|
Abstract:
|
Statistical analysis and inference form a bedrock for providing sound evidence in advancing medicine. Although the P-value based frequentist inference has been popular, it has notable limitations in synthesizing evidence. In contrast, Bayesian approach provides a natural and consistent framework to incorporate additional information, both external and internal to the trial, into the calculation of the posterior probability, e.g., using the power prior or the hierarchical commensurate prior. Bayesian network analysis offers a useful tool for evaluating both direct and indirect effects in comparing treatments. Bayesian hierarchical model allows borrowing information across subgroups to increase the study efficiency. Bayes factor gives an appealing alternative to P-value in testing hypothesis. Applying the decision theoretical approach, informed decision can be made to address difficult questions such as the efficacy/toxicity tradeoff and the cost/benefit analysis in public health policy making. Key steps in prior elicitation, utility formulation, and sensitivity analysis will be discussed. Bayesian approach sheds a new light and offers a promising alternative in drug development.
|
