improve the appropriate use of imaging. Specifically, we excluded studies that were designed to examine the accuracy of imaging as a diagnostic test.

We classified interventions implemented in the included studies according to the taxonomy of professional, organisational, financial and regulatory interventions developed by the Cochrane Effective Practice and Organisation of Care (EPOC) Group (EPOC 2007). Table 1 provides details of the relevant interventions for this review from the EPOC taxonomy. Under the EPOC category of "Professional Interventions", we added an intervention type of "Patient Mediated". A patient mediated intervention is one where the intervention is directed at the patient with the aim of changing the behaviour of the patient's healthcare provider, for example patient education (materials or verbal).

Types of outcome measures

The outcomes considered for inclusion in this review were related to the objective of this review to determine the effects of interventions that aim to improve the appropriate use of imaging for musculoskeletal conditions. We thus included studies that used a valid method to measure provider performance of appropriate use of imaging, for example, number of people referred for x‐ray, or clinical outcomes, or both.

Primary outcomes

(a) Provider performance of appropriate use of imaging, for example, number of people referred for x‐ray. 
 (b) Clinical outcomes (for example fracture or pain).

When either of the primary outcomes have been reported, then the secondary outcomes were reported if available.

Secondary outcomes

(a) Health practitioner intentions about imaging behaviour. 
 (b) Health practitioner or general population knowledge and attitudes about imaging. 
 (c) Organisational or process level outcomes, for example, imaging costs, measurements of usability or extent of the intervention.

All time points of the outcomes reported were included.

Search methods for identification of studies

See: Effective Practice and Organisation of Care Group methods used in reviews.

The following electronic databases were searched: 
 (a) The EPOC Specialised Register (and the database of studies awaiting assessment) was reviewed in June 2007 (see SPECIALISED REGISTER under GROUP DETAILS) 
 (b) The Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2007, Issue 2) 
 (c) MEDLINE (January 1966 to June 2007), EMBASE (January 1980 to June 2007) and CINAHL (January 1982 to June 2007).

The search strategy for the MEDLINE database is shown in Appendix 1. The search strategy shown in the Appendix was added to the standard EPOC search terms (see: Effective Practice and Organisation of Care Group methods used in reviews) in order to capture relevant studies. This search strategy was adapted for the other electronic databases searched (CENTRAL, EMBASE, CINAHL and the EPOC Specialised Register) using the appropriate controlled vocabulary where applicable.

In addition to the electronic searches, we searched reference lists of included studies and relevant reviews identified, we undertook citation searches of all included studies, we contacted authors of included studies regarding any further published or unpublished work, and we contacted other experts in the field of effective professional practice for any relevant studies that they may be aware of.

Data collection and analysis

Selection of studies

Two review authors (SF and HB) independently screened all titles and abstracts of retrieved studies. We ordered the full text of studies that met the initial selection criteria or those for which we could not determine eligibility from the title and abstract. SF and HB then independently assessed the full text for inclusion status. Any disagreements on inclusion status were resolved with discussion with the other two review authors (SG and RB). If any studies were located in the search strategy that were carried out by one of the authors of this review, the paper was sent to an independent review author to assess whether it met the inclusion criteria for this review.

Data extraction and management

SF undertook data abstraction of each of the included studies using a modified EPOC Data Collection Checklist. The data extraction form was pilot tested on two included studies to minimise misinterpretation. A second review author (HB) independently assessed the methodological quality and extracted results of included studies as recommended by the EPOC group (Bero 2008). When necessary, we approximated results from graphical representations in the reports of the included studies. If SF and HB disagreed in their assessment of methodological quality or data extraction, this was resolved by discussion and consensus, followed by discussion with a third review author if necessary.

Assessment of risk of bias in included studies

We used two methods to undertake risk of bias assessment of the included studies; the EPOC recommended method (Bero 2008) and the method outlined in the Cochrane Handbook for Systematic Reviews of Interventions (version 5.0.0). In order to be consistent, for included parallel group trials (cluster RCTs, individual RCTs and non‐randomised CCTs) we renamed the EPOC classification for the overall quality rating from ‘protection against bias’ to ‘risk of bias’, and assigned an overall quality rating of ‘high’, ‘moderate’ or ‘low’ risk of bias (criteria as outlined in the protocol). As planned, we undertook a separate assessment of interrupted time series (ITS) studies according to the EPOC criteria, but did not assign an overall rating for these.

The Cochrane Handbook risk of bias assessment was not proposed in the published review protocol because this methodology was not developed at that time. We carried out this assessment for parallel group trials using the criteria: (1) sequence generation; (2) allocation concealment; (3) blinding; (4) incomplete outcome data; (5) selective outcome reporting; and, (6) other potential sources of bias, for example, protection against contamination of the intervention or significant baseline differences between groups. Each of the risk of bias criteria were judged as ‘Yes’ (indicating low risk of bias), ‘No’ (indicating high risk of bias), or ‘Unclear’ (indicating either lack of information or uncertainty over the potential for bias).

Measures of treatment effect

Although we planned to pool the results of studies if possible, we were unable to undertake any formal meta‐analysis because no two studies were sufficiently similar in terms of condition studied, interventions evaluated and outcomes measured. Hence, we used statistical methods proposed by Grimshaw and colleagues to guide our data analysis and presentation (Grimshaw 2004).

We reported the primary results for each study in natural units extracted from the reports of included studies. We attempted to present a consistent method of reporting results of included studies by calculating standardised effect sizes. In cases where there were insufficient data to calculate effect sizes, we presented results of studies in the form reported by the original study authors. The desired direction of effect differed between studies. Some aimed for an increase in a behavioural outcome (for example, increase in bone mineral density (BMD) test ordering for suspected osteoporosis), while others aimed for a decrease in a behavioural outcome (for example, decreased ordering of lumbar x‐rays for low back pain). To avoid confusion, we have standardised the effect size so that a positive difference between post‐intervention percentages represents an increase in compliance with appropriate use of imaging, regardless of the desired direction of effect.

For studies reporting dichotomous outcomes we calculated a standardised effect size, which is the difference between the intervention group and the control group post‐intervention percentage compliance with evidence‐based recommended imaging. If baseline levels were available, we calculated an absolute adjusted risk difference (ARD) which adjusts for important differences in baseline measures between groups. We defined an important difference in baseline levels where the absolute difference between the groups before the intervention was more than 10%, or where the baseline difference was similar to that found post‐intervention.

We calculated the ARD as follows:

Adjusted risk difference (ARD) = [risk of non‐compliance (intervention control) pre‐intervention] minus [risk of noncompliance (intervention control) post‐intervention]

A positive ARD means that compliance improved more in the intervention group than in the control group, for example an ARD of 0.10 indicates an absolute improvement in appropriate imaging of 10%. This method of calculating the ARD has been used previously in Cochrane reviews (Doumit 2007; Jamtvedt 2006), and is adopted by the Cochrane EPOC group as recommended methodological practice. For studies that reported more than one outcome measure related to musculoskeletal imaging, we calculated the ARD for all musculoskeletal imaging outcomes and then identified the value that represented the median ARD between measures (Grimshaw 2004).

For studies reporting continuous outcomes we calculated the relative percentage change, that is, the percent improvement relative to the post intervention average in the control group. We also calculated standardised mean differences (SMD), as defined by EPOC, for continuous outcomes where possible. We calculated the SMD as follows:

SMD = [mean of intervention group post‐intervention minus mean of control group post‐intervention] divided by control standard deviation pre‐intervention

In many previously published cluster RCTs, practitioners or practices were randomised but the statistical analyses were of individual patient data. Analysis of a cluster RCT that does not account for the clustering effect of patients attending the same practitioner or practice has what is termed a ‘unit of analysis error’ (UAE) (Ukoumunne 1999a). In these studies, the point estimate of effect is unlikely to be biased, but confidence intervals are likely to be overly narrow and P values are likely to be artificially low, increasing the chance of misleading conclusions due to spurious statistically significant findings (Ukoumunne 1999). If we judged a cluster RCT included in this review as having a UAE, we reported the relevant point estimate of the study but not the statistical significance, and indicated in the data tables that a UAE was possible.

In many previously published interrupted time series (ITS) studies, inappropriate statistical analyses were performed leading to spurious statistical significance (Ramsay 2003). When we judged that incorrect analysis of an included ITS study had been performed, we reanalysed ITS comparisons using time‐series regression methods as recommended by Ramsay and colleagues (Ramsay 2003) and calculated, where possible, the difference in slopes (pre‐ and post‐intervention) or a level effect, or both. Data were derived from tables of results or from graphics presented in the original studies. Where information on individual values was reported graphically, we derived the data by importing a digital image of the graph into a graphics software package and measuring each data point manually by applying grid lines and reading off the corresponding values. 

We reported the synthesis of results of the included studies classified by condition (osteoporosis, low back pain, knee pain, other musculoskeletal conditions) in the following structured format:

1. Evaluations of interventions against ‘no‐intervention’ control groups

2. Evaluations of interventions against ‘other intervention’ control groups

To provide a description of the size of effect (absolute difference across post‐intervention measures) for imaging dichotomous measures, the following terms were used, suggested by Grimshaw and colleagues (Grimshaw 2004):

• ‘small’ to describe effect sizes ≤ 5%

• ‘modest’ to describe effect sizes > 5% and ≤ 10%

• ‘moderate’ to describe effect sizes >10 and ≤ 20%

• ‘large’ to describe effect sizes >20%.

We planned to explore heterogeneity using the following subgroup analyses:

1. Type of intervention (educational, organisational, financial, regulatory)

2. Type of health professional (medical practitioner, dentist, nurse, allied health professional)

3. Musculoskeletal condition under study

4. Whether the intervention was based on theory or not

Results

Description of studies

Selection of studies for inclusion

Figure 1 describes the process from searching to study inclusion. We identified a total of 4072 non‐duplicate potentially relevant citations from electronic databases (MEDLINE, EMBASE, CINAHL, CENTRAL and the EPOC Register). Reference lists and contact with experts yielded an additional 22 potentially relevant citations and citation searches of included studies an additional 25 potentially relevant citations, for a total of 4119 records. After two researchers (SF and HB) independently screened the title and abstract of these studies, we excluded 3923 records and obtained 196 studies for full text review. Twenty eight of these studies met our inclusion criteria and their details are described in the Characteristics of included studies table. Twelve (43%) of the studies included in this review were not identified by our search of electronic databases, highlighting the need to use multiple searching methods when undertaking a review in this area. Studies initially appearing to meet the eligibility criteria but which we subsequently excluded are reported in the Characteristics of excluded studies table along with our first reason for exclusion. Three ongoing studies were identified; one from a published trial protocol (Bessette 2008) and two are being undertaken by some of the review authors (McKenzie 2008). These studies are described in the Characteristics of ongoing studies table.

1.

Characteristics of study design and setting

Sixteen included studies were cluster RCTs, six were individual RCTs, one was a non‐randomised CCT and five were ITS studies. Sixteen of the 22 RCTs were two arm trials and six had more than two arms: two compared three arms and four compared four arms.

All of the included studies were based in developed countries, with 15 based in the USA, seven in the UK, four in Canada and two in The Netherlands. All of the studies evaluating interventions directed at healthcare providers were targeted at medical practitioners, with no other healthcare providers represented, for example allied healthcare providers. Twenty four (86%) of the included studies evaluated interventions delivered in primary care settings alone, one study described an intervention delivered in both primary care and a specialist rheumatology setting, two studies delivered interventions in a hospital setting (one in‐patient care, one out‐patient care), and one study delivered an intervention in a specialist rheumatology setting alone.

Conditions managed and targeted imaging behaviours

Fifteen included studies involved the management of osteoporosis, or patients at risk of osteoporosis, and the targeted imaging behaviour was to increase BMD test ordering. Six studies involved the management of low back pain; in four of these studies the targeted behaviour was the reduction of ordering of lumbar x‐rays and in two the targeted behaviour was the reduction of ordering of all types of imaging of the lumbar spine (that is, CT scan, MRI, as well as plain x‐rays). Two studies involved the management of low back pain and knee pain; the intervention in one of these studies targeted the behaviour of a reduction of ordering lumbar and knee x‐rays and the other study targeted the reduction of lumbar and knee MRI. The remaining five included studies involved the management of multiple conditions and the targeted behaviour in all of these was reduced ordering of different types of imaging (for example, plain x‐rays of the chest, spine, limbs and joints; abdominal, renal, pelvic and testicular ultrasound examinations).

Characteristics of interventions delivered

The 28 studies evaluated 36 separate comparisons covering a wide range of intervention types, including professional, organisational and patient mediated interventions. Table 2 summarises the different types of interventions evaluated in each study. Twenty three studies evaluated professional interventions alone, two studies evaluated organisational interventions alone, one study evaluated a professional intervention in combination with an organisational and a financial intervention, and two studies evaluated a combination of professional and organisational interventions. More detail of the interventions delivered in each study is provided in the Characteristics of included studies table.

