Effectiveness of Therapies for Heart Failure with Mid-Range Ejection Fraction

Project Summary / Abstract Evidence from trials has established that 4 separate classes of guideline-directed medical therapies (GDMT) for heart failure with reduced ejection fraction (HFrEF) together reduce mortality by nearly two-thirds. Conversely, for heart failure with mid-range ejection fraction (HFmrEF), trials have produced only weaker

evidence, resulting in ambiguous guidelines. Our long-term goal is to build the largest echocardiography registry in the world to generate comparative effectiveness evidence for questions for specific populations in heart failure where trials have not been able to provide strong evidence. The overall objectives in this

application are to build the registry, test mathematical assumptions required for the application of strong causal inference methods developed in economics, and then apply the strongest methods possible to measure treatment effects for GDMT in HFmrEF. The central hypothesis is that “real-world” treatment decisions will

differ at guideline-suggested thresholds of left ventricular ejection fraction (LVEF), and patients on either side of those immediate LVEF thresholds will be otherwise similar (“as good as random”), allowing the application of regression discontinuity methods to measure treatment effects. We believe these methods will demonstrate

reduced mortality with GDMT for patients with HFmrEF. The rationale for this proposal is that (1) LVEF cutoffs in guidelines are semi-arbitrary in the sense that LVEF exists on a physiological continuum but guideline cutoffs are based on strict thresholds, and (2) our preliminary data demonstrates expected discontinuities in

treatment frequency at influential LVEF thresholds for other HF therapies (defibrillators) in real-world practice. As such, the conditions for “as good as random” likely exist in small ranges around relevant LVEF thresholds, allowing the application of regression discontinuity methods. Our central hypothesis will be tested with 2

specific aims. In Aim #1, we will use natural language processing to aggregate data from echocardiography reports from 11 hospitals creating the largest echocardiography registry in the world, allowing clinical adjudication of both clinical information and echocardiographic images. We will then systematically test

mathematical assumptions required for regression discontinuity. In Aim #2, we will create robust estimates of each of the effects of each GDMT medication class on mortality in HFmrEF with fuzzy RDD and alternative methods, including propensity score methods. The feasibility of propensity score methods does not depend on

the Aim 1 analyses, so at least 1 of the 2 proposed methods will be feasible. We therefore expect that these results will upgrade evidence from the current class 2b (“usefulness is unknown”) in heart failure guidelines for 3 of the 4 GDMT medication classes. We think the proposed research is innovative because it combines

strong causal inference methods to observational data with the adjudication standards of trials to fill in unanswered gaps in evidence. The research proposed here is significant because it will upgrade the current ambiguous treatment guidelines for HFmrEF in ways that will reduce mortality for this growing population.