Does accounting for seizure frequency variability increase clinical trial power?

Area of Science:

• Neurology
• Clinical Trials
• Biostatistics

Background:

• Seizure frequency variability impacts placebo responses in randomized controlled trials (RCTs).
• High variability can lead to drug misclassification and reduced statistical power.
• Traditional methods like the 50%-responder rate (RR50) may not adequately address this variability.

Purpose of the Study:

• To investigate a novel method, Z_V, that directly incorporates seizure frequency variability into RCT analysis.
• To compare the predictive accuracy and statistical power of Z_V against the traditional RR50.

Main Methods:

• Assessed two models: traditional 50%-responder rate (RR50) and variability-corrected score (Z_V).
• Tested Z_V on three epilepsy datasets (SeizureTracker, Human Epilepsy Project, NeuroVista).
• Simulated 200 trials for various sample sizes using an independent dataset to determine statistical power.

Main Results:

• Z_V demonstrated higher prediction accuracy (91-100%) compared to RR50 (42-80%).
• Simulated RCTs using Z_V analysis achieved over 90% power at N=100 per arm.
• The traditional RR50 method required N=200 per arm to achieve similar power.

Conclusions:

• The Z_V method shows potential to increase statistical power in epilepsy RCTs.
• Z_V offers a more accurate and powerful approach to analyzing seizure frequency data in clinical trials.