2. Intervention types used in each study.

Intervention category1	Intervention method2	No. of studies	Studies3
Professional	Distribution of educational materials	20	Curtis 2007; Dey 2004; Eccles 2001; Feldstein 2006; Hollingworth 2002; Jackson 2005; Kerry 2000; Majumdar 2004; Majumdar 2008; Matowe 2002; Oakeshott 1994; Prihar 2008; Robling 2002; Rozental 2008; Schectman 2003; Solomon 2004; Solomon 2007; Solomon 2007a; Verstappen 2003
Professional	Reminders	10	Baker 1987; Boyd 2002; Eccles 2001; Feldstein 2006; Feldstein 2007; Lafata 2007; Majumdar 2004; Majumdar 2008; Solomon 2007a; Stock 1998
Professional	Audit and feedback	8	Curtis 2007; Eccles 2001; Kerry 2000; Robling 2002; Schectman 2003; Solomon 2004; Verstappen 2003; Winkens 1995
Professional	Educational meetings	4	Robling 2002; Schectman 2003; Solomon 2004; Verstappen 2003
Professional	Educational outreach visits	3	Dey 2004; Solomon 2007; Solomon 2007a
Professional	Patient mediated4	7	Feldstein 2006; Gardner 2005; Lafata 2007; Majumdar 2004; Majumdar 2008; Prihar 2008; Schectman 2003; Solomon 2007; Solomon 2007a
Financial	Provider incentives	1	Feldstein 2007
Organisational	Case management	4	Dey 2004; Feldstein 2007; Majumdar 2007; Rozental 2008
Organisational	Clinical multi‐disciplinary team	1	Rossignol 2000

1. Category of intervention as classified by the EPOC taxonomy EPOC 2007
2. See Table 1 for definition of each intervention
3. Some studies used more than one intervention type and are listed against each category
4. Intervention directed at the patient by way of patient education (materials or verbal) with the aim of changing the behaviour of the patient’s healthcare provider

Studies comparing an intervention to 'no‐intervention' control

Twenty‐two of the included studies (representing 26 comparisons) evaluated an intervention versus a 'no‐intervention' or 'usual care' control, and Table 3 describes the different components of these interventions. Distribution of educational materials was the most frequent single intervention evaluated against a no‐intervention control, included in four comparisons, followed by patient‐mediated interventions included in three comparisons, and reminders used in two comparisons. Distribution of educational materials was also the intervention strategy used most frequently as part of multifaceted interventions compared to a no‐intervention control, followed by patient mediated interventions and audit and feedback.

3. Intervention combinations compared to a no‐intervention control group.

Intervention combinations	No. of comparisons	Study ID
Single component interventions:
Audit and feedback	1	Winkens 1995
Distribution of educational materials	4	Hollingworth 2002, Jackson 2005, Matowe 2002, Oakeshott 1994
Patient mediated	3	Gardner 2005, Lafata 2007, Solomon 2007
Reminder	2	Baker 1987, Stock 1998
Organisational	1	Rossignol 2000
Multifaceted interventions:
Two intervention components
Audit and feedback + distribution of educational materials	2	Curtis 2007, Kerry 2000
Distribution of educational materials + educational outreach	1	Solomon 2007
Distribution of educational materials + patient mediated	1	Prihar 2008
Distribution of educational materials + reminder	1	Feldstein 2006
Patient mediated + organisational	1	Majumdar 2007
Patient mediated + reminder	1	Lafata 2007
Three intervention components:
Audit and feedback + distribution of educational materials + educational meeting	2	Solomon 2004, Verstappen 2003
Distribution of educational materials + educational outreach + organisational	1	Dey 2004
Distribution of educational materials + educational outreach + patient mediated	1	Solomon 2007
Distribution of educational materials + patient mediated + reminder	2	Majumdar 2004, Feldstein 2006
Organisational + reminder + financial	1	Feldstein 2007
Four intervention components:
Distribution of educational materials + educational outreach + patient mediated + reminder	1	Solomon 2007a

Studies that compared an intervention to a different intervention control

Five included studies (representing ten comparisons) evaluated interventions versus a control comprising a different intervention type and Table 4 describes these interventions. Two of these interventions were single component interventions and there were seven different combinations of multifaceted interventions. All of the comparison interventions were single component interventions.

4. Intervention combinations compared to controls receiving a different intervention.

Study intervention combinations	Control interventions	No. of comparisons	Study ID
Single component interventions
One reminder type	A different reminder type	1	Boyd 2002
Organisational	Distribution of educational materials	1	Rozental 2008
Multifaceted interventions
Two intervention components:
Audit and feedback + distribution of educational materials	Distribution of educational materials	2	Eccles 2001, Robling 2002
Distribution of educational materials + educational meeting	Distribution of educational materials	1	Robling 2002
Distribution of educational materials + reminder	Distribution of educational materials	1	Eccles 2001
Three intervention components:
Audit and feedback + educational meeting + distribution of educational materials	Distribution of educational materials	1	Robling 2002
Audit and feedback + distribution of educational materials + reminder	distribution of educational materials	1	Eccles 2001
Distribution of educational materials + patient mediated + reminder	Patient mediated	1	Majumdar 2008
Four intervention components:
Audit and feedback + distribution of educational materials + educational meeting + patient mediated	Patient mediated	1	Schectman 2003

Intervention rationale and fidelity

Only a few of the included studies reported a rationale for the choice of intervention evaluated. Nineteen (68%) provided no rationale. Four studies included reference to previous studies of effectiveness of interventions as the only rationale for the intervention chosen (Feldstein 2006; Lafata 2007; Solomon 2004; Verstappen 2003). Three studies reported that the intervention was designed to address barriers to change (Lafata 2007; Majumdar 2007; Solomon 2007), but only limited information was provided of the methods used to measure these barriers, what the barriers were, and how the interventions were matched to these barriers. In one study, reference was made to interviews conducted with healthcare professionals to direct the choice of the intervention evaluated (Majumdar 2007), however details of the results of these interviews were not reported and no related publication could be identified. Four studies indicated that the intervention, or a component of the intervention, was based on theory (Curtis 2007; Dey 2004; Solomon 2007; Solomon 2007a) (details of the reported theories are provided in the Characteristics of included studies table).

Twenty three (82%) of the included studies provided no details on whether or not intervention fidelity was evaluated, that is, whether or not the intervention was implemented as planned. Five studies provided limited information on intervention fidelity (Curtis 2007; Eccles 2001; Solomon 2004; Solomon 2007; Solomon 2007a). Any reported fidelity assessment is described in the Characteristics of included studies table.

Cost

Only three included studies considered economic outcomes (Majumdar 2007; Majumdar 2008; Robling 2002). Two of these studies reported a cost of the intervention per patient (in a sample of patients) (Majumdar 2007; Majumdar 2008) and the other study reported cost of the intervention per health practitioner (Robling 2002).

Risk of bias in included studies

Our assessment of risk of bias according to the Cochrane Handbook for Systematic Reviews of Interventions for each of the parallel group trials can be found in the Characteristics of included studies table. Figure 2 demonstrates a graphical representation of the judgements about each of the risk of bias items, presented as percentages across all included studies. ITS studies are represented by the blank spaces on the graph because these studies were not judged according to these criteria. Just under half of the included trials addressed incomplete outcome data and reported adequate sequence generation. Blinded outcome assessment was undertaken in approximately one quarter of the trials, approximately one fifth reported adequate allocation concealment and one quarter were judged to be free of other bias. Only approximately 10% of the trials were judged to be free of selective reporting.

2.

Methodological quality graph: review authors' judgements about each methodological quality item presented as percentages across all included studies.

ITS studies were not judged by this risk of bias assessment method and are represented by the blank space on each bar. See Table 6 for risk of bias assessment of ITS studies

Our assessment of methodological quality according to the EPOC method of each of the RCTs and the CCT is outlined in Table 5, with studies ordered by the degree of risk of bias. We judged four studies to be 'low' risk of bias, 16 'moderate' risk of bias and three 'high' risk of bias.

5. Methodological quality of controlled trials according to the EPOC method.

Study ID	Allocation concealment	Blinded assessment	Follow‐up	Baseline measure	Reliable primary outcome	Protection against contamination	Summary*
Eccles 2004	DONE	DONE	DONE	ND	NC	DONE	Low
Majumdar 2007	DONE	DONE	DONE	DONE	NC	ND	Low
Majumdar 2008	DONE	DONE	DONE	DONE	NC	DONE	Low
Oakeshott 1994	DONE	DONE	DONE	DONE	NC	DONE	Low
Boyd 2002	DONE	NC	ND	NC	NC	DONE	Moderate
Dey 2004	DONE	ND	DONE	DONE	NC	DONE	Moderate
Feldstein 2006	NC	DONE	DONE	DONE	NC	ND	Moderate
Gardner 2005	NC	NC	DONE	NC	NC	NC	Moderate
Kerry 2000	DONE	NC	DONE	DONE	NC	DONE	Moderate
Lafata 2007	DONE	NC	DONE	NC	NC	DONE	Moderate
Majumdar 2004	ND	DONE	DONE	ND	NC	NC	Moderate
Prihar 2008	DONE	ND	NC	DONE	NC	NC	Moderate
Robling 2002	DONE	NC	NC	DONE	NC	DONE	Moderate
Rossignol 2000	NC	DONE	NC	ND	NC	ND	Moderate
Rozental 2008	NC	NC	DONE	NC	NC	ND	Moderate
Schectman 2003	DONE	NC	ND	ND	ND	DONE	Moderate
Solomon 2007	DONE	NC	DONE	DONE	NC	DONE	Moderate
Stock 1998	NC	NC	DONE	NC	NC	NC	Moderate
Verstappen 2003	DONE	NC	DONE	DONE	NC	DONE	Moderate
Winkens 1995	DONE	NC	NC	DONE	NC	DONE	Moderate
Curtis 2007	NC	NC	DONE	DONE	NC	DONE	High
Solomon 2004	NC	NC	NC	DONE	NC	NC	High
Solomon 2007a	NC	NC	NC	DONE	NC	NC	High

* Low risk of bias if the first three criteria are scored as DONE; moderate if one or two of the first three criteria are scored as NOT CLEAR (NC) or NOT DONE (ND); and high if more than two of the first three criteria are scored as NOT CLEAR or NOT DONE

The assessment of methodological quality for the included ITS studies is outlined in Table 6. In none of the ITS studies was the intervention clearly independent of other changes, that is, it was not clear whether other events during the study period could have affected the results. In two out of the five included ITS studies was an appropriate statistical analysis undertaken. None of the reports of these studies described the reason for the number of pre‐ and post‐intervention data points measured. Only two studies reported that blinded assessment of outcomes was undertaken. No overall score is calculated for ITS studies, however considering that the majority of the quality items were judged as 'not clear' or 'not done', we judged the overall quality of the included ITS studies as poor.

6. Methodological quality of interrupted time‐series studies according to the EPOC method.

Study	1	2	3	4	5	6	7	8
Baker 1987	ND	ND	ND	ND	DONE	ND	NC	ND
Feldstein 2007	NC	DONE	NC	DONE	DONE	NC	DONE	NC
Hollingworth 2002	NC	ND	NC	NC	DONE	DONE	NC	NC
Jackson 2005	NC	ND	NC	ND	DONE	DONE	NC	NC
Matowe 2002	NC	DONE	NC	DONE	DONE	NC	NC	NC

NC: Not clear; ND: Not done.

1. The intervention is independent of other changes
2. Data were analysed appropriately
3. Reason for the number of points pre‐ and post‐intervention given
4. Shape of the intervention effect was specified
5. Intervention unlikely to affect data collection
6. Blinded assessment of primary outcome(s)
7. Completeness of data set
8. Reliable primary outcome measure(s).

Note: no summary score is calculated for ITS studies.

Effects of interventions

We were unable to undertake any formal meta‐analysis because no two studies were sufficiently similar in terms of condition studied, interventions evaluated and outcomes measured. We have presented some of the results as forest plots to allow visual interpretation of the size of effect for the comparisons, but no pooling of results was undertaken for these. In many studies there was no reporting of baseline imaging performance, and hence for these studies we were unable to calculate ARD for dichotomous measures and SMD for continuous measures (as defined by EPOC; see Data collection and analysis for formulae).

Although originally planned, we did not undertake a meta‐regression analysis to estimate the effects of different interventions. Consistent with the findings of a previous systematic review of guideline implementation strategies, we found the number of different combinations of multifaceted interventions evaluated precluded this type of analysis Grimshaw 2004. In addition, this analysis was not possible due to the number of studies with unit of analysis errors. After further advice from the EPOC group we concluded that this analysis would not be informative above and beyond the methods we were already using. However, we did perform an analysis of the ‘dose‐response’ effect of an increasing number of components in multifaceted interventions. For this analysis we could only utilise dichotomous outcomes because data were commonly missing for continuous measures.

We originally planned an examination of whether the effects of interventions varied depending on different study factors. For the factors of type of healthcare professional targeted and the setting of the study these analyses were not possible because only medical practitioners were targeted in all of the included studies, and 25 of the 28 included studies were in a primary care setting, with insufficient studies in the non‐primary care setting to compare these results against. We were able to perform some exploratory analyses of varying effects if the intervention was based on theory, the type of behaviour targeted (decreased or increased use of imaging) and by degree of risk of bias in the study.

Osteoporosis studies

Fifteen studies (involving 19 comparisons) included patients with osteoporosis or people at risk of developing osteoporosis. Seven of these studies were for secondary prevention of osteoporosis, that is, included patients had a pre‐existing fracture, and eight studies were for primary prevention of osteoporosis. Twelve of these studies were conducted in the USA and three in Canada. Eight of these studies were cluster RCTs, five were individual RCTs, one was a non‐randomised CCT and one was an ITS study. Twelve studies were conducted in primary care alone, one was conducted in both primary care and rheumatology practice, one in rheumatology practice alone and one in an inpatient care setting. The desired imaging outcome in all of these studies was an increase in BMD test ordering.

Osteoporosis: evaluations of interventions compared to no‐intervention control groups

Of the studies that included patients with osteoporosis, or people at risk of developing osteoporosis, 12 studies (involving 16 comparisons) evaluated single or multifaceted interventions compared to a no‐intervention control. The majority of these studies evaluated professional interventions and most reported significant improvements in BMD test ordering.

Osteoporosis: results from RCT and CCT comparisons

Dichotomous imaging outcomes: Fourteen comparisons reported dichotomous imaging outcomes, including six cluster RCT comparisons (Curtis 2007; Lafata 2007; Prihar 2008; Solomon 2004; Solomon 2007; Solomon 2007a), three individual RCT comparisons (Feldstein 2006; Gardner 2005; Majumdar 2007) and one CCT comparison (Majumdar 2004). Appendix 2 shows the range of interventions that were evaluated and the imaging outcome data reported in each of these studies. All but one of these comparisons were of professional interventions versus control, with an organisational intervention evaluated in one study (Majumdar 2007).

