Sequential methods for comparative effectiveness experiments: point of care cliical trials

Abstract

The goal of comparative effectiveness research (CER) is to support evidence-based choices of treatments. Currently the majority of randomized trials for CER are designed to demonstrate superiority, which often require large sample size because the effect sizes between treatments in current use are typically small to moderate and there are usually more than two treatments to be compared. We propose an alternative group sequential design for such setting. Instead of testing superiority, we aim to select high quality treatments that are within a small distance from the best treatment. The basic idea is to eliminate non-promising treatments at interim analyses that cannot be much better than the currently observed best treatment, based on generalized likelihood ratio tests. This approach can also be used for guideline implementation and for phase II selection trials. Comparative Effectiveness Research (CER) involves the comparison of med- ical, surgical, and other treatments on their ability to benefit patients. The word 'effectiveness' conveys the sense that the treatments are to be compared on their effects as they are actually used by clinicians and received by patients, in pa- tient populations selected by diagnosis and other relevant considerations, but otherwise not restricted by race, sex, economic condition, insurance coverage, likelihood of successful adherence to treatment, and the like. Effectiveness con- trasts with 'efficacy', a term usually denoting the presence or absence of a desired effect in somewhat ideal circumstances of adherence to treatment and often in more highly selected populations (those without serious co-morbid conditions, in a restricted range of severity, or otherwise selected for homogeneity). CER is often approached by observational methods, including analysis of claims data or registry data. For example, Stukel et al. (2007) describe several attempts to use