Filling Two Major Gaps in the Analysis of Heterogeneity of Treatment Effects for Patient-Centered Outcomes Research
Ravi Varadhan, PhD
Johns Hopkins University
Accelerating PCOR and Methodological Research
Given the same treatment, some people receive benefit, some experience harm, and others are unaffected. This project studies heterogeneity of treatment effect (HTE), which means to study how the same treatment can have different effects on different people. This issue is very important to making personal healthcare decisions, but there are major challenges to knowing when the differences in treatment effects are real. The approach used most often, subgroup analysis, can produce misleading findings.
The Bayesian approach to study HTE can reduce the chance of misleading findings by allowing the use of prior information and prior beliefs of different stakeholders about the likelihood of treatment effect varying across people. However, the Bayesian approach has been underused in HTE analysis for three main reasons: lack of understanding about the role and impact of prior information and belief, lack of guidance on how to understand the prior distribution, and lack of easy-to-use software to conduct analysis.
Another basic problem in the analysis of HTE is that it depends on the outcome scale used to study treatment effect (e.g., ratios or differences). It is not clear which outcome scale is better or whether it is useful to study HTE in terms of all scales.
The Patient-Centered Outcomes Research Institute’s (PCORI) Methodology Committee, in its recent report, identified these two issues, underutilization of Bayesian approach and choice of effect scale, as having a high priority for patient-centered outcomes research.
In this proposal, we aim to fill these two critical gaps in the analysis of HTE.
1. To encourage Bayesian analysis of HTE
a. To develop recommendations on how to study HTE using Bayesian statistical models
b. To develop a user-friendly, free, validated software for Bayesian methods for HTE analysis
2. To develop recommendations about the choice of treatment effect scale for the assessment of HTE in PCOR
The main products of this study will be: (i) recommendations or guidance on how to do Bayesian analysis of HTE in PCOR, (ii) software to do the Bayesian methods, (iii) recommendations or guidance on choosing appropriate treatment effect scale for HTE analysis in PCOR, and (iv) demonstration of our products using data from large comparative effectiveness trials. These products will influence current practice and lead to meaningful improvement in patient health, well-being, or quality of care.