Abstract 4347325: Plasma Metabolomics and Machine Learning Identify Causal Metabolic Contributors to Incident Heart Failure in 38,628 MGB Biobank Participants

Abstract

Background: The growing prevalence of cardiometabolic risk factors creates a need for deeper understanding of the metabolic pathways that promote cardiac dysfunction and heart failure (HF) to identify novel biomarkers and therapeutic targets. Objective: We sought to identify potentially causal metabolic contributors to the development of HF in the MGB Biobank, which has metabolomics data and a validated machine learning- and natural language processing-based algorithm for HF case ascertainment. Methods: We included MGB Biobank participants with metabolomics data and no history of HF. The primary exposures were plasma levels of 43 metabolites measured using a commercial H 1 nuclear magnetic resonance platform. The primary outcome was incident HF, ascertained by a validated machine learning- and natural language processing-based algorithm (area-under-the-curve: 0.92; positive predictive value: 0.90). We used multivariable Cox proportional hazards regression to quantify the associations between a 1-SD difference in metabolite level and the time to incident HF. We tested the potential causality of each metabolite-heart failure association using three different Mendelian randomization methods and genome-wide association summary statistics from the UK Biobank (metabolomics) and HERMES consortium (HF). Results: The final analytical cohort included 38,628 individuals (mean age: 63 years; 56% women). A total of 375 incident heart failure events occurred over a median follow-up of 7 (5-9) years (1.5 [1.4-1.7] events per 1,000 person-years). Higher plasma levels of glutamine and phenylalanine associated with higher risk of incident HF while higher levels of total phospholipids, cholines, phosphoglycerides, phosphatidylcholines and docosohexaenoic acid were associated with lower HF risk. Higher levels of polyunsaturated fatty acids (driven by linoleic acid, an omega-6 fatty acid) and sphingomyelins were associated with lower incident HF risk only in the subgroup with diabetes (P-interaction < 0.05 for each). Mendelian randomization analysis provided evidence of causal relationships in all three models for total polyunsaturated fatty acids, total omega-3 fatty acids, docosahexaenoic acid and linoleic acid with incident HF in the general population. Conclusions: Dysregulated fatty acid metabolism, primarily docosohexaenoic acid and linoleic acid, is causally associated with a higher risk of incident heart failure.