Introduction

Top Abstract Introduction Methods Results Discussion References

Each year about one million people die and about 10 million are seriously injured on the world's roads.¹ The World Health Organization has indicated that, for people aged 3-35 years, road traffic crashes are now the leading cause of death and disablement. The global economic burden of road traffic crashes is estimated at $500bn (£300bn, 500bn).² Most of the casualties are in low and middle income countries, and most are vulnerable road users: pedestrians, cyclists, and riders of two wheeled motor vehicles. Children as pedestrians are particularly vulnerable, and pedestrian injuries account for most of the 280 000 childhood road deaths each year. ^{1 3 4} Elderly pedestrians constitute another particularly vulnerable group.⁴

In the prevention of pedestrian injuries, educational measures to teach pedestrians how to cope with the traffic environment are considered to be an essential component of any strategy, and pedestrian education has been recommended in high, middle, and low income countries.² Because the resources available for road safety are limited, a key question for road safety policy concerns the relative effectiveness of different prevention strategies. The aim of this systematic review of randomised controlled trials was to quantify the effectiveness of safety education programmes for pedestrians in improving their knowledge, attitudes, and behaviour and, most importantly, in preventing pedestrian-motor vehicle collisions.

	Methods

Top Abstract Introduction Methods Results Discussion References

Identification of trials
We aimed to identify all randomised controlled trials of road safety education programmes for pedestrians of all ages. We also included community based interventions such as media awareness campaigns and parental education programmes. We excluded studies where safety education of pedestrians was confounded by another intervention and studies that tried to modify the behaviour of drivers towards pedestrians.

We identified trials by computerised searches of the Cochrane Injuries Group specialised register, Cochrane Controlled Trials Register, Transport, Medline, Embase, ERIC, PsychLit, Spectr, and the World Health Organization's database on the internet; by checking the reference lists of relevant reviews, books, and articles; by contacting authors of relevant papers; by use of the citation analysis facility of the Science Citation Index and Social Sciences Citation Index; and by contacting relevant professionals, organisations, and voluntary agencies. No methodological filters were used, and we made no language restrictions and repeated searches with key words translated into French, German, Italian, Spanish, Dutch, and Danish.

Outcome measures and data extraction
Two reviewers independently extracted data on pedestrians' injuries, behaviour, attitude, and knowledge; pedestrian-motor vehicle collisions; methods of randomisation; and numbers lost to follow up. We assessed trial quality using the method proposed by Schulz.⁵ Disagreements were resolved by discussion with a third reviewer. When the method used to conceal allocation of intervention was not clearly reported we contacted the study author, if possible, for clarification.

Data analysis and statistical methods
Wherever possible we performed an intention to treat analysis. Meta-analysis was not considered appropriate because of the differences across studies in the types of interventions and the types of outcomes. We calculated effect estimates with RevMan version 4.1 and report these as relative risks (95% confidence intervals) for dichotomous outcomes (relative probability of presenting the measured outcome in trained pedestrians compared with non-trained ones) and as standardised mean difference (95% CI) for continuous outcomes. If the variance for the change score was not presented and could not be obtained from the authors, we ascribed a value using a correlation factor between pretest and post-test scores of r=0.50. ^{6 7} We report the post-test data or the change between pretest and post-test when available, grouped by age categories and by type of outcomes. Outcomes are expressed as "positive" expected behaviour, attitude, or knowledge, so that a relative risk of >1 and a standardised mean difference of >0 represent a beneficial effect of the intervention programme.

In the included studies, training was provided either directly to the target population (direct education) or by training "intermediate" educators such as parents or teachers (indirect education). The way safety education is provided and the age of the target group are potential effect modifiers, but we did not explore their influence because we did not perform a meta-analysis.

For cluster randomised trials, we calculated an "effective sample size" if the intra-cluster coefficient was available.⁸ We excluded studies in which there were less than five randomised clusters because, in order to analyse at the individual level, one would have to assume that there is no clustering of individual responses within the community, which is almost always untenable.⁹

	Results

Top Abstract Introduction Methods Results Discussion References

We identified 13 899 studies, of which 674 (5%) were potentially relevant based on the title or abstract of the report. After a full text review, we identified 15 trials that met our inclusion criteria,^10-23 two of which are reported in the same document.²² Table 1 shows the basic characteristics of these trials.

The methodological quality of the included trials was generally poor. The method of allocation concealment was adequate in only three trials, ^{13 17 18} outcome assessment was blinded in eight, ^{10 11 14-16 21-23} and in most of the studies large numbers of participants were lost to follow up. The participants were children in 14 of the studies and institutionalised adults in one.¹⁶ No trial focused on the other vulnerable pedestrian group, elderly people. All trials were conducted in high income countries. Eight studies involved the direct education of study participants, ^{10 12 15 16 18 19 21 23} and seven involved the use of parents ^{13 14 22} or teachers ^{11 17 20} as educators. Outcomes were measured before and after the intervention in 12 studies ^{10 12-17 20-23} and only after the intervention in three studies. ^{11 18 19} None of the trials assessed the effect of safety education on the occurrence of pedestrian injury, but five assessed the effect on observed behaviour, ^{14 16 18 19 22} one assessed reported behaviour,¹⁷ six assessed attitude, ^{10 12 19 21-23} and five assessed knowledge. ^{11 13 15 17 20}

