Does home visiting prevent childhood injury? A systematic review of randomised controlled trialsBMJ 1996; 312 doi: https://doi.org/10.1136/bmj.312.7022.29 (Published 06 January 1996) Cite this as: BMJ 1996;312:29
- a Department of Community Paediatric Research, Montreal Children's Hospital, McGill University Faculty of Medicine, Montreal, Quebec, Canada
- b Departments of Pediatrics and of Epidemiology and Biostatistics, McGill University Faculty of Medicine
- c Division of Clinical Epidemiology, Royal Victoria Hospital, Montreal
- Correspondence to: Dr Ian Roberts, Department of Epidemiology and Biostatistics, Institute of Child Health, London WC1N 1EH.
- Accepted 30 November 1995
Objective: To quantify the effectiveness of home visiting programmes in the prevention of child injury and child abuse.
Design: Systematic review of 11 randomised controlled trials of home visiting programmes. Pooled odds ratios were estimated as an inverse variance weighted average of the study specific odds ratios.
Setting: Randomised trials that were available by April 1995.
Subjects: The trials comprised 3433 participants.
Results: Eight trials examined the effectiveness of home visiting in the prevention of childhood injury. The pooled odds ratio for the eight trials was 0.74 (95% confidence interval 0.60 to 0.92). Four studies examined the effect of home visiting on injury in the first year of life. The pooled odds ratio was 0.98 (0.62 to 1.53). Nine trials examined the effect of home visiting on the occurrence of suspected abuse, reported abuse, or out of home placement for child abuse. Because of the potential for bias in outcome reporting in these studies, pooled effect estimates were not calculated.
Conclusions: Home visiting programmes have the potential to reduce significantly the rates of childhood injury. The problem of differential surveillance for child abuse between intervention and control groups precludes the use of reported abuse as a valid outcome measure in controlled trials of home visiting.
Home visiting has been proposed as a prevention strategy, but the results from randomised trials are conflicting
This systematic review of randomised trials shows that home visiting can substantially reduce rates of child injury
No consistent effect on child abuse was found, but differential surveillance for child abuse between visited groups and control groups is an important weakness in many trials
The role of health visitors and non-professionals in the prevention of child injury deserves further attention
Home visiting programmes have long been advocated for improving the health of disadvantaged children. In Britain home visits by health visitors are considered to have a key role in accident prevention because of the advice given during the visits on child development and home safety.1 In the United States home visiting has been promoted primarily for the prevention of child abuse and neglect.2 In 1991 the United States Advisory Board on Child Abuse and Neglect called for the establishment of a universal programme of home visiting in an attempt to stem the increase in numbers of child abuse reports.3
Over the past two decades several randomised trials have examined the effect of home visiting programmes on the occurrence of child abuse and other child health outcomes. The results of these trials, however, have been conflicting. Although several published articles have reviewed the evidence from randomised trials,4 5 none of these satisfies the methodological criteria that have been proposed for scientific overviews.6 To quantify the effect of home visiting programmes on the occurrence of child injury and abuse we conducted a systematic review.
Inclusion criteria—We included studies in the systematic review if they met all three of the following criteria: (a) the assignment of the study participants to the intervention or control group had to be random or quasi-random—for example, alternate record numbers; (b) the study intervention had to include one or more postnatal home visits; and (c) the study had to address the outcomes of child injury (unintentional or intentional).
Identification of relevant trials—We identified trials by a computerised literature search of Medline (January 1966 to April 1995) and Embase (January 1975 to April 1995). We also searched the social sciences citation index for articles referencing randomised trials of home visiting. Key terms used for searching included social support, family support, home (and health) visitors, home (and health) visitation, child abuse, and child neglect. We reviewed the references of all relevant papers found in the searches, as well as those of review articles and textbooks. Because home visiting is often encountered in the context of the prevention of child abuse, a hand search was conducted of the Journal of Child Abuse and Neglect (from 1977 1(1) through to 1995 19(3)). We contacted the authors of identified papers and experts in the field and asked about any published or unpublished work that they might be aware of. To access studies not formally published, such as research reports and abstracts, we searched relevant conference proceedings. If studies met the first two inclusion criteria but did not report outcomes of child injury or abuse we asked the authors to provide any unpublished data on child injury.
