Which patient groups benefit from antihypertensive medications, statins, and low-dose aspirin for the primary prevention of atherosclerotic cardiovascular disease: A regression discontinuity approach in large electronic health record data

• Applicant: Professor Dr. Till Bärnighausen, Ph.D.
• Subject Area: Public Health, Healthcare Research, Social and Occupational Medicine
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 504369372

Randomized controlled trials have shown that antihypertensive treatment, statins, and low-dose aspirin can be effective for the primary prevention of atherosclerotic cardiovascular disease. Despite their existence for decades, there are still many unanswered questions on their optimal use, their effectiveness for large parts of the population, and their side effects.This project will test the feasibility and validity of the regression discontinuity design to estimate the causal effect of these treatments on mortality and the incidence of atherosclerotic cardiovascular disease using electronic health record (EHR) data. Regression discontinuity is a novel study design in clinical research that exploits exogenous variation in the treatment probability induced by threshold rules that exist in clinical guidelines to estimate the effect of treatment on outcomes. Intuitively, if the only reason why being just above or just below the threshold matters is that this circumstance determines if a patient is treated or not, then the impact of treatment will be the difference in outcomes between observations just above and just below the threshold.The enormous size of EHR data would allow for two innovations that have the potential for regression discontinuity to become a crucial tool in the movement towards more individually tailored cardiovascular treatment. First, regression discontinuity could not only be used to establish the average causal effect of antihypertensive treatment, statin therapy, and low-dose aspirin therapy in a “real-life” patient population but, in combination with machine learning, it also establishes a way to rigorously examine how treatment effects vary within these populations by a large array of patient characteristics. Specifically, EHR data would allow for sufficient statistical power to examine highly granular patient subgroups consisting of all permutations of important patient characteristics, such as combinations of various co- and multi-morbidities, body weight, age groups, and ethnicities. Second, employing rigorous methods for multiple hypothesis testing, regression discontinuity in EHR data can be used to causally test for a vast number of side effects – whether therapeutic or adverse – of medications and other treatments. For our analyses, we will use the Aurum and Gold dataset of the Clinical Practice Research Datalink (CPRD) data, which is large-scale EHR data from the UK. We will repeat all analyses using the Optum Clinformatics® Data Mart (CDM) version 8, containing data from the US, as well as Danish nation-wide EHR data.

• DFG Programme: Research Grants
• International Connection: Denmark, United Kingdom
• Cooperation Partners: Professorin Dr. Justine Davies; Dr. Anant Jani; Professor Dr. Laust Mortensen; Dr. Simon Sawhney