Bayesian Estimation Improves Prediction of Outcomes After Epilepsy Surgery

Area of Science:

• Neurology
• Biostatistics
• Medical Research Methodology

Background:

• Epilepsy surgery aims for seizure freedom, but study power is often insufficient.
• Underpowered studies may overestimate treatment effects, complicating clinical decisions.

Purpose of the Study:

• To estimate the statistical power of studies predicting seizure freedom post-epilepsy surgery.
• To compare effect size exaggeration between traditional and Bayesian methods in underpowered studies.

Main Methods:

• Data extracted from a Cochrane meta-analysis on epilepsy surgery outcomes.
• Statistical power and effect size exaggeration were calculated for included studies.
• Bayesian estimation of odds ratios was used for comparison.

Main Results:

• Median statistical power across studies was low, at 14%.
• Studies with median sample size or less (n≤56) and significant results exaggerated effect sizes by 5.4 times.
• Bayesian odds ratio estimation attenuated this exaggeration to 1.6 times.

Conclusions:

• Bayesian estimation of odds ratios effectively reduces the overestimation of effect sizes in underpowered epilepsy surgery studies.
• This method can enhance the interpretation of results from studies with small sample sizes, crucial for advancing epilepsy treatment research.