Statistical issues in the assessment of the evidence for an interaction between factors in epilepsy trials

We examine the common clinical belief that there is an interaction between epilepsy type and the two standard anti-epileptic drugs, valproate and carbamazepine, using data from several randomized clinical trials. Epilepsy type is not always easy to define, and three possible reclassifications are investigated to see whether misclassification of epilepsy type within the trials has potentially masked such an interaction. Regression modelling is employed to investigate whether heterogeneity between trial results can be explained by patient factors. Our work suggests that uncertainty in epilepsy type classification should be recognized in future studies. We also generate the hypothesis that the interaction of drug effect with age may reflect the perceived interaction with epilepsy type. We suggest that in any context where misclassification is likely, it is worth considering the use of an explicit 'unclassified' group, and investigating whether additional covariates can answer the substantive question.