Abstracts

Rationale: The choice of primary endpoint to support the regulatory approval of an antiepileptic drug as add-on therapy for focal epilepsy is not agreed upon. Recently, more sophisticated statistical methodologies have been applied to analyze seizure outcomes, including the application of parametric statistical methods after transformation of seizure frequencies. While these methodologies have advantages, there are also critical considerations when applying these methods. The purpose of this analysis was to evaluate the impact of alternative standardizations of partial onset seizure (POS) frequency when analyzing log-transformed outcome. Methods: Seizure data from two randomized, double-blind, placebo-controlled trials (N01252/N01253) evaluating the efficacy and tolerability of adjunctive brivaracetam (BRV) in patients with uncontrolled focal epilepsy were used. After completing an 8-week prospective Baseline Period, eligible subjects were randomized to receive placebo or BRV doses ranging from 5 mg/day to 100 mg/day (equally divided b.i.d. dosing) without up-titration during a 12-week Treatment Period. Analysis of covariance (ANCOVA) was applied to analyze log (x 1) transformed Treatment Period seizure frequency, with adjustment for stratification variables and log-transformed baseline seizure frequency. Analyses were produced for POS frequency standardized to 7-day and 28-day durations. Treatment effects are characterized as percent reduction over placebo after back-transformation of effect estimates from the ANCOVA. Results: Table 1 presents results for the ANCOVA analyses of POS frequency standardized to a 7-day and 28-day duration. The magnitudes of p-values are comparable for the two methods and conclusions regarding statistical significance are not altered. However, the effect estimates (percent reductions over placebo) differ for the two methods. The magnitude of effect estimates for the analysis of 7-day adjusted frequency appear to be inconsistent with secondary variables, such as median percent reduction in POS frequency from Baseline (Table 2). Furthermore, an evaluation of the underlying statistical methodology supports the conclusion that the analysis of 7-day adjusted seizure frequency underestimates the magnitude of the treatment effects, and the analysis of 28-day adjusted seizure frequency is statistically valid and provides effect estimates that more appropriately characterize the treatment effects for these studies. Conclusions: These results illustrate the impact of the choice of statistical methodology on the perceived magnitude of treatment effect when applying a parametric statistical method to analyze log-transformed seizure outcome. The duration over which seizure frequencies are standardized may have a notable impact. UCB-sponsored (NCT00464269; NCT00490035)