• international collaboration

Unintentional injury to children aged under 5 caused 428 300 deaths globally in 1990. Not surprisingly, the majority are in the developing world, with 740 deaths annually per million of the population aged less than 5, compared with 249 per million a year in developed countries. However, injuries account for 15.6% of deaths after the perinatal period in this age group in the developed regions, compared with 4.1% in developing regions.¹ Thus, although in absolute terms injuries are a problem that needs to be tackled worldwide, proportionally they are a greater problem in the developed world.

Either way, the size of the problem means there is a clear need for initiatives of proven value that can be used to reduce mortality.² The greater prosperity and relative importance of childhood injury in the developed world mean that this is where robust research is most likely to be performed. However, studies that included developing countries could inform a global injury prevention strategy.

Young children, are less competent than adults in assessing and avoiding risk; thus society has an obligation to develop initiatives to reduce their injury rate. Such initiatives can be broadly divided into:

(1) Community based interventions, that affect mainly the environment outside the home, or safety of items brought into the home, such as encouraging (or enforcing), the use of child seats in cars or improving the safety of children's toys.

(2) Educational initiatives targeted at parents to improve safety behaviour such as avoidance of burns and scalds and the correct use of child seats in cars.

Although some public policy initiatives have been tested in a comparative manner either using historical data or similar areas with no intervention as controls, few such initiatives have been evaluated using randomised controlled designs.³ Subjecting the introduction of public policy initiatives to the rigour of a controlled trial, possibly using a no intervention control district or districts, presents considerable difficulties.⁴ If a small number of districts were used in such studies it would be difficult to generalise from the results or even to be certain that any differences found were due to the intervention. The additional problems involved in ensuring adequate randomisation and an element of blinding mean that it is unlikely that such studies will be performed.

Educational and support programmes, however, can and should be subjected to rigorous scientific evaluation before they are introduced on a large scale. Roberts et al, in a systematic review of eight trials, calculated the pooled odds ratio of the effect of home visiting to prevent unintentional injury to be 0.74 (95% confidence interval 0.6 to 0.92).⁵ All but one of these studies selected mothers of children perceived to be at high risk of injury and all provided labour intensive interventions. Although the rate of injury among children at high risk means there is, in relative terms, a large potential benefit from such interventions, most children are not at high risk.

To achieve a significant absolute reduction in the rate of injury any programme should be universal. It is unlikely that resources will be available to fund individual professional or non-professional support, of the type analysed by Roberts, for all new mothers. A more realistic scenario is the provision of an injury prevention programme aimed at all pregnant women or parents of preschool children.

This could include additional classes, leaflets, or other educational material provided during antenatal care, the postpartum period, or routine child health surveillance. Additional support, or the provision of safety equipment, could also be targeted at families perceived to be at higher risk based perhaps, on home accident risk audits.

Existing studies have failed to demonstrate any convincing injury reductions from educational initiatives. However, an effect on parental behaviour, an interim outcome measure, has been shown.³ One review reported such positive effects in five out of seven randomised controlled trials.⁶

Nevertheless, educational and support programmes are an attractive approach to reducing childhood injury, not least because if effective they would be relatively economical. The failure of existing studies to show a reduction in injury rates may reflect inadequate sample size rather than an ineffective intervention.

In future such trials may be well advised to concentrate on those under 5 for the following reasons.

(1) A home education programme will have less impact on older children as they become more autonomous, less under direct parental control and, therefore, more likely to be injured outside the home.

(2) Younger children are injured more often which makes it, statistically, easier to demonstrate any benefit.

(3) Parents of younger children will be more accessible for an education and support programme than those of older children.

(4) If a cluster randomisation is used, increasing the number of individuals in each cluster will provide little statistical advantage (see below).

In the developed world one in 4000 of children aged less than 5 will have a fatal accident each year. Accordingly a trial to demonstrate a reduction in mortality as the end point would have to be unrealistically large. It would, however, be possible to test, in a randomised controlled trial, the hypothesis that:

An intensive parental educational and support programme will reduce hospital admissions, due to injury in the first five years of life by one fifth.

UK figures show that 2% of children aged under 5 will be admitted to hospital with an injury annually; one quarter of these will be moderate or severe.⁷ Thus allowing for repeat admissions, about 9.6% of children will have been admitted to hospital with an injury at least once by their fifth birthday. To show a reduction in admissions by one fifth to 7.7% (significant at the 5% level and with a power of 80%), for five years of intervention and data collection, 3463 children in each group would be required.

Randomisation by family would be inappropriate because unintentional exposure of the control group to the intervention could occur if they attended the same health clinics as the intervention group,

This means that randomisation by “health care unit” is preferable (cluster randomisation). Unfortunately, because some of the differences between groups attending different health care facilities will be due to factors other than the intervention, it will be necessary to apply an inflation factor to the sample size calculation.⁸ The inflation factor for the effect of randomising by cluster is given by the formula:−1+(ñ−1)ρ (ñ is the average number of individuals in a cluster and ρ is the intracluster correlation coefficient (ICC) a measure of the extent to which the behaviour of individuals in a cluster is affected by their cluster membership). A value of one for the ICC means that all the individuals in the cluster behave identically manner, a value of zero means that their behaviour is unaffected by cluster membership. The cluster effect can be reduced by matching the treatment and control centres for known potential confounding variables, but it cannot be avoided completely.

In the UK 98% of the population are registered with a general (family) practice. Services are free at the point of use, and provide, either directly or indirectly, most community maternal and child health services. Assuming a low ICC (ρ) of 0.01, and an average cluster size (ñ) of 100 children (the number born annually into a typical British general practice) a doubling of the sample size is required. If the true value of ρ is larger the sample needed would increase greatly and vice versa if smaller clusters were used.

This sample size exercise demonstrates that to perform a trial of this nature in the UK at least 70 general practices would need to be involved for five years. A trial of this magnitude is costly and unlikely to be repeated. It should be designed therefore, to be as robust and as generalisable as possible. If such a trial was performed in a single country the results could not necessarily be applied internationally. A collaborative study, involving centres in a number of different countries with different economies, health care systems, and injury risk profiles might, therefore, be preferable. If a mixture of developed and developing regions were included robust conclusions could be drawn for the main trial end points and substudies of national interest could also be performed. However, international differences will affect both the injury rate and the likelihood of a child being admitted to hospital after an injury. These differences will make calculating the sample size more complex and could increase the sample size required.

The education package should be tailored to the different risks in participating countries. For example, drowning in a swimming pool is commoner in southern compared with northern Europe, while the reverse is true for burns from domestic heaters. What would be tested is, the pragmatic hypothesis that an education and support package will reduce injuries, rather than explanatory hypotheses identifying which part or parts of the package has the effect.

Although international cooperation increases the complexity of such a trial, if positive results are found there would be a mandate to institute the programme worldwide. If negative results are found we will know conclusively that the solution to this problem lies elsewhere. The European Union through its BIOMED II programme, provides concerted action funding to allow European collaborators to meet and produce detailed trial protocols to be submitted to funding bodies in individual countries. Similar pump-priming from a supranational agency, such as the World Health Organisation would allow potential collaborators to produce a robust trial design. It is time for international cooperation to design a definitive multicentre trial to decide if an intensive parental educational and support programme will reduce the global burden of childhood injury.