The effect of any type of intervention versus a no‐intervention control in osteoporosis studies ranged from –1.9% to 51.0%, with a median absolute improvement in BMD test ordering, of any type of intervention to change practice, of +10% (interquartile range (IQR) 0.0 to +27.7). Five of the comparisons (three from one study) reported a non‐significant deterioration in BMD test ordering (Curtis 2007; Solomon 2007), the significance of one comparison was not reported (Gardner 2005) and seven comparisons reported a significant improvement in BMD test ordering (Feldstein 2006; Lafata 2007; Majumdar 2004; Majumdar 2007; Solomon 2007a). The remaining comparison had a potential unit of analysis error (Prihar 2008). Two comparisons reported baseline BMD test ordering levels (Prihar 2008; Solomon 2004). Both were cluster RCT comparisons and the difference in absolute change from baseline was 7.7% improvement in BMD test ordering in one comparison (distribution of educational materials plus patient mediated intervention versus control) and 3.8% deterioration in BMD test ordering in the other comparison (audit and feedback, distribution of educational materials, plus educational meeting versus control).

Continuous imaging outcomes: One comparison of reminders versus a no‐intervention control reported a continuous imaging outcome, the number of BMD tests ordered per month (Stock 1998). Baseline levels were not reported so SMD could not be calculated. The comparison reported a significant improvement in BMD test ordering of +81% (P value = 0.002) with use of a reminder. Appendix 3 shows the data from this study.

Exploratory analysis of different intervention components:Table 7 presents an exploratory analysis of the comparisons of the three most common components included in interventions to improve the use of BMD test ordering in the osteoporosis studies, compared to no‐intervention control groups that reported dichotomous imaging data. These results should be interpreted with caution considering the small number of comparisons available and the indirect nature of the comparisons. The median effect for interventions that included a patient mediated component was a moderate absolute improvement in BMD test ordering, whereas the median effect of those that did not include a patient mediated component was small. Also, six of the seven comparisons that reported a significant improvement in BMD test ordering included a patient mediated component in their intervention whereas only one of the five comparisons that showed a non‐significant effect included a patient mediated component. For reminder interventions, the median effect was a large improvement in BMD test ordering if included, but small if not included. This suggests that the effect was more likely to be positive and larger if patient mediated or reminder interventions were a component of the intervention. When interventions included a component with distribution of educational materials the effect was small and the IQR range extended to both a negative and positive effect. It is not clear if the effect varies whether distribution of educational materials are a component of the intervention or not.

7. Summary of comparisons of most common intervention components in osteoporosis studies reporting dichotomous imaging data.

Intervention	No. of comparisons	Median absolute effect size	Interquartile range	Range
Patient mediated component
Included	10	+12.80	+3.8 to 27.7	–0.2 to 51.0
Not included	4	–0.90	–1.9 to 17.3	–1.9 to 34.6
Reminder component
Included	5	+27.7	+18.1 to 34.6	+3.8 to 44.8
Not included	9	+0.7	–0.2 to 10.6	–1.9 to +51.0
Distribution of educational materials component
Included	9	+3.8	–0.2 to 27.7	–1.9 to 44.8
Not included	5	+15.0	+10.6 to 18.1	+0.7 to 51.0

Osteoporosis: clinical outcomes

Two osteoporosis studies that included participants with wrist or hip fractures reported clinical outcomes at six months (Majumdar 2007; Majumdar 2008) (see data from these studies in Appendix 4 and Appendix 5). One study found no significant change in additional fracture or death (Majumdar 2007), and the other study found no significant change in health status (SF‐12) (Majumdar 2008).

Osteoporosis: cost outcomes

For the two included studies that reported cost outcomes, the organisational intervention cost US$50 per patient (Majumdar 2007) and the professional intervention cost US$9.64 per patient (Majumdar 2008). Neither study reported a cost‐effectiveness analysis.

Osteoporosis: results from ITS comparison

One ITS comparison evaluated a multifaceted intervention consisting of an organisational and reminder intervention, then an additional financial intervention in Phase 2 of the intervention period (Feldstein 2007). The study report provided time‐series regression of the change in slope of improvement of osteoporosis management (composite outcome including BMD testing and osteoporosis medication prescription). The comparison reported a significant improvement, with the probability of appropriate osteoporosis management increasing on average 3.1% (P value <0.001) every 2 months during the post‐intervention period, but reported that the additional financial intervention did not produce added benefit. Reanalysis of the imaging data only showed a significant improvement in the slope for BMD test ordering, but not for change in level, suggesting the intervention had no immediate effect but there was an effect over time. Data from this study are described in Appendix 6.

Osteoporosis: evaluations of interventions compared to another intervention

Three comparisons evaluated interventions compared to another intervention and all outcomes for these were dichotomous. The range of interventions that were evaluated and the data from these studies are provided in Appendix 7. One of the comparisons was a cluster RCT and the other two comparisons were individual RCTs. All three comparisons observed improvements in BMD test ordering compared to the control intervention. The cluster RCT comparison of two types of reminders found an absolute improvement in BMD test ordering of +7.1% (95% confidence interval (CI) –8.0 to 22.0) but had a potential unit of analysis error (Boyd 2002). The other two comparisons were reported as statistically significant. One compared a multifaceted intervention consisting of professional interventions (including a patient mediated intervention) to a patient mediated intervention alone, and found an absolute improvement in BMD test ordering of +34.0% (95% CI 23.0 to 45.0) (Majumdar 2008). The remaining study compared an organisational intervention to distribution of educational materials and found an absolute improvement in BMD test ordering of +62.0% (95% CI 41.0 to 83.0) (Rozental 2008).

Osteoporosis studies: summary

The majority of the 12 studies that evaluated interventions versus no‐intervention controls to improve the management of people with osteoporosis, or people at risk of osteoporosis, observed significant improvements in BMD test ordering. However, it is not clear which intervention, or intervention combination, is likely to have the largest effect. The effect of any type of intervention (consisting primarily of professional interventions) compared to no‐intervention controls was modest (absolute improvement in BMD test ordering +10%, IQR 0.0 to +27.7).

For comparisons of interventions compared to another type of intervention for osteoporosis management, one RCT comparison of a multifaceted intervention consisting of professional interventions (including a patient mediated intervention) versus a patient mediated intervention alone demonstrated a large effect (+34.0%, 95% CI 23.0 to 45.0, absolute improvement in BMD test ordering). Another RCT comparison of an organisational intervention versus distribution of educational materials also demonstrated a large effect (+62.0%, 95% CI 41.0 to 83.0, absolute improvement in BMD test ordering).

Evaluation of individual intervention components revealed some suggestions for future research. It appears that interventions for the management of osteoporosis that included a patient mediated or reminder component tended to show larger effects than interventions that did not include these components. However, these results should be interpreted with caution due to their indirect nature. The effect of the intervention distribution of educational materials is unclear.

Two RCTs that evaluated an organisational intervention for improved BMD testing in osteoporosis observed large effects. One was a RCT of an organisational plus a patient mediated intervention compared to no‐intervention control, and one was a RCT of an organisational intervention compared to distribution of educational materials. These findings were supported by an ITS study of an organisational, reminder and financial intervention which also demonstrated an improvement in BMD test ordering.

There were no significant changes in healthcare outcomes observed in the few studies that included these measures. However these studies may have been underpowered or had too short a follow‐up period to detect a change.

Low back pain studies

Twelve studies (involving 16 comparisons) included people with low back pain. Six of these studies included only people with low back pain, two studies included people with low back pain or knee pain and four studies included people with low back pain or other numerous conditions, for example, neck pain, or patients referred for an x‐ray of the chest or limbs and joints. Seven of these studies were conducted in the UK, three in the USA, one in Canada and one in The Netherlands. Seven of these studies were cluster RCTs, one was an individual RCT and four were ITS studies. Eleven out of the 12 studies were conducted in primary care and the other in outpatient care. In 11 of these studies, lumbar radiography was one of the imaging outcomes, and in one study the imaging outcome was lumbar MRI. The desired imaging outcome in all of these studies was a decrease in imaging.

Low back pain: evaluations of interventions compared to a no‐intervention control group

Of the studies that included people with low back pain, five studies (involving six comparisons) evaluated single or multifaceted interventions compared to no‐intervention controls.

Low back pain: results from RCT comparisons

Dichotomous imaging outcomes: Three comparisons reported dichotomous imaging outcomes, including the proportion of low back pain participants to receive a lumbar x‐ray or the proportion of lumbar x‐rays that were concordant with evidence‐based guidelines. The range of interventions that were evaluated and the available data for these comparisons are reported in Appendix 8. Two of these comparisons were cluster RCT comparisons (Dey 2004; Oakeshott 1994) and one was an individual RCT comparison (Rossignol 2000). The effect of any type of intervention aiming to change professional practice compared to a no‐intervention control for low back pain imaging behaviour ranged from –1.4% to +11.3% and the median effect was +9.3% absolute improvement in the appropriate ordering of lumbar imaging. Two of the three comparisons reported a non‐significant effect, and for the comparison with four imaging outcomes, three of these outcomes reported a non‐significant effect and one a significant improvement in lumbar imaging, but the sample sizes were small.

Continuous imaging outcomes: Three low back pain comparisons of an intervention versus a no‐intervention control reported continuous imaging outcomes, the number of spinal x‐rays ordered. Appendix 9 shows the range of interventions that were evaluated and the available imaging outcome data reported in each of these studies. Two of the comparisons were cluster RCTs from the same research group (Kerry 2000; Oakeshott 1994). One cluster RCT comparison of distribution of educational materials versus control reported a 45.2% relative improvement in use of appropriate imaging and a SMD of 0.40 (<0.05) (Oakeshott 1994). It was not possible to abstract comparable data for the two other cluster RCTs. One was a comparison of audit and feedback plus distribution of educational materials versus no‐intervention control and only total numbers of x‐rays were reported rather than mean and standard deviation data. This study reported a significant 20% absolute improvement in the use of appropriate imaging of the lumbar spine (95% CI 3 to 37) (Kerry 2000). The remaining cluster RCT comparison of audit and feedback versus no‐intervention control did not report musculoskeletal imaging outcome data separate to non‐musculoskeletal outcome data (Winkens 1995). The comparison observed a significant improvement in appropriate lumbar imaging (P value = 0.004), but the size of this effect was not reported.

Low back pain: results from ITS comparisons

There were four ITS studies, three of which evaluated distribution of educational materials (Hollingworth 2002; Jackson 2005; Matowe 2002), and the other evaluated reminders (Baker 1987). The range of interventions that were evaluated and the data from these comparisons are presented in Appendix 10. One ITS study was correctly analysed with time‐series regression and found no significant difference in lumbar spine radiography referrals, neither in absolute change in referral or in change in referral trend after distribution of a guideline (Matowe 2002). An additional comparison reported that there was no evidence that referrals for radiography of the lumbar spine had decreased with distribution of clinical guidelines (Hollingworth 2002) and reanalysis of this comparison using time‐series regression observed no significant change in both the level and slope. Another comparison reported no evidence of a trend in increasing compliance after distribution of a guideline (Jackson 2005). Reanalysis observed no significant change in level, and a significant deterioration in lumbar imaging, indicated by an increase in slope (P value = 0.1). The remaining comparison reported a 47% decrease in the number of lumbar radiographic examinations after a reminder intervention (Baker 1987). Reanalysis of this comparison using time‐series regression observed a significant change in level (P value = 0.001) but not in slope (P value =0.251), indicating a sustained effect of the reminder intervention.

Low back pain: clinical outcomes

Only one included low back pain study reported patient level clinical outcomes and are presented in Appendix 11 (Rossignol 2000). This comparison of an organisational intervention versus a no‐intervention control observed a non‐significant change in pain (aim was to decrease pain) at three and six months, with relative percentage change of 32.1% and 78.9% and SMD of 0.17 and 0.49, respectively.

Low back pain: evaluations of interventions compared to another intervention

Of the studies that included patients with low back pain, four studies (involving 11 comparisons) evaluated single or multifaceted interventions compared to a different single intervention control. Three of these comparisons (one study) presented outcome data for knee and lumbar MRI combined, and did not present lumbar MRI outcomes separately (Robling 2002).

Dichotomous imaging outcomes: Seven cluster RCT comparisons evaluated interventions versus a different intervention and reported dichotomous outcomes (Eccles 2001; Schectman 2003; Robling 2002). The range of interventions that were evaluated and the data from these comparisons are presented in Appendix 12. Six out the seven comparisons reported a non‐significant deterioration in low back pain imaging outcomes (Eccles 2001; Robling 2002), and one comparison had a potential unit of analysis error (Schectman 2003). The absolute deterioration in imaging behaviour among these comparisons ranged from –8.3% to +1.0% (IQR –8.0% to –0.7%).

Continuous imaging outcomes: Three comparisons of different interventions compared to another intervention from the one cluster RCT reported the continuous imaging outcome of radiograph requests per 1000 patients, and data is presented in Appendix 13 (Eccles 2001). All comparisons were significant (P value < 0.05), demonstrating a median effect of 23.1% relative improvement in lumbar radiographs ordered and a median SMD of 0.38 (range +0.20 to 0.41). The authors of this study recommended caution in the interpretation of these data due to baseline imbalance between the study groups.

Low back pain: cost outcomes

Only one low back pain study included cost outcomes (Robling 2002). This study compared three professional interventions, with audit and feedback plus education the most costly (£3578), but with no evidence of benefit.

Low back pain: summary

The majority of the 12 studies concerning low back pain included distribution of educational materials as an intervention component and the majority observed no significant improvement in appropriate imaging. Only one of the included low back pain studies reported healthcare outcomes and this observed a non‐significant change in pain at three and six months follow‐up for an organisational intervention.

Three studies (representing three comparisons) evaluated various interventions versus no‐intervention controls and reported dichotomous outcomes. The median effect of any intervention versus a no‐intervention control was +9.3% (range –1.4% to +11.3%) absolute improvement in the appropriate ordering of lumbar imaging, with the majority of the outcomes observing a non‐significant effect. For studies reporting continuous outcomes, one cluster RCT comparison of distribution of educational materials versus a no‐intervention control observed a significant improvement in use of appropriate imaging with a +45.2% relative improvement and a SMD of +0.40, but the authors of this study recommended caution in the interpretation of these data due to baseline imbalance between the study groups. One cluster RCT compared audit and feedback plus distribution of educational materials to no‐intervention control and reported a 20% absolute improvement in the use of appropriate imaging of the lumbar spine (95% CI 3 to 37). One cluster RCT comparison observed a significant improvement in appropriate lumbar imaging (P value = 0.004), but the size of this effect was not reported and appropriate data could not be extracted.

