Combined MRI lesions and relapses as a surrogate for disability in multiple sclerosis—Sormani et al., Neurology 77, 2011

Abstract

This excellent paper assesses and finds strong evidence for the score. At the trial baseline, the treated patients (triangles) would potential utility of a novel composite endpoint in phase III trials for interferon or biosimilars in relapsing remitting (RR) multiple sclerosis (MS) patients. Composite endpoints are often used, particularly in cardiovascular trials, in order to increase trial power at a given duration, or where no single endpoint is clinically sufficient; but in such contexts composite outcomes can be difficult to interpret, particularly when treatment effects differ for the component measures, or when these differ in clinical importance. These disadvantages do not affect this application, since the composite endpoint is designed both as a surrogate for a single clinical endpoint and to predict treatment effect on this single clinical measure. The endpoint combines the number of clinical relapses and the number of MRIdetected new lesions, both observed over one year trial duration, and is assessed as a surrogate to detect the treatment response observed at two years in a standard measure of clinical disability, the Expanded Disability Status Scale (EDSS); the surrogate offers the possibility of shorter duration trials assessing treatment as effectively as longer trials using EDSS outcome. The authors used data from a trial of interferon in RRMS (560 patients) to determine whether the treatment effect on the surrogate at one year explained the treatment effect on EDSS at two years, using a method based on established surrogacy criteria. These criteria require an effective surrogate in this context to have some but not necessarily strong correlation with disability; however, the treatment effect on the surrogate must be strongly correlated with the treatment effect on disability. The idea is illustrated in the figure, which uses a simulated example with (for ease of illustration) a continuous disability measure and a single surrogate (dichotomised EDSS was used by the authors, and correspondingly logistic rather than linear regression, but the same principles apply). The two parallel lines are fitted regression lines representing in each trial group the relationship between disability after two years and surrogate after one year. The fairly wide dispersion of data points around the fitted lines illustrates a modest correlation between surrogate and disability: a patient’s disability score cannot be precisely predicted by their surrogate