Neurological scales as endpoints in stroke studies. Aspects of statisticalanalysis

Even after publication of ECASS Il, the latest paper in a series of large, placebo-controlled studies on thrombolysis in acute ischaemic stroke, there is still uncertainty as to what the best clinical endpoint(s) is (are) in trial design for reliably identifying significant differences between treat ment groups. If the expected treatment difference as measured by a neurolog ical outcome scale like the Modified Rankin Scale corresponds more to a shi ft in dispersion ton average a majority of patients profits greatly) rather than to a shift in location ion average each patient profits much), then t he power of the odds ratio test is much higher than that of the Wilcoxon te st and therefore the clinical outcome parameters should be dichotomised. Wi th respect to the time window of 0-6 hrs from symptom onset of an acute isc haemic stroke, for example, a dichotomisation of 0-2 vs. 3-6 for the Modifi ed Ran kin Scale is reasonable. In the case of multiple endpoints, a global (multivariate) test should be used, but the correlation between these endp oints must not be too high, which means that the various manifestations of the stroke disease should be considered.