Each research group used different tools to measure outcomes, and the delay for the post-test measurement varied from less than one month to eight months. Six trials measured the effect of safety education at different times after the intervention. ^{14 17 21-23} The effect of the intervention was lower in the later follow up period for 18 of the 24 behavioural outcomes, ^{14 22} for two of the four attitude outcomes,^21-23 and for the two knowledge outcomes.¹⁷

In some studies, the post-test conditions varied and influenced the results. For example, Limbourg and Gerber¹⁴ reported that 5-6 year old children given safety education were, at five months after intervention, more likely to stop and look at the line of vision when crossing roads than controls (relative probability 1.79 (95% confidence interval 1.18 to 2.72) for children without distraction). However, when the children were distracted by racing with another child the relative probability increased to 2.80 (1.39 to 5.64).

Table 2 shows the most pertinent outcomes and only the longest period to post-test measurements. (More detailed results are available in the Cochrane Library.) Overall, the effect of safety education on pedestrian behaviour varied considerably. The relative probability of trained pedestrians behaving correctly compared with controls ranged between 1.63 and 2.13 for the selected outcomes in table 2 but varied overall between 0.49 (control group performed better than trained group) and 9.29 for all the studies and outcomes (data not shown). Safety education improved pedestrians' attitude and intentions (with standardised mean differences ranging from 0.17 to 1.28) and their knowledge about road safety when outcomes were measured before and after intervention (standardised mean differences from 0.16 to 2.39), but for dichotomous outcomes the range of effect was wide (relative probability ranging from 0.72 to 1.66) (data not shown).

	Discussion

Top Abstract Introduction Methods Results Discussion References

Despite a thorough search in several databases in many languages and by contact with various interested parties, we could not identify good evidence of effectiveness of safety education for adult pedestrians and only limited evidence for child pedestrians. None of the included trials assessed the effect of safety education on the occurrence of pedestrian injury, but six trials assessed the effect on observed behaviour. Some of these trials showed evidence of behavioural change after safety education, but for various reasons it is difficult to predict what effect this might have on pedestrian injury risk.

Firstly, we cannot be sure that the observed behaviour is causally related to the occurrence of pedestrian injury. For example, Nishioka et al¹⁸ considered that slowing down or stopping before crossing a road to be the safe response. However, even if this behavioural change, observed in a simulated traffic environment, was repeated in a real traffic situation it is difficult to estimate what effect it would have on injury risk. Once a child has established that a road is clear, it may be safer to run across before another vehicle approaches because it reduces the time of exposure to risk. Secondly, assuming that the measured behaviours are causally related to risk of pedestrian injury, we have no reliable information about the size of this effect, and so we cannot predict how much a given behavioural change will reduce the risk of injury. Finally, there is uncertainty about the extent to which the observed behavioural changes persist over time, although the apparent declines may have been due to chance alone.

Limitations of review
Certain methodological issues could have an important bearing on the validity of our findings. In particular, publication and other selection biases may have resulted in the over-representation of studies showing promising intervention effects. This is especially likely in the context of road safety, where a large proportion of the available research information is published in the grey literature of road safety research organisations. Most of the statistical methods that can be used to assess the possibility of publication bias require the use of meta-analysis and so could not be used in this systematic review.

Although we made considerable efforts to identify all eligible trials, published and unpublished irrespective of language, we cannot exclude the possibility of selection bias. The validity of the inferences from any systematic review depends on the quality of the included studies, and in this case many of the studies were of poor quality. It has been shown that inadequate allocation concealment, lack of blinding of outcome assessment, and large losses to follow up can result in the overestimation of intervention effects in randomised controlled trials,⁵ and many of these methodological weaknesses were present in the included trials.

Several included studies were conducted more than 10 years ago, and so their relevance to the current situation is open to question. Walking habits and the pedestrian environment have dramatically changed during the past two decades. All the included trials compared groups that were in the same surroundings, allowing the effect of the intervention to be isolated. Another limitation of this study is that we could not identify any randomised controlled trial conducted in low and middle income countries.

Implications of results
The Global Road Safety Partnership strongly recommends road safety education of children worldwide.²⁴ Our review indicates that there is no reliable evidence supporting the effectiveness of pedestrian education for preventing injuries in children and inconsistent evidence that it might improve their behaviour, attitudes, and knowledge. While the value of safety education of pedestrians remains in doubt, environmental modification and the enforcement of appropriate speed limits may be more effective strategies to protect children from road traffic.

Conclusions
Pedestrian safety education can improve children's knowledge of the road crossing task and can change observed road crossing behaviour, but whether this reduces the risk of pedestrian-motor vehicle collision is unknown. No trial focused on the other vulnerable road users, elderly pedestrians. None of the trials was conducted in low and middle income countries.

Large scale, randomised controlled trials with injury outcomes (or end points that are likely to predict injury outcomes, such as near misses) are needed to establish the effectiveness of safety education of pedestrians. Although some existing trials showed evidence of behavioural change after safety education, these changes cannot be assumed to decrease pedestrian injury risk.