Abstract:
|
Evaluating risks (toxicity/tolerability/safety) and gains (activity/efficacy/effectiveness) separately often gives misleading results. When higher gains correlate with either more or fewer risks, significant results may be either irrelevant, or a comprehensive evaluation may yield more significant results, respectively.
Overall benefit [5-7] can be assessed from pairwise comparisons of data profiles. Profile A is better than B if all outcomes are at least as good and at least one is better. If A is better for some outcomes and worse for others, A and B cannot be ordered. Nonetheless, all profiles can be scored using either u-statistics or the marginal likelihood. These scores are invariant to scale transformations and intrinsically valid. No better profile can get a lower score. The approach is based on the same mathematical foundation, as well-known methods for censored observations[3,4]. Although computationally intensive, it has been successfully applied to a variety of clinical trials [1,2].
[1]CancRsrch 2001,61:6451 [2]Blood 1995,86:2815 [3]Biometrika 1973,60:267 [4]StatistNeerl 1983,37:69 [5]AJPH 88:671 [6]JASA 1988,83:1163; 1992,87:258; [7]AJPH 1998,88:590,972; 1999,89:109
|