Charles Manski, Northwestern University

Using Limited Trial Evidence to Choose Drug Dosage when Efficacy and Toxicity Increase with Dose

Abstract:  In medical treatment with drugs, it has become standard to base dosage on evidence obtained in randomized trials. Yet the trials performed to obtain drug approval rarely study how patient outcomes vary with dosage. They specify one dose of a new drug and compare it with an established therapy or placebo. The standard approach to analysis of trials views each trial arm as qualitatively different, but efficacy and toxicity both increase with dosage. A further problem is the practice of studying efficacy and toxicity separately, even though optimization of care requires balanced attention to both. This paper develops methodology to credibly use limited trial evidence to choose dosage when efficacy and toxicity increase with dose. I suppose that dosage is an integer choice variable t ∊ (0, 1, 2, . . . , T), where T is a specified maximum dose. Let the objective be to maximize expected patient welfare. The central difficulty addressed in the paper is that trial evidence on outcomes is available for only K dose levels, where K < T + 1. Then the population distribution of dose response is partially rather than point identified. Choice of dosage to maximize expected welfare is infeasible, but a clinician or public-health planner can make a reasonable choice under ambiguity. I study decision making using the minimax-regret criterion.