Pan Submits PCORI Methods Application

Kudos to Wei Pan and his entire team for the submission of their PCORI Methods application entitled "Leveraging Electronic Health Records to Better Understand Sensitivity and Robustness of Comparative Effectiveness Research to Unmeasured Confounders." This proposal requests funding for a three-year period with a start date of June 1, 2019. 

The ability to control unmeasured factors has been a long-standing issue in observational studies, including most of comparative effectiveness research (CER) studies on patient-centered outcomes. The concerns about unmeasured factors often lead to researchers' doubt about their CER findings, which can create healthcare providers' uncertainty about medical procedures or treatments they recommend to their patients, and in turn, patients' skepticism about the effectiveness of treatments they receive. The current sensitivity analyses of unmeasured factors, unfortunately, fail to fully control this issue because the techniques in those sensitivity analyses mainly rely on a hypothetical range of impact of unmeasured factors and it is often difficult to translate the subjectively specified impact to clinical context. In other words, the assessment of the degree of control for unmeasured factors is based on unknown information and, thus, may unnecessarily increase the uncertainty about CER findings. In contrast, the robustness to unmeasured factors can use known information from large observational datasets, including those derived from electronic health records (EHRs) in order to assess the ability to resist the impact of unmeasured factors. Methods using robustness are less mature, but have the possibility to increase the confidence in, as opposed to uncertainty about, CER findings.

Addressing this methodological gap will respond to the Cycle 2 2018 Methods PFA's priority of "Methods to improve the design and conduct of PCOR/CER studies in circumstances limiting the use of RCTs" in the Methods to Improve Study Design Research Area of Interest. The overarching goal of this proposed methodological research is to increase healthcare providers' confidence when recommending treatments to patients and, in turn, strengthen patients' belief in the effectiveness of treatments they receive. Specifically, we will (1) develop a statistical value of robustness to unmeasured factors (R-value) using rich clinical information stored in large observational data such as EHRs; and compile a statistical tool for computing R-values and disseminate the tool through public sources; (2) verify the efficacy of R-values and set a benchmark for interpreting R-values under various data conditions through a simulation study; and (3) demonstrate the clinical applicability and interpretability of R-values through a real-world CER using EHRs on treatment effects on mortality and rehospitalization of patients with heart failure.

The expected outcomes of this proposed research will be the development and dissemination of a ready-to-use statistical tool for researchers to assess R-values for their CER studies. This tool will enable them to better understand the robustness of their CER study results to unmeasured factors based on known information, or whether their CER is robust enough to unmeasured factors, by using clinically rich EHRs.