Tangency Outcomes for Clinical Trials
Year of Publication
Oden, N; Sharma, G
Society for Clinical Trials 35th Annual Meeting
In substance-use trials, treatment may affect not only abstinence, but also quality of life, cognitive function, criminal behavior, socioeconomic status, etc. We propose to consider the outcome of a trial to be a weighted average of these terms that discriminates treatment groups well. We also show how to use these weights to test treatment effect. To search for optimal weights, we adapt Markowitz’s financial portfolio theory to clinical trials Our simulations compare 60 scenarios with respect to (1) the “no short-sales” tangency portfolio weights, (2) the estimated global minimum-variance portfolio weights, (3) the “apparently most powerful” t-test, and (3) the simple average. The scenarios investigated by simulation vary the number of and correlation between outcomes, and the relationship between outcome means and standard deviations. Similar to the classical investment advice that one should own many uncorrelated stocks, the simulations suggest that the Markowitz “no shortsales” tangency portfolio gives weights that, under a permutational hypothesis, have the appropriate test size and improve t-test power over tests of single outcomes when there are many weakly-correlated outcomes. The methods investigated are robust to departures from normality.