Parenting interventions for the prevention of unintentional injuries in childhood

Abstract

Background

Parent education and training programmes can improve maternal psychosocial health, child behavioural problems, and parenting practices. This review assesses the effects of parenting interventions for reducing child injury.

Objectives

To assess the effects of parenting interventions for preventing unintentional injury in children 18 years of age and younger and for increasing possession and use of safety equipment and safety practices by parents.

Search methods

We searched the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE, BIOSIS Preview, PsycINFO, Sociological Abstracts, Social Science Citation Index, CINAHL, ProQuest Dissertations and Theses, ERIC, DARE, ASSIA, Web of Science, SIGLE, and ZETOC. We also handsearched abstracts from the World Conferences on Injury Prevention and Control and from the journal Injury Prevention. The searches were conducted in January 2011.

Selection criteria

We included randomised controlled trials (RCTs), non‐randomised controlled trials (non‐RCTs), and controlled before and after studies (CBAs), which evaluated parenting interventions administered to parents of children 18 years of age and younger and reported outcome data on injuries to children (unintentional or unspecified intent) and possession and use of safety equipment or safety practices (including the Home Observation for Measurement of the Environment (HOME) scale, which contains an assessment of home safety) by parents. Parenting interventions were defined as those with a specified protocol, manual, or curriculum aimed at changing knowledge, attitudes, or skills covering a range of parenting topics.

Data collection and analysis

Studies were selected, data were extracted, and quality was appraised independently by two authors. Pooled relative risks (RRs) were estimated using random‐effects models.

Main results

Twenty‐two studies were included in the review: 16 RCTs, two non‐RCTs, one partially randomised trial that contained two randomised intervention arms and one non‐randomised control arm, two CBA studies, and one quasi‐randomised controlled trial. Seventeen studies provided interventions comprising parenting education and other support services, 15 of which were home visiting programmes and two of which were paediatric practice‐based interventions. Two provided solely educational interventions. Nineteen studies recruited families who were from socio‐economically disadvantaged populations who were at risk of adverse child outcomes or people who may benefit from extra support, such as single mothers, teenage mothers, first‐time mothers, and mothers with learning difficulties. Ten RCTs involving 5074 participants were included in the meta‐analysis, which indicated that intervention families had a statistically significant lower risk of injury than control families (RR 0.83, 95% confidence interval (CI) 0.73 to 0.94). Sensitivity analyses undertaken that included only RCTs at low risk of various sources of bias found the findings to be robust to including only those studies at low risk of detection bias in terms of blinded outcome assessment and attrition bias in terms of follow‐up of fewer than 80% of participants in each arm. When analyses were restricted to studies at low risk of selection bias in terms of inadequate allocation concealment, the effect size was no longer statistically significant. Several studies found statistically significantly fewer home hazards or a greater number of safety practices in intervention families. Of ten studies reporting scores on the HOME scale, data from three RCTs were included in a meta‐analysis, which found no evidence of a difference in quality of the home environment between treatment arms (mean difference 0.57, 95% CI ‐0.59 to 1.72). Most of the studies reporting home safety practices, home hazards, or composite home safety scores found statistically significant effects favouring intervention arm families. Overall, when Grading of Recommendations Assessment, Development, and Evaluation guidelines (GRADE) were used, the quality of the evidence was rated as moderate.

Authors' conclusions

Parenting interventions, most commonly provided within the home using multi‐faceted interventions, are effective in reducing child injury. Fairly consistent evidence suggests that they also improve home safety. This evidence relates mainly to interventions provided to families from disadvantaged populations, who are at risk of adverse child health outcomes, or whose families may benefit from extra support. Further research is required to explore mechanisms by which these interventions may reduce injury, to identify the features of parenting interventions that are necessary or sufficient to reduce injury, and to assess the generalisability of these findings to different population groups.

Plain language summary

Parenting programmes for the prevention of unintentional injuries in childhood  

This review looked at whether parent education and training programmes (called parenting programmes) help parents provide a safer home environment and reduce injuries in children. We undertook the review because several factors that can increase the risk of unintentional injury to children may be helped by parents' attending parenting programmes, such as mothers' mental health and child behaviour problems. In addition, injuries are more likely to occur when parents are unable to predict a child’s ability to perform tasks such as climbing furniture or opening locks. Parenting programmes may help parents develop realistic expectations of their child's behaviour based on age and stage of development. Therefore, we wanted to assess whether parenting programmes reduced the risk of unintentional injuries in children, and whether parents provided a safer home environment by using more items of safety equipment, such as stair gates, and by adopting safe practices such as keeping medicines out of reach.

Through searches of databases and web sites, we found 22 randomised and non‐randomised studies that evaluated the effects of parenting programmes on childhood injuries or home safety. Fifteen of these were home visiting programmes that provided a range of support services, as well as parent education or training. These programmes were usually provided to families who were disadvantaged, whose children were considered to be at risk of poor health, or who may benefit from extra support.

We pooled the results from 10 randomised controlled trials, which included a total of 5074 children, and found that children from families who had completed the parenting programmes had sustained fewer injuries than those from families who had not attended the programmes. We pooled the results from three randomized controlled trials that measured home safety using the Home Observation for Measurement of the Environment (HOME) tool. The results from these three studies, which included a total of 368 children, showed no difference in HOME scores between families receiving parenting programmes and those not receiving these programmes. Overall, the quality of the studies was reasonable.

We conclude that parenting programmes are effective in reducing unintentional injury in children and can improve home safety, particularly in families who may be considered 'at risk,' such as some teenage or single mothers. It would be worthwhile for health and social care providers to make parenting programmes available to families.

Summary of findings

Summary of findings for the main comparison. Parenting interventions compared to usual care for the prevention of unintentional injuries in childhood.

Parenting interventions compared to usual care for the prevention of unintentional injuries in childhood
Patient or population: parents of children 18 years of age and younger Settings: Intervention: parenting interventions Comparison: usual care
Outcomes	Illustrative comparative risks* (95% CI)		Relative effect (95% CI)	No of participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Usual care	Parenting interventions
Medically attended or self‐reported injuries ‐ RCTs only	Low1		RR 0.83 (0.73 to 0.94)	5074 (10 studies)	⊕⊕⊕⊝ Moderate2
	69 per 1000	57 per 1000 (50 to 65)
	High1
	237 per 1000	197 per 1000 (173 to 223)
*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio.
GRADE Working Group grades of evidence: High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.

¹ Two RCTs recruited families considered at low risk of unintentional injury (low birth weight and premature infants (IHDP) and families with newborns up to 4 weeks of age (Minkowitz(a)). The remaining eight RCTs recruited families considered at high risk of unintentional injury (defined as having at least one risk factor for unintentional injury).
 ² Half or more of the RCTs were susceptible to bias in terms of allocation concealment and/or outcome assessment.

Background

Description of the condition

In industrialised countries, injuries are the leading cause of death in childhood, accounting for 40% of all child deaths between the ages of 1 and 14 years (UNICEF 2001). In the UK, more than 160 children die from an injury (ONS 2009) each year. In addition, more than 100,000 hospital admissions (The NHS Information Centre 2007) and more than two million attendances at accident and emergency departments resulting from child injuries occur each year (DTI 2003). A steep social class gradient in child injury mortality (Roberts 1997; Edwards 2006) and morbidity (Hippisley‐Cox 2002) has been noted, along with evidence that the gradient in injury mortality may be widening (Roberts 1996).

Description of the intervention

Parenting programmes are aimed at strengthening the quality of the parent‐child relationship by improving parenting practices; aspects of parental functioning such as mental health, attributions, and cognitions; the child's emotional and behavioural adjustment; and family functioning more generally. These programmes can be provided in a variety of settings such as at home, in clinics, and in other community venues. Several recent systematic reviews have shown that they are effective in helping with behaviour problems in infants and toddlers (Barlow 2002) and in 3‐ to 10‐year‐old children (Barlow 2000), and in improving maternal psychosocial health in the short term, including reducing scores measuring anxiety and depression and enhancing self esteem (Barlow 2003) and self confidence. Evidence also shows that they are effective in improving outcomes for both teenage mothers and their children (Coren 2002).

How the intervention might work

Parenting programmes may reduce child injury rates through several mechanisms. First, evidence shows that maternal psychological well‐being is associated with childhood unintentional injury. Higher unintentional injury rates have been found amongst children whose mothers are defined as having a psychiatric disorder (Brown 1978) or suffering from depression (Beautrais 1981; Harris 1994; O'Connor 2000) or anxiety (Bradbury 1999); as having experienced a greater number of life events, such as separation from their partner, recent bereavement, or moving the household (Beautrais 1982; O'Connor 2000); or as having higher levels of stress (Harris 1994). A large US cohort study found that depressed mothers were significantly less likely to engage in safety practices such as use of a child car seat or electric socket covers, and were less likely to keep an emetic agent for the treatment of poisoning in the home (McLennan 2000). It is therefore possible that parenting programmes may help reduce childhood unintentional injuries by improving maternal psychological health. Second, child behavioural problems, including aggressive or overactive behaviour and attention‐deficit hyperactivity disorder (ADHD) (Miller 2004; Rowe 2004; Lam 2005), are associated with increased unintentional injury rates (Bijur 1986; Bijur 1988a; Bijur 1988b; Bussing 1996), and parenting programmes, by reducing child behavioural problems, may help to reduce such injuries. Third, injuries can occur when parents are unable to predict the child’s ability to perform tasks such as climbing or opening locks, or when parents expect children to understand and remember instructions aimed at keeping them safe from injury (Smithson 2011). Parenting programmes teach parents realistic expectations that are appropriate for a child’s age and developmental stage, thus potentially reducing the risk of injury (Sanders 2002; Hunt 2003). Data from a systematic review of home visiting programmes provide some evidence to support these suggested mechanisms; home visiting programmes were found to be effective in improving parenting, maternal psychosocial health, and child behavioural problems, and in reducing childhood unintentional injuries (Elkan 2000; Kendrick 2000). More recent work suggests that positive parenting (e.g. frequent praise, playing with children) is associated with a protective effect on injury rates (Schwebel 2004; Soubhi 2004). It is therefore possible that parenting programmes may help prevent childhood unintentional injury by developing particular styles of parenting. We are therefore interested in assessing the effectiveness of parenting programmes in preventing childhood injury.

Why it is important to do this review

We were unable to find any systematic reviews examining the effect of parenting programmes on the prevention of childhood unintentional injury for the original review; thus this review was undertaken. A recent search failed to find any recent reviews published since our original Cochrane review.

Objectives

The primary objective of the review is to evaluate the effectiveness of parenting programmes in preventing unintentional injury in childhood.

The secondary objective of the review is to evaluate the effectiveness of parenting programmes in increasing possession and use of home safety equipment and in improving parental safety practices.

Methods

Criteria for considering studies for this review

Types of studies

We included individually and cluster randomised controlled trials, non‐randomised controlled trials, and controlled before and after studies.

Types of participants

We included studies in which the intervention was provided for parents of children 18 years of age and younger.

Types of interventions

Studies evaluating the effectiveness of individual and group‐based parenting interventions were eligible for inclusion irrespective of the theoretical basis underpinning the programme. Parenting programmes are defined as interventions with a specified protocol, manual, or curriculum aimed at changing knowledge, attitudes, or skills covering a range of topics relevant to parenting. The control group in studies should not be provided with a parenting intervention.

Types of outcome measures

Studies were eligible for inclusion if they reported the primary outcome of:

• Self‐reported or medically attended unintentional injury or injury of unspecified intent.

or secondary outcomes of:

• Possession and use of home safety equipment.

• Safety practices (e.g. storage of hazardous objects and substances, safe use of baby walkers, avoidance of unsafe hot water temperature) and composite measures of safety practices, including the Home Observation for Measurement of the Environment (HOME) Inventory. The Infant‐Toddler HOME Inventory, designed for use with children from birth to 3 years of age, consists of six subscales: emotional and verbal responsivity of the primary caregiver (items 1 to 11); avoidance of restriction and punishment (items 12 to 19); organisation of the physical and temporal environment (items 20 to 25), including "the child’s play environment appears safe and free of hazards"; provision of appropriate play materials (items 26 to 34); parental involvement with the child (items 35 to 40); and opportunities for variety in daily stimulation (items 40 to 45) (Caldwell 2003; Totsika 2004). A higher score indicates a more appropriate home environment for child development.

Intentional injury outcomes were excluded as these are the focus of another Cochrane review (Barlow 2006). Parent training interventions for children with ADHD, which are the subject of a protocol for a Cochrane review (Zwi 2011), were included because injury is not an outcome listed in the protocol.

Search methods for identification of studies

The search strategy was selected to identify randomised and non‐randomised studies using terms to identify parenting programmes, injuries, safety equipment, and safety practices.

Electronic searches

We searched the following databases, with no language restrictions, from the date of inception to January 2011 (from 2005 to 2011 for the update):

• Cochrane Central Register of Controlled Trials (CENTRAL), Issue 1, 2011.

• Cochrane Database of Systematic Reviews (CDSR).

• MEDLINE (Ovid SP), 1950 to January 2011.

• EMBASE (Ovid SP), 1980 to January 2011.

• BIOSIS Previews (was Biological Abstract), 1969 to January 2011.

• PsycINFO (Ovid), 1806 to January.

• Sociological Abstracts (was Sociofile) (CSA), 1952 to January 2011.

• CINAHL (EBSCO), 1982 to January 2011.

• ProQuest Dissertations and Theses (was Dissertation Abstracts), 1743 to January 2011.

• ERIC, 1966 to January 2011.

• Database of Abstracts of Reviews of Effectiveness, 1994 to January 2011.

• ASSIA, 1987 to January 2011.

• ISI Web of Science: Social Sciences Citation Index (SSCI), 1970 to January 2011.

• ISI Web of Science: Conference Proceedings Citation Index‐Social Science & Humanities (CPCI‐SSH), 1990 to January 2011.

• SIGLE (ceased March 2005).

• ZETOC, 1993 to January 2011.

We also searched a range of websites, including:

• Injury Prevention Research Center at the Centers for Disease Control and Prevention (USA) (http://www.cdc.gov/injury/ ‐ searched December 2010).

• National Institute for Health and Clinical Excellence (NICE) (http://www.nice.org.uk/ ‐ searched December 2010).

• Public Health website (UK) (http://www.dh.gov.uk/en/index.htm ‐ searched December 2010).

• Children's Safety Network (USA) (http://www.childrenssafetynetwork.org/ ‐ searched December 2010).

• International Society for Child and Adolescent Injury Prevention (International) (http://iscaip.net/iscaip/ ‐ searched December 2010).

• Child Accident Prevention Trust (UK) (http://www.capt.org.uk/ ‐ searched December 2010).

• Injury Control Resource Information Network (USA) (http://www.injurycontrol.com/icrin/research.htm ‐ searched December 2010).

• National Injury Surveillance Unit (Australia) (http://www.nisu.flinders.edu.au/ ‐ searched December 2010).

• Injury Prevention Web and SafetyLit (USA) (http://www.safetylit.org/ ‐ searched December 2010).

• Barnado's Policy and Research Unit (UK) (http://www.barnardos.org.uk/what_we_do/policy_research_unit.htm ‐ searched December 2010).

• NCH (UK) (http://www.actionforchildren.org.uk/ ‐ searched December 2010).

• National Children's Bureau (UK) (http://www.ncb.org.uk/ ‐ searched December 2010).

• Children in Wales (UK) (http://www.childreninwales.org.uk/index.html ‐ searched December 2010).

• Homestart (UK) (http://www.home‐start.org.uk/homepage ‐ searched December 2010).