One out of four ITS studies observed a significant improvement in imaging outcomes, but the other three observed no effect. The study that showed a significant effect evaluated reminders and was undertaken in the 1980s, whereas the non‐significant ITS comparisons evaluated distribution of educational materials and were undertaken in the 2000s.

Seven cluster RCT comparisons reporting dichotomous outcomes evaluated interventions versus a different intervention control. Five of the seven comparisons reported a non‐significant deterioration in low back pain imaging outcomes (Eccles 2001; Robling 2002), and one comparison had a potential unit of analysis error. For all comparisons of an intervention versus a different intervention reporting dichotomous outcomes, the absolute deterioration in lumbar imaging behaviour ranged from –8.3% to +1.0% (IQR –8.0% to –0.7%). Three cluster RCT comparisons of interventions versus a different intervention from the same study reported continuous outcomes. All comparisons observed significant improvements in lumbar imaging outcomes, but baseline imbalance precludes confident conclusions from this data.

Overall, no firm conclusions can be drawn about the most effective interventions to improve imaging behaviour for the management of low back pain. The most common intervention evaluated, distribution of educational materials, showed varying effects when compared to no‐intervention controls, and it is currently unclear whether this intervention is effective or not for changing low back pain imaging behaviour. For other interventions, no conclusions can be drawn due to only a few studies available showing variable effects.

Knee pain studies

Two cluster RCTs (representing six comparisons) evaluated interventions for improving imaging for people with knee pain (Eccles 2001; Robling 2002), and the desired imaging outcome was a decrease in imaging. Both of these studies were conducted in primary care settings in the UK. Three of these comparisons of different combinations of professional interventions (including audit and feedback, educational meetings and distribution of educational materials) versus distribution of educational materials were from one study that reported low back pain and knee pain outcomes combined, and these have already been discussed in the low back pain section of this review (Robling 2002). Data from this study are presented in Appendix 12. All three intervention comparisons observed a non‐significant effect (Robling 2002).

The three other cluster RCT comparisons were from one study (Eccles 2001). Some data from this study has already been discussed in the low back pain section of this review, but because data were reported separately for low back pain and knee pain, the knee pain data are discussed in this section. These comparisons were of  multifaceted interventions (including audit and feedback, reminders and distribution of educational materials) versus distribution of educational materials and the interventions evaluated and data from these comparisons are reported in Appendix 14 (dichotomous outcomes) and Appendix 15 (continuous outcomes). None of the dichotomous outcomes of the proportion of knee x‐rays concordant with guidelines were significant.

The same cluster RCT also reported continuous outcomes of the number of knee radiographs requested per 1000 patients. All comparisons observed a statistically significant improvement in imaging behaviour. The range of relative percentage improvement was +10.0% to 25.8%, and the range of SMD was +0.18 to 0.46. The authors of this study recommended caution in the interpretation of these results due to baseline imbalance between the study groups.

Knee pain studies: summary

No firm conclusions can be drawn for various professional interventions to improve the appropriate use of imaging in knee pain. No studies were found of interventions compared to no‐intervention control. Two cluster RCTs provide no clear effect of different intervention combinations, including audit and feedback, distribution of educational materials and reminders.

Studies of other conditions

Five studies (involving five comparisons) included participants with musculoskeletal conditions other than osteoporosis, low back pain or knee pain. One study included people with degenerative joint disease (Verstappen 2003) and in the remaining four studies, the conditions included were not specified, because the focus was on radiographs ordered of the limbs and joints rather than any specific musculoskeletal conditions (Kerry 2000; Matowe 2002; Oakeshott 1994; Winkens 1995). Three of these studies were conducted in the UK and two in the Netherlands. Four of these studies were cluster RCTs and one was an ITS study. The desired imaging outcome in all of these studies was a decrease in imaging.

Other conditions: results from RCT comparisons

Four cluster RCTs for other conditions were included and all data and interventions from these studies are shown in Appendix 14 and Appendix 15.

One cluster RCT was an incomplete block design and compared a multifaceted intervention (including distribution of educational materials, educational meetings and audit and feedback) for a group of conditions (including degenerative joint diseases as the only musculoskeletal condition) to a no‐intervention control (Verstappen 2003). Improved ordering of imaging for degenerative joint diseases was reported via two continuous outcomes, the number of x‐rays ordered and the number of inappropriate x‐rays ordered. Change in both outcomes was not significant, with an average relative percentage improvement of 25.4% and an average SMD of +0.31.

One cluster RCT was a comparison of distribution of educational materials versus a no‐intervention control (Oakeshott 1994). The dichotomous outcome of the percentage of limb and joint x‐ray requests conforming to a guideline was an absolute change from baseline of 6.7%, and the absolute improvement was 5.2% (not significant). The same cluster RCT comparison also reported a non‐significant effect of the continuous outcome of number of x‐rays requested (relative percentage change of 38.6%, SMD +0.27).

One cluster RCT comparison of audit and feedback plus distribution of educational materials versus no‐intervention control reported a non‐significant +10% (95%CI –5.0 to +25.0) improvement in ordering of the number of limb and joint x‐rays requested (mean and standard deviation not reported) (see Appendix 15) (Kerry 2000). It was not possible to abstract data for the remaining cluster RCT comparison (Winkens 1995), which did not report musculoskeletal imaging outcomes separate to non‐musculoskeletal outcomes (except for the lumbar imaging outcome which was described in the low back pain section previously).

Other conditions: results from ITS comparisons

One ITS study of distribution of educational materials for the imaging of other musculoskeletal conditions was correctly analysed with time‐series regression and found no significant difference in musculoskeletal imaging referrals, neither in absolute change in referral, or in change in referral trend, after distribution of a guideline (Matowe 2002).

Other conditions: summary

All of the included studies conducted in people with musculoskeletal conditions other than osteoporosis, low back pain and knee pain observed no significant change in appropriate imaging outcomes. The interventions distribution of educational materials, educational meetings and audit and feedback were not shown to be effective for changing imaging ordering behaviour in other musculoskeletal conditions.

Additional analyses

In these analyses we used only those comparisons that reported dichotomous outcome measures. This represents a subset of 64% of the included studies.

Do the effects of interventions vary with different aspects of study characteristics?

We performed exploratory analyses to determine whether different study characteristics varied the effects of the interventions. The results of these analyses are presented in Table 8.

8. Effect sizes of dichotomous measures with varying aspects of study characteristics.

Study characteristic	No. of comparisons (n studies)*	Median effect size	Interquartile range	Range
Imaging behaviour targeted:
Decreased use of imaging	17 (13)	–2.4	–8.0 to 0.0	–12.0 to +8.3
Increased use of imaging	19 (15)	+10.6	+0.7 to +34.0	–1.9 to +62.2
Theory used in intervention development:
Yes	6 (4)	–0.8	–1.8 to +0.7	–1.9 to +3.8
No	30 (24)	+8.8	–1.2 to +30.9	–12.0 to +62.2
Study risk of bias:
High	3 (3)	0.0	–1.9 to +3.8	–1.9 to +3.8
Low	33 (25)	+7.1	–1.8 to +27.7	–12.0 to +62.2

* Data not drawn from all included studies, only those that reported dichotomous outcomes

Some studies targeted a decreased use of imaging (for example, decreased number of lumbar x‐rays ordered) and other studies targeted an increased use of imaging (for example, increased use of BMD test ordering). We examined whether this different type of targeted behaviour modified, on average, the effect of the interventions. The median effect size for comparisons that targeted a decreased use of imaging was –2.4 (IQR –8.0 to 0.0), compared to +10.6 (IQR +0.7 to 34.0) for comparisons that targeted an increased use of imaging. It appears that, on average, if the intervention targeted an increase in the use of imaging, then effects tended to be larger than if the intervention targeted a decrease in the use of imaging but this result should be interpreted with caution due to the analysis being dominated by the osteoporosis and low back pain studies.

We also attempted to determine if the effects of interventions varied depending on whether or not the intervention was based on theory and also compared studies that had a high risk of bias to those that did not. For these two exploratory analyses, due to the imbalance between the proportion of studies that did and did not have the characteristics of interest, these analyses should be interpreted with caution and no firm conclusion can be drawn. It cannot be determined if the effect varies with the use of theory to develop interventions, or between studies with a high or low risk of bias.

Does the effectiveness of multifaceted interventions increase with the number of intervention components?

Table 9 reports the median effect sizes of all interventions by number of components comprising the interventions in the study group and in the control group. Figure 3 illustrates the spread of effect sizes for increasing the number of intervention components, using box plots. Visually, there appeared to be no relationship between effect size and number of components in the interventions. There was no evidence of a relationship between the number of interventions used in the study group and the effect size (Kruskal‐Wallis test, P value = 0.48). Quantile regression analysis also indicated that there was no evidence of an increased effect size by increasing the number of components (coefficient = –2.51, 95% CI –11.58 to 6.56, P value = 0.57).

9. Summary of median effect sizes for intervention evaluations by number of interventions in the study and control arms.

No. of components in study arm	Median absolute effect size and IQR across comparisons (no. of comparisons)
	No. of components in control arm
	0	1
1	+9.43, +4.48 to 12.80 (n = 11)	+34.65, +7.10 to 62.20 (n = 2)
2	13.75, –1.80 to +34.60 (n = 7)	–5.35, –10.15 to –1.20 (n = 4)
3	0.0, –0.20 to +27.70 (n = 7)	–7.60, –8.0 to +34.0 (n = 3)
4	+3.80 (n = 1)	+0.73 (n = 1)

3.

Effect sizes of multifaceted interventions by number of components in intervention study group

Due to small number of studies and skewed data, median does not equal mean and hence median line does not appear in the centre of the boxes

Discussion

We included 28 studies assessing a range of interventions. Twenty three studies evaluated professional interventions alone, two studies evaluated organisational interventions alone, two studies evaluated a combination of professional and organisational interventions and one study evaluated a professional intervention in combination with an organisational and a financial intervention. Fifteen studies involved the management of osteoporosis, six studies the management of low back pain, two studies the management of low back pain and knee pain, and five studies involved the management of multiple conditions, including various musculoskeletal conditions.

Summary of main results

In order to improve appropriate imaging for osteoporosis, it is not clear which intervention, or intervention combination, was most effective. Our exploratory analyses observed that patient mediated, reminder, and organisational interventions have most potential, however these interventions require more testing in high quality studies. For low back pain studies, no intervention evaluated to date produced convincing effects for improving the use of lumbar imaging. There were insufficient studies of high quality for improving imaging behaviour in the management of low back pain that evaluated patient mediated or reminder components. Interventions rigorously tested to improve the use of imaging for low back pain have been predominantly distribution of educational materials and few observed significant benefit. No conclusions can be made about other interventions for improving the use of imaging in low back pain. For other musculoskeletal conditions no firm conclusions can be drawn.

Overall completeness and applicability of evidence

We were unable to determine, under what circumstances and contexts, which interventions to improve the use of imaging in musculoskeletal conditions are most effective. Multiple problems precluded this analysis including poor reporting, inadequate intervention detail, heterogeneity among interventions and inappropriate primary analyses. We were therefore unable to conduct robust subgroup analyses of the potential effect modifiers of different interventions for different musculoskeletal conditions. The diversity of interventions, outcomes and conditions make overall conclusions about the effects of interventions evaluated difficult to draw. When grouped by condition, pooling by intervention type was not possible because very few studies evaluated the same intervention combination. Only conclusions about intervening in general can be made. Further, most of the included studies were conducted in well resourced developed countries, and in primary care settings, so conclusions drawn may only apply to similar settings.

Contemporary implementation research is increasingly recognising the need for interventions to be designed with consideration of theory and barriers to change (Eccles 2005; ICEBeRG 2006). Very few studies in this review evaluated theory‐based interventions or interventions directed at specific barriers to change. Also, fidelity evaluation of interventions in the included studies was rare but is essential to ensure that intervention effects can be fully understood (Bellg 2004).

Only three included studies reported clinical outcomes. Although the main topic of the review was a process outcome, imaging ordering behaviour, we reported healthcare outcomes where available. There is some debate about the need to measure healthcare outcomes in evaluative implementation research in instances when there is a high degree of certainty that the targeted behaviour will lead to the healthcare outcome (Hakkennes 2006). Future studies should include healthcare outcomes when evidence is not available to support that the change in practitioner behaviour will lead to improved patient outcomes. This will require larger studies for longer follow‐up periods, and greater resources would be needed.

Only three included studies considered cost outcomes, and none of these studies reported a cost‐effectiveness analysis. Future studies in this area should include a well designed economic analysis.

Quality of the evidence

Perhaps the most important limitation of the review is the quality of the included studies, many of which had methodological weaknesses and poor reporting that diminish the certainty of our conclusions. Although only three studies were judged as having ‘high’ risk of bias, many studies had methodological weaknesses. Common methodological omissions in the conduct or reporting of these studies included lack of adequate allocation concealment, lack of blinded outcome assessment and lack of protection against intervention contamination. We judged the overall quality of the five included ITS studies as poor. Future studies should be of high quality and well reported to address previous methodological shortcomings.

Potential biases in the review process

Formulating a search strategy for a review with such broad condition and intervention inclusion was challenging, highlighted by almost half of the included studies not identified by our initial electronic database search. This highlights the importance of not relying solely on electronic databases for reviews in this field. While our search strategy of using multiple search methods identified further studies, it is possible that other important studies were not located. In addition, only one of the included osteoporosis studies was published pre‐2002. Although we only identified one ongoing osteoporosis study, this appears to be an area of research with more studies being conducted more recently and for this review to remain relevant it will need to be updated relatively soon. In light of this we will update the search of this review within 12 months of publication, and update the review with any new studies we identify.

