Bayesian methods for optimizing medical therapy given at multiple stages
Medical therapy often consists of multiple stages, with a treatment chosen by the physician in each stage based on the patient’s current history of treatments and clinical outcomes. Yet, most statistical methods for treatment evaluation focus on evaluating the treatments given at a single stage, which is of limited use to practicing physicians, since it does not reflect the actual therapeutic process, and can lead to sub-optimal treatment.
I am interested in applying Bayesian methods to identify optimal multi-stage treatment regimens, and thus improve treatment guidelines in mutiple-stage settings.
Clinical trial design in settings with multiple primary outcomes
Medical outcomes often are complex and multivariate, so physicians routinely select each patient’s treatment based on consideration of risk-benefit tradeoffs between desirable and undesirable clinical outcomes. Yet, conventional designs for randomized comparative trials seldom reflect this aspect of medical practice. Rather, most designs in clinical trial protocols are based on one outcome, identified as “primary,” with all other outcomes given the nominal status of “secondary.”
I am interested in designing clinical trials with decision rules that better reflect actual medical practice by accounting for all clinically relevant outcomes. This typically requires developing a practical decision rule and a flexible Bayesian model that will be used to assess the decision rule as the data accrue.
Incorporating additional sources of data into an analysis or clinical trial
Medical research is inherently sequential, and thus there often is an additional source(s) of data available that may be relevant to an analysis, or even a clinical trial. Because this additional source of information may be severely biased, it is usually ignored or very conservatively incorporated into the analysis.
I am interested in developing methods that facilitate incorporating additional data sources into an analysis or clinical trial such that inference is not severely affected when the primary and additional data sources are in conflict, but inference is more precise when the primary and additional data sources are in agreement. Using these methods in a clinical trial can increase power and/or reduce sample size requirements.