Empowering Research on Epilepsy Surgery Outcomes

Low statistical power is a recognized problem in many fields. We performed a systematic review to determine the median statistical power of studies of epilepsy surgery outcomes. We performed a PubMed search for studies reporting epilepsy surgery outcomes for the years 1980-2020, focusing on studies using stereoelectroencephalography (SEEG). We extracted patient count data for comparisons of surgical outcome between two groups, based on a reported prognostic factor. We defined a clinically meaningful difference as the difference in seizure freedom for MRI positive (66.9%) versus negative (45.5%) from the largest study found. Based on 69 studies of surgery outcomes in patients undergoing SEEG, the median sample size was 38 patients, and the median statistical power was 24%. This implies at least a 17% chance a study with a significant result is false, assuming 1:1 pre-test odds. Results from simulation studies suggest that, if a typical SEEG study finds a significant effect, then the median observed effect size will be more than double the true effect size. We conclude that studies of epilepsy surgery outcomes using SEEG are often statistically underpowered, which limits the reproducibility and reliability of the literature. We discuss how statistical power could be improved. SHORT SUMMARYWe performed a systematic review to determine the median statistical power of studies of epilepsy surgery outcomes, focused on stereoelectroencephalography. We extracted patient count data for comparisons of outcomes between two groups. We defined a clinically meaningful difference as the prognostic value of a normal versus abnormal MRI. Based on 69 studies, the median sample size was 38 patients, and the median statistical power was 24%. Underpowered studies will overestimate the size of true effects and are more likely to report false positive results. We discuss how statistical power, and thus reproducibility and reliability of results, can be improved.