Data extraction and study appraisal—We extracted the following data from each study: strategy for allocation concealment, number of randomised participants, duration of follow up, loss to follow up, blinding of outcome assessment, and the professional background of the home visitor (health or welfare professional or non-professional). We evaluated the quality of the trial using a modification of Prendiville's criteria.7 With this approach trials are scored from 1 to 3 (1=poorest score, 3=best score) on three important aspects of study methodology: control of confounding at entry (adequacy of allocation concealment); control of selection bias (extent to which analyses are based on all randomised participants); and control of information bias in assessing outcome (blinding of observers). While the original criteria assigned a score of 3 for random assignment by telephone and 2 for using opaque sealed envelopes, we assigned a score of 3 for using either of these methods. Trials that assigned subjects to treatment by using methods intended to reduce the risk of foreknowledge of allocation but which were not as secure as random assignment by telephone or use of opaque sealed envelopes scored 2. Trials in which the authors did not report the method of allocation concealment (and were unable to provide further details or could not be contacted) and trials using alternate record numbers or other similar strategies scored 1. If a published report contained insufficient information for us to assess the quality of the trial, we asked the authors to provide further details. Two assessors performed the data extraction independently, with agreement on methodological criteria evaluated with weighted (kappa).8 Each point of disagreement was settled by collaborative review.
Statistical methods—The measure of association, the odds ratio, was calculated directly for studies in which injury was expressed in binary (yes/no) form, with the variance estimated by Wolf's method.9 For studies in which injury occurrence was allowed to be multiple and expressed as an incidence density, the odds ratio was estimated on the assumption of a Poisson distribution, with the probability of a participant having at least one event being given by 1-e-ID, where ID is the incidence density. Pooled odds ratios were then calculated as an inverse variance weighted average of the study specific odds ratios.
The combined search strategies identified 33 trials meeting the first two inclusion criteria (randomised trials of postnatal home visiting).10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 Eleven of these trials (with 3433 participants) reported outcome data on injury or abuse, or on both.10 11 12 13 14 15 16 17 18 19 42 One of the eleven trials was published as an abstract only42; the author of this report was contacted, but the relevant outcome data were not available for inclusion in the review. Of the remaining 10 trials, one reported no differences in the occurrence of accidents,13 and in another injury outcome data had been collected but not reported.19 In both of these trials the authors gave us the relevant data. The authors of 13 of the 22 trials meeting the first two inclusion criteria but not reporting outcome data on injury or abuse responded to our request for information on unpublished injury outcomes. As a result of this process one further trial was identified that met all three inclusion criteria.43 Eleven trials were therefore identified that had outcome data on injury or abuse, or both.
Table 1 shows the scores for the quality of methodology for the trials included in the systematic review. The weighted (kappa) for agreement between the two assessors was 0.94 for adequacy of allocation concealment, 0.51 for the extent to which analyses were based on all randomised participants, and 0.78 for blinding. The mean scores for the unintentional injury outcomes were: adequacy of allocation concealment, 2.4; extent to which analyses were based on all randomised participants, 1.9; blinding, 1.5.
Table 2 shows the data for the eight trials that examined the effect of home visiting on the occurrence of childhood injury. Six of the eight trials reported a lower incidence of injury in the group that received home visits. One study reported three injury outcome measures, representing three different time periods of follow up. For this study, the overall injury rates and odds ratios were calculated for the entire (four year) follow up period (odds ratio 0.74 (95% confidence interval 0.55 to 0.99)). The pooled odds ratio for injury for the eight trials (figure) was 0.74 (0.60 to 0.92). Four studies examined the effect of home visiting on injury occurrence in the first year of life only. The pooled odds ratio was 0.98 (0.62 to 1.53).