Searching other resources

We handsearched abstracts from the 1st to 10th World Conferences on Injury Prevention and Control and the table of contents for the journal Injury Prevention, from the first publication to January 2011. We also handsearched reference lists of articles identified through database searches and bibliographies of systematic and non‐systematic reviews. Injuries, safety equipment, or safety practices may have been secondary outcome measures in clinical studies; therefore, we attempted to contact authors of studies excluded because of the outcomes they reported to ascertain if they had measured, but not reported, outcomes relevant for our review.

To identify unpublished studies, we searched the following sources:

• The Cochrane Library.

• Current Controlled Trials.

• National Research Register (NRR) (up to September 2007, the date of its closure).

• UK Clinical Research Network Study Portfolio.

No restrictions by language or publication status were applied.

Data collection and analysis

Selection of studies

A two‐stage screening process was undertaken. Two review authors independently scanned titles and abstracts of articles to identify articles that could be retrieved in full. In cases of disagreement between review authors, the decision was made by a third review author. The full article was retrieved for those articles retained at this stage. Two review authors independently assessed selected articles using a standard form listing inclusion criteria, with disagreements dealt with by referral to a third review author.

Data extraction and management

Data extraction was undertaken independently by pairs of review authors (DK, CM, TS, LY) using a pre‐tested data extraction form. For each study, we extracted data on the following: age of participants; country where study was set; whether participants were considered at risk of non‐accidental injury; type of intervention (e.g. group‐based, individual); aim of intervention in terms of changing parenting knowledge, parenting attitudes, or parenting skills, and in preventing non‐accidental injury, improving child behaviour, and improving maternal psychosocial health; who delivered the intervention and where; the numbers of contacts and sessions; and the length of each contact/session and over what time it occurred. Data on the study population were extracted, such as living in a deprived area, age of the mother, years/level of maternal education, single parenthood, and ethnic group. We also extracted data on study design, the numbers of participants recruited for intervention and control groups, the number of intervention group parents who did not receive the intervention, and the number who completed the study.

In terms of primary outcomes (medically attended injuries) and secondary outcomes (possessions and use of safety equipment) of interest, we extracted data on the numbers of participants in the intervention and control groups and the numbers with the outcome of interest. For the primary outcome, these data were extracted at child, family, or cluster level, as appropriate to the study design. Where data were presented as the mean number of injuries over a period of time (m), we estimated the probability of a participant's not having an injury by assuming that the occurrence of injuries followed a Poisson distribution using e^‐m, and from this estimated the numbers of children who did and did not experience at least one injury.

For the secondary outcome of HOME score, we extracted the mean score and the standard deviation (SD) for the intervention and control groups. For cluster randomised controlled trials, we extracted the intraclass correlation coefficients (ICCs) for the primary and secondary outcome measures, if reported. We also extracted data on study outcomes, and we extracted data on the period over which outcomes were measured.

If key data were not available in the published reports, we contacted study authors to obtain missing information.

Assessment of risk of bias in included studies

Critical appraisal of included studies was undertaken independently by two review authors, who covered the following sources of bias for randomised controlled trials (RCTs):

• Random sequence generation (selection bias).

• Allocation concealment (selection bias).

• Blinding (performance bias and detection bias).

• Blinding of participants and personnel (performance bias).

• Blinding of outcome assessment (detection bias).

• Incomplete outcome data (attrition bias).

• Selective reporting (reporting bias).

• Other bias.

The following sources of bias were assessed for non‐randomised studies:

• Participant selection (selection bias).

• Blinding (performance bias and detection bias).

• Blinding of participants and personnel (performance bias).

• Blinding of outcome assessment (detection bias).

• Incomplete outcome data (attrition bias).

• Selective reporting (reporting bias).

• Risk of bias due to confounding: Was the distribution of confounders assessed between treatment arms? If so, do treatment arms appear similar in terms of confounders?

• Other bias.

Review authors gave a brief description of possible sources of each type of bias and rated the risk of bias as high, low, unclear, or unknown. Disagreement between review authors was dealt with by referral to a third review author.

Measures of treatment effect

Pooled relative risks and 95% confidence intervals have been used for binary outcome measures and mean differences, and 95% confidence intervals for continuous outcome measures.

Unit of analysis issues

We adjusted the reported treatment effect in cluster allocated studies reporting binary outcomes as numerators and denominators, unadjusted for clustering, using the ICC of the study if available, otherwise using the ICC of similar cluster randomised studies. We adjusted numerators and denominators by using the design effect, which was calculated from the ICC and the average cluster size.

Dealing with missing data

We assessed missing data and dropouts for each included study and reported the number of participants included in the final analysis as a proportion of all participants in each study. Reasons for missing data are provided in the narrative summary, where available, and we assessed the extent to which the results of the review could be altered by the missing data. The extent to which studies conformed to an intention‐to‐treat analysis and the effects of this on the results were also assessed.

Assessment of heterogeneity

Statistical tests of homogeneity were undertaken using Chi² tests (with significance defined as a P value < 0.1) and the I² statistic. The I² statistic describes the percentage of total variation across studies caused by heterogeneity rather than by chance. A value of 0% indicates no observed heterogeneity, and larger values show increasing heterogeneity; substantial heterogeneity is considered to exist when I² > 50% (Higgins 2005). Pooled relative risks (RRs) and 95% confidence intervals (CIs) have been estimated for the binary outcome of occurrence of at least one self‐reported or medically attended injury, with random‐effects models used to allow for heterogeneity.

Assessment of reporting biases

Publication bias was assessed for the primary analysis with the use of a funnel plot and Egger's test (using Stata, version 11).

Data synthesis

Pooled RRs and 95% CIs were estimated for occurrence of at least one self‐reported or medically attended injury using data from included RCTs. Pooled mean differences and 95% CIs were estimated for HOME scale scores using data from included RCTs. Random‐effects models were used to allow for and to quantify the degree of statistical heterogeneity present between individual studies. When clinically homogenous studies were insufficient to be combined in a meta‐analysis, or when findings were derived from non‐RCTs, the results were combined in a narrative review. We produced a "summary of findings table" for the primary outcome of injuries and assessed the evidence using Grading of Recommendations Assessment, Development, and Evaluation guidelines (GRADE).

Subgroup analysis and investigation of heterogeneity

No subgroup analyses have been undertaken.

Sensitivity analysis

Sensitivity analyses have been undertaken that included only RCTs considered to be at low risk of selection bias in terms of adequate allocation concealment, detection bias in terms of blinded outcome assessment, and attrition bias as the result of follow‐up of less than 80% of participants in each arm. Sensitivity analyses have also been undertaken by excluding one study when there was some uncertainty as to the extent to which the intervention was based on a protocol, manual, or curriculum (Fergusson 2005); and by including longer term (seven years), but less complete, follow‐up data from one study (Johnson 2000) whose short‐term follow‐up data (Johnson 1993) were included in the main analysis. Finally, one study reported a range of injury types (bruises, burns, scratches, and unspecified injury, separately), and the most commonly reported outcome (bruises) was chosen for inclusion in the main analysis; sensitivity analyses were undertaken for the other outcomes (Armstrong 2000).

Results

Description of studies

Twenty‐two studies from 30 articles were included in the review and are described in the table 'Characteristics of included studies' (Characteristics of included studies; Table 2). The process of study selection is documented in Figure 1 (Figure 1). Several authors report results from the same study at different follow‐up time points in separate papers: two papers presented results from one RCT at child's first birthday (Johnson 1993) and at 7‐year follow‐up (Johnson 2000); two articles presented results from one RCT at 12 (Barlow 2007) and 36 months (Barlow 2008) postnatally; two papers presented results from one RCT at two years (Olds 1986) and at 25‐ to 50‐month follow‐up (Olds 1994); two papers presented results from one RCT at four months (Armstrong 2000) and at 18 months (Fraser 2000) postnatally; two papers presented results from a non‐RCT at one week and three months postpartum (Johnston 2004) and at 30 months postpartum (Johnston 2006; and two papers presented results from one RCT at two years (Duggan 1999) and three years (Duggan 2004) postrecruitment. Similarly, several authors report results from the same study in more than one paper: Results from one RCT collected at the time of the child's second birthday are presented in two papers (Caldera 2007)(Duggan 2007); results from one RCT collected at 1, 12, and 24 months postpartum are presented in two papers (Kemp 2008; Kemp 2011); and results from one RCT collected annually on the child's birthday for five years are presented in two papers (St Pierre 1999) (Goodson 2000). One paper presents the results from both an RCT (Minkovitz 2003 (a)) and a CBA (Minkovitz 2003 (b)).

1. Summary of included studies: participant and intervention characteristics.

Study	First author, year First author, year (second paper) Study design	Study population	Age of child at start of programme	Length of programme (months)	Planned total contacts and plan of contacts	Actual total visits	Delivery setting	Delivery mode	Outcomes
1	Armstrong 2000 Fraser 2000 RCT	EITHER one of: physical domestic violence; childhood abuse of either parent; sole parenthood; ambivalence to the pregnancy. OR three of: maternal age < 18 years; unstable housing; financial stress; < 10 years' maternal education; low family income; social isolation; history of mental health disorder; drug/alcohol abuse; domestic violence	birth	6	approx 12 (weekly for 6 weeks, fortnightly until 3 months, monthly until 6 months)	6‐13	home	individual	injuries HOME score
2	Barlow 2007 Barlow 2008 RCT	vulnerable women using a range of criteria (e.g. mental health problems, domestic violence, drug/alcohol abuse)	prenatal	18	approx 72 (weekly visits)	mean 41.2	home	individual	HOME score
3	Caldera 2007 Duggan 2007 RCT	"At‐risk" families: risks included maternal mental health, maternal substance abuse, partner violence	50% prenatal 50% birth	24	approx 32 (weekly for first 6 to 9 months)	year 1: mean 22.1 (range 15.4‐26.4) year 2: mean 20.0 (range 11.7 ‐27.3)	home	individual	injuries HOME score
4	Culp 2007 cluster non‐RCT	first‐time mothers living in rural community	prenatal	15	approx 40 (weekly in 1st month, fortnightly for remainder of pregnancy, weekly for 1st 3 months postpartum, fortnightly from 3 to 12 months postpartum)	mean 10.9 prenatal mean 20.7 0‐12 months	home	individual	injuries HOME safety*
5	Duggan 1999 Duggan 2004 RCT	families classified as at risk of child abuse and neglect	< 1 month	24	approx 45 (weekly visits decreasing to quarterly)	mean 22 (in year 1)	home	individual	injuries HOME score
6	Emond 2002 cluster CBA	first‐time parents from deprived areas	prenatal	11	approx 8 (antenatal, birth, 3 weeks postnatal, then every 5 weeks until 8 months)	not reported	home	individual	injuries HOME safety*
7	Feldman 1992 RCT	mentally retarded mothers	1‐23 months	2	as many as necessary for mother to learn skills (weekly)	mean 7.7 (range 2‐29)	home	individual	HOME safety*
8	Fergusson 2005 RCT	2 or more of the following risk factors: age of parents, social support, planning of pregnancy, parental substance abuse, financial situation, family violence	< 3 months	36	not reported	not reported	home	individual	injuries
9	Gutelius 1977 RCT	primigravidas, unmarried, women aged 15‐19 years, mothers to black babies, from deprived areas	prenatal	36	approx 64 routine visits (at least 9, 6, and 4 in years 1, 2, and 3, respectively); cognitive stimulation visits (at least 18, 12, and 8 in years 1, 2, and 3, respectively) group sessions (16)	at least 64	home and elsewhere	individual and group	injuries
10	Hardy 1989 quasi‐RCT	inner‐city mothers of black babies, aged 18 years +	shortly after birth (< 10 days old)	24	approx 10 (first at 7‐10 days, then 2‐3 weeks before well‐child clinic appointments at 2, 4, 6, 9, 12, 15, 18, 21, and 24 months)	not reported	home	individual	injuries
11	IHDP 1990 RCT	mothers of low birth weight babies	birth	36	approx 116 (weekly year 1, fortnightly years 2 and 3, bimonthly group meetings years 2 and 3)	not reported	home and elsewhere	individual and group	injuries
12	Johnson 1993 Johnson 2000 RCT	first‐time mothers living in deprived areas	birth	12	approx 14 (once a month for first year + birth and 6‐week visit)	Of 127 mothers, 82 (65%) received at least 10 visits, 34 (27%) received between 5 and 9 visits, and 11 (9%) received fewer than 5 visits	home	individual	injuries
13	Johnston 2004 Johnston 2006 cluster RCT	pregnant women	prenatal	7	approx 7 (3 prenatal home visits + 4 postnatal home and office visits)	not reported	home and elsewhere	individual and group	HOME safety*
14	Kemp 2008 Kemp 2011 RCT	1 or more of the following risk factors: age < 19 years, current probable distress (assessed as an Edinburgh Depression Scale (EDS) score of 10+), lack of emotional and practical support, later antenatal care (20 weeks+ gestation), major stressors in the past 12 months, current substance misuse, current or history of mental health problem or disorder, history of abuse in mother's own childhood, history of domestic violence	prenatal	27	approx 27 (antenatally: at least every second week; postnatally: weekly until 6 weeks; fortnightly until 12 weeks; monthly to 6 months; bimonthly until 2 years)	mean 16.3 (range 0‐52 visits)	home	individual and group	HOME score
15	Kitzman 1997 RCT	African‐American women, no previous births, with at least 2 risk factors: unmarried, < 12 years' education, unemployed	prenatal	27	average of 7 (range 0‐18) home visits during pregnancy, and 26 (range 0‐71) visits from birth to child's second birthday	During pregnancy: mean 7 (range 0‐18) During 2 years postpartum: mean 26 (range 0‐71)	home	individual	injuries HOME score
16	Koniak‐Griffin 2003 RCT	pregnant women aged 14‐19 years, first‐time mothers	prenatal	15	max 17 (2 prenatal and 15 postnatal)	prenatal: mean 2.13 (SD 0.77) postnatal: mean 10.35 (SD 3.04)	home	individual and group	injuries HOME score
17	Larson 1980 partial RCT	18‐35 years, working class income, high school graduation or less	prenatal	18	group A: 1 prenatal, 1 postpartum hospital visit, 4 visits 1‐6 weeks, 5 visits 6 weeks‐15 months group B: 7 visits 6 weeks‐6 months, 3 visits 6‐15 months	not reported	home	individual	injuries HOME score
18	Llewellyn 2003 RCT	parents with intellectual disability	younger than 5 years	3	10 weekly	mean 11.5	home	individual	HOME safety*
19 20	Minkovitz 2003a, RCT Minkovitz 2003b, cluster CBA	parents of newborns	newborns < 4 weeks	36	6 home visits in first 3 years + parent groups	mean home visits 2 in 2.5 years + family made mean 11 well‐child visits	home and elsewhere	individual and group	injuries HOME safety*
21	Olds 1986 Olds 1994 RCT	one of the following characteristics that predispose to infant health and developmental problems: age < 19 years; single parent; low socioeconomic status	prenatal	27	intervention 1: free transport to prenatal and well‐child visits (0 home visits); intervention 2: free transport to prenatal and well‐child visits + fortnightly home visits during pregnancy (6 home visits); intervention 3: free transport to prenatal and well‐child visits + fortnightly home visits during pregnancy + postnatal weekly visits from birth‐6 weeks, every 2 weeks from 6 weeks‐4 months, every 3 weeks from 4 months‐14 months; every 4 weeks from 14‐20 months; every 6 weeks from 20‐24 months (39 home visits)	mean visits: intervention 2‐9 intervention 3‐23	home and elsewhere	individual	injuries HOME safety* HOME score
22	St Pierre 1999 Goodson 2000 RCT	woman pregnant or had a baby < 1 year and a family income at or below poverty guidelines	Child < 1 year	60	approx 78 home visits (fortnightly visits from ages birth‐3 years)	2‐4 visits per month. One third of families were in programme for 5 years, 15% for 4 years, 34% for 1 ‐3 years, 18% < 1 year	home and elsewhere	individual and group	HOME score

*Home safety: includes home safety practices and/or use of home safety equipment and/or home hazards.

