Evaluation of Strategies for the Phase II to Phase III Progression in Treatment Discovery
Sanchez, Brittany Jo
MetadataShow full item record
The goal of clinical research is to improve the health of the population through the prevention, diagnosis, and treatment of disease. Clinical trials are essential for reliably evaluating a proposed treatment to determine whether it should be adopted into clinical practice. Current standards involve the evaluation of a new treatment through several phases of investigation. After preliminary evaluations of the safety and ethics of further study, promising treatments are studied in preliminary screening trials and then ultimately large, confirmatory trials. Although well developed, the "treatment discovery process" is lengthy, expensive, and has low success rates for treatments even at confirmatory phases of the investigation. Improvements to trial design and implementation are necessary for better achieving the goals of clinical research. In this research, we consider the progression of studies for investigating a new treatment, and discuss strategies in a framework that encompasses the period from the start of preliminary Phase II studies to the completion of the confirmatory Phase III studies. Using a general notational framework for evaluating new treatments, we examine optimality criteria for a strategy that best addresses the often competing goals of science, ethics, and efficiency. These optimality criteria include not only the standard frequentist operating characteristics of type I error and power and the standard Bayesian criteria of positive and negative predictive values, but also the efficiency considerations of the number of new treatments identified in a setting with limited resources. We parameterize the Phase II and Phase III designs using frequentist type I error and power in such a way as to attain high Bayesian positive predictive value (PPV). We then explore the impact specific choices of those design parameters have on the number of effective and ineffective treatments identified with constrained resources. We illustrate how allowing for early trial termination for efficacy or futility with a group sequential design (GSD) within Phase II and/or Phase III improves efficiency in terms of the number of subjects used on average for identifying effective therapies. Other methods for improving efficiency by eliminating the time spent between Phase II and Phase III have been proposed. A "seamless" Phase II/III trial design is one that combines the Phase II screening stage with the Phase III confirmatory stage. We consider how a single sequential design differs from the optimal approach of independent stages. We then explore how the traditional approach of adapting hypotheses at the end of Phase II fits in with the newer adaptive methods. We discuss how powering of Phase III based on Phase II results mimics adaptive sample size re-estimation / re-powering of study and does not offer improvement beyond that of GSDs. Bias in the estimate of the treatment effect is a result of the lack of precision of small samples inherent in Phase II studies and at early interim analyses. We investigate how such bias can be addressed with adjustment methods. We then examine differences between conducting subgroup analyses when there exist homogeneous versus heterogeneous effects and how inflation of the type I error can be controlled in this setting and in the setting of considering multiple summary measures. In our research, we thus demonstrate that the optimal Phase II to Phase III progression defined by an acceptable PPV and a maximal number of effective treatments can be identified for an anticipated prevalence and hypothesized resources by a parameterization of type I error and power at Phase II. We recognize that several approaches lead to the same optimality criteria, and that the chosen strategy will depend on individual objectives of clinical researchers, trial sponsors, regulatory agencies, patients on study, and those who might benefit from new knowledge about treatments being studied.
- Biostatistics