PP11-83: Exploration of Statistical Methods for Seizure Frequency in Epilepsy Trials

Poster Presenter

Xiangmin Zhang

Statistician
FDA
United States

Objectives

In late phase clinical trials for anti-epileptic drugs, seizure frequency is a common efficacy outcome. In our research, we aimed to evaluate various statistical models for seizure frequency using empirical data and simulated data and use the results to inform future trial design and analysis.

Method

Our empirical database comprised of registration efficacy trials in regulatory approvals for epilepsy in the US in the past decade. Statistical models, including analysis of covariance and Poisson models, were applied to empirical data and simulated data.

Results

We compared the p-values and AICs under different models using the empirical data and found that the Poisson models generally did not perform well. We also found from simulation studies that the Poisson models may not control type I errors, even when the over-dispersion has been accounted for.

Conclusion

Using Poisson models to fit seizure data may be questionable for epilepsy trials. We recommend investigating the diagnostics of model fitting to avoid using the “wrong” models to make decisions. Disclaimer: This poster reflects the views of the authors and should not be construed to represent FDA’s views or policies.