1.

Quorum flow chart detailing process of study selection for all studies included in the review.

Types of studies

Sixteen (73%) included studies were RCTs (Gutelius 1977; Olds 1986; IHDP 1990; Feldman 1992; Johnson 1993; Kitzman 1997; Duggan 1999; St Pierre 1999; Armstrong 2000; Koniak‐Griffin 2003; Llewellyn 2003; Minkovitz 2003 (a); Fergusson 2005; Barlow 2007; Caldera 2007; Kemp 2011), two (9%) were non‐RCTs (Johnston 2004; Culp 2007), one (5%) was a partially randomised study with two randomised intervention arms and one non‐randomised control arm (Larson 1980), two (9%) were CBA studies (Emond 2002; Minkovitz 2003 (b)), and one (5%) was a quasi‐RCT (Hardy 1989). Four studies used clustered allocation (Emond 2002; Minkovitz 2003 (b); Johnston 2004; Culp 2007). Thirteen studies (59%) were from the United States, three from Australia (14%), two (9%) each from Canada and England, and one (5%) each from Ireland and New Zealand.

Types of participants

Fifteen of the studies recruited socio‐economically disadvantaged participants (Gutelius 1977; Larson 1980; Olds 1986; Hardy 1989; Johnson 1993; Kitzman 1997; Duggan 1999; St Pierre 1999; Armstrong 2000; Emond 2002; Koniak‐Griffin 2003; Fergusson 2005; Barlow 2007; Caldera 2007; Kemp 2011); five of these studies specifically recruited participants considered to be at risk of child abuse or neglect (Olds 1986; Duggan 1999; Armstrong 2000; Barlow 2007; Caldera 2007), and two specifically recruited young mothers (15 to 19 years of age) (Gutelius 1977; Koniak‐Griffin 2003). Two studies recruited participants with a learning disability (Feldman 1992; Llewellyn 2003), one study recruited mothers who were considered at risk of poor coping as a parent (Kemp 2011), one study recruited mothers of low birth weight premature infants (IHDP 1990), two studies recruited first‐time mothers ( Johnson 1993; Culp 2007), and three studies recruited consecutive newborns from a range of paediatric practices (Minkovitz 2003 (a); Minkovitz 2003 (b); Johnston 2004).

Types of interventions

Seventeen studies evaluated multi‐faceted home visiting programmes aimed at improving a range of child and often maternal health outcomes (Gutelius 1977; Larson 1980; Olds 1986; Hardy 1989; IHDP 1990; Johnson 1993; Kitzman 1997; Duggan 1999; St Pierre 1999; Armstrong 2000; Emond 2002; Koniak‐Griffin 2003; Fergusson 2005; Barlow 2007; Caldera 2007; Culp 2007; Kemp 2011). Three studies evaluated paediatric practice‐based multi‐faceted interventions, aimed at improving a range of child health outcomes, all of which included some home visits (Minkovitz 2003 (a); Minkovitz 2003 (b); Johnston 2004), but these were not the main method of delivery of the intervention. Two studies provided solely educational interventions in the home (Feldman 1992; Llewellyn 2003). The 20 studies evaluating multi‐faceted interventions provided both parenting education and a range of other support services, including 12 that helped the family with solving a variety of problems (Kitzman 1997; Larson 1980; IHDP 1990; Johnson 1993; St Pierre 1999; Emond 2002; Minkovitz 2003 (a); Minkovitz 2003 (b); Johnston 2004; Fergusson 2005; Barlow 2007; Kemp 2011) and 12 that facilitated access to child healthcare (Gutelius 1977; Larson 1980; Olds 1986; IHDP 1990; Kitzman 1997; Duggan 1999; Minkovitz 2003 (a); Minkovitz 2003 (b); Johnston 2004; Caldera 2007) or other community services (Olds 1986; Hardy 1989; Duggan 1999; Armstrong 2000).

All studies provided the intervention to individual parents. In addition, four studies provided opportunities for peer support from other parents (St Pierre 1999; Minkovitz 2003 (a); Minkovitz 2003 (b); Kemp 2011) and one for enhanced informal support from family and friends (Olds 1986); and five studies provided parenting education to groups of parents (Gutelius 1977; IHDP 1990; St Pierre 1999; Koniak‐Griffin 2003; Johnston 2004), which, as a consequence, also provided opportunities for peer support. One of the two studies that provided solely educational interventions (Llewellyn 2003) included a control group that received home visits in which the visitor discussed parents' experiences of raising their children but without the educational intervention, thus allowing the effect of the home visit as opposed to that of the education to be assessed (Llewellyn 2003).

Types of outcome measures

Sixteen studies reported medically attended or self‐reported injury (Gutelius 1977; Larson 1980; Olds 1986; Hardy 1989; IHDP 1990; Johnson 1993; Kitzman 1997; Duggan 1999; Armstrong 2000; Emond 2002; Koniak‐Griffin 2003; Minkovitz 2003 (a); Minkovitz 2003 (b); Fergusson 2005; Caldera 2007; Culp 2007). Of these, one did not provide numerators and denominators (Emond 2002). Data from one study (IHDP 1990) for use in the meta‐analysis were taken from data reported in a systematic review (Roberts 1996b). Data from another study (Culp 2007) were calculated from unpublished data. Kitzman (Kitzman 1997) reported the incidence of five types of hospital encounters for injuries and ingestions, including the number of outpatient visits and the number of hospitalizations; data on the total number of healthcare encounters for injuries and ingestions were used in the meta‐analysis for this review.

One study presented maternal reports of medically attended injuries in the first and second years of life (Duggan 1999), and a separate article from the same study reported hospitalisations in the first three years of life (Duggan 2004), which are not mutually exclusive outcomes. Data from the first two years of life were included in the meta‐analysis as the number of events for this outcome was higher.

Seven studies reported a range of safety outcomes, including use of socket covers (Emond 2002; Minkovitz 2003 (a); Minkovitz 2003 (b); Johnston 2006) and stair gates (Emond 2002; Johnston 2006), lowering of hot water heater temperature (Minkovitz 2003 (a); Minkovitz 2003 (b); Culp 2007), use of cabinet locks and knowing the number to call if a child ingests harmful substances (Minkovitz 2003 (a); Minkovitz 2003 (b); Johnston 2006), accessibility of poisons (Olds 1986), presence of stickers on poisonous substances (Minkovitz 2003 (a); Minkovitz 2003 (b)), having a functional smoke alarm (Minkovitz 2003 (a); Minkovitz 2003 (b)), and adhering to sleep safety practices (Feldman 1992; Johnston 2004). Two studies reported a range of home hazards using different tools (Olds 1986; Llewellyn 2003), and one study reported use of a home safety index at 3 and 30 months (Johnston 2004; Johnston 2006), based on the presence of a functioning smoke alarm, regular and correct use of a car seat, absence or safe storage of firearms in the house, and, for the 3‐month data collection period only, following safe sleep practices (Johnston 2004).

Ten studies (Larson 1980; Olds 1986; Kitzman 1997; Duggan 1999; St Pierre 1999; Armstrong 2000; Koniak‐Griffin 2003; Barlow 2007; Caldera 2007; Kemp 2011) measured the quality of the home environment using the HOME Inventory, one subscale of which measures organisation of the environment in relation to child development and safety. Of the ten studies, a total of eight reported total HOME scores, three reported "organisation of the home environment" subscale scores, and two reported scores for subscales that were irrelevant to the outcomes of this review (Table 3). The HOME score was most commonly measured at 12 months. One study measured the quality of the home environment by using the Massachusetts Home Safety Questionnaire (Culp 2007).

2. Reporting of HOME scores: total and subscales.

	Author, year	Study design	HOME score reported and when measured		Comments (findings reported in text of the review)
			Total HOME score	Organisation of the home environment
1	Armstrong 2000	RCT	6 months 12 months	6 weeks 4 months 12 months
2	Barlow 2007	RCT	12 months 36 months	12 months 36 months	SD not reported, obtained for 12‐month data through personal communication
3	Calder 2007	RCT	24 months		Reports: 1. % with HOME score < or equal to 33 2. total HOME score, no SD reported
4	Duggan 1999	RCT			does not report total HOME score or organisation of the environment subscale
5	Kemp 2011	RCT		12 months 24 months	mean and SE reported unclear whether 12‐ or 24‐month data reported
6	Kitzman 1997	RCT	12 months 24 months		mean, mean difference, and 95% CI reported adjusted for maternal psychological resources, household income, and poverty level of census tract at 24 months. No SD reported
7	Koniak‐Griffin 2003	RCT	12 months 24 months		results from ANCOVA reported
8	Larson 1980	partially randomised	6 weeks 6 months 12 months 18 months		means and statistical significance reported. No SD reported
9	Olds 1986	RCT	34 months 46 months		no SD reported
10	St Pierre 1999	RCT	48 months		Only data at 48 months reported

Risk of bias in included studies

Studies were assessed for quality using the criteria described in the table 'Characteristics of included studies' (Characteristics of included studies). In terms of selection bias, 10 (63%) of the 16 RCTs had low risk as the result of adequate random sequence generation and seven (44%) because of adequate allocation concealment (Figure 2; Figure 3). Although 15 (94%) of the 16 RCTs were judged to be at high risk of performance bias, only five (31%) were judged to be at high risk of detection bias. Six (38%) of the 16 RCTs had a high risk of attrition bias, and five (31%) were judged as being at high risk of selective reporting bias.

2.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

3.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all 22 included studies.

Effects of interventions

See: Table 1

Medically attended or self‐reported injury: RCTs

Eleven RCTs reported medically attended or self‐reported injury. One did not provide outcome by treatment group (Koniak‐Griffin 2003); thus results from ten RCTs were used in a meta‐analysis. Findings from this (Analysis 1.1) show that the intervention arm families had a statistically significantly lower risk of injury than control arm families (RR 0.83, 95% CI 0.73 to 0.94; Chi² = 9.23, 9 degrees of freedom (df), P = 0.42; I² = 2%). There did not appear to be any evidence of publication bias amongst the ten RCTs in the primary analysis (Egger's test regression coefficient = ‐0.65 (standard error (SE) 0.49), P = 0.22) (Figure 4). The study reported by Duggan (Duggan 2004), which measured maternal reports of medically attended injuries and medical records of hospitalisations for trauma, found 1.5% of the 342 children in the intervention group hospitalised for trauma in the first three years of life compared with 1.7% of the 231 children in the control group (statistical significance not reported). One article described two studies that evaluated the same paediatric clinic‐based multi‐faceted intervention (Minkovitz 2003 (a);Minkovitz 2003 (b)) using an RCT and a CBA. The randomised sites showed no statistically significant difference in emergency department (ED) use for injuries (odds ratio (OR) 0.94, 95% CI 0.65 to 1.34, P = 0.09). Although the Koniak‐Griffin study (Koniak‐Griffin 2003) reported numbers of visits to the ED by treatment group, the authors state that most of these visits were needed for illness, and that although only eight children were treated for injuries and accidents, this was not reported by treatment group.

1.1. Analysis.

Comparison 1 Intervention versus Control (n/N in clustered studies adjusted for clustering), Outcome 1 Medically attended or self‐reported injuries ‐ RCTs only.

4.

Funnel plot of comparison: 1 Intervention versus Control (n/N in clustered studies adjusted for clustering), outcome: 1.1 Medically attended or self‐reported injuries ‐ RCTs only.

&&One study (Johnson 2000) reported medically attended injury at 7‐year follow‐up from the Community Mothers trial (Johnson 1993). Investigators reported that 10 of 38 children in the intervention group and 17 of 38 children in the control group had had an accident that required a visit to the hospital (RR 0.59, 95% CI 0.31 to 1.11, P = 0.09). Unfortunately, two thirds of participants had been lost to follow‐up by seven years, but a sensitivity analysis was undertaken that included the longer term but not the shorter term results, as these were encompassed in the longer term results. The findings were robust to the use of the longer term results (RR 0.83, 95% CI 0.74 to 0.94, P = 0.52).

Sensitivity analyses were undertaken first by using each of the injury outcomes reported in a previous trial (Armstrong 2000). The findings were robust to the varying definitions of injury (burns RR 0.85, 95% CI 0.75 to 0.96; scratches RR 0.85, 95% CI 0.75 to 0.96; unspecified injury RR 0.84, 95% CI 0.74 to 0.95). Findings were also robust (RR 0.85, 95% CI 0.75 to 0.97) to excluding one study in which the extent to which the intervention was based on a protocol, manual, or curriculum was unclear (Fergusson 2005).

Finally, sensitivity analyses were undertaken for the primary analysis, including only RCTs at low risk of various sources of bias. The findings were robust to including only those studies at low risk of detection bias in terms of blinded outcome assessment (RR 0.83, 95% CI 0.71 to 0.97) (Kitzman 1997; Duggan 1999; Armstrong 2000; Caldera 2007) and at low risk of attrition bias in terms of follow‐up of less than 80% of participants in each arm (RR 0.80, 95% CI 0.70 to 0.93) (Gutelius 1977; IHDP 1990; Johnson 1993; Kitzman 1997; Duggan 1999; Armstrong 2000; Fergusson 2005; Caldera 2007). When analyses were restricted to studies at low risk of selection bias in terms of inadequate allocation concealment (RR 0.83, 95% CI 0.67 to 1.04) (Johnson 1993; Kitzman 1997; Duggan 1999; Armstrong 2000; Minkovitz 2003 (a)), the relative risk was similar but the 95% CIs were wider and the effect size was no longer statistically significant, possibly as a result of the small number of studies included in this subgroup analysis. Overall, when GRADE was used, the quality of the evidence was rated as moderate (Table 1).

Medically attended or self‐reported injury: non‐RCTs

Six non‐RCTs reported medically attended or self‐reported injury. Two of the non‐RCTS used clustered allocation (Minkovitz 2003 (b); Culp 2007) and reported numerators and denominators for ED use for injury‐related causes, allowing us to adjust for clustering by using an ICC of 0.017 (Kendrick 1999), and these were rounded to the nearest integer. Two studies that evaluated the same paediatric clinic‐based multi‐faceted intervention (Minkovitz 2003 (a); Minkovitz 2003 (b)) by using an RCT and a CBA, found a statistically significant reduction in ED use for injuries only in the CBA study sites (OR 0.67, 95% CI 0.49 to 0.90, P = 0.02). After adjustments were made for clustering, it was noted that 32 visits to the ED for injury related for 368 children in the intervention arm, and 40 were reported for 342 children in the control arm (Minkovitz 2003 (b)). One study (Larson 1980) found that 64 children in the intervention arm suffered 42 accidents compared with 32 accidents suffered by 41 children in the control arm－a statistically significant difference (P < 0.1). A second study also found statistically significantly fewer intervention arm mothers than mothers in the control arm reporting injuries to their children in the preceding 12 months after adjustments were made for confounding factors and for clustering (OR 0.54, 95% CI 0.33 to 0.88, P = 0.022) (Emond 2002). In contrast, another study (Hardy 1989) reported 8 (6%) observations of sustained closed head trauma for 131 children in the intervention arm and 15 (11%) for 132 children in the control arm. This difference was not statistically significant. Similarly, the Culp study (Culp 2007) did not find a statistically significant difference between intervention and control arms. After adjustment for clustering, 7 visits were made to the ED for the care of 103 children in the intervention arm, and 11 visits were made for 86 children in the control arm (statistical significance not reported).

Home safety outcomes

Studies reported home safety practices and hazards using a variety of methods and scales.

