Kudos to Wei Pan, associate professor, Robert Mentz, of Duke University School of Medicine, and their entire team for the submission of their NIH R01 application entitled: "A Holistic Approach to Handling Unmeasured Confounding Bias in Electronic Health Records Focused on Pharmacological Treatments for Patients with Heart Failure and Preserved Ejection Fraction." This proposal requests funding for a four-year period with a start date of July 1, 2022. 

Heart disease is the leading cause of death in the United States. Approximately 50% of heart failure patients have preserved ejection fraction (HFpEF) where patients’ heart muscle contracts normally by conventional assessment, but the ventricles do not relax sufficiently. The other half of heart failure patients have reduced ejection fraction (HFrEF) where patients’ heart muscle does not contract effectively. Effects of pharmacological treatments (e.g., angiotensin-converting enzyme inhibitors [ACEIs] / angiotensin receptor blockers [ARBs], beta-blockers, mineralocorticoid receptor antagonists [MRAs]) on mortality and rehospitalization for patients with HFrEF have been well documented but less definitive for patients with HFpEF. The inconclusive findings in patients with HFpEF are not only attributed to the heterogeneity of patient population and the complex pathophysiology of HFpEF, but also the inadequacy of methodological approaches that failed to appropriately handle unmeasured confounding bias in the data. 

To this end, we propose a research project with the goal of establishing a holistic approach for handling unmeasured confounding bias in big data such as electronic health records (EHRs) by assessing both sensitivity and robustness and adopting the holistic approach to improve personalized pharmacological treatments for heterogeneous patients with complex HFpEF. 

There are three specific aims: (1) To advance robustness analysis by developing a novel robustness value (R-value) for assessing robustness to unmeasured confounding bias in EHRs and compare the efficacy of R-values with that of five existing representative sensitivity analyses through a Monte Carlo simulation study; (2) To establish and adopt a holistic approach (i.e., robustness analysis with R-values plus the five representative sensitivity analyses) to a real-world example of head-to-head comparisons of effects of pharmacological treatments (e.g., ACEIs/ARBs, beta-blockers, and MRAs) on mortality and rehospitalization for heterogeneous patients with HFpEF in EHRs utilizing propensity score methods for multiple treatments; and (3) To compile a statistical R package for conducting robustness analysis with R-values and explore the feasibility of developing an application programming interface (API) that links between EHRs and R packages for holistically assessing both sensitivity and robustness to unmeasured confounding bias in EHRs. 

The proposed research is both innovative and significant because robustness analysis with the novel R-values makes extensive use of rich information in EHRs, and the holistic assessment of both sensitivity and robustness to unmeasured confounding bias will reduce uncertainty about validity of research findings and, thus, enhance confidence in clinical decision making.