In this proposal we will focus on two aspects primarily, a) Eliciting prior information from heterogeneous sources such as past similar clinical trials from patients/clinicians/stakeholders opinion; and b) Incorporating this prior in both two-arm and three-arm NI trial design. In a three-arm NI trial when it is ethically acceptable a placebo arm is added along with experimental and active comparator. While this ease out many stringent assumptions of two-arm NI trial, it has its own challenges. The outline of the proposal is following.
Initially we will develop prior elicitation both from quantitative past clinical trial data and patients/stakeholders/clinicians opinion about different aspects of treatment choices and disease through a questionnaire. We will develop formal statistical method for combining heterogeneous priors from multiple sources which will be then used in the NI trial design via meta-analysis framework. Next we will develop Bayesian hypothesis testing methods for two-arm and three-arm NI trials when the outcomes are categorical in nature. Current PCORI-PFA makes special emphasis on the uses of Bayesian methods for this type of categorical “outcome scale for THE analysis (e.g., risk difference, risk ratio, log of odds-ratio)” (pg. 3). Classical testing methods for NI trials can be broadly divided into two approaches, a) fraction margin and b) fixed margin based approach. We intend to develop Bayesian methods to cover both. We would like to perform extensive simulation study to benchmark performance of the developed methods.
Finally we will perform an exploratory sensitivity analysis to understand the effect of different priors on NI trial design. We would like to note our existing work on Bayesian models for NI testing with continuous outcome shows promising performance when compared to its frequentist counterpart. Finally, all the approach will be implemented and evaluated with free statistical software (Rpackage 21) rendering greater accessibility to the clinical research community interested in performing Bayesian CER. The software will be made available, at no cost, and clinicians will be educated via Continuing Medical Education (CME) courses.