Home Observation for Measurement of the Environment (HOME) scores

Ten studies reported total HOME or subscale scores (Table 3). One study (Duggan 1999) reported only results from subscales that were irrelevant to this review. Authors most often reported HOME scores at 12 months (six of ten studies) but frequently reported insufficient detail for the study data to be included in a meta‐analysis. Data on total HOME scores at 12 months from three RCTs (Armstrong 2000; Koniak‐Griffin 2003; Barlow 2007) were included in a meta‐analysis. Data from one RCT (Koniak‐Griffin 2003) were received in a personal communication. Results from the meta‐analysis show no statistically significant difference in total HOME scores between intervention and control arm families (mean difference 0.57, 95% CI ‐0.59 to 1.72, Chi² = 0.41, 2 df, P = 0.82; I² = 0%) (a higher HOME score represents a more enriched home environment for the child). Another trial (Armstrong 2000) also reported organisation of the home environment subscale scores and found a statistically significant difference favouring the intervention arm (mean score intervention arm 5.70 (SD 0.77) vs mean score control arm 5.11 (SD 1.16), P < 0.05). An additional trial (Barlow 2008) reported that no statistically significant differences were found between home visiting and control groups on the HOME Inventory scale at 36 months postnatally (mean score intervention arm = 24.76 (no SD reported), mean score control arm = 23.45 (no SD reported), P = 0.98).

Of the six studies not included in the meta‐analysis that reported total HOME scores or organisation of the environment subscale scores, two found statistically significant differences favouring intervention arm families (Larson 1980; Kitzman 1997). Another group (Kitzman 1997) found a statistically significant effect of the intervention on the overall HOME score at 24 months (total HOME score intervention arm 32.3 vs control arm 30.9 (SDs not reported), mean difference ‐1.3, 95% CI ‐2.2 to ‐0.4, P = 0.003). An additional trial (Larson 1980) found that mothers in intervention group A (antenatal and postnatal programme) had statistically significantly higher overall HOME scores compared with those in group B (postnatal programme only) or in the control arm at three of the four assessment points (mean group A at six weeks, six months, 12 months, and 18 months: 29.3, 35.2, 40.1, and 41.2; vs mean group B: 25.8, 33.7, 37.8, and 38.6; vs mean control arm: 26.7; 33.2; 37.8; and 39.0 (no SD reported); P values < 0.001; < 0.005; < 0.017; and < 0.041).

Four studies found no statistically significant difference between treatment arms, although one study found statistically significant differences only amongst distressed mothers (Kemp 2011). Another group (Caldera 2007) found no statistically significant difference in total HOME scores at 24‐month follow‐up (mean score intervention arm 36.7 vs 35.9 control arm (SD not reported), P = 0.10). Fewer intervention families than control families had extremely poor total HOME scores (i.e. less than or equal to 33) (20% vs 31%, P < 0.001). Group scores did not differ statistically significantly on any HOME subscale. One trial (Kemp 2011) reported organisation of the home environment subscale scores and found no statistically significant difference between groups (mean score intervention arm 4.92 (SE 0.08), mean score control arm 4.84 (SE 0.08), mean difference 0.09, 95% CI ‐0.13 to 0.30, P = 0.43). This group (Kemp 2011) also reported scores for mothers with more than one risk factor (mean score intervention arm 4.96 (SE 0.10), mean score control arm 4.68 (SE 0.11), mean difference 0.27 (95% CI ‐0.03 to 0.57), P = 0.07) and for distressed mothers (mean score intervention arm 5.00 (SE = 0.10), mean score control arm 4.60 (SE=0.10), mean difference 0.40 (95% CI 0.11 to 0.69), P = 0.01). It is not clear whether reported results were measured at 12 or 24 months. Another study (Olds 1994) reported no statistically significant differences in total HOME scores at both 34 (mean score intervention arm 39.08, mean score control arm 39.03, mean difference ‐0.05, 95% CI ‐1.92 to 1.84) and 46 months (mean score intervention arm 39.66, mean score control arm 39.67 (no SD reported), mean difference 0.01, 95% CI ‐1.66 to 1.67). One trial (St Pierre 1999) presented total HOME scores measured at four years and found no statistically significant difference between groups (mean score intervention arm 32.55 (SD 9.46), mean score control arm 33.03 (SD 9.45), P value reported only as "non‐significant").

Safety practices

Seven studies reported specific safety practices or use of items of safety equipment; 5 studies found effects favouring intervention arm families.

One study (Olds 1994) found no effect of the intervention on the extent to which mothers reported that they kept poisonous substances out of reach of their children or used child restraints in cars. Another study (Feldman 1992) measured sleep safety but reported findings only for all child care skills combined. The intervention arm had a statistically significant higher score for all child care skills at follow‐up than was reported for the control arm (intervention arm mean 88.1% skills correct (no SD reported) vs control arm mean 60.6% skills correct (no SD reported), P < 0.001). Other investigators (Emond 2002) reported that statistically significantly more mothers in the intervention arm used electric socket covers (OR adjusted for confounders and clustering 1.92, 95% CI 1.16 to 3.17, P = 0.019) and safety gates (data not reported) than control arm mothers.

One study (Johnston 2004) found no statistically significant difference in the use of safe‐sleep practices at 3 months postpartum between treatment groups (intervention arm 80.1% vs control arm 80.3%, adjusted RR 1.02, 95% CI 0.98 to 1.05). Another study (Johnston 2006), reporting on the same participants as were described in an earlier study (Johnston 2004) at 30 months postpartum, found that intervention arm families were statistically significantly more likely to use stair gates than were control arm families (I 33.2% vs C 30.2%, RR 1.19, 95% CI 1.15 to 1.23, P < 0.05) and to have the local poison control centre number accessible (I 95.8% vs C 90.4%, RR 1.08, 95% CI 1.03 to 1.12, P < 0.05) but were statistically significantly less likely to use safety latches on cabinets (I 66.8% vs C 77.9%, RR 0.88, 95% CI 0.83 to 0.93, P < 0.05). No statistically significant difference between groups was noted for use of covers on electric outlets (I 92.9% vs C 92.3%, RR 1.00, 95% CI 0.98 to 1.03).

One group (Minkovitz 2003 (a)), reporting results from an RCT, found no statistically significant difference in the proportion of families who lowered the temperature on water heaters (intervention arm 64.4% vs control arm 60.4%, P = 0.11) or who used safety latches on cabinets (intervention arm 63.3% vs control arm 61.8%, P = 0.34). However, a statistically significant difference was noted in the proportion of families who used covers on electric sockets (intervention arm 91.9% vs control arm 88.8%, P = 0.04). Results from the CBA (Minkovitz 2003 (b)) found no statistically significant difference in the proportion of families who lowered the temperature on water heaters (intervention arm 56.8% vs control arm 56.3%, P = 0.82), used covers on electric sockets (intervention arm 90.5% vs control arm 89.5%, P = 0.46), or used safety latches on cabinets (intervention arm 63.5% vs control arm 62.5%, P = 0.62). Another group (Culp 2007) reported that intervention arm families were more likely to have hot water adjusted to a safe temperature and electric cords beyond a child's reach (figures and P values not reported).

Home hazards

Two studies reported measures of home hazards.

The first (Olds 1994) reported statistically significantly fewer observed hazards in the home at both 34 months (mean intervention arm 0.22 vs control arm 0.38 (no SD reported), P = 0.04) and 46 months (mean intervention arm 0.21 vs control arm 0.46 (no SD reported), P = 0.003) amongst intervention arm families than control arm families.

The other study (Llewellyn 2003) compared the Home Learning Programme with home visits without parental education, with lesson only booklets without any face‐face education and with usual care. They found that parents in the Home Learning Programme group identified statistically significant more dangers within the home (mean 76.25 (SD 10.64)) than those who had received home visits without parental education (mean 54.82 (SD 15.78)) or those who had received only usual care (two groups: mean first control group 55.70 (SD 8.06), mean second control group 57.33 (SD 19.22), P < 0.001); suggesting that the beneficial effects were attributable to the parenting intervention rather than the home visit, but care must be taken in interpreting this finding as it relates only to one study. Parents in the Home Learning Programme group also identified statistically significantly more dangers within the home (mean 76.27 (SD 13.67)) than those in the lesson booklet only group (mean 62.0 (SD 12.53), P < 0.001), suggesting a greater effect of face‐to‐face education written information. The Home Learning Programme group also identified a statistically significantly greater number of precautions to reduce the risk of injury (mean 78.85 (SD 17.24)) than those who had received home visits without parental education (mean 48.91 (SD 15.36)), or those who had received usual care (two groups, mean first control group 47.10 (SD 13.76), mean second control group 45.33 (SD 13.87), P < 0.001), again suggesting a beneficial effect of parental education above that achieved from the home visit. Parents in the Home Learning Programme group identified statistically significantly more precautions to reduce the risk of injury (mean 85.27 (SD 21.12)) than those in the lesson booklet only group (mean 54.29 (SD 17.06), P < 0.001), again suggesting greater effects of face‐to‐face education than written information. Finally, parents in the Home Learning Programme group implemented a statistically significantly greater number of precautions to reduce the risk of injury (mean 88.09 (SD 34.92)) than those in the lesson booklet only group (mean 57.50 (SD 11.48), P < 0.001), again suggesting that face‐to‐face educational visits had a greater effect than written information.

Composite home safety measures

One study (Johnston 2004) reported a composite home safety measure, the Home Safety Index, which was developed for the study and comprised the sum of binary responses to six items (maximum possible score 7) on car seat use, safe storage of firearms, functioning smoke detectors, scald prevention activities, and safe infant sleep practices. At 3 months postpartum, intervention arm families had a statistically significantly higher mean score (reflecting safer practices) than control arm families (intervention arm mean 6.28 (SD 0.89) vs control arm mean 6.10 (SD 1.11), difference between the means 0.10, 95% CI 0.02 to 0.17). This difference was due mainly to differences in gun storage practices.

Another group (Johnston 2006), reporting on the same families discussed in the earlier trial (Johnston 2004) but at 30 months postpartum, used the same tool minus the sleep safety practices. Intervention arm families were statistically significantly more likely to report safe practices for all items than were control arm families (intervention arm 86.2% vs control arm 72.1%, RR 1.19, 95% CI 1.09 to 1.28, P < 0.05).

A later trial (Culp 2007) assessed home safety using the Massachusetts Home Safety Questionnaire. This scale assessed both safety practices and use of safety equipment. At 12 months, families in the intervention group had statistically significantly safer homes (mean (M) 38.1, SD 2.4) than did control group families (M 36.9, SD 2.6, t(261) 3.9, P = 0.0001).

Discussion

Principal findings

We have found that parenting interventions, most commonly provided on a one‐to‐one basis in the home as part of multi‐faceted interventions to improve a range of child (and often maternal health) outcomes during the first two years of a child's life, are effective in reducing self‐reported or medically attended injury amongst young children. This finding was consistent across studies with little evidence of statistical heterogeneity between effect sizes. This finding was also robust to most aspects of study quality and study design. All but one of the studies contributing to this analysis evaluated multi‐faceted home visiting programmes. However, one study, in which the intervention was provided primarily in paediatric primary care, found a similar effect size to that from the home visiting programmes, but this was confined to a non‐randomised part of the study. Only two studies reported effects of parenting interventions comprising solely educational interventions, neither of which measured injury outcomes or was included in the meta‐analyses.

In terms of home safety, parenting interventions appeared to have a greater effect on home safety practices and reduction of hazards than on HOME scores. Meta‐analysis of HOME scores found no statistically significant difference between treatment arms. Only three studies were included in this meta‐analysis. Three studies not included in the meta‐analysis found statistically significant differences in total HOME scores or organisation of the environment subscale scores. Three studies reporting home safety using alternative composite scores found that intervention arm families had statistically significantly safer homes. Five of the seven studies assessing home safety practices and use of safety equipment found statistically significant effects favouring intervention arm families. The two studies reporting home hazards found fewer hazards in the homes of intervention arm families. Therefore, fairly consistent evidence indicates that parenting interventions can have a positive effect on both home safety and childhood injury rates. Only one study used a control group that received home visits but without also providing a parenting intervention (Llewellyn 2003), thus allowing the effect of the intervention to be assessed above any effect of the home visit per se. Findings in this report suggested that the beneficial effects were attributable to the parenting intervention rather than to the home visit, but care must be taken in interpreting this finding because it relates to only one study.

Strengths and weaknesses of this systematic review

Our search strategy included searching a large number of bibliographic databases and grey literature and handsearching some conference abstracts and journals. However, our search terms included injury and home safety outcome terms, and these may have been secondary outcomes in some studies; therefore, this may have led to some studies being missed in our searches. However, we attempted to contact the authors of all studies excluded on the basis of outcomes to ascertain whether they had measured any outcomes relevant to our review. Of the seven papers excluded on the basis of lack of relevant outcomes, the authors of two confirmed that they did not assess unintentional injury, two authors were untraceable, and three did not respond. No evidence of publication bias was found, although the number of studies included in this assessment was fairly small (10), hence the funnel plot and Egger's test should be interpreted with caution. The analysis adjusted for cluster allocated studies, and sensitivity analyses were undertaken to test assumptions regarding the potential for bias; uncertainty as to the extent to which the intervention was based on a protocol, manual, or curriculum; and follow‐up period and injury type. Findings were robust to these assumptions.

The parenting interventions included in our review were complex interventions, and only a minority of studies were explicit about the theoretical basis of the intervention or hypothesised about why it may have resulted in a reduction in childhood injuries (Olds 1986; Kitzman 1997; Minkovitz 2003 (a); Minkovitz 2003 (b); Johnston 2004). Meta‐analyses for home safety outcomes other than HOME scores were not possible because of the variety of tools and subscales used. The meta‐analysis of HOME scores did not find a statistically significant difference between treatment arms but included only three studies, and hence will have had limited power. In addition, the meta‐analysis was restricted to the total HOME score containing six domains, only one of which measured home safety.

The generalisability of these findings is limited by the study populations, which mainly consisted of families considered to be 'at risk' of adverse child health outcomes. In addition, all included studies came from high‐income countries, so the findings may not be generalisable to low‐ or middle0income countries. All studies provided the intervention to individual parents, and whilst several also included some parents groups, none of the studies delivered the intervention primarily to groups of parents, hence findings may not be generalisable to group‐based parenting interventions. Similarly, most studies provided the intervention mainly within the home, so the findings may not be generalisable to parenting interventions provided outside the home.

Strengths and weaknesses of included studies

All reviews are dependent on the quality of reporting in the included studies and the availability and willingness of study authors to respond to requests for information. It did not come as a surprise that the more recently published studies, especially RCTs, tended to be reported more comprehensively. Most studies described the content of the intervention in sufficient detail and described and reported injury outcomes, enabling data to be extracted for meta‐analysis. Three cluster allocated studies reported findings adjusted for clustering (Emond 2002;Minkovitz 2003 (b); Johnston 2004). Most studies used parental reports of injuries, which may be subject to biased reporting, particularly because blinding participants to treatment arm allocation is not possible with interventions such as these. However, there did not appear to be a consistent relationship between self‐reported injury or that verified by medical records and effect size. Safety outcomes were reported less consistently, with a minority of studies reporting whether a statistically significant difference was found, but not reporting effect sizes for some safety outcomes (Olds 1986; Feldman 1992; Emond 2002; Culp 2007). Some studies reported overall Home Observation for Measurement of the Environment scores but not the subscale most relevant to child safety (Larson 1980; Olds 1986; Kitzman 1997; St Pierre 1999; Armstrong 2000; Koniak‐Griffin 2003; Barlow 2007; Caldera 2007), or an overall score for the Massachusetts Home Safety Questionnaire but not subscale scores for safety practices and use of safety equipment (Culp 2007), or scores for all child care skills combined but not separate sleep safety scores (Feldman 1992). It is possible that improvements in the safety subscales were not reflected in improvements in overall scores (Armstrong 2000). The quality of studies was variable, with half or more of the RCTs included in the meta‐analysis being susceptible to bias in terms of allocation concealment and/or outcome assessment. However, despite this, sensitivity analyses demonstrated little impact on the results of excluding studies without blinded outcome assessment. Excluding studies without adequate allocation concealment resulted in a similar effect size, but the effect was no longer statistically significant, possibly because of lack of power. Only two studies included in the meta‐analysis reported high attrition rates.

