Sliding dichotomy compared with fixed dichotomization of ordinal outcome scales in subarachnoid hemorrhage trials

In randomized clinical trials of subarachnoid hemorrhage (SAH) in which the primary clinical outcomes are ordinal, it has been common practice to dichotomize the ordinal outcome scale into favorable versus unfavorable outcome. Using this strategy may increase sample sizes by reducing statistical power. Authors of the present study used SAH clinical trial data to determine if a sliding dichotomy would improve statistical power.

Available individual patient data from tirilazad (3552 patients), clazosentan (the Clazosentan to Overcome Neurological Ischemia and Infarction Occurring After Subarachnoid Hemorrhage trial [CONSCIOUS-1], 413 patients), and subarachnoid aneurysm trials (the International Subarachnoid Aneurysm Trial [ISAT], 2089 patients) were analyzed. Treatment effect sizes were examined using conventional fixed dichotomy, sliding dichotomy (logical or median split methods), or proportional odds modeling. Whether sliding dichotomy affected the difference in outcomes between the several age and neurological grade groups was also evaluated.

In the tirilazad data, there was no significant effect of treatment on outcome (fixed dichotomy: OR = 0.92, 95% CI 0.80-1.07; and sliding dichotomy: OR = 1.02, 95% CI 0.87-1.19). Sliding dichotomy reversed and increased the difference in outcome in favor of the placebo over clazosentan (fixed dichotomy: OR = 1.06, 95% CI 0.65-1.74; and sliding dichotomy: OR = 0.85, 95% CI 0.52-1.39). In the ISAT data, sliding dichotomy produced identical odds ratios compared with fixed dichotomy (fixed dichotomy vs sliding dichotomy, respectively: OR = 0.67, 95% CI 0.55-0.82 vs OR = 0.67, 95% CI 0.53-0.85). When considering the tirilazad and CONSCIOUS-1 groups based on age or World Federation of Neurosurgical Societies grade, no consistent effects of sliding dichotomy compared with fixed dichotomy were observed.

There were differences among fixed dichotomy, sliding dichotomy, and proportional odds models in the magnitude and precision of odds ratios, but these differences were not as substantial as those seen when these methods were used in other conditions such as head injury. This finding suggests the need for different outcome scales for SAH.