A Review of the Performance of Tests Used to Establish Whether There is a Drug Effect in Dose-Response Studies

Abstract

Within the pharmaceutical industry the assessment of dose-response is a critical part of a drug development program. When analyzing dose-response studies in clinical research one of the most important questions to be addressed is whether there is a drug effect since only once this is established can other analyses be performed to assess the nature of the dose-response relationship. In this paper the performance of eight tests for establishing a drug effect will be compared for a variety of different patterns of results. It will be shown that specialized tests conditioning on the a priori ordering of responses are robust to a variety of different patterns of results, and therefore, should be used. The more traditional approaches used by statisticians in the pharmaceutical industry based on analysis of variance lead to a loss of power.