Trichotomous prognosis-based outcome analysis: an efficient and clinically relevant method for analysis of acute stroke trials

Background Stroke trial outcomes are usually classified in ordered functional categories, e.g. the modified Rankin Scale (mRS). Using a single cut-off to distinguish “good” and “poor” outcome is statistically inefficient and could miss effects in subgroups where few patients can be expected to achieve good outcome. “Ordinal shift analysis”, using ordinal logistic regression to estimate a common odds ratio for all possible outcome category thresholds, is more efficient but could still obscure clinically important differences in response among prognostic subgroups. An alternative “prognosis based outcome” (PBO) approach defines separate, clinically meaningful good and poor outcome thresholds (GOTs and POTs) within each prognostic stratum. The strata are then combined in an overall trichotomous comparison of “good”/“intermediate”/“poor” outcomes.