Findings in relation to previous research

Although no existing systematic reviews have examined the effects of parenting interventions on child injury, our findings are consistent with those of two previous meta‐analyses examining the effects of home visiting programmes on child injury (Roberts 1997; Elkan 2000).

Potential explanations for the findings

The authors of included studies suggest that a reduction in injuries may have occurred via a range of mechanisms, including increasing the ability of parents to manage minor injuries without medical help (Kitzman 1997), improving the quality of child care provided (Olds 1986; Kitzman 1997), and providing parental 'guidance' (Olds 1986) on home safety (Olds 1986; Culp 2007) and on the greater belief that children must be protected to "succeed in school, work, and mainstream society" (Olds 1986). Our review suggests that parenting interventions are likely to improve home safety, but other plausible explanations can be offered for why parenting interventions may reduce childhood injuries. All studies included in the primary meta‐analysis were aimed at improving a range of child (and often maternal) health outcomes. Six of these studies reported statistically significant improvements in child behaviour (Gutelius 1977; Olds 1986; IHDP 1990; Fergusson 2005; Johnston 2006; Caldera 2007); three reported statistically significant less punitive discipline practices amongst intervention group parents (Kitzman 1997; Duggan 1999; Fergusson 2005); five reported statistically significantly increased or improved mother‐child interaction (Gutelius 1977; Olds 1986; Johnson 1993; Armstrong 2000; Johnston 2004); and two reported statistically significant improvements in maternal psychological well‐being (Johnson 1993; Johnston 2004). It is therefore possible that the reduction in childhood injuries may result from improvements in child behaviour, more effective supervision or discipline practices, or greater or more positive interactions between mother and child, all of which may be associated with improved maternal psychological well‐being. A recent large systematic review of home safety education and the provision of safety equipment found strong evidence that these increased home safety practices and behaviours, along with some evidence that they reduced childhood injury (Kendrick 2012). None of the studies included in our review specifically provided or fitted home safety equipment, and it is plausible that combining parenting interventions with the provision and fitting of safety equipment may further enhance their effect on childhood injury. Understanding how parenting interventions work and which components of often complex interventions are necessary or sufficient to reduce childhood injury is important for designing effective and efficient services for children and parents.

Authors' conclusions

Implications for practice.

Parenting interventions, most commonly provided within the home as part of a multi‐faceted intervention to improve a range of child (and often maternal health) outcomes, are effective in reducing self‐reported or medically attended unintentional injury. Fairly consistent evidence also suggests that they improve home safety. This evidence relates mainly to interventions provided to families 'at risk' of adverse child health outcomes, including those 'at risk' of child abuse and neglect. Health and social care providers should make home visiting programmes available to such families as part of their injury prevention and wider child and maternal health strategies. Such provision is also likely to have a range of other beneficial effects for maternal and child health.

Implications for research.

A series of research questions remain to be answered: whether parenting interventions delivered outside the home have positive effects on childhood injury; whether parenting educational interventions, as opposed to multi‐faceted interventions, delivered within or outside the home and aimed at improving a range of parenting practices are effective in reducing childhood injury; whether group‐based parenting interventions are effective in reducing childhood injury; whether providing home safety education and safety equipment further increases the effectiveness of parenting interventions in reducing child injury; and finally, whether parenting interventions are effective in reducing child injury when delivered to families not considered to be 'at risk' of adverse child health outcomes.

What's new

Date	Event	Description
29 May 2013	Amended	Copy edits made.

History

Protocol first published: Issue 2, 2006
 Review first published: Issue 4, 2007

Date	Event	Description
8 August 2012	New citation required and conclusions have changed	The review has been updated with data from seven new studies. We have now included a meta‐analysis of HOME scores. The results and conclusions have changed (minor).
31 January 2011	New search has been performed	The search for studies was updated to January 2011. Seven new studies are included.

Appendices

Appendix 1. Search strategy

Cochrane Injuries Group Specialised Register (searched 24 Jan 2011): 57 records (limit: 2005 to 2011)

1.      (child* or infan* or toddl* or pre‐school* or preschool* or "pre school" or young* or adolesc* or pediat* or paediat* or minor* or boy* or girl* or teen* or adolesc* or baby or babies)

2.      ((accident* AND  prevent*) or safety or (safe* AND  device*) or (safe* AND  equipment*) or (infan* AND  equipment*) or (protective AND  device*) or injur* or wound* or accident* or fracture* or poison* or suffocat* or asphyx* or drown* or burn* or scald* or lacer* or contus* or (smok* AND  inhal*))

3.      (parent* AND  program*) or (parent* AND  train*) or (parent* AND  educat*) or (parent* AND  promot*) or (parent* AND  skill*) or (parent* AND  intervent*) or (parent* AND  group*) or (parent* AND  support*) or ("parent‐child relations" or "parent‐child interaction" or  "object attachment")

4.      1 and 2 and 3

Cochrane Central Register of Controlled Trials 2011, issue 1 (The Cochrane Library): 101 records

#1        ((parent* near3 program*) or (parent* near3 train*) or (parent* near3 educat*) or (parent* near3 promot*) or (parent* near3 skill*) or (parent* near3 intervent*) or (parent* near3 group) or (parent* near3 support) or (parent‐child relations or parent‐child interaction or object attachment)):ti, ab

#2        ((accident* near3 prevent*) or safety or (safe* near3 device*) or (safe* near3 equipment*) or (infan* near3 equipment*) or (protective near3 device*) or injur* or wound* or accident* or fracture* or poison* or suffocat* or asphyx* or drown* or burn* or scald* or lacer* or contus* or (smok* near3 inhal*)):ti, ab

#3        (child* or infan* or toddl* or pre‐school* or preschool* or pre school* or young* or adolesc* or pediat* or paediat* or minor* or boy* or girl* or teen* or adolesc* or baby or babies):ti, ab

#4        (#1 AND #2 AND #3)

MEDLINE (Ovid SP) 1950 to 2011 week 2 (limit: 2005 to 2011): 274 records

1. (infan$ or child$ or teen$ or adolesc$ or minor$ or toddl$ or bab$).mp. 

2. ((parent$ adj3 program$) or (parent$ adj3 train$) or (parent$ adj3 educat$) or (parent$ adj3 promot$) or (parent$ adj3 skill$) or (parent$ adj3 intervent$) or (parent$ adj3 group) or (parent$ adj3 support) or (parent‐child relations or parent‐child interaction or object attachment)).mp.

3. ((accident$ adj3 prevent$) or safety or (safe$ adj3 device$) or (safe$ adj3 equipment$) or (infan$ adj3 equipment$) or (protective adj3 device$) or injur$ or wound$ or accident$ or fracture$ or poison$ or suffocat$ or asphyx$ or drown$ or burn$ or scald$ or lacer$ or contus$ or (smok$ adj3 inhal$)).mp.  

4. (randomised controlled trial or randomized controlled trial or random allocation or double blind method or clinical trial or control group or evaluat$ or intervent$ or comparative study).mp.

5. 1 and 2 and 3 and 4

EMBASE (Ovid SP) 1980 to 2011 week 2 (limit:2005 to 2011): 406 records

1.      ((parent* adj3 program*) or (parent* adj3 train*) or (parent* adj3 educat*) or (parent* adj3 promot*) or (parent* adj3 skill*) or (parent* adj3 intervent*) or (parent* adj3 group) or (parent* adj3 support) or (parent‐child relations or parent‐child interaction or object attachment)).mp.

2.      ((accident* adj3 prevent*) or safety or (safe* adj3 device*) or (safe* adj3 equipment*) or (infan* adj3 equipment*) or (protective adj3 device*) or injur* or wound* or accident* or fracture* or poison* or suffocat* or asphyx* or drown* or burn* or scald* or lacer* or contus* or (smok* adj3 inhal*)).mp.

3.      (randomised controlled trial or randomized controlled trial or random allocation or double blind method or clinical trial or control group or evaluat* or intervent* or comparative study).mp.

4.      (child* or infan* or toddl* or pre‐school* or preschool* or pre?school* or young* or adolesc* or pediat* or paediat* or minor* or boy* or girl* or teen* or adolesc* or baby or babies).mp.

5.      1 and 2 and 3 and 4

6.      5

7.      limit 6 to yr="2005 ‐ 2010"

 ISI Web of Science: Social Sciences Citation Index (SSCI) 1970 to January 2011 (limit:2005 to 2011) and ISI Web of Science: Conference Proceedings Citation Index‐ Social Science & Humanities (CPCI‐SSH) –1990 to January 2011(limit:2005 to 2011): 185 records

 1.      (child* or infan* or toddl* or pre‐school* or preschool* or pre school* or young* or adolesc* or pediat* or paediat* or minor* or boy* or girl* or teen* or adolesc* or baby or babies)

2.      ((accident* SAME prevent*) or safety or (safe* SAME device*) or (safe* SAME equipment*) or (infan* SAME equipment*) or (protective SAME device*) or injur* or wound* or accident* or fracture* or poison* or suffocat* or asphyx* or drown* or burn* or scald* or lacer* or contus* or (smok* SAME inhal*))

3.      (parent* SAME program*) or (parent* SAME train*) or (parent* SAME educat*) or (parent* SAME promot*) or (parent* SAME skill*) or (parent* SAME intervent*) or (parent* SAME group) or (parent* SAME support) or (parent‐child relations or parent‐child interaction or object attachment)

4.      1 AND 2 AND 3

5.      (randomised OR randomized OR randomly OR random order OR random sequence OR random allocation OR randomly allocated OR at random OR randomized controlled trial) 

6.      (controlled clinical trial OR controlled trial OR clinical trial OR placebo)

7.      ((singl* OR doubl* OR trebl* OR tripl*) SAME (blind* OR mask*))

8.      5 OR 6 OR 7

9.      (human*)

10.  8 AND 9

11.  4 AND 10

 CINAHL (EBSCO) 1982 to 24 Jan 2011 (limit:2005 to 2011): 213 records

 S1   (child* or infan* or toddl* or pre‐school* or preschool* or pre school* or young* or adolesc* or pediat* or paediat* or minor* or boy* or girl* or teen* or adolesc* or baby or babies) 

S2   ((accident* N3 prevent*) or safety or (safe* N3 device*) or (safe* N3 equipment*) or (infan* N3 equipment*) or (protective N3 device*) or injur* or wound* or accident* or fracture* or poison* or suffocat* or asphyx* or drown* or burn* or scald* or lacer* or contus* or (smok* N3 inhal*))

 S3   (parent* N3 program*) or (parent* N3 train*) or (parent* N3 educat*) or (parent* N3 promot*) or (parent* N3 skill*) or (parent* N3 intervent*) or (parent* N3 group) or (parent* N3 support) or (parent‐child relations or parent‐child interaction or object attachment)

 S4   S1 and S2 and S3

ASSIA and ERIC

(infan* or child* or teen* or adolesc* or minor* or toddl* or bab*) and ((parent* within 3 program*) or (parent* within 3 train*) or (parent* within 3 educat*) or (parent* within 3 promot*) or (parent* within 3 skill*) or (parent* within 3 intervent*) or (parent* within 3 group) or (parent* within 3 support*) or ((parent‐child relations) or (parent‐child interaction) or (object attachment))) and ((accident* within 3 prevent*) or safety or (safe* within 3 device*) or (safe* within 3 equipment*) or (infan* within 3 equipment*) or (protective within 3 device*) or injur* or wound* or accident* or fracture* or poison* or suffocat* or asphyx* or drown* or burn* or scald* or lacer* or contus* or (smok* within 3 inhal*)) and (((randomised controlled trial) or (randomized controlled trial) or (random allocation)) or ((double blind method) or (clinical trial) or (control group)) or (evaluat* or intervent* or (comparative study)))

ProQuest dissertation and thesis

(infan* or child* or teen* or adolesc* or minor* or toddl* or bab*) AND (parent*) AND (injur* or accident* or wound*) AND (randomi* controlled trial or random* allocation) OR (double blind method) OR (clinical trial or control group) OR (comparative stud* or evalua* or intervent*) AND PDN(>1/1/2005) AND PDN(<12/31/2011)

BIOSIS Preview

Topic=(infan* OR child* OR teen* OR adolesc* or minor* OR toddl* OR bab*) AND Topic=(parent* program* OR parent* train* OR parent* educat* OR parent* promot* OR parent* skill* OR parent* intervent* OR parent* group OR parent* support* OR (parent‐child relations OR parent‐child interaction OR object attachment)) AND Topic=(injur* OR accident* OR wound* OR fracture* OR safe* OR protect* OR poison* OR suffocat* OR asphyx* OR drown* OR burn* OR scald* OR lacer* OR contus*) AND Topic=((randomised controlled trial OR randomized controlled trial OR random allocation) OR (double blind method OR clinical trial OR control group) OR (evaluat* OR intervent* OR comparative study))

Databases=PREVIEWS Timespan=2005‐2011

Sociological Abstracts

#1  (infan* or child* or teen* or adolesc* or minor* or toddl* or bab*) and (((((parent‐child relations) or (parent‐child interaction)) or ((parent‐child relations) or (parent‐child interaction)) or (object attachment)) and ((accident* or injur* or wound* or accident* or fracture* or poison* or suffocat* or asphyx* or drown* or burn* or scald* or lacer* or contus*) or (safe* or protective*) or (smok* within 3 inhal*)) and (((randomised controlled trial) or (randomized controlled trial) or (random allocation)) or ((double blind method) or (clinical trial) or (control group)) or (evaluat* or intervent* or (comparative study)))

Zetoc

general: child or infant or baby and parent.

general: parenting intervention and injury.

Data and analyses

Comparison 1. Intervention versus Control (n/N in clustered studies adjusted for clustering).

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Medically attended or self‐reported injuries ‐ RCTs only	10	5074	Risk Ratio (M‐H, Random, 95% CI)	0.83 [0.73, 0.94]

Comparison 2. Total HOME scores.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Total HOME scores	3	368	Mean Difference (IV, Random, 95% CI)	0.57 [‐0.59, 1.72]

2.1. Analysis.

Comparison 2 Total HOME scores, Outcome 1 Total HOME scores.

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Armstrong 2000.

Methods	Randomised controlled trial.
Participants	Families of newborns at inner‐city obstetric hospital, with English literacy skills sufficient to complete study questionnaire and any one of the following: physical domestic violence, child abuse of either parent, single parent, ambivalence to pregnancy; or 3 or more of the following: maternal age <18 years, 3 or more household moves in last 2 years, financial stress, < 10 years' maternal education, low income, social isolation, history of mental health disorder, drug or alcohol abuse in either parent, domestic violence other than physical abuse.
Interventions	Intervention arm: home visits weekly for first 6 weeks, 2 weekly from 6 weeks to 3 months of age, monthly from 3 to 6 months of age. Home visits comprised manualised programme aimed at enhancing parents' relationship with the infant and positively influencing ability to provide a safer, more nurturing, and healthier environment for their child. Focus of home visits was to establish trust with family, enhancing parental self‐esteem and confidence, providing anticipatory guidance for normal child development problems such as sleeping or crying behaviour problems, promoting preventive child health care, and facilitating access to community services. In addition, brief social work intervention was offered to families experiencing partnership or grief difficulties. Home visits were provided by child health nurse. Control arm: conventional community child health services, which include offer of one home visit from child health nurse and child health centre visits by appointment.
Outcomes	Parent‐reported injury, including bruises, scratches, burns, and unspecified injury. HOME inventory, including subscales "organization of the home environment" and "provision of play materials."
Notes	Allocation concealment: by clerical officer not involved in determining eligibility for study using computer‐generated random number list. Detection bias: outcome assessors blinded to treatment arm allocation. Attrition bias: At start of study: Intervention arm n = 90. Control arm n = 91. At 4‐month follow‐up: Intervention arm n = 80/90 (88.9%). Control arm n = 80/91 (87.9%). Data from 12‐month follow‐up reported in Fraser 2000.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"A random number table was computer generated."
Allocation concealment (selection bias)	Low risk	Allocation was by "a clerical officer not involved in determining eligibility to determine intervention status."
Blinding of participants and personnel (performance bias) All outcomes	High risk	Unable to blind participants and study personnel.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Data collected by a researcher naive to the intervention status of the participants."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Attrition rate of 12%.
Selective reporting (reporting bias)	Low risk	All outcomes reported.
Other bias	Low risk	Not reported whether intention‐to‐treat analysis undertaken. Appears to be free of other biases.

