Effect Size Estimates for the ESCAPE Trial: Proportional Odds Regression Versus Other Statistical Methods

Area of Science:

• Neurology
• Biostatistics
• Clinical Trials

Background:

• Ordinal outcomes, like the modified Rankin Scale (mRS), are standard in acute stroke trials.
• Various regression models exist for analyzing continuous, binary, or ordinal data.
• No consensus has been reached on the optimal statistical method for analyzing ordinal stroke trial data.

Purpose of the Study:

• To compare the statistical power of different regression models for analyzing ordinal outcomes in acute stroke trials.
• To identify the most effective statistical approach based on predicted treatment effects across the modified Rankin Scale (mRS).

Main Methods:

• Utilized data from the Interventional Management of Stroke-III and Prolyse in Acute Cerebral Thromboembolism II (PROACT-2) trials.
• Included patients with specific baseline criteria (ASPECTS > 5, M1 occlusion, adequate collaterals, short treatment times).
• Employed Monte Carlo simulations to compare the statistical power of regression models under various data scenarios.

Main Results:

• Binary logistic regression demonstrated higher statistical power when treatment benefits were concentrated at one end of the mRS.
• Proportional odds regression showed greater power when treatment effects were predicted across both ends of the mRS.

Conclusions:

• The distribution of mRS scores in both treatment and control groups impacts the power of statistical models.
• Selecting the best primary analysis method requires careful consideration of the expected outcome distribution across the mRS spectrum.