Bayesian Estimation Improves Prediction of Outcomes after Epilepsy Surgery

Area of Science:

• Medical research
• Statistics in medicine

Background:

• Underpowered studies, common in medical research, inflate statistically significant findings.
• This exaggeration is particularly problematic in fields like epilepsy surgery, affecting treatment outcome predictions.

Purpose of the Study:

• To quantify statistical power and magnitude error in epilepsy surgery outcome studies.
• To evaluate the effectiveness of Bayesian estimation in mitigating effect size exaggeration in underpowered research.

Main Methods:

• Analysis of statistical power and effect size magnitude in published epilepsy surgery studies.
• Comparison of traditional odds ratio estimation with Bayesian methods for underpowered study results.

Main Results:

• Median statistical power across studies was only 14%.
• Studies with median sample size or less (n<=56) and significant results exaggerated true effect size by a factor of 5.4 (median OR 9.3 vs. true OR 1.7).
• Bayesian estimation of odds ratio attenuated exaggeration to a factor of 1.6 (median OR 2.7 vs. true OR 1.7).

Conclusions:

• Bayesian estimation of odds ratio effectively reduces the overestimation of effect sizes in underpowered studies.
• This statistical approach can enhance the accuracy of patient counseling regarding seizure freedom probabilities after epilepsy surgery.