There were four included studies where the conditions included were not specified, because the focus was radiographs ordered of the spine, limbs and joints rather than any specific musculoskeletal condition (Kerry 2000; Matowe 2002; Oakeshott 1994; Winkens 1995). We assumed that the imaging tests were for musculoskeletal conditions, although it was not possible to determine this from the study report. One of these studies did not contribute data to the review because appropriate data could not be extracted from the study report (Winkens 1995). The other three studies contributed data to the low back pain and the other musculoskeletal conditions section, so it is possible that this limitation could have an impact on the conclusions of the review for these conditions.

At the protocol stage we planned to include CCTs and ITS studies because, for some interventions in this context, a RCT is not feasible. If feasible to conduct, RCTs are the strongest research design to evaluate intervention effects for the improvement of imaging behaviour, but data from non‐RCTs can provide useful information. However, it has been argued that if a well conducted RCT is available to answer an intervention research question then non‐RCTs should not be included in a systematic review of interventions due to their high risk of bias (Reeves 2008). In this review, an ITS study for osteoporosis management provided data on a financial intervention that was not evaluated in any RCTs (Feldstein 2007). If we excluded non‐RCTs at the protocol stage we would not have included this study in the review and this data would have been overlooked. Also, other included non‐RCTs evaluated instances when guidelines had been disseminated to all health practitioners in a geographic region and a RCT would not have been possible (Kerry 2000; Matowe 2002; Hollingworth 2002). This argument also extends to CBA studies, which we did exclude from the review. It is possible that a review that included CBA studies would have different conclusions.

Agreements and disagreements with other studies or reviews

The most comprehensive systematic review to date of all guideline implementation strategies across all health conditions included studies to mid‐1998 and was conducted by Grimshaw and colleagues (Grimshaw 2004). This review found educational materials had moderate effects, patient mediated (or directed) interventions observed moderate to large effects, especially when targeting preventive services, and reminder interventions had moderate effects. Our targeted review has supported these findings for the effectiveness of reminder and patient mediated interventions, in the context of appropriate imaging in osteoporosis. Grimshaw and colleagues’ conclusion that the intervention distribution of educational materials had a modest effect across all conditions differs to our findings for the included low back pain studies where we found no significant improvement for this intervention. Another more recent Cochrane systematic review examined printed educational materials across all clinical conditions, and included 23 studies (Farmer 2008). Only one study was included that evaluated printed educational materials for low back pain management, but this study did not report imaging outcomes so we did not include it in our review (Hazard 1997). This other review concluded that when compared to no intervention, printed educational materials slightly improved process outcomes. It appears from our review that clinician imaging behaviour with respect to low back pain does not readily change with dissemination of educational materials alone. Further studies to improve the use of imaging in low back pain management should consider evaluating other types of professional or organisational interventions, and have a strong rationale for the intervention choice.

Previous reviews have drawn different conclusions about the cumulative effect of increasing number of intervention components. Our review is in agreement with Grimshaw et al who found no increased effect with increasing number of intervention components (Grimshaw 2004), although three previous systematic reviews have found multifaceted interventions to be more effective than single interventions (Davis 1995; Davis 1997; Wensing 1998). In the conduct of these other reviews, how the review authors coded the included interventions and the analytical methods used were not explicitly reported. In our review, and that of Grimshaw et al (Grimshaw 2004), we coded all intervention components and used explicit methods to determine the effect size for each study, where possible. However, our analysis about the cumulative effect should be interpreted with caution because many of the comparisons were indirect. There were few head‐to‐head comparisons of the different interventions and it would not be appropriate to conclude that an intervention, single or multifaceted, is more effective than another based upon an indirect comparison unless appropriate methods were used. Considering that many of the included studies did not provide details on the reasons for the interventions chosen, it is not clear if the investigators chose intervention components for specific reasons or due to an assumption that ‘the more that is done the greater the chance of effect’. Our findings, and that of Grimshaw et al (Grimshaw 2004), support a hypothesis that careful planning to overcome specific barriers to change is needed to design effective interventions in this area. A Cochrane systematic review including 15 studies of tailored interventions to overcome identified barriers to change concluded that more research is needed in this area (Cheater 2005), so this hypothesis needs further testing in well designed studies.

For the study characteristics of interest (the use of theory and degree of risk of bias), due to the imbalance between the proportion of studies that did and did not have the characteristics, no firm conclusions can be drawn. However, for the analysis of type of imaging behaviour targeted, there was reasonable balance in the number of studies that aimed for an increase in imaging and the number that aimed for a decrease. Our analysis suggests that it may be more difficult for an intervention to reduce an imaging behaviour that is already established than it is to increase an imaging behaviour that is underused. This may explain some of the results where we found that osteoporosis studies (aiming for an increase in imaging behaviour) most often demonstrated a positive effect, whereas low back pain studies (aiming for a decrease in imaging behaviour) most often demonstrated no effect. On the other hand, this analysis is driven primarily by the osteoporosis and low back pain studies so this finding may be due to factors inherent in the management of osteoporosis and low back pain rather than due to increasing or decreasing behaviours per se.

Two previous systematic reviews examined the effect of interventions to improve the use of BMD test ordering in osteoporosis (Elliot‐Gibson 2004; Morris 2004). Both reviews were unable to determine which intervention was most effective for improving BMD test ordering in osteoporosis. These reviews were undertaken prior to the publication of the majority of the osteoporosis RCTs we have identified. Therefore, there is very little overlap in included studies between our review and the previous reviews and it is difficult to make comparisons between findings. Further, these two reviews included CBA studies, which we excluded from our review, and one review included an RCT which we excluded for methodological reasons (Wroe 2000).

Authors' conclusions

Implications for practice.

Interventions which can improve imaging practices in musculoskeletal conditions could have great benefit and result in both improved health outcomes and reduced healthcare costs. This review identified 28 studies that evaluated a range of interventions including multiple combinations of complex interventions designed to improve imaging in musculoskeletal conditions. For osteoporosis, most professional interventions were found to result in a modest improvement of bone mineral density test ordering of approximately 10%. If a patient mediated or reminder component was part of the intervention the effect tended to be larger. Firm conclusions about other intervention types, for example, organisational interventions or audit and feedback, were not possible due to the small number of studies identified. For low back pain, the majority of the 12 included studies evaluated distribution of educational materials and most observed no significant improvement in appropriate imaging. No recommendations for practice can be made about other musculoskeletal conditions.

Implications for research.

Future research is needed to identify interventions that are effective in improving imaging for musculoskeletal conditions in particular settings. Our review has identified some interventions that show some promise. For osteoporosis, it is not clear which intervention, or combination of interventions, will be most effective. To date, patient mediated, reminder, and organisational interventions have each demonstrated promising results, but require further evaluation in rigorous studies. For low back pain, it appears that distribution of educational materials alone is not effective for improving imaging behaviour, while studies evaluating other interventions were few. Only a few studies evaluated theory‐based interventions directed at specific barriers to change, and further research should be directed towards evaluating these approaches in high quality studies. Further work is also needed to determine the most effective and cost‐effective combination of intervention components for particular settings and conditions and fidelity evaluation of interventions is essential to ensure that intervention effects can be fully understood.

History

Protocol first published: Issue 3, 2006
 Review first published: Issue 1, 2010

Date	Event	Description
30 April 2008	Amended	Converted to new review format.

Appendices

Appendix 1. MEDLINE search strategy

These search terms specific for this review were added to the standard EPOC search terms (see: Effective Practice and Organisation of Care Group methods used in reviews).

1. exp back pain/

2. neck pain/

3. Shoulder pain/

4. Tennis Elbow/

5. exp Tendinopathy/

6. Whiplash Injuries/

7. Sciatica/

8. Intervertebral Disk Displacement/

9. (pain adj3 (neck or back or shoulder? or elbow? or forearm? or wrist? or hand? or arm? or hip? or knee? or ankle? or leg? or foot or feet)).tw.

10. (epicondylitis or tendonitis or tendinitis or bursitis or synovitis or sprain? or strain?).tw.

11. (Whiplash or sciatica).tw.

12. exp Joint Diseases/

13. exp Spinal Diseases/

14. exp Spondylarthritis/

15. exp Arthritis, Rheumatoid/

16. exp osteoarthritis/

17. exp Osteoporosis/

18. (arthriti$ or osteoarthriti$ or osteoporo$ or bone loss$).tw.

19. ankle/ or hip/ or knee/

20. elbow/ or wrist/ or shoulder/

21. elbow joint/ or exp hand joints/ or hip joint/ or knee joint/ or sacroiliac joint/ or shoulder joint/

22. exp cervical vertebrae/ or intervertebral disk/ or lumbar vertebrae/ or thoracic vertebrae/

23. exp Back/

24. exp Spine/

25. (spine or spinal).tw.

26. or/1‐25

27. exp Diagnostic Imaging/

28. exp Radiology/

29. Radiology Department, Hospital/

30. (MRI or x‐ray$ or xray$ or myelogra$ or ultrasound or ultrasonography or imaging or radiograph$ or radiolog$ or tomography or (bone adj scan) or CT or (CAT adj scan)).tw.

31. or/27‐30

32. 26 and 31

Appendix 2. Osteoporosis studies with intervention versus no‐intervention control, imaging outcomes, dichotomous data

Study	Intervention	Outcome	Int pre (%)1	C pre (%)2	Int post n/N (%)3	C post n/N (%)3	Relative % change (post)	Int vs C absolute change from pre (%)	ARD5	Absolute % change (post)	P value6
Curtis 2007	A&F + ed mat	BMD test	‐	‐	90/472 (19.1)	100/477 (21.0)	‐	‐	‐	‐1.9	0.48
Feldstein 2006	Ed mat + rem	BMD test	‐	‐	40/101 (39.6)	5/101 (5.0)	‐	‐	‐	34.6	<0.01
	Ed mat + pt med + rem	BMD test	‐	‐	36/110 (32.7)	5/101 (5.0)	‐	‐	‐	27.7	<0.01
Gardner 2005	Pt med	BMD test	‐	‐	12/40 (30.0)	6/40 (15.0)	‐	‐	‐	15.0	NR7
Lafata 2007	Pt med	BMD test	‐	‐	720/3367 (21.4)	313/2901 (10.8)	‐	‐	‐	10.6	<0.001
	Pt med + rem	BMD test	‐	‐	1181/4086 (28.9)	313/2901 (10.8)	‐	‐	‐	18.1	<0.001
Majumdar 2004	Ed mat + pt med + rem	BMD test	‐	‐	34/55 (61.8)	8/47 (17.0)	‐	‐	‐	44.8	<0.001
Majumdar 2007	Organisational + pt med	BMD test	‐	‐	88/110 (80.0)	32/110 (29.0)	‐	‐	‐	51.0	<0.001
Prihar 2008	Ed mat + pt med	BMD test	(6.3)	(4.6)	174/918 (19.0)	57/595 (9.6)	97.9	12.7 vs 5.0	7.7	9.4	UAE
Solomon 2004	A&F + ed mat + ed meet	BMD test	(9.0)	(5.2)	13/168 (7.8) 8	16/205 (7.8) 8	0.0	‐1.2 vs 2.6	‐3.8	0.0	0.9
Solomon 2007	Ed mat + EO	BMD test	‐	‐	183/3574  (5.1)	224/3268 (6.9)	‐	‐	‐	‐1.8	NS
	Pt med	BMD test	‐	‐	249/3274 (7.6)	224/3268 (6.9)	‐	‐	‐	0.7	NS
	Ed mat + EO + pt med	BMD test	‐	‐	223/3339 (6.7)	224/3268 (6.9)	‐	‐	‐	‐0.2	NS
Solomon 2007a	Ed mat + EO + pt med + rem	BMD test	‐	‐	126/997 (12.6)	86/976 (8.8)	‐	‐	‐	3.8	0.01

1. Intervention group pre‐intervention compliance; 2. Control group pre‐intervention compliance; 3. Intervention group post‐intervention compliance; 4. Control group post‐intervention compliance; 5. Adjusted risk difference = [risk of non‐compliance (intervention ‐ control) pre‐intervention] ‐ [risk of noncompliance (intervention ‐ control) post‐intervention]; 6. P value reported by study authors; 7. NR: P value not reported for this outcome; 8. data estimated from graph in publication

A&F: audit and feedback; BMD: bone mineral density; C: control group; ed mat: distribution of education materials; ed meet: educational meeting; EO: educational outreach; Int: intervention group; NR: not reported; NS: not significant; pt med: patient mediated; rem: reminder; UAE: potential unit of analysis error; vs: versus

Appendix 3. Osteoporosis study with intervention versus no‐intervention control, imaging outcome, continuous data

Study	Intervention	Outcome	Int pre mean (SD)1	Int post mean (SD)2	C pre mean (SD)3	C post mean (SD)4	Post diff5	Relative % change6	SMD7	P value8
Stock 1998	Reminder	BMD tests ordered per month	‐	1.30 (1.21)	‐	0.72 (0.71)	0.58	80.56	N/A	0.002

1. Intervention group pre‐intervention mean (standard deviation); 2. Intervention group post‐intervention mean (standard deviation); 3. Control group pre‐intervention mean (standard deviation); 4. Control group post‐intervention mean (standard deviation); 5. Difference between post‐intervention means; 6. Relative percentage change post‐intervention = ((Int post mean/C post mean)/C post mean) X 100; 7. Standardised mean difference (EPOC defined) = (Int post mean/ C post mean)/ C post SD; 8. P value reported by study authors

Appendix 4. Osteoporosis study, health care outcomes, dichotomous data

Study	Intervention comparison	Outcome	Int n/N (%)	Controls n/N (%)	Absolute % change (post)	P value
Majumdar 2007	Organisational + pt med vs control	Additional fracture	2/110 (1.8)	2/110 (1.8)	0.0	NS
		Death	3/110 (2.7)	2/110 (1.8)	0.9	NS

Appendix 5. Osteoporosis study, health care outcomes, continuous data

Study	Intervention comparison	Outcome	Int pre mean (SD)	Int post mean (SD)	C pre mean (SD)	C post mean (SD)	Post diff	Relative % change	SMD	P value
Majumdar 2008	Ed mat + pt med + rem vs pt med	Health status: SF‐12 (mental)	N/A	54.9 (7.6)	N/A	53.4 (7.4)	‐	‐	‐	NS
		Health status: SF‐12 (physical)	N/A	47.5 (9.2)	N/A	46.8 (9.7)	‐	‐	‐	NS

See Appendix 3 for legend

Appendix 6. Osteoporosis, interrupted time series study, imaging outcome

Study	Intervention	Outcome	Mean pre (SD)	Mean post	Mean post minus mean pre	Relative % change pre to post	SMD pre to post	mean change in level (P value)	mean change in slope (P value)
Feldstein 2007	Organisational + reminder + financial	BMD test	72	129.5	57.5	79.9	N/A	16.8 (p=0.07)	5.6 (p <0.0001)

Appendix 7. Osteoporosis studies with intervention versus a different intervention control, imaging outcomes, dichotomous data

Study	Comparison	Outcome	Int pre (%)	C pre (%)	Int post n/N (%)	C post n/N (%)	Relative % change (post)	Int vs C absolute change from pre (%)	ARD	Absolute % change (post)	P value
Boyd 2002	Rem (extended letter) vs rem (short letter)	BMD test	‐	‐	29/78 (37.2)	25/83 (30.1)	‐	‐	‐	7.1	UAE
Majumdar 2008	Ed mat + pt med + rem vs pt med	BMD test	‐	‐	71/137 (51.8)	24/135 (17.8)	‐	‐	‐	34.0	<0.001
Rozental 2008	Organisational vs ed mat	BMD test	‐	‐	25/27 (92.6)	7/23 (30.4)	‐	‐	‐	62.2	<0.001

See Appendix 2 for legend.

