|
Abstract:
|
Clinical research has failed to routinely translate evidence based practices into clinical care. The field of Dissemination and Implementation science (D&I) has emerged to study methods that increase the systematic uptake of research findings and bridge the research-practice gap. D&I science needs solutions for pragmatic trial designs, and mechanistic models that help evaluate questions on why does a treatment work and for whom does it work. Probabilistic programming languages (PPLs) allow for the encoding of probability models with the ability to go through an iterative process of increasing complexity. They offer a unified framework for accounting for uncertainty and utilize state of the art methods for inference. In this presentation three increasingly complex examples that apply JAGS and Stan (two popular Bayesian PPLs) are used. First, power is estimated using Monte Carlo simulation for a stepped wedge clinical trial. Second, the model is extended with a mediator and the direct and indirect effects of a hypothetical intervention are estimated. Third, the model is extended to handle interactions which can be estimated on multiple scales (additive and multiplicative).
|