Barlow 2007.

Methods	Randomised controlled trial.
Participants	Vulnerable women who were identified during pregnancy by community midwives attached to one of 40 participating GP practices using a range of demographic and socio‐economic criteria (i.e. risk factors) such as mental health problems, domestic violence, drug and alcohol abuse, and housing problems.
Interventions	18 months of weekly visits starting during the second trimester of pregnancy from a Health Visitor trained in understanding the process of helping, with skills of relating to parents effectively and knowledge of methods of promoting parent‐infant interaction using the Family Partnership Model. Parents in both control and intervention groups received the standard help that was available to such families. Women not wanting to be randomly assigned or without a working understanding of English were excluded.
Outcomes	Home Observation for Measurement of the Environment (HOME) Inventory was used to measure the quality of the home environment at 12 and 36 months postnatal.
Notes	Attrition bias: At the start of the study: Intervention group n = 68. Control group n = 63. At 12 months: Intervention group n = 62 (91%). Control group n = 58 (92%). Later follow‐up data reported in Barlow 2008 (unpublished): At 36 months: Intervention group n = 51 (77%). Control group n = 46 (72%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not reported.
Allocation concealment (selection bias)	Low risk	Use of "sequentially numbered sealed opaque envelopes."
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants and those delivering intervention not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Data were collected, coded, and analysed by researchers who had not been involved in recruitment and were therefore blind to the intervention group."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Low risk at 12 months with attrition <10%, high risk at 3 years with attrition 26%.
Selective reporting (reporting bias)	Low risk	All outcomes reported.
Other bias	Low risk	Intention‐to‐treat analysis undertaken. Appears to be free of other biases.

Caldera 2007.

Methods	Randomised controlled trial.
Participants	Families identified as being at risk according to a protocol. Families who screen positive are assessed for risk using Kempe's Family Stress Checklist, and those scoring greater than or equal to 25 are eligible for the HFAK. Half the families are identified prenatally, and half at the time of the child's birth. To be eligible for inclusion, participants could not have previously enrolled in HFAK and had to speak English well enough to complete study activities.
Interventions	Voluntary intensive, long‐term (3 to 5 years) home visiting. Visits are offered weekly for the first 6 to 9 months, with the frequency of visits decreasing as family functioning improves. Home visitors are to provide information, make referrals to community resources, help parents prepare for developmental milestones, screen and refer for developmental delay, and promote child environmental safety. They are to support positive parent‐child interaction via role modelling and reinforcement of positive interaction and parental empathy. Use of parenting curricula is encouraged but no specific curriculum is required. Home visitors are to encourage parents to establish a medical home for child health care and to support parents in crises. They are encouraged to use the Individual Family Support Plan (IFSP) as a tool for teaching problem solving around family‐initiated goals.
Outcomes	Home Observation for Measurement of the Environment (HOME) Inventory was used to measure the quality of the home environment. Number of injuries requiring medical care.
Notes	Attrition bias: At the start of the study: Intervention = 162. Control = 163. At 24 months, interviews obtained with : Intervention = 138 (85%). Control = 140 (86%). For number of injuries, data available only for those with complete medical records: Intervention = 131. Control = 137.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Use of "table of random numbers."
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants and those delivering intervention not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Trained research staff blinded to family group assignment" collected baseline and follow‐up data.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Attrition rate less than 20%.
Selective reporting (reporting bias)	Low risk	No evidence.
Other bias	Low risk	Intention‐to‐treat analysis undertaken. Appears to be free of other biases.

Culp 2007.

Methods	Non‐randomised controlled trial.
Participants	First‐time mothers recruited before 28th week of pregnancy.
Interventions	Home visitors provided information, made referrals to community services, screened and referred for developmental delay, and promoted child environmental safety. Home visiting used an individualized manualised curriculum delivered by professional home visitors with college degrees in child development. The three major foci were maternal health, child health and safety, and family functioning and parenting. Mothers received weekly home visits during the first month after enrolment, biweekly for the remainder of their pregnancy, weekly for the first 3 months postpartum and biweekly for months 3 to 12 postpartum. The health curriculum during the prenatal phase included a minimum of four visits for sharing information on nutrition, alcohol, smoking, foetal growth and development, labour and delivery, and family planning. The health curriculum from birth to 12 months focused on effects of second‐hand smoke on infant growth and development, family planning, immunisations, infant nutrition, and health food preparation. A special emphasis was placed on teaching household safety when the infants were crawling through the end of the intervention period at 12 months.
Outcomes	Massachusetts Home Safety Questionnaire: 42 items. Number of hospital visits and emergency room visits.
Notes	Attrition bias: At start of study: Intervention = 205. Control = 150. At 12 months: Intervention = 156 (76%). Control = 107 (71%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation to intervention/control (selection bias)(For non randomised and CBA studies)	Unclear risk	Control sites selected from counties with similar risk and demographic characteristics.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants and those delivering intervention not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Not reported.
Incomplete outcome data (attrition bias) All outcomes	High risk	Attrition rate of 26%.
Selective reporting (reporting bias)	Low risk	Appears to report all outcomes.
Other bias	Low risk	Intention‐to‐treat analysis undertaken. Appears to be free of other biases.
Risk of bias due to confounding (For non randomised and CBA studies)	Low risk	Assessment of maternal characteristics at recruitment and in treatment arms appears similar.

Duggan 1999.

Methods	Randomised controlled trial.
Participants	Families with newborns identified as at risk for child abuse or neglect using the Family Stress Checklist, which included items such as substance use, poor mental health, domestic violence, history of abuse as a child, unrealistic expectations of child, unwanted pregnancy, and other risk of poor bonding.
Interventions	Intervention arm: Home visiting programme for 3 to 5 years designed to help families cope with the challenges of child rearing (the "Healthy Start Programme"). Programme is aimed at improving family functioning, preventing child abuse, and promoting child health. Programme commences with weekly visits, which are reduced in frequency as family functioning improves. Intervention included helping families to identify and build on strengths to improve family functioning, role modelling of problem‐solving skills, linking families with needed services, providing parenting education and modelling of effective parent‐child interaction. Programme sites were allowed flexibility in which parenting curricula they delivered. Control arm: no "Healthy Start Programme".
Outcomes	Parent‐reported injury requiring medical care in first 2 years of life. Hospitalisations for trauma in first 3 years of life.
Notes	Selection bias: allocation using predetermined table of random numbers. Detection bias: outcome assessors blinded to treatment group allocation. Attrition bias: At start of study: Intervention arm n = 373. Control arm n = 270. At 2‐year follow‐up: Intervention arm n = 309 (82.8%). Control arm n = 225 (83.3%). Data from 3‐year follow‐up published in Duggan et al 2004. At 3‐year follow‐up: Intervention arm n = 342 (92%). Control arm n = 231 (86%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Table of random numbers."
Allocation concealment (selection bias)	Low risk	"Evaluation staff informed EID worker."
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Interviewers did not know group status.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Attrition rate 18% at 2‐year follow‐up and 11% at 3‐year follow‐up.
Selective reporting (reporting bias)	Low risk	Does not appear to:
Other bias	Low risk	Intention‐to‐treat analysis undertaken. Other biases not apparent.

Emond 2002.

Methods	Controlled before‐and‐after study (cluster allocated).
Participants	First‐time parents in disadvantaged areas.
Interventions	Intervention arm: home visits by specially trained health visitor antenatally, at 10 days and 3 weeks postnatally, and every 5 weeks until 8 months of age. Home visiting programme comprised part of the "Child Development Programme," which is based on the principles of empowerment and is aimed at improving the health and development of young children by supporting and advising mothers during first phase of parenting using specially designed written materials and cartoons (First Parent Health Visitor Scheme (FPHVS)). Control arm: conventional health visiting.
Outcomes	Use of electric socket covers and stair gates, accidents in the last 12 months.
Notes	3 health visiting bases already delivering the FPHVS constituted the intervention group; 4 health visiting bases matched on social, economic, and demographic profiles were chosen as comparison bases. Only data from prospective part of the study included in this review. Detection bias: unclear whether outcome assessors were blinded to treatment group allocation. Attrition bias: At start of study: Intervention arm = 205. Control arm = 254 At 2‐year follow‐up: n = 368 (80%). Number not given by treatment arm, but authors report no significant difference in attrition rate between arms.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation to intervention/control (selection bias)(For non randomised and CBA studies)	High risk	Intervention areas already receiving FPHVS. Control areas chosen to closely match intervention areas on key demographic variables.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Not clear.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Attrition rate of 20%.
Selective reporting (reporting bias)	High risk	Does not report all outcomes.
Other bias	Unclear risk	Intention‐to‐treat analysis not reported.
Risk of bias due to confounding (For non randomised and CBA studies)	High risk	Distribution of confounders between treatment arms was assessed. Some differences between arms were noted on confounders.

Feldman 1992.

Methods	Randomised controlled trial.
Participants	Mothers with low IQ at risk of child neglect, with children aged 1 to 23 months of age.
Interventions	Intervention arm: home‐based weekly parent training programme focussing on teaching basic child care skills (e.g. preparing feeds, bottle feeding, crib safety, sleep safety, toilet training). Training consisted of verbal instruction, specially designed picture books depicting each stage of the child care task, modelling of each step by trainer, and feedback on mother's performance, plus coupons for small gift items when mothers demonstrated 80% of trained skills correctly. Mean duration of training: 7.7 weeks. Training provided by psychology or early childhood education graduates, trained to deliver the programme. Control arm: no training.
Outcomes	Demonstrated sleep safety tasks.
Notes	Selection bias: allocation to treatment group by picking numbers out of a hat. Detection bias: outcome assessors not blinded to treatment group allocation. Attrition bias: At start of study: intervention arm n = 11. Control arm n = 11. At follow‐up (immediately post training or at next home visit): Intervention arm n = 11 (100%). Control arm n = 11 (100%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Randomly assigned" but no reports on how generated.
Allocation concealment (selection bias)	High risk	"Made up slips of paper with numbers on them, folded them and put then into a hat, shook the hat, reached in the hat (without looking), and picked a slip out."
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Primary observers and reliability checkers were not told of the between‐group experimental design."
Incomplete outcome data (attrition bias) All outcomes	Low risk	For immediate posttest, attrition was 0%. For later follow‐up, attrition rate of 28%.
Selective reporting (reporting bias)	Low risk	No evidence.
Other bias	Unclear risk	Intention‐to‐treat analysis not reported. Other biases not apparent.

Fergusson 2005.

Methods	Randomised controlled trial.
Participants	Families with children younger than 3 months of age facing stress and difficulty, defined as having two or more risk factors, identified from screening tool covering parental age, planning of pregnancy, social support, substance use, financial situation, and family violence.
Interventions	Intervention arm: home visiting programme (the "Early Start Program"), number of visits not specified, provided for a mean of 24 months. Intervention involved use of a social learning model approach, which included assessment of family needs, development of positive partnership between family support worker and client, collaborative problem solving to devise solutions to family challenges, and provision of support and mentoring and advice to assist families to mobilise their strengths and resources. Home visits provided by family support workers (nurses or social workers). Degree to which the intervention involved the use of a protocol, manual, or curriculum is unclear. Control arm: no "Early Start Program."
Outcomes	Hospital attendance for accident/injury or accidental poisoning in first 3 years of life, obtained from medical records.
Notes	Selection bias: allocation using computer‐generated random number list. Detection bias: unclear whether outcome assessors not blinded to treatment group allocation. Attrition bias: At start of study: Intervention arm n = 220. Control arm n = 223. At 36‐month follow‐up: Intervention arm = 184 (84%). Control arm = 207 (93%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Computer‐generated sequence of random numbers."
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	High risk	"Research interviewers were aware of the treatment group to which the family was assigned."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Follow‐up rate of 88%.
Selective reporting (reporting bias)	Low risk	Does not appear so.
Other bias	Low risk	Intention‐to‐treat analysis undertaken.

Gutelius 1977.

Methods	Randomised controlled trial.
Participants	Unmarried primigravidas 15 to 18 years of age in low‐income areas.
Interventions	Intervention arm: extensive child health supervision with emphasis on counselling and anticipatory guidance (based on Baby and Well Child Care, by Benjamin Spock), plus well‐child care provided by paediatrician and nurse using mobile coach, which visited child's home, making 9, 6, and 4 visits in 1st, 2nd, and 3rd years of life. Cognitive stimulation programme provided by nurse home visits (10, 8, and 6 visits in 1st, 2nd, and 3rd years of life). Programme covered age‐appropriate forms of visual, auditory, tactile, and motor types of stimulation, with particular emphasis on language development. Group sessions for parents (16 over 1 year) to discuss child‐rearing practices and other child and family problems. Control arm: conventional care, including one home visit by nurse and paediatrician in neonatal period and referral to well‐baby clinic.
Outcomes	Number of toxic ingestions in first 3 years of life.
Notes	Selection bias: assigned using random numbers. Detection bias: outcome assessors not blinded to treatment group allocation. At start of study: Intervention arm n = 47. Control arm n = 48. At 3‐year follow‐up: Intervention arm n = 44 (94%). Control arm n = 45 (94%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Assigned by a series of random numbers," but it is not clear how it is generated.
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants and personnel not blinded.
Blinding of outcome assessment (detection bias) All outcomes	High risk	Project nurse and paediatrician collected information on control and experimental series.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Attrition rate of 6%.
Selective reporting (reporting bias)	High risk	Mainly significant findings reported.
Other bias	Unclear risk	Intention‐to‐treat analysis not reported. Other bias not apparent.

Hardy 1989.

Methods	Quasi‐randomised controlled trial.
Participants	Inner‐city black mothers 18 years of age and older, with infants weighing > 2000 g.
Interventions	Intervention arm: 10 home visits from 7 to 10 days to 24 months of age. Curriculum plus single‐issue pamphlets aimed at developing parenting and child care skills, including topics such as well‐child care, sick child care, feeding, clothing, and safety. Anticipatory guidance was given, developmental milestones were discussed, and suggestions were made for enhancing development. Referral to social worker or educator for psychosocial issues. Home visits provided by specially trained local ex‐resident working under supervision of education specialist and social worker. Control arm: conventional medical, developmental, and social assessments with referral to other services as indicated. No home visits or attendances at child development centres or at parent groups.
Outcomes	Closed head trauma, obtained from maternal report verified by hospital records.
Notes	Selection bias: alternate allocation based on odd (intervention arm) and even (control arm) medical record numbers. Detection bias: unclear whether outcome assessors were blinded to treatment group allocation. Risk of bias due to confounding: treatment arms comparable in terms of % families eligible for Medicaid and % noted in medical records to be at "high risk" of illness, developmental problems, neglect, or abuse. Authors state that age range of children is similar in both groups, but no figures are given. Attrition bias: At start of study: Intervention arm n = 143. Control arm n = 147. At 23‐month follow‐up: Intervention arm n = 131 (92%). Control arm n = 132 (90%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation to intervention/control (selection bias)(For non randomised and CBA studies)	Low risk	Based on odd/even medical record number. "Infants whose medical history numbers ended in even digits served as control subjects."
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants and personnel not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Not reported.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Attrition rate of 9%.
Selective reporting (reporting bias)	Low risk	Appears all outcomes are reported.
Other bias	Low risk	Intention‐to‐treat analysis undertaken. Appears to be free of other biases.
Risk of bias due to confounding (For non randomised and CBA studies)	Low risk	Distribution of confounders was assessed between treatment arms. Arms appear similar.