Appendix 8. Low back pain studies with intervention versus no‐intervention control, imaging outcomes, dichotomous data

Study	Intervention	Outcome	Int pre (%)	C pre (%)	Int post n/N (%)	C post n/N (%)	Relative % change (post)	Int vs C absolute change from pre (%)	ARD	Absolute % change (post)	P value
Dey 2004	Ed mat + EO + organisational	Number referred for x‐ray	(20.4)	(21.1)	158/1049 (15.1)	156/1138 (13.7)	10.2	5.3 vs 7.4	‐2.1	‐1.4	NS
Oakeshott 1994	Ed mat	X‐ray requests (by practice) conforming to guidelines (spine)	(39.5)	(34.7)	(44.8)*	(33.5)*	33.7	5.3 vs ‐1.2	6.5	11.3	NS
Rossignol 2000	Organisational**	Number received imaging test (x‐ray) 3 months	(61.1)	(53.6)	7/47 (14.9)	14/48 (29.2)	49.0	46.2 vs ‐24.4	21.8	14.3	NS
		Number received imaging test (MRI, CT, mylogram) at 3 months	(22.0)	(23.2)	6/47 (12.8)	19/48 (39.6)	67.7	9.2 vs ‐16.4	25.6	26.8	<0.01
		Number received imaging test (x‐ray) 3 to 6 months	(61.1)	(53.6)	4/44 (9.1)	6/45 (13.3)	31.6	52.0 vs 40.3	11.7	4.2	NS
		Number received imaging test (MRI, CT, mylogram) 3 to 6 months	(22.0)	(23.2)	6/44 (13.6)	6/45 (13.3)	2.3	8.4 vs 9.9	1.5	‐0.3	NS
					Median effect size from Rossignol 2000					9.3

See Appendix 2 for legend. * only percentages reported; ** clinical multidisciplinary team

Appendix 9. Low back pain studies with intervention versus no‐intervention control, imaging outcomes, continuous data

Study	Comparison	Outcome	Int pre mean (SD)	Int post mean (SD)	C pre mean (SD)	C post mean (SD)	Post diff	Relative % change	SMD	P value
Kerry 2000*	A&F + ed mat	Number of x‐rays requested (spine)	2655	2181	2171	2221	‐	‐	‐	‐
Oakeshott 1994	Ed mat	X‐rays requested (spine)	3.3 (3.7)	1.7 (1.9)	3.5 (3.5)	3.1 (3.5)	1.4	45.16	0.400	<0.05

See Appendix 3 for legend. * Mean not reported, only total number of x‐rays

Appendix 10. Low back pain, interrupted time series studies, imaging outcomes

Study	Intervention	Outcome	Mean pre (SD)	Mean post (SD)	Mean post minus mean pre	Relative % change pre to post	SMD pre to post	mean change in level (P value)	mean change in slope (P value)
Baker 1987	Reminder	Lumbosacral radiographs ordered	111.1 (20.4)	64.9 (10.1)	46.2	41.6	2.26	‐4.3 (p=0.001)	‐1.7 (p=0.251)
Hollingworth 2002	Ed mat	Lumbosacral radiographs ordered	1133.0 (50.0)	1208.7 (111.5)	‐75.7	‐6.7	‐1.51	‐121.5 (p=0.167)	6.8 (p=0.776)
Jackson 2005	Ed mat	Lumbosacral radiographs ordered	15.7 (5.9)	18.4 (2.7)	‐2.7	‐17.5	‐0.47	8.7 (p=0.181)	4.9 (p=0.1)
Matowe 2002	Ed mat	Total x‐ray examinations*	2463.8	2420	43.8	1.8	N/A	NS	NS

* included non‐musculoskeletal imaging

Appendix 11. Low back pain study health care outcomes, continuous data

Study	Intervention	Outcome	Int pre mean (SD)	Int post mean (SD)	C pre mean (SD)	C post mean (SD)	Post diff	Relative % change	SMD	P value
Rossignol 2000	Organisational vs no‐intervention control	Pain (3 months)	48.5 (20.4)	14.4 (27.7)	52.4 (20.7)	10.9 (24.0)	3.5	32.1	0.169	NS
		Pain (6 months)	48.5 (20.4)	22.9 (29.3)	52.4 (20.7)	12.8 (27.0)	10.1	78.9	0.488	NS

See Appendix 3 for legend.

Appendix 12. Low back pain studies with intervention versus a different intervention control, imaging outcomes, dichotomous data

Study	Interventions compared	Outcome	Int pre (%)	C pre (%)	Int post n/N (%)	C post n/N (%)	Relative % change (post)	Int vs C absolute change from pre (%)	ARD	Absolute % change (post)	P value
Eccles 2001	A&F + ed mat vs ed mat	X‐ray concordant with guideline (lumbar)	‐	‐	64/181 (35.3)	120/275 (43.6)	‐	‐	‐	‐8.3	NS
	Reminder + ed mat vs ed mat	X‐ray concordant with guideline (lumbar)	‐	‐	35/85 (41.2)	120/275 (43.6)	‐	‐	‐	‐2.4	NS
	A&F + ed mat + reminder vs ed mat	X‐ray concordant with guideline (lumbar)	‐	‐	89/247 (36.0)	120/275 (43.6)	‐	‐	‐	‐7.6	NS
Robling 2002	Ed meet + ed mat vs ed mat	MRI concordant with guideline (lumbar and knee combined)	‐	‐	30/38 (79)	42/53 (79)	‐	‐	‐	0.0	NS
	A&F + ed mat vs ed mat	MRI concordant with guideline (lumbar and knee combined)	‐	‐	28/42 (67)	42/53 (79)	‐	‐	‐	‐12.0	NS
	A&F + ed meet + ed mat vs ed mat	MRI concordant with guideline (lumbar and knee combined)	‐	‐	35/49 (71)	42/53 (79)	‐	‐	‐	‐8.0	NS
Schectman 2003*	A&F + ed mat + ed meet + pt med vs pt med	Lumbar x‐ray (% of patients received)	(31.0)	(21.0)	(19.0)	(18.0)	5.6	12.0 vs 3	9.0	‐1.0	UAE
		Lumbar CT or MRI  (% of patients received)	(7.6)	(5.6)	(5.6)	(7.1)	21.1	2.0 vs ‐1.5	3.5	1.5	UAE
		Lumbar x‐ray not consistent with guideline	(14.5)	(8.2)	(8.1)	(8.6)	5.8	6.4 vs ‐0.4	6.8	0.5	UAE
		Lumbar CT or MRI not consistent with guideline	(5.7)	(3.5)	(3.5)	(5.4)	35.2	2.2 vs ‐1.9	4.1	1.9	UAE
			Median effect size for Schectman 2003							1.0

See Appendix 2 for legend. * only percentages reported

Appendix 13. Low back pain studies with intervention versus a different intervention control, imaging outcomes, continuous data

Study	Comparison	Outcome	Int pre mean (SD)	Int post mean (SD)	C pre mean (SD)	C post mean (SD)	Post diff	Relative % change	SMD	P value
Eccles 2001	A&F vs ed mat	Radiograph requests per 1000 patients (lumbar)	7.24 (4.8)	5.97 (4.2)	7.53 (4.1)	6.8 (4.3)	0.83	12.21	0.202	<0.05
Eccles 2001	Rem vs ed mat	Radiograph requests per 1000 patients (lumbar)	7.31 (5.2)	5.14 (3.7)	7.53 (4.1)	6.8 (4.3)	1.66	24.41	0.405	<0.05
Eccles 2001	A&F + rem vs ed mat	Radiograph requests per 1000 patients (lumbar)	8.30 (5.1)	5.23 (3.7)	7.53 (4.1)	6.8 (4.3)	1.57	23.09	0.383	<0.05

See Appendix 3 for legend. * Mean not reported, only total number of x‐rays

Appendix 14. Studies of other conditions, imaging outcomes, dichotomous data

Study	Interventions compared	Outcome	Int pre (%)	C pre (%)	Int post n/N (%)	C post n/N (%)	Relative % change (post)	Int vs C absolute change from pre (%)	ARD	Absolute % change (post)	P value
Eccles 2001	A&F + ed mat vs ed mat	X‐ray concordant with guideline (knee)	‐	‐	52/240 (21.7)	83/328 (25.3)	‐	‐	‐	‐3.6	NS
	Reminder + ed mat vs ed mat	X‐ray concordant with guideline (knee)	‐	‐	26/85 (30.6)	83/328 (25.3)	‐	‐	‐	5.3	NS
	A&F + reminder + ed mat vs ed mat	X‐ray concordant with guideline (knee)	‐	‐	70/252 (27.8)	83/328 (25.3)	‐	‐	‐	2.5	NS
Oakeshott 1994	Ed mat vs control	X‐ray requests by practice conforming to guidelines (limbs & joints)	(85.7)	(87.2)	(88.8)*	(83.6)*	6.2	3.1 vs ‐3.6	6.7	5.2	NS

See Appendix 2 for legend. * no numbers, only percentages reported

Appendix 15. Studies of other conditions, imaging outcomes, continuous data

Study	Comparison	Outcome	Int pre mean (SD)	Int post mean (SD)	C pre mean (SD)	C post mean (SD)	Post diff	Relative % change	SMD	P value
Eccles 2001	A&F + ed mat vs ed mat	Radiograph requests per 1000 patients (knee)	7.03 (5.1)	6.32 (4.0)	6.67 (3.9)	7.02 (3.6)	0.7	9.97	0.179	<0.05
Eccles 2001	Ed mat + rem vs ed mat	Radiograph requests per 1000 patients (knee)	7.18 (5.0)	5.22 (3.6)	6.67 (3.9)	7.02 (3.6)	1.8	25.64	0.462	<0.05
Eccles 2001	A&F + ed mat + rem vs ed mat	Radiograph requests per 1000 patients (knee)	9.34 (6.1)	5.21 (3.7)	6.67 (3.9)	7.02 (3.6)	1.81	25.78	0.461	<0.05
Kerry 2000*	A&F + ed mat vs control	Number of xrays requested (limbs & joints)	4275	4253	3646	3986	‐	‐	‐	NS
Oakeshott 1994	Ed mat vs control	Xrays requested (limbs & joints)	4.0 (4.6)	2.7 (2.6)	5.5 (6.3)	4.4 (6.6)	1.7	38.6	0.270	NS
Verstappen 2003	Ed mat + ed meet + AF vs control	Number of x‐rays ordered	72 (43)	58 (37)	54 (38)	49 (36)	‐9.0	18.4	0.237	NS
		Number of inappropriate x‐rays ordered^	50 (34)	41 (26)	36 (26)	31 (22)	10.0	32.3	0.385	NS

See Appendix 3 for legend. * Mean not reported, only total number of x‐rays; ^ includes non‐musculoskeletal imaging (chest radiography) and musculoskeletal data not reported separately

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Baker 1987.

Methods	Type of targeted behaviour: decrease in test ordering (lumbar x‐rays) Study design: ITS Country: USA
Participants	Setting: hospital Number of providers: not reported; Number of tests: 1443 in control year and 759 in experimental year Condition: acute LBP
Interventions	Professional intervention (reminder: note on x‐ray referral form)
Outcomes	Professional practice: number of x‐rays ordered; proportion of appropriate x‐rays Patient level: none
Notes	Justification for intervention: not reported Intervention fidelity: not reported

Boyd 2002.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: cluster RCT Country: USA
Participants	Setting: primary care 149 providers, 258 patients Condition: osteoporosis detected by heel ultrasound
Interventions	1. Professional intervention (reminder: extended letter to physician about patient’s risk of osteoporosis) 2. Professional intervention (reminder: short letter about patient’s risk of osteoporosis)
Outcomes	Professional practice: number of patients contacted by physicians following distribution of reminders; number of patients having a BMD test within 6 months; prescription of osteoporosis medication Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	"randomly assigned" is only text used
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	High risk	Outcomes for only 63% of patients followed up
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Unclear risk	Potential unit of analysis error

Curtis 2007.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: cluster RCT Country: USA
Participants	Setting: rheumatology and primary care 153 providers, 949 patients Condition: taking oral glucocorticoids (high likelihood of osteoporosis)
Interventions	1. Professional intervention (audit and feedback + distribution of educational materials) 2. Control
Outcomes	Professional practice: medication for osteoporosis or a BMD test within 12 months after the intervention Patient level: none
Notes	Justification for intervention type: reported that intervention based on “adult learning principles” but no further information provided Intervention fidelity: percentage of physicians that completed online modules was measured; per protocol analysis showed significant benefit in BMD testing for clinicians who completed all intervention modules
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	From report: “Block randomization was used to balance the number of intervention vs control physicians randomized over time.”
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Low risk	No missing outcome data
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Low risk	Contamination avoided by practice randomisation

Dey 2004.

