|
Abstract
|
We congratulate Chapman et al for their paper on prevention of dog bites in children.1 The study was a cluster randomised trial; schools were allocated to intervention or control groups, and data were collected at the level of the individual schoolchild. The analysis did, however, not account for cluster randomisation. The assumption of independence of observations required for standard statistical tests, such as the chi-square test used by Chapman et al, is violated in cluster randomised trials. It is not unreasonable to assume that the actions of the children from one school towards the dog were more similar to each other than to the actions of children from another school.
Standard statistical methods that do not account for cluster effects in data from cluster randomised trials will result in the overestimation of the significance of an intervention. For dichotomous outcomes, Donner and Klar recommend a two sample t test based on cluster level event rates when the numbers of clusters in each arm of the trial number is less than 10.2 Analysing the data ignoring clustering gives a difference between the groups of 70% (P<0.0001; 95% confidence interval 62% to 77%). Reanalysis of the school level event rates using a t test with equal variances gave a mean difference of 70% (P=0.0018; 36% to 100%). This showed that although the evidence is still in favour of the intervention to prevent a bite, it is not as strong as when clustering is ignored. A revision of the CONSORT statement for reporting of cluster randomised trials should help researchers avoid the potential pitfalls of such unit of analysis errors.3
1. Chapman S, Cornwall J, Righetti J, Sung L. Preventing dog bites in children: randomised controlled trial of an educational intervention. BMJ. 2000;320:1512–1513.
2. Donner A, Klar N. Statistical considerations in the design and analysis of community intervention trials. J Clin Epidemiol. 1996;49:435–439.
3. Melbourne D, Campbell MK. Extending the CONSORT statement to cluster randomised trials: for discussion. Stat Med 2001;20(3):489-496.
Keywords: Animal Bites; Dog Bites
Language: en |