A single treatment may have a different effect on different patients. In particular, some patients may benefit from a given treatment while others do not. Often, some of the variation in effect among patients can often be explained by characteristics of those patients that are observable before treatment. Widespread acknowledgment of treatment effect variation due to observable patient characteristics has increased the health science community's interest in a broad field referred to as personalized or precision medicine. Among the aims of precision medicine are identifying the set of treatments that would benefit a given patient, and conversely, identifying the population of patients who would benefit from a given treatment. We treat the latter problem in the context of clinical trials run by treatment developers (e.g., pharmaceutical companies), with special attention paid to interactions between those developers and the relevant regulatory agencies (e.g., the US Food and Drug Administration). The primary difficulty in estimating the benefiting population in such settings is controlling the frequency with which at least one type of patient is incorrectly determined to benefit, and doing so in a way that does not render the approach excessively conservative. As a motivating application throughout this dissertation, we consider a battery of related clinical trials of treatments for Alzheimer's disease carried out by the pharmaceutical company AbbVie. These trials contain a small number of continuous and binary baseline patient characteristics that may influence the treatment effect. We apply standard and more novel regression models to the supplied data and develop methods of inference to accommodate the varied features of the datasets, such as nonlinear effects, multiple important endpoints, more than two treatments, and regions of the covariate space that are sparse in observations or lacking common support among treatment arms. We also discuss topics in practical implementation of these methods. Our approaches yield reliable and easily interpretable inferences regarding the population that benefits from treatment.
University of Minnesota Ph.D. dissertation. May 2017. Major: Biostatistics. Advisor: Bradley Carlin. 1 computer file (PDF); x, 134 pages + 1 supplementary zip file.
Credible Subgroups: Identifying the Population that Benefits from Treatment.
Retrieved from the University of Minnesota Digital Conservancy,
Content distributed via the University of Minnesota's Digital Conservancy may be subject to additional license and use restrictions applied by the depositor.