Methods	Type of targeted behaviour: decrease in test ordering (lumbar x‐rays) + referrals + prescribing Study design: cluster RCT Country: UK
Participants	Setting: primary care 24 practices, 2187 patients Condition: acute LBP
Interventions	1. Professional intervention (educational outreach visit + distribution of educational materials) + organisational intervention (case management: access to a fast‐track physiotherapy service + access to a back clinic) 2. Standard practice control group
Outcomes	Professional practice: rate of referral for lumbar spine x‐ray within 3 months; number of sickness certificates issued; number prescribed opioids or muscle relaxants; number referred to secondary care; number referred to physio or educational program Patient level: none
Notes	Justification for intervention type: previous studies cited for effectiveness of educational outreach; theory ‘elaboration likelihood model of persuasion’ used for outreach visits Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	Minimisation with a random element
Allocation concealment?	Low risk	Central allocation
Blinding? All outcomes	High risk	From report: "Owing to financial constraints, only one research assistant was employed, and blind outcome assessment was not possible."
Incomplete outcome data addressed? All outcomes	Low risk
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Low risk

Eccles 2001.

Methods	Type of targeted behaviour: decrease in test ordering (lumbar and knee x‐rays) Study design: cluster RCT Country: UK
Participants	Setting: primary care 247 practices Condition: acute LBP or knee pain
Interventions	1. Professional intervention (distribution of educational materials + audit and feedback) 2. Professional intervention (distribution of educational materials + reminders) 3. Professional intervention (distribution of educational materials + audit and feedback + reminders) 4. Distribution of educational materials (guideline)
Outcomes	Professional practice: number of lumbar or knee radiographs requested per 1000 patients for 2 years Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: in part. Measured attachment rate of educational reminder messages to x‐ray reports
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	Computer‐generated random‐number tables
Allocation concealment?	Low risk	Performed centrally by study statistician
Blinding? All outcomes	Low risk	Outcome measure was objective and recorded by radiology departments
Incomplete outcome data addressed? All outcomes	Low risk
Free of selective reporting?	Low risk
Free of other bias?	High risk	Some baseline imbalance between study groups

Feldstein 2006.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: RCT Country: USA
Participants	Setting: primary care 15 practices, 159 providers, 327 patients Condition: women aged 50 to 89 who had suffered a fracture (any type) and therefore high likelihood of osteoporosis
Interventions	1. Professional intervention (reminders: electronic medical record message about patient’s risk of osteoporosis + distribution of education materials) 2. Professional intervention (reminders + distribution of education materials) + patient mediated (education materials) 3. Standard practice control group
Outcomes	Professional practice: proportion of study population who received medication for osteoporosis or a BMD test within 6 months after the intervention Patient level: regular physical activity; total caloric expenditure; total calcium intake; patient satisfaction
Notes	Justification for intervention type: previous studies showing electronic medical records effective in other clinical conditions Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	From report: “A computer random‐number generator seeded by date and time once at the start of the study generated the random sequence”
Allocation concealment?	Unclear risk	From report: “The study statistician randomized and assigned participants to the study groups”, but not specified if study statistician involved in recruitment process
Blinding? All outcomes	Low risk	From report: “The study analyst assessing the outcomes was blinded to the treatment groups”
Incomplete outcome data addressed? All outcomes	Low risk
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	High risk	No protection against contamination.

Feldstein 2007.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: ITS Country: USA
Participants	Setting: primary care 15 practices, 255 providers, 3588 patients Condition: people with a fracture and high likelihood of osteoporosis
Interventions	Phase 1: Professional intervention (reminder: electronic medical record message about patient’s risk of osteoporosis) + organisational intervention (case management) Phase 2: above + financial intervention (provider incentives)
Outcomes	Professional practice: proportion of patients who received BMD test within 6 months post‐fracture; proportion who had osteoporosis medication prescribed Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported

Gardner 2005.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: RCT Country: USA
Participants	Setting: intervention provided to patients in hospital, imaging outcome measured against primary care provider 80 patients Condition: hip fracture in elderly (>65 age) and high likelihood of osteoporosis
Interventions	1. Patient mediated (education materials + 15 minute talk) + list of questions to take to primary care physician 2. Control: unrelated patient education
Outcomes	Professional practice: proportion of patients who had BMD test performed within 6 months; bisphosphonate therapy prescribed Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	From report: “The patients were randomly enrolled into two groups by means of sealed envelopes, which were divided equally into ‘control group’ and ‘study group’ designations”
Allocation concealment?	Unclear risk	Reported that envelopes were used but not reported if they were opaque and sequentially numbered
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Low risk	4 from control (died); 4 from intervention (died) and 2 more lost to follow‐up considered to not have had BMD testing
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	High risk	No protection against contamination; patient directed intervention and outcomes self‐reported by patient.

Hollingworth 2002.

Methods	Type of targeted behaviour: decrease in test ordering (lumbar x‐rays) Study design: ITS Country: UK
Participants	Setting: primary care Number of practices and providers not reported. Analysed 2100 x‐ray referrals Condition: LBP (although only implied because no patients contacted or enrolled)
Interventions	Professional intervention (distribution of educational materials)
Outcomes	Professional practice: number of primary care referrals for radiography of the lumbar spine Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported

Jackson 2005.

Methods	Type of targeted behaviour: decrease in test ordering (lumbar x‐rays) + referral + prescribing Study design: ITS Country: USA
Participants	Setting: primary care Over 10 million patient encounters; number of physicians not reported Condition: acute LBP
Interventions	Professional intervention (distribution of educational materials)
Outcomes	Professional practice: number of radiographs ordered for LBP; medication recommended (acetaminophen, NSAIDs, muscle relaxant, narcotic); physiotherapy referral Patient level: none
Notes	Justification for intervention type: not applicable because measuring impact of release of guidelines Intervention fidelity: not reported

Kerry 2000.

Methods	Type of targeted behaviour: decrease in test ordering (x‐rays, various) Study design: cluster RCT Country: UK
Participants	Setting: primary care 69 practices, 175 providers, 43,778 radiological requests Condition: people potentially requiring an x‐ray of chest, spine or limbs and joints
Interventions	1. Professional intervention (distribution of educational materials + audit and feedback) 2. No intervention control group
Outcomes	Professional practice: number of x‐rays requested (chest, limbs & joints, spine) within 12 months Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	From report: “Practices were randomly allocated to an intervention or a control group using a stratified randomization”
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Unclear risk	Not reported
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Low risk

Lafata 2007.

Methods	Type of targeted behaviour: increase in test ordering (BMD testing) + prescribing Study design: cluster RCT Country: USA
Participants	Setting: primary care 15 practices, 123 providers, 10,354 patients Condition: women 65 to 89 years of age and high likelihood of osteoporosis
Interventions	1. Patient mediated (education materials) 2. Patient mediated (education materials) + professional intervention (reminder) 3. Standard practice control group
Outcomes	Professional practice: proportion of patients receiving BMD testing within 12 months; prescription of an osteoporosis medication Patient level: none
Notes	Justification for intervention type: previous studies for different clinical conditions Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	From report: “Within stratum, clinics were allocated to the three arms using a random numbers table”
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Low risk	Missing outcome data balanced in numbers across intervention groups, with similar reasons for missing data across groups
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Low risk

Majumdar 2004.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: non‐randomised CCT Country: Canada
Participants	Setting: primary care 101 providers, 102 patients Condition: fragility fracture of wrist with high likelihood of osteoporosis
Interventions	1. Professional intervention (distribution of educational materials + reminders) + patient mediated (educational materials + verbal education) 2. Standard care
Outcomes	Professional practice: proportion of patients who had BMD test ordered within 6 months; osteoporosis medication prescribed Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	High risk	From report: “For 1 month at a time, in sequential order, the intervention was ‘on’ at 1 emergency department while it was ‘off’ at the other.”
Allocation concealment?	High risk	Not reported
Blinding? All outcomes	Low risk	From report: “All outcomes were ascertained without knowledge of allocation status.”
Incomplete outcome data addressed? All outcomes	Low risk	No missing outcome data
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	High risk	Trial stopped early “because of overwhelming intervention efficacy and concerns related to continuing to enrol patients into the ‘usual care’ group.”; no protection against contamination

Majumdar 2007.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: RCT Country: Canada
Participants	Setting: hospital (inpatient) 220 patients Condition: people 50 years and older with hip fracture undergoing surgical fixation (high likelihood of osteoporosis)
Interventions	1. Organisational intervention (case management: case manager, profession not reported + scheduling BMD test) + patient education (provided in hospital about BMD testing and medication) 2. Standard practice control group
Outcomes	Professional practice: proportion of patients who had BMD testing within 6 months; bisphosphonate therapy Patient level: recurrent fractures; admission to hospital; death Cost: intervention cost per patient
Notes	Justification for intervention type: literature reviews and qualitative in‐depth interviews with health professionals Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	From report: “Patients who met inclusion and exclusion criteria and provided informed consent were randomized (allocation‐concealed) to either the case manager intervention or usual care.”
Allocation concealment?	Low risk
Blinding? All outcomes	Low risk	From report: “investigators and analysts... were masked to allocation status at all times.”
Incomplete outcome data addressed? All outcomes	Low risk
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Low risk

Majumdar 2008.

Methods	Type of targeted behaviour: general management of a problem Study design: RCT Country: Canada
Participants	Setting: primary care 266 providers, 272 patients Condition: 50 years or older and distal forearm fracture (high likelihood of osteoporosis)
Interventions	1. Professional intervention (distribution of educational materials + reminder) + patient mediated (education and counselling via telephone) 2. Patient mediated (distribution of educational materials)
Outcomes	Professional practice: proportion of patients who had received BMD test; prescription of osteoporosis medication; composite measure of quality of guideline‐concordant or “appropriate” care Patient level: health status (SF‐12); osteoporosis‐related quality of life; wrist‐related functional outcomes; osteoporosis‐related knowledge; satisfaction with care Cost: intervention cost per patient
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk
Allocation concealment?	Low risk	From report: “Allocation was concealed by application of variable block sizes and by use of a secure, centralized, Internet‐based, computer‐generated randomization system”
Blinding? All outcomes	Low risk	From report: “Research nurses collected outcomes data without knowledge of allocation status”
Incomplete outcome data addressed? All outcomes	Low risk	All included patients accounted for
Free of selective reporting?	Low risk	All outcomes listed in Methods were reported. Comment: Probably done.
Free of other bias?	Low risk

Matowe 2002.

Methods	Type of targeted behaviour: decrease in test ordering (x‐rays, various) Study design: ITS Country: UK (Scotland)
Participants	Setting: primary care Number of practices and providers not reported. Analysis included 117 747 imaging requests Condition: not reported. Imaging ordered was analysed, not condition. Imaging included x‐rays (spinal, chest, limbs, sinus) and ultrasound (abdominal, kidney, ureters, bladder, pelvis, testicular)
Interventions	Professional intervention (distribution of educational materials)
Outcomes	Professional practice: number of x‐ray referrals Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported

Oakeshott 1994.

Methods	Type of targeted behaviour: decrease in test ordering (x‐rays, various) Study design: cluster RCT Country: UK
Participants	Setting: primary care 62 practices. Analysis was of 2578 x‐ray examinations Condition: not specified as data collected only at test level, related to x‐rays of chest, limbs and joints, and spine
Interventions	1. Professional intervention (distribution of educational materials) 2. Control (not specified)
Outcomes	Professional practice: number of radiology requests within 9 weeks; percentage of radiology requests that conform to guidelines Patient: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	From report: “Practices were stratified by number of partners and number of radiographic examinations requested, and randomized into two groups.”
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Low risk	Blinding for conformity with guidelines. No blinding for other outcomes, but not likely to be influenced by lack of blinding
Incomplete outcome data addressed? All outcomes	Low risk	No missing outcome data
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Low risk

Prihar 2008.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) Study design: cluster RCT Country: USA
Participants	Setting: primary care 2 practices, 54 providers Condition: women 65 years and older high likelihood of osteoporosis
Interventions	1. Patient mediated (educational materials) + professional intervention (distribution of educational materials: poster) 2. No intervention control group
Outcomes	Professional practice: proportion of patients who underwent BMD testing within 2 months Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported Reported in letter to the editor and limited information available. Some additional information provided by first author Potential unit of analysis error
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	Coin tossing
Allocation concealment?	Low risk	Likely. Only two clinics were randomised
Blinding? All outcomes	High risk	From author correspondence: “Outcome of BMD screening was a completed bone density test.”
Incomplete outcome data addressed? All outcomes	Unclear risk	From author correspondence: “No, there were no women which could not be followed up unless some women went of our health system to get a bone density which we will not be able to know”
Free of selective reporting?	Low risk	From author correspondence: “No other outcomes were measured”
Free of other bias?	Unclear risk	Protection against contamination: not reported

Robling 2002.

Methods	Type of targeted behaviour: decrease in test ordering (lumbar and knee MRI) Study design: cluster RCT Country: UK
Participants	Setting: primary care 30 practices, 182 MRI requests Condition: people who potentially require MRI for knee or lumbar problems
Interventions	1. Professional intervention (distribution of educational materials + educational meetings, practice based) 2. Professional intervention (audit and feedback) 3. Professional intervention (1 + 2) 4. Control group: distribution of educational materials
Outcomes	Professional practice: proportion of MRI requests that are in concordance with guideline (length of follow‐up not clear) Patient level: none Cost: intervention cost
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	From report: “Randomization...was performed using a random numbers table”
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Unclear risk	Not reported
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Unclear risk	Not reported

Rossignol 2000.

