Integrating Causal Inference, Evidence Synthesis, and Research Prioritization Methods
John B. Wong, MD
Tufts Medical Center
Accelerating PCOR and Methodological Research
This proposal seeks to advance methods for doing comparative effectiveness research (CER) by combining causal inference (how we decide if something is beneficial or not), meta-analysis (how we combine evidence), and research prioritization (how we should decide what research to do) in the evaluation of outcomes important to patients. We examine the quality and degree of relevance of evidence that is based on observational studies (OBS) and/or randomized, controlled trials (RCTs) considering varying levels of evidence detail (published summaries of a study or having de-identified individual patient data). Randomized trials provide the most unbiased estimates of therapeutic benefit but are expensive and time-consuming. Observational studies, in contrast, can be more contemporary and estimate benefit occurring in routine “real-world” practice, but may be biased because physicians (or patients) choose particular treatments (selection bias) or because of particular patient or physician characteristics affecting treatment selection and outcome (confounding). To establish the benefits and harms of therapeutic interventions based on OBS and/or RCTs with varying levels of evidence detail, we seek to inform decision making and outcomes (survival) that matter to people, highlighting comparisons (in this case, medical therapy or coronary revascularization for coronary heart disease). Our long-term objective is to evaluate and enhance current methodologies to improve CER by (1) examining the differences in patient populations and results of RCTs and OBS analyses, (2) developing and expanding ways of combining evidence to integrate OBS and RCT data, and (3) assessing the value of reducing uncertainty in the current state of knowledge as a foundation for improved and efficient CER. We aim to answer the following questions: How do patient populations in RCT and OBS studies differ; does the estimated treatment efficacy differ among RCT and OBS populations; would various OBS-based prediction models yield the results observed in different populations? Can our statistical approaches anticipate and resolve differences between RCT and OBS analyses, and can they be used to explore subgroups of patients or individuals with a unique set of conditions? We will use individual patient-level data from RCTs and from a detailed longitudinal clinical OBS dataset to extend methodologies for (1) comparing and understanding OBS and RCT results by using state-of-the-art statistical methods, (2) combining different sources of evidence at different levels of detail, and (3) assessing the “value of information” to determine when gathering more information through additional research would be an efficient use of societal resource. We expect these analyses to result in improved methods for conducting CER that will enhance the validity and credibility of CER.