New statistical method for analyzing time to first seizure

248 words) Introduction: Time to first seizure is a common outcome in antiepileptic drug (AED) studies. Previous studies have typically failed to find statistically significant differences between carbamazepine (CBZ) and valproate (VPS). We re-analysed a meta-analysis comparing CBZ and VPS monotherapy with new powerful statistical methods which incorporate baseline seizure rate information. Methods: Individual patient data were available on 1,265 patients from a meta-analysis of 5 trials. The outcome measure was time to first seizure after randomisation, adjusted for background variables and baseline seizure rate. Results: We found strong evidence of an interaction between treatment and epilepsy type, and between treatment and age. For generalized onset seizures VPS was statistically significantly better than CBZ: VPS delayed the first seizure after treatment 58%, 52%, 44% and 36% longer than CBZ for individuals aged 10, 20, 30 or 40, respectively. For partial onset seizures in individuals older than 30, CBZ was significantly better then VPS; CBZ delayed the time to first seizure by 9%, 25%, 44% and 66% longer than VPS for individuals aged 20, 30, 40 or 50, respectively. Discussion: The results show clear age-varying differences between the effectiveness of CBZ and VPS for generalized onset and partial onset seizures, which have not been identified in previous studies using standard statistical methods. In future trials of AED monotherapy or add-on where time to first or N seizure is an outcome, methodology which can incorporate baseline seizure rate information would allow more powerful comparisons between treatments or between treatment and placebo.