Table 3 shows the data for the nine trials that examined the effect of home visiting on the occurrence of suspected abuse, reported abuse, or out of home placement for child abuse. In four trials the frequency of occurrence of abuse was lower in the visited group. In five trials the frequency of occurrence was higher in the visited group. Substantial heterogeneity of the odds ratios was found across the studies. The potential for bias in the outcome reporting was considered to be a serious threat to validity in all nine studies. Specifically, the presence of the home visitor may have resulted in an increased surveillance for child abuse and hence an increase in the number of reports of abuse. If present, this bias would have resulted in an apparent increased incidence of abuse in the visited group. Pooled effect estimates were therefore not calculated.
Although home visiting is unlikely to be associated with adverse effects, the widespread implementation or intensification of home visiting programmes may have important resource implications. Our meta-analysis of the results from eight randomised trials shows a significant preventive effect of home visiting on the occurrence of childhood injury.
The effect of home visiting on the occurrence of child abuse varied across studies in both magnitude and direction. This may have been the result of bias in the assessment of child abuse outcomes. A report of child abuse entails the occurrence and discovery of an injury, as well as an attribution of intent. In several of the primary studies the information leading to the report of child abuse was provided by the home visitor, raising the possibility of “surveillance bias.” Differential surveillance for child abuse between the intervention and control groups would almost certainly result in a substantial underestimation of any beneficial effect of home visiting programmes on the occurrence of child abuse, possibly to the extent of reversing the direction of the apparent effect. Indeed, the usefulness of reported abuse as an outcome measure in trials of home visiting deserves reconsideration.
Publication bias is one of the most important potential threats to the validity of systematic reviews. Such bias may arise if certain outcome data are selectively omitted from published reports because the results fail to reach significance. To avoid this type of bias we wrote to the authors of all identified randomised trials of home visiting programmes, asking them to provide any unpublished outcome data on injury or abuse (one further trial was identified by this approach). The authors of nearly half of the studies meeting the first two inclusion criteria, however, could not be traced. These were predominantly small studies and so would make a comparatively minor impact on the overall result. Funnel plots can be used to estimate the extent of publication bias, but because their use is limited to meta-analyses that have enough trials to allow a funnel shape to be visualised, this approach is not helpful in this review.44
A recurring issue in the context of systematic reviews is the extent to which the interventions examined are sufficiently comparable for the results from the studies to be combined. The effectiveness of home visiting may depend on its timing, duration, and intensity. Nevertheless, for unintentional childhood injuries no clear heterogeneity was seen in the effect across studies.
Because most of the trials included in this review used non-professional home visitors, the question of the relative effectiveness of professional versus non-professional home visiting remains unanswered. The observed effect of home visiting on child injury is consistent with a generic effect of home based maternal support. In Britain a programme of home visiting is provided by health visitors. Current health visiting programmes, however, should not be assumed to achieve the effects on childhood injury that are implied by the results of this systematic review. Firstly, the experimental home visiting may have been more intense than that which is typically provided by health visitors. Secondly, in all but one of the trials the intervention was targeted at groups considered to be at increased risk for adverse child health outcomes. This may restrict the extent to which the results are generalisable to programmes of universal health visiting.
The Health of the Nation strategy established child accident prevention as a national priority. Few injury prevention interventions, however, have been shown to reduce injury rates in randomised controlled trials. Given the results of this systematic review, the effectiveness of home visiting by health visitors and non-professional support agencies in preventing childhood injury deserves further examination.
We thank the authors of the original articles who replied to inquiries and provided data not included in their published articles. We specifically thank Drs Marie McCormick and Pamela Klebanov for conducting new analyses of the data from the infant health and development programme.
Funding IR was supported by an overseas fellowship from the Health Research Council of New Zealand. SS is a research scholar of the Fonds de la Recherche en Sante in Quebec.
Conflict of interest None.