The difficulty in predicting the most suitable dichotomy for mRS, dependence on case mix, and failure to incorporate information on potential harm at other levels of the scale each renders this counterintuitive; ordinal analysis is statistically more powerful,
is more robust to variation in case mix among trials, and better expresses the importance to clinicians and patients of functional gains at any level of mRS. If a second measure should be used, it would be reasonable to require that it simply tracks with mRS in a predicted manner and shows changes in a similar direction, but it would not be reasonable to require a similar level of statistical significance for a less powerful and less relevant measure. -
http://stroke.ahajournals.org/content/43/4/1163.long