IHDP 1990.

Methods	Randomised controlled trial.
Participants	Low birth weight (<=2500 g) premature (<=37 weeks' gestation) infants.
Interventions	Intervention arm: weekly home visits in first year of life and two weekly visits from 1 to 3 years of age. Home visits focused on two curricula: (1) Curriculum emphasising cognitive, linguistic, and social development. (2) Curriculum designed to provide a systematic approach to help parents manage self‐identified problems. Home visits provided by non‐professionals. Plus attendance at child development centres from age 12 to 36 months and two monthly parent groups from age 12 months providing information on child rearing, health, and safety as well as parent support (the "Infant Health Development Program"). Control arm: usual care.
Outcomes	Maternal reports of injuries not requiring hospital care.
Notes	Selection bias: allocation by independent computer programme. Detection bias: outcome assessors not blinded to treatment arm allocation. At start of study: Intervention arm n = 377. Control arm n = 608. At 24‐month follow‐up: Intervention arm n = 345 (92%). Control arm n = 551 (91%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not reported.
Allocation concealment (selection bias)	Unclear risk	Not clear, reportedly performed by the National Study Office.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	High risk	Blinded for cognitive, behavioural, and growth assessments.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Follow‐up rate of 91%.
Selective reporting (reporting bias)	Low risk	Does not appear so.
Other bias	Unclear risk	Intention‐to‐treat analysis not reported.

Johnson 1993.

Methods	Randomised controlled trial.
Participants	First‐time mothers living in a deprived area.
Interventions	Intervention arm: monthly home visits by community mother to provide support and encouragement to first‐time parents in rearing children using the "Child Development Programme," including modules on educational development, language development, and cognitive development. Control arm: conventional public health nurse input, including home visits at birth and six weeks, and at other times as required.
Outcomes	Maternal reports of injuries.
Notes	Selection bias: randomisation using sealed opaque envelopes. Detection bias: outcome assessors not blinded to treatment group allocation. Attrition bias: At start of study: Intervention arm n = 141. Control arm n = 121. At 12‐month follow‐up: Intervention arm n = 127 (90%). Control arm n = 105 (87%). Later follow‐up data reported in Johnson 2000. At 7‐year follow‐up: Intervention arm n = 38 (27%). Control arm n = 38 (31%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Table of random numbers.
Allocation concealment (selection bias)	Low risk	"Cards were sealed in consecutively marked envelopes."
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	High risk	Not blinded.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Follow‐up rate of 89%.
Selective reporting (reporting bias)	Low risk	Appears to report all outcomes.
Other bias	Unclear risk	Intention‐to‐treat analyses not reported. Other biases not apparent.

Johnston 2004.

Methods	Non‐randomised controlled trial.
Participants	Pregnant women of less than 22 weeks' gestation receiving care from a participating clinic; younger than 45 years, English speaking, and planning to use a study clinic for paediatric care.
Interventions	Intervention: Healthy Steps for Young Children programme (HS) beginning prenatally (PP + HS) or postnatally (HS). HS consists of risk reduction activities and universal components, including developmental screening, anticipatory guidance, and follow‐up services offered to all families receiving care. HS participants received postnatal home visits, developmental advice and parent‐initiated telephone support, developmental assessments conducted in tandem with scheduled well‐child care, the Reach Out and Read literacy program, and other risk‐based screening services and parenting classes delivered by HS specialists. PP participants also received three home visits when they were at approximately 20, 27, and 34 weeks' gestation. PP participants received home visits to help parents create a safe, knowing, and welcoming environment for their newborn, and screening and intervention for targeted risk factors such as smoking, depression, and domestic violence. Control: usual paediatric care.
Outcomes	Safety practices including safe sleep practices and Home Safety Index comprising items on car seat use, safe storage of firearms, functioning smoke detector, scald prevention activities, and safe infant sleeping practices. Results at 3 months postpartum.
Notes	Selection bias: allocation to treatment groups not described. Detection bias: unclear whether outcome assessors were blinded to treatment group. Attrition bias: At start of study: Intervention arm n = 304 (PP + HS = 151, HS = 152). Control arm n = 136. At 3‐month follow‐up: Intervention arm n = 275 (PP + HS = 139 (92.7%), HS = 136 (90.1%)). Control arm n = 121 (89.7%). Later follow‐up described in Johnston 2006. At 30‐month follow‐up, follow up of intervention arm PP + HS = 122 (81%), HS = 117 (77%). Control arm = 104 (76%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation to intervention/control (selection bias)(For non randomised and CBA studies)	Unclear risk	Does not report how intervention and treatment clinics chosen.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants and personnel not blinded. Participants may have been unaware of control/intervention clinics as clinics were geographically distant.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Not reported.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Follow‐up rate of 91% at 3 months. At 30‐month follow‐up, follow‐up of intervention arm PP + HS = 122 (81%) , HS = 117 (77%). Control arm = 104 (76%).
Selective reporting (reporting bias)	Low risk	Does not appear to be selective reporting.
Other bias	Low risk	Intention‐to‐treat analysis undertaken.
Risk of bias due to confounding (For non randomised and CBA studies)	High risk	Distribution of confounders was assessed between treatment arms, and significant differences were noted between arms on maternal education and age.

Kemp 2011.

Methods	Randomised controlled trial.
Participants	Mothers reporting one or more of the following risk factors: aged younger than 19 years, current probable distress (assessed as an Edinbugh Dpression Scale (EDS) score of 10 or more), lack of emotional and practical support, late antenatal care (after 20 weeks' gestation), major stressors in the past 12 months, current substance misuse, current or history of mental health problem or disorder, history of abuse in mother's own childhood, and history of domestic violence.
Interventions	Women received an average of 16.3 (range 0 to 52) visits, each of 60 to 90 minutes' duration, by a child health nurse, commencing on average at 26 weeks' gestation (range 12 to 40) and continuing to the child's second birthday (average duration of participation in the programme to child age 57.0 weeks (range 0 to 122): 82% visited prenatally, 95% visited in the first year postnatally, and 53% visited in the second year). All women received usual antenatal midwifery, obstetric, and birthing services.
Outcomes	HOME score at 12 and 24 months. Subscale data provided on "organisation of environment" and provision of appropriate play materials." Data collected at baseline and at 1, 6, 12 18, and 24 months postnatally.
Notes	Attrition bias: At baseline: Intervention arm n = 111. Control arm n = 97. At 1‐month follow‐up: Intervention arm n = 111 (100%). Control arm n = 97 (100%). At 12‐month follow‐up: Intervention arm n = 89 (80%). Control arm n = 80 (83%). At 18‐month follow‐up: Intervention arm n = 86 (78%). Control arm n = 75 (77%). At 24‐month follow‐up: Intervention arm n = 85 (77%). Control arm n = 69 (71%). Further data published in Kemp et al 2008: At 36‐month follow‐up: Intervention arm n = 72 (65%). Control arm n = 58 (60%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	A permuted block design was used to randomly allocate mothers to the intervention or comparison group. Blocks varied in size from 0 to 6. Within each weekly block, a random selection of cases to receive intervention was made using the Statistical Package for the Social Sciences (SPSS).
Allocation concealment (selection bias)	Low risk	Women were given "a sealed envelope that contained information advising them of their group assignment."
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not possible.
Blinding of outcome assessment (detection bias) All outcomes	High risk	Outcome data were collected by an RA, who was initially blind to group allocation; however, participants commonly would reveal their group allocation during the data collection process.
Incomplete outcome data (attrition bias) All outcomes	High risk	At 12 months, follow‐up rate of 82%. At 24 months, follow‐up rate of 74%.
Selective reporting (reporting bias)	High risk	Unclear which data point HOME scores were used for the analysis.
Other bias	Low risk	Intention‐to‐treat analysis.

Kitzman 1997.

Methods	Randomised controlled trial.
Participants	Primiparous African‐American women < 29 weeks' gestation, with at least two of the following: unmarried, < 12 years' education, unemployed.
Interventions	Intervention arm 1: free transportation to prenatal care visits. Intervention arm 2: intervention arm 1 + developmental screening and referral services for child at 6, 12, and 24 months of age (this group acted as the control group for assessing the effect of postnatal home visiting). Intervention arm 3: intervention arm 2 + prenatal home visits + 1 hospital + 1 home postpartum visit. Intervention arm 4: intervention arm 3 + 26 postnatal home visits from birth to 2 years of age. (This group acted as the intervention group for assessing the effect of postnatal home visiting.) Home visits comprised specific curriculum covering parental education regarding understanding infants' communication signals, enhancing interest in playing with children in ways that promote emotional and cognitive development, and creating households that are safer for children. Help also was provided to mothers to clarify goals and solve problems related to completing education, finding work, and planning future pregnancies. Home visits provided by nurses.
Outcomes	Total number of health care encounters ascertained from medical records, for injuries/ingestions. HOME inventory.
Notes	Selection bias: allocation by independent computer programme. Detection bias: outcome assessors blinded to treatment arm allocation. Data from intervention arms 2 (control) and 4 (intervention) used in paper to evaluate effects of home visits and have been used in meta‐analyses. Attrition bias: At start of study: Control arm n = 515. Intervention arm n = 228. At 24 months' follow‐up: Control arm n = 481 (93%). Intervention arm n = 216 (95%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Use of “computer program that randomised individual women.”
Allocation concealment (selection bias)	Low risk	Performed remotely to participants and study personnel. Allocation “procedure concealed the randomisation from individuals directly involved with participants.”
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not possible.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Interviews and abstractions of medical and social service records were conducted by staff members who were unaware of the women’s treatment assignment."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Attrition rate of 6%.
Selective reporting (reporting bias)	Low risk	Reports all outcomes.
Other bias	Low risk	Intention‐to‐treat analysis undertaken. Appears to be free of other biases.

Koniak‐Griffin 2003.

Methods	Randomised controlled trial.
Participants	Adolescents 14 to 19 years of age at 26 weeks' gestation or less, having their first child and planning to keep the infant.
Interventions	Early Intervention Program (EIP) provided to young mothers through home visits by public health nurses and designed to influence maternal and child health. Included a maximum of 17 home visits: 2 prenatal and 15 postpartum, 1.5 to 2 hours each. The mean number of visits prenatally was 2.13 (SD 0.77) and postnatally 10.35 (SD 3.04). Control mothers received Traditional Public Health nursing care, which consisted of 1 home visit prenatally and 1 home visit postnatally. The mean number of visits prenatally was 1.02 (SD 0.26) and postnatally 1.09 (SD 0.42).
Outcomes	Emergency department visits for injury but not reported by treatment group. Total HOME scores. Data collected at intake, 6 weeks after birth, and 6, 12, 18, and 24 months postpartum for all outcomes except for HOME score, which was assessed at 12 months' and 24 months' postpartum.
Notes	Attrition bias: Mothers enrolled: Intervention arm (EIP) n = 56. Control arm (TPHNC) n = 45. No information on 43 dropouts (excluded from numbers above). Data on SD for HOME scores received from a personal communication (15th April 2012).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Using a computer‐based program."
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Participants and nurses not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Interviews conducted by Public Health nurses who were not involved in the intervention and were blind to group assignment.
Incomplete outcome data (attrition bias) All outcomes	High risk	No data on 43 dropouts. Data given only on 101 who took part.
Selective reporting (reporting bias)	High risk	Does not report all data for injuries or HOME scores.
Other bias	Low risk	Appears to be free of other biases.

Larson 1980.

Methods	Partially randomised controlled trial.
Participants	Pregnant women, 18 to 35 years of age, income below poverty line, high school education or less, no significant illness in pregnancy, no history of psychiatric hospitalisation, and normal delivery of full‐term infant; discharged from hospital within 5 days with no major congenital defects.
Interventions	Group A: prenatal home visit, plus postnatal hospital visit, plus 9 postnatal home visits from 6 weeks to 15 months of age. Home visits followed protocol consisting of counselling and advice related to general caretaking (e.g. feeding, sleeping, clothing, bathing, accident prevention, encouragement to take part in well‐child care), mother‐infant interaction (e.g. encouragement of frequent reciprocal interaction), social status (reviewing mother's relationship with child's father, support systems, and any areas of exceptional stress and concern), and child development (e.g. reviewing child's developmental competence, suggesting activities to promote child's capabilities). Home visits provided by psychology graduates. Group B: Postnatal home visits as for group A (10 visits from 6 weeks to 15 months of age). Group C: No home visits.
Outcomes	Cumulative accident rates obtained from maternal reports. HOME Inventory.
Notes	Selection bias: allocation to groups B and C by random assignment. Allocation to group A sequential after random assignment to groups B and C. Detection bias: outcome assessors blinded to treatment arm allocation. Risk of confounding: authors state no significant difference between treatment groups in terms of sex, parity, income, parental education, marital status, planned pregnancy, prenatal care, presence of father at delivery, or use of private medical care. Attrition bias: At start of study: Group A n = 35. Group B n = 36. Group C n = 44. Follow‐up when child 18 months of age: Group A = 26 (74%). Group B = 27 (75%). Group C = 37 (84%). For the purposes of meta‐analysis, groups A and B were combined as the intervention group, as a parenting programme had been during the postnatal period. Group C was used as the control group.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Not reported.
Allocation to intervention/control (selection bias)(For non randomised and CBA studies)	Unclear risk	Not reported.
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Home observers were blinded.
Incomplete outcome data (attrition bias) All outcomes	High risk	Attrition rate of 22%.
Selective reporting (reporting bias)	Low risk	Appears complete.
Other bias	Unclear risk	Intention‐to‐treat analysis not reported.
Risk of bias due to confounding (For non randomised and CBA studies)	Low risk	Distribution of confounders was assessed between treatment arms, and the arms appear similar.

Llewellyn 2003.

Methods	Randomised controlled trial.
Participants	Parents with intellectual disability whose first language was English, who were the primary caregivers for children younger than 5 years of age.
Interventions	4 groups with 3 intervention phases: Group 1: Home Learning Programme (HLP) comprising 10 lessons with accompanying booklets provided by parent educator at home, covering understanding sickness and health, when to call the doctor, how to take a child's temperature and check respiration, common life‐threatening emergencies, and home safety. HLP delivered in intervention phase 1. Group 2: home visits to discuss everyday experiences of raising children without educational input, delivered in intervention phase 1. HLP delivered in intervention phase 2. Group 3: usual care delivered in intervention phase 1. Mailed lesson booklets + telephone contact with parent educators to check progress with reading booklets and implementing information in booklets with no face‐to‐face education, delivered in intervention phase 2. HLP delivered in intervention phase 3. Group 4: Standard community services delivered in intervention phase 1. HLP delivered in intervention phase 2.
Outcomes	Number of home dangers identified; number of precautions identified to deal with home dangers; total number of precautions taken by parents to deal with home dangers observed using checklist.
Notes	Selection bias: allocation by project manager using table of random numbers. Detection bias: outcome assessors blinded to treatment group allocation. Attrition bias: At start of study: n = 57 not specified by treatment arm. At 3‐month follow‐up: Group 1 n = 20. Group 2 n = 11. Group 3 n = 10. Group 4 n = 4.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Table of random numbers."
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Trained parent assessors who were blind to each parent's group."
Incomplete outcome data (attrition bias) All outcomes	High risk	Attrition rate of 29%.
Selective reporting (reporting bias)	Unclear risk	Does not appear so.
Other bias	Unclear risk	Intention‐to‐treat analysis not reported. Other biases not apparent.