Methods	Type of targeted behaviour: general management of a problem Study design: RCT Country: Canada
Participants	Setting: primary care 110 patients Condition: LBP
Interventions	1. Organisational intervention (clinical multidisciplinary team) 2. Usual care
Outcomes	Professional practice: proportion of patients who received lumbar imaging (x‐ray, CT, MRI or myelogram) within 6 months Patient level: return to work; function; health care consumption; satisfaction
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	Computer generated
Allocation concealment?	Unclear risk	Reported that envelopes used but not reported if they were opaque
Blinding? All outcomes	Low risk	All necessary researchers were blinded
Incomplete outcome data addressed? All outcomes	Unclear risk	No reason for missing data reported (20 patients (18%) lost to follow up)
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	High risk	Baseline differences in x‐ray rates between intervention groups; study underpowered

Rozental 2008.

Methods	Type of targeted behaviour: general management of a problem Study design: RCT Country: USA
Participants	Setting: primary care Number of providers not reported, 50 patients Condition: patients with fragility fracture of distal radius and probable osteoporosis
Interventions	1. Case management (treating orthopaedic surgeon ordered BMD test then forwarded results to primary care physician) 2. Professional intervention (distribution of educational materials: treating orthopaedic surgeon sent guideline to primary care physician)
Outcomes	Professional practice: proportion of patients who received BMD test within 6 months; osteoporosis treatment discussed; osteoporosis treatment administered Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	From report: “Patients were randomized to one of two interventions” was only information reported
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Low risk	No missing data
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	High risk	No protection against contamination. Patients randomised but intervention directed at providers, therefore providers could have patients in both intervention groups

Schectman 2003.

Methods	Type of targeted behaviour: general management of a problem Study design: cluster RCT Country: USA
Participants	Setting: primary care 14 practices, 106 providers, 4066 patients Condition: acute LBP
Interventions	1. Professional intervention (distribution of educational materials + educational meeting (including local opinion leader) + audit and feedback) 2. Patient mediated (education materials: pamphlet and video) 3. 1 + 2 4. No intervention control group Results presented as (1 + 3) vs (2 + 4).
Outcomes	Professional practice: proportion of lumbar plain x‐rays, CT or MRI consistent with guideline within 12 months; subspecialty referral; physiotherapy referral Patient level: beliefs about care, satisfaction with care, clinical outcome measures
Notes	Justification for intervention type: stated that needed multi‐factorial approach Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	From report: “Clinician practices were stratified by affiliation (academic vs nonacademic) and then, using sealed envelopes, randomized by an investigator to 4 groups in a 2 × 2 factorial design”
Allocation concealment?	Unclear risk	Envelopes used but not described as numbered or opaque
Blinding? All outcomes	Unclear risk	Chart audit, but no indication if assessors blinded to allocation or not
Incomplete outcome data addressed? All outcomes	Unclear risk	Not explicitly reported if all charts audited
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	High risk	Only 50% follow up of professionals; baseline imbalance for radiologic and specialty services; decision about grouping of analysis made post‐hoc after demonstrated no effect for patient education; potential unit of analysis error

Solomon 2004.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: cluster RCT Country: USA
Participants	Setting: rheumatology 1 practice, 21 providers, 373 patients Condition: rheumatoid arthritis and taking oral glucocorticoids (high likelihood of osteoporosis)
Interventions	1. Professional intervention (distribution of educational materials + educational meeting + audit and feedback) 2. No intervention control group
Outcomes	Professional practice: number of BMD tests ordered within 6 months; number of prescription medications for osteoporosis Patient level: none
Notes	Justification for intervention type: cite previous studies showing multifaceted interventions better than meetings Intervention fidelity: reported only that 9 of 10 clinicians attended meeting, otherwise no reporting of fidelity No raw data presented. Data read from histograms
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	From report: “To create a balanced study patient population, we stratified rheumatologists based on their patient volume and randomly assigned them to either intervention or control groups.”
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Author who measured outcome was not involved in intervention, but did not explicitly report if outcome measures were blinded
Incomplete outcome data addressed? All outcomes	Unclear risk	Not reported
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Unclear risk	Protection against contamination: not reported

Solomon 2007.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: cluster RCT Country: USA
Participants	Setting: primary care 828 providers, 13,455 patients Condition: people with high likelihood of osteoporosis and high risk of future fracture
Interventions	1. Professional intervention (distribution of educational materials + educational outreach) 2. Patient mediated (education materials) 3. 1 + 2 4. No intervention control group
Outcomes	Professional practice: number of patients who began osteoporosis medication or had BMD test within 12 months; any medication use; physical therapy utilisation; home visit Patient level: medication adherence; fracture of wrist, humerus or hip; attitudes, beliefs, knowledge and self‐reported behaviours
Notes	Justification for intervention type: patient intervention based on theory (Stages of Change Model and Witte’s Extended Parallel Processing Model); academic detailing: cite studies that have shown effectiveness Intervention fidelity: in part. Measured proportion of providers who received intervention
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	Random number generator
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Low risk	No missing data
Free of selective reporting?	High risk	Protocol outcomes not reported: any medication use; physical therapy utilisation; home visit; medication adherence; patient attitudes, beliefs, knowledge and self‐reported behaviours
Free of other bias?	High risk	Contamination of intervention possible because randomisation occurred at provider level, not practice level

Solomon 2007a.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) + prescribing Study design: cluster RCT Country: USA
Participants	Setting: primary care 434 providers, 1973 patients Condition: people with high likelihood of osteoporosis
Interventions	1. Professional intervention (distribution of educational materials + educational outreach + reminder) + patient mediated (education via automated telephone call) 2. Control group (not specified)
Outcomes	Professional practice: proportion of patients who received BMD test within 10 months; osteoporosis medications prescribed Patient level: fractures
Notes	Justification for intervention type: educational outreach based on theory (“Principles of academic detailing”) Intervention fidelity: in part. Reported on percentage of physicians who received outreach and percentage of patients who received telephone call
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	Not reported
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Unclear risk	Not reported
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Unclear risk	Not reported

Stock 1998.

Methods	Type of targeted behaviour: increase in test ordering (BMD test) Study design: cluster RCT Country: USA
Participants	Setting: primary care 68 providers Condition: people with high likelihood of osteoporosis
Interventions	1. Professional intervention (reminder: long narrative clinical reports on BMD test) 2. Usual care control group: short technical reports on BMD test
Outcomes	Professional practice: number of BMD tests ordered (length of follow‐up not clear); physician understanding of test results; referrals to specialists; further testing to rule out secondary causes of osteoporosis; pharmacologic treatment of osteoporosis Patient level: none
Notes	Justification for intervention type: not reported Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	Report states: “computer‐generated allocation sequence”
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Unclear risk	Not reported
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Unclear risk	Not reported

Verstappen 2003.

Methods	Type of targeted behaviour: modifying test ordering (various) Study design: cluster RCT (balanced incomplete block design) Country: The Netherlands
Participants	Setting: primary care 26 practices, 174 providers Condition: degenerative joint complaints (plus other conditions not relevant for this review, including cardiovascular disease/hypertension, upper/lower abdominal complaints, chronic obstructive pulmonary disease, asthma, fatigue)
Interventions	1. Professional intervention (distribution of educational materials + audit and feedback + educational meetings) 2. Same intervention but for a different group of conditions Due to study design and that musculoskeletal conditions only included in one arm of trial, this study represented as multifaceted intervention vs a no‐intervention control
Outcomes	Professional practice: total number of requested diagnostic tests within 6 months; number of inappropriate tests ordered Patient level: none
Notes	Justification for intervention type: cite previous studies about effectiveness of multifaceted interventions Intervention fidelity: not reported
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	From report: “...randomization was performed centrally with Duploran, a random number program”
Allocation concealment?	Low risk
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Low risk	No missing outcome data
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Low risk

Winkens 1995.

Methods	Type of targeted behaviour: modifying test ordering (various) Study design: cluster RCT (balanced incomplete block design) Country: The Netherlands
Participants	Setting: primary care 79 providers Condition: various conditions potentially requiring diagnostic tests, including cervical smear, electrocardiography, endoscopy, allergy tests, radiography (chest, cervical spine, pelvis, knees, ankles, sinuses) and ultrasound (kidneys, liver/biliary tract)
Interventions	1. Professional intervention (audit and feedback for one set of tests) 2. Professional intervention (audit and feedback for another set of tests) Due to study design and that musculoskeletal conditions only included in one arm of trial, this study represented as single intervention vs a no‐intervention control
Outcomes	Professional practice: number of diagnostic tests ordered within 2 years; diagnostic tests concordant with guideline Patient level: none
Notes	Justification for intervention type: not clear Intervention fidelity: not clear Data not available. Data not presented for individual tests
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Unclear risk	Not reported
Allocation concealment?	Unclear risk	Not reported
Blinding? All outcomes	Unclear risk	Not reported
Incomplete outcome data addressed? All outcomes	Unclear risk	Not reported
Free of selective reporting?	Unclear risk	Insufficient information provided
Free of other bias?	Unclear risk	Not reported

BMD: bone mineral density; CCT: controlled clinical trial; CT: computerised tomography; ITS: interrupted time series; LBP: low back pain; MRI: magnetic resonance imaging; NSAIDs: non‐steroidal anti‐inflammatory drugs; RCT: randomised controlled trial; UK: United Kingdom; USA: United States of America

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Ashe 2004	Patient directed intervention and no objective outcomes
Berwick 1986	Controlled before‐after study (CBA)
Bliuc 2006	Patient directed intervention and no objective outcomes
Broadhurst 2007	CBA
Buchbinder 2001	CBA
Casez 2006	CBA
Charalambous 2002	CBA
Che 2006	CBA
Chevalley 2002	CBA
de vos Meiring 1990	CBA
Deyo 2000	ITS study with only two pre‐intervention measurements
Freeborn 1997	CBA
Gallagher 1998	ITS study with only one pre‐ and two post‐intervention measurements
Geusens 2008	CBA
Greenfield 1975	Change of imaging behaviour was not a primary objective of the study
Hawker 2003	CBA with historical controls
Jousimaa 2002	Not specific for musculoskeletal conditions and data not available
Kane 2006	ITS study with only one pre‐ and two post‐intervention measurements
Kerry 2002	Participants randomised to x‐ray or not, and no imaging outcomes
Majumdar 2007a	Follow up of included study (Majumdar 2004), but all participants received the intervention so not controlled
Marchiori 2000	Abstract only, not enough information provided to make a decision. Contact with authors but unable to provide more information
Martin 1980	Not musculoskeletal condition specific
Moskowitz 2000	ITS study with only two pre‐ and three post‐intervention measurements
Pazirandeh 2002	CBA
Polinski 2006	CBA
Robling 1999	Knowledge only outcome
Solomon 2006	Patient directed intervention and no objective outcomes
Streeten 2006	CBA
Stross 1980	Change of imaging behaviour was not a primary objective of the study
Torgerson 1997	Randomised to BMD or not, no imaging outcomes
Wroe 2000	No control group for outcome of BMD test received

First reason for study exclusion listed

Characteristics of ongoing studies [ordered by study ID]

Bessette 2008.

Trial name or title	Recognizing osteoporosis and its consequences in Quebec (ROCQ)
Methods	Type of targeted behaviour: general management of osteoporosis Study design: RCT Country: Canada
Participants	Setting: primary care 672 patients (sample size calculation) Condition: women (>50 years) with fragility or traumatic fracture
Interventions	1. Patient mediated (educational video + written educational materials) + professional intervention (distribution of educational materials + reminder) 2. Patient mediated (written educational materials ) + professional intervention (distribution of educational materials + reminder) 3. No intervention control
Outcomes	BMD testing; initiation of osteoporosis pharmacological treatment; fragility fractures; health care resource utilisation; satisfaction; mortality
Starting date	June 2003
Contact information	Louis Bessette, email: rocq@crchul.ulaval.ca
Notes

McKenzie 2008.

Trial name or title	The IMPLEMENT study
Methods	Type of targeted behaviour: decrease in test ordering (lumbar x‐rays) Study design: cluster RCT Country: Australia
Participants	Setting: primary care (general practice) 92 general practices, 2300 patients (sample size calculation) Condition: adults with non‐specific low back pain
Interventions	1. Professional intervention (educational meetings + distribution of educational materials) 2. Professional intervention (distribution of educational materials)
Outcomes	GP level: lumbar x‐rays ordered; giving advice to stay active; any imaging referral; fear‐avoidance beliefs; measurement of behavioural constructs Patient level: low back pain disability (Roland‐Morris Disability Questionnaire); pain; fear‐avoidance beliefs; assessment of Quality of Life; health service utilisation
Starting date	June 2006
Contact information	Simon French, email: s.french@unimelb.edu.au
Notes	Trial Registration: Australian New Zealand Clinical Trials Registry ACTRN012606000098538 (date registered 14/03/2006). Website: http://www.cochrane.org.au/projects/implement.php

Differences between protocol and review

We modified some of the planned methods documented in the protocol in response to piloting and advances in the methods for systematic reviews. As noted in the main text of the review, we modified the search strategy, we changed the methods of assessment of risk of bias, and changed some of the methods of data analysis.

Contributions of authors

SF, RB and SG conceived and designed the review. SF and HB screened search results and SG and RB acted as arbitrators when disagreement arose. SF and HB extracted data from included studies. SF led the interpretation and write up of the results of the review, and SG and RB provided detailed comments and guidance on different aspects of the review.

Sources of support

Internal sources

• Australasian Cochrane Centre, Monash Institute of Health Services Research, Monash University, Australia.

External sources

• National Health and Medical Research Council (NHMRC) Public Health PhD Scholarship, Australia.

  Simon French undertook this review as part of his PhD, which was supported by a NHMRC Scholarship

Declarations of interest

Three of the review authors (SF, SG and RB) are involved in a trial that, when completed, will be considered for inclusion into this review. This is a cluster RCT of a strategy to implement the Australian acute low back pain guidelines into general practice, and one of the primary outcomes is to reduce the ordering of x‐rays. The expected completion date of this trial is the end of 2009. The study details are available in the table of Characteristics of ongoing studies.

No other conflicts of interest are known.

New