Minkovitz 2003 (a).

Methods	Randomised controlled trial.
Participants	Families with newborns up to 4 weeks of age. Families were excluded if the newborn was to be adopted or fostered, or was too ill to make an office visit in the first 4 weeks of life, if mother did not speak English or Spanish, or if the family intended to leave the practice within 6 months.
Interventions	Intervention arm: Healthy Steps Programme for the first 3 years of life, which included extended well‐child office visits (average 11 in first 2.5 years of life) to address questions and concerns about child development and behaviour and to promote positive parent‐child interactions; home visits (average < 2 in first 2.5 years of life) to inform parents about fostering intellectual and emotional development; telephone line to answer questions about child development; parent groups to offer social support and interactive learning sessions and practice in problem solving; and written information for parents emphasising importance of prevention and health promotion. Programme delivered by paediatricians and Healthy Steps specialists (nurses, nurse practitioners, social workers, and early childhood educators). Programme planned and implemented and process evaluated using the PRECEDE/PROCEED model. Control arm: conventional paediatric care.
Outcomes	Maternal reports of emergency department use in last year for injury‐related causes. Safety practices, including lowered temperature on water heater, use of socket covers, use of cabinet safety catches, functional smoke detectors, placed stickers on bottles of poisonous substances; knew number to call if child swallowed something harmful.
Notes	These are the results from the randomisation sites from the Minkovitz 2003 study. Selection bias: allocation using computer‐generated assignment sequence and sealed opaque envelopes. Detection bias: unclear whether outcome assessors were blinded to treatment group allocation. Attrition bias: At start of study (randomised part): Intervention arm n = 1133. Control arm n = 1102. At 30‐ to 33‐month follow‐up (randomised part): Intervention arm n = 832 (73%). Control arm n = 761 (69%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Computer‐generated assignment."
Allocation concealment (selection bias)	Low risk	"Concealed in sealed envelopes."
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Not reported.
Incomplete outcome data (attrition bias) All outcomes	High risk	Attrition rate of 29%.
Selective reporting (reporting bias)	Low risk	Reporting appears complete.
Other bias	Low risk	Intention‐to‐treat analysis undertaken. Other biases not apparent.

Minkovitz 2003 (b).

Methods	Controlled before and after study (cluster allocated).
Participants	Families with newborns up to 4 weeks of age. Families were excluded if the newborn was to be adopted or fostered, or was too ill to make an office visit in the first 4 weeks of life; if mother did not speak English or Spanish, or if the family intended to leave the practice within 6 months.
Interventions	Intervention arm: Healthy Steps Programme for the first 3 years of life, which included extended well‐child office visits (average 11 in first 2.5 years of life) to address questions and concerns about child development and behaviour and to promote positive parent‐child interactions, home visits (average < 2 in first 2.5 years of life) to inform parents about fostering intellectual and emotional development, telephone line to answer questions about child development, parent groups to offer social support and interactive learning sessions and practice in problem solving, and written information for parents emphasising importance of prevention and health promotion. Programme delivered by paediatricians and Healthy Steps specialists (nurses, nurse practitioners, social workers and early childhood educators). Programme planned and implemented and process evaluated using the PRECEDE/PROCEED model. Control arm: conventional paediatric care.
Outcomes	Maternal reports of emergency department use in last year for injury‐related causes. Safety practices, including lowered temperature on water heater, use of socket covers, use of cabinet safety catches, functional smoke detectors, placed stickers on bottles of poisonous substances, knew number to call if child swallowed something harmful.
Notes	These are the results from the CBA sites reported in the same paper as for Minkovitz 2003 (a). Attrition bias: At start of study: Intervention arm n = 1830. Control arm n = 1500. At 30‐ to 33‐month follow‐up: Intervention arm n = 1189 (65%). Control arm n = 955 (64%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation to intervention/control (selection bias)(For non randomised and CBA studies)	Unclear risk	Does not describe how sites were chosen.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not blinded.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Not reported.
Incomplete outcome data (attrition bias) All outcomes	High risk	Attrition rate of 64%.
Selective reporting (reporting bias)	Low risk	Reporting appears complete.
Other bias	Low risk	Intention‐to‐treat analysis undertaken. Other biases not apparent.
Risk of bias due to confounding (For non randomised and CBA studies)	High risk	Distribution of confounders was assessed between treatment arms, and arms differ on several potential confounding variables.

Olds 1986.

Methods	Randomised controlled trial.
Participants	Primiparous women < 30 weeks' gestation, and at least one of the following: < 19 years of age, single parents, low socio‐economic status.
Interventions	Intervention arm 1: usual care + free transportation to prenatal and well‐child visits. Intervention arm 2: intervention arm 1 + 9 home visits during pregnancy. Intervention arm 3: intervention arm 2 + postnatal home visits weekly from birth to 6 weeks of age; 2 weekly visits from 6 weeks to 4 months of age; 3 weekly visits from 4 to 14 months of age; 4 weekly visits from 14 to 20 months of age, and 6 weekly visits from 20 to 24 months of age. Home visits comprised detailed curriculum covering parental education regarding infant development aimed at improving parental behaviour that affects the child's well‐being (e.g. understanding infant's temperament, crying behaviour, need for responsive care giving, physical health care needs, managing common health problems); enhancing informal support through encouraging family and friends to help with household responsibilities and child care, and linking family with community health and social care services (e.g. well‐child care, vocational training programmes, mental health counselling, legal aid, nutritional supplementation programmes).Home visits provided by nurses. Control arm: usual care, no transportation to prenatal or well‐child visits, no home visits.
Outcomes	Emergency department visits for injuries or ingestions. Home hazards assessed by observers at 34 and 46 months (included chipped or flaking paint, sharp objects, danger of burns, dangerously placed objects posing a risk of falls). Self‐reported presence and accessibility of poisonous substances and use of child car restraints.
Notes	Selection bias: allocation using deck of cards, drawn by participants. Detection bias: outcome assessors blinded to treatment arm allocation. During the study, the control arm and intervention arm 1 were combined to form the control arm when no differences were found in use of routine prenatal and well‐child care. Data from intervention arm 3 and the combined control arm used in article to evaluate effects of home visits and have been used in the meta‐analyses. Attrition bias: At start of study: Control arm n = 184. Intervention arm n = 116. At 1‐year follow‐up: Control arm n = 136 (74%). Intervention arm n = 87 (75%). At 2‐year follow‐up: Control arm n = 121 (66%). Intervention arm n = 75 (65%). Later follow‐up described in Olds 1994: Assessed at 25 to 50 months. At 25‐ to 50‐month follow‐up: Control arm n = 137 (74%). Intervention arm n = 84 (72%).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	Unclear.
Allocation concealment (selection bias)	High risk	"Women drew their treatment assignments from a deck of cards,” some deviations.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Unable to blind participants and study personnel.
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Except in a small number of cases in which participating women inadvertently disclosed their treatment assignments, all interview and medical record data were collected by staff members who were unaware of the families’ treatment assignments.”
Incomplete outcome data (attrition bias) All outcomes	High risk	“Attrition varied from 15% to 21%.”
Selective reporting (reporting bias)	High risk	Data not reported.
Other bias	Low risk	Intention‐to‐treat analyses not reported. Appears to be free of other biases.

St Pierre 1999.

Methods	Randomised controlled trial.
Participants	Families with incomes below the federal poverty level and with unborn children or children younger than 1 year of age.
Interventions	The Comprehensive Child Development Programme was not conceived as a home visiting programme, but it used home visits as the primary means for delivering case management and early childhood education. Short‐ and long‐term programme effects were expected for children and parents through the delivery of educational, health, and social services tailored to each family. Case managers were expected to conduct home visits to each family at least twice per month that would last for 1 hour. In practice, visits were conducted 3 to 4 times a month and lasted 30 to 90 minutes, depending on the family's needs. The model called for services to be provided to all families continuously from the time the family entered the programme, before the child was 1 year old, until the child entered school.
Outcomes	Data were collected mainly at the child's 2nd, 3rd, 4th, and 5th birthdays, with some data collected at 18 and 30 months of age. HOME score assessed at 18 or 24 months of age and at 36 months of age; however, results given only for 4‐year follow‐up.
Notes	Attrition bias: At the start of the study: Intervention arm n = 2213. Control arm n = 2197. For HOME scores at 4‐year follow‐up: Intervention arm n = 1321 (59.7%). Control arm n = 1423 (64.8%). See also article by Goodson et al (2000).
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Each site implemented a random assignment procedure." "Most sites used stratified random assignment, using stratifiers such as ethnicity, age of mother, and service sites."
Allocation concealment (selection bias)	Unclear risk	Not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Not possible.
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Child testers were blind to the family's group assignment at the start of study.
Incomplete outcome data (attrition bias) All outcomes	High risk	At the 5‐year assessment, 74% of intervention families and 78% of control families participated in the evaluation.
Selective reporting (reporting bias)	High risk	Early data not reported,
Other bias	Low risk	Other biases not apparent.

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Akinbami 2001	Does not report study design of interest: no control group.
Babul 2007	Does not report intervention of interest.
Baron‐Epel 2006	Does not report study design of interest: no control group.
Beirens 2006	Article unobtainable.
Bond 2006	Does not report study design of interest: no control group.
Conroy 1994	Does not report study design of interest: no control group.
Culp 1998	Does not report study design of interest: no control group.
Cupples 2011	Does not report intervention of interest.
Dawson 1989	Does not report intervention of interest.
Donnellan 1981	Does not report outcomes of interest.
Ertem 2006	Does not report intervention of interest.
Feldman 2004	Does not report study design of interest: no control group.
Fitzpatrick 1997	Does not report study design of interest: a prospective study.
Gershater‐Molko 2002	Does not report study design of interest: no control group.
Gershater‐Molko 2003	Does not report study design of interest: no control group.
Gray 1979	Does not report intervention of interest.
Guyer 2000	Does not report outcomes of interest.
Hedges 2005	Does not report outcomes of interest.
Hemmo‐Lotom 2006	Article unobtainable.
Huxley 1993	Does not report study design of interest: retrospective control group.
Jackson 2009	Does not report study design of interest: developing, as opposed to testing, an intervention.
Johnson 2009	Does not report outcomes of interest.
Jordaan 2006	Does not report study design of interest: developing a measurement tool.
Jouriles 2010	Does not report outcomes of interest.
Kluger 2000	Does not report study design of interest: not testing an intervention.
Lealman 1983	Does not report intervention of interest.
McAuley 2004	Does not report intervention of interest.
Odendall 2009	IDoes not report intervention of interest.
Powell 2004	Does not report outcomes of interest.
Quraishi 2005	Does not report study design of interest: not testing an intervention.
Sharma 2006	Article unobtainable.
Smith 1984	Does not report outcomes of interest.
Subhi 2009	Does not report study design of interest: no control group.
Swart 2008	Does not report intervention of interest.
Taban 2001	Does not report study design of interest: no control group.
Trudeau 2010	Does not report intervention of interest.
Tsoumakas 2009	Does not report study design of interest: a descriptive study.
Van As, 2006	Article unobtainable.

Differences between protocol and review

1. Types of outcomes measures: We also reported scores on the HOME scale, thus section 3 of “Types of outcomes measures” now reads:

"Safety practices (e.g. storage of hazardous objects and substances, safe use of baby walkers, avoidance of unsafe hot water temperature) and composite measures of safety practices, including the Home Observation for Measurement of the Environment (HOME) Inventory. The Infant‐Toddler HOME Inventory, designed for use with children from birth to 3 years of age, consists of six subscales: emotional and verbal responsivity of the primary caregiver (items 1 to 11); avoidance of restriction and punishment (items 12 to 19); organisation of the physical and temporal environment (items 20 to 25), including "the child’s play environment appears safe and free of hazards"; provision of appropriate play materials (items 26 to 34); parental involvement with the child (items 35 to 40); and opportunities for variety in daily stimulation (items 40 to 45) (Caldwell 2003; Totsika 2004). A higher score indicates a more appropriate home environment for child development."

In addition, in this update, we had sufficient data to undertake a meta‐analysis of HOME scores.

2. Websites searched: We also searched “Public Health website (UK).”

3. Quality assessment: In the protocol, we stated that we would assess the quality of non‐randomised studies using the tool developed by Reisch and colleagues (Reisch, 1989). However we did not use the Reisch tool, and thus this section now reads as below:

“The following sources of bias were assessed for non‐randomised studies:

• Participant selection (selection bias).

• Blinding (performance bias and detection bias).

• Blinding of participants and personnel (performance bias).

• Blinding of outcome assessment (detection bias).

• Incomplete outcome data (attrition bias).

• Selective reporting (reporting bias).

• Risk of bias due to confounding: Was the distribution of confounders assessed between treatment arms? If so, do treatment arms appear similar in terms of confounders?

• Other bias.”

4. Quality assessment: In addition, we have added the following sentence to this section:

“Disagreement between review authors was dealt with by referral to a third review author.”

5. Measures of treatment effect: We stated in the protocol that we would pool results and present them as relative risks and 95% CIs for the binary outcomes. We have adjusted this sentence to read:

 “Pooled relative risks and 95% confidence intervals have been used for binary outcome measures and mean differences, and 95% confidence intervals for continuous outcome measures.”

6. Data analysis/synthesis: We stated in the protocol, “if there are sufficient studies, we will pool results and present them as relative risks and 95% CIs for the binary outcomes of possession of items of safety equipment, safety practices, and occurrence of at least one self‐reported medically attended injury.” However in the review, we pooled results for self‐reported medically attended injuries and HOME scores, and thus we have written the following sentence in the full review:

“Pooled relative risks and 95% CIs were estimated for occurrence of at least one self‐reported or medically attended injury. Pooled mean differences and 95% CIs were estimated for HOME scale scores.”

7. Primary and secondary analyses: In the protocol, we state, “As the review includes both randomised and non‐randomised studies, the primary analysis will be based on randomised studies, with a secondary analysis including both randomised and non‐randomised studies.” However we have not combined RCTs and non‐RCTs in a single meta‐analysis.

Thus in the review under “Data synthesis,” we have written, “Pooled RRs and 95% CIs were estimated for occurrence of at least one self‐reported or medically attended injury using data from included RCTs.” and “When clinically homogenous studies were insufficient to be combined in a meta‐analysis, or when findings were derived from non‐RCTs, the results were combined in a narrative review.” 

8. Sensitivity analyses: We stated in the protocol, “sensitivity analyses would be undertaken for individual aspects of the study quality, as discussed in the section on quality assessment.” In the final review, we give more precise detail, stating the following: “Sensitivity analyses have been undertaken that included only RCTs considered to be at low risk of selection bias in terms of adequate allocation concealment, detection bias in terms of blinded outcome assessment, and attrition bias as the result of follow‐up of less than 80% of participants in each arm.”

Contributions of authors

DK had the original idea for the review. DK, CM, JM, and SS‐B drafted the study protocol. LY and CM undertook the searches. LY, CM, TS, and DK undertook data extraction. CM undertook the analyses. CM and DK drafted the final review.

Sources of support

Internal sources

• University of Nottingham, UK.

• University of Warwick, UK.

External sources

• Department of Health, UK.

  • This Cochrane update was funded by the National Institute for Health Research Health Technology Assessment (NIHR HTA) Programme (project number 09/02/02) and a summary will be published in Health Technology Assessment.
  • The views and opinions expressed therein are those of the authors and do not necessarily reflect those of the HTA programme, NIHR, NHS or the Department of Health.

Declarations of interest

None known.

Edited (no change to conclusions)