Serum Metabolites and Heart Failure Risk: MESA and RS

Background The role of circulating metabolites in heart failure (HF) mechanisms and their clinical utility remain unclear. We aimed to examine the associations between serum metabolites and incident HF and assess their performance in improving HF risk prediction. Methods A total of 23 571 serum metabolomic spectral variables were measured using untargeted proton nuclear magnetic resonance spectroscopy. Participants from MESA (Multi‐Ethnic Study of Atherosclerosis; discovery cohort) and RS (Rotterdam Study; replication cohort) with metabolomic data and without prevalent HF at baseline were included. Cause‐specific proportional hazards models were used to estimate the associations between metabolomic features and incident HF, with false discovery rate–adjusted P values for multiple comparisons. We further assessed how adjusting for cardiovascular risk factors influenced these associations. Correlations among metabolites were tested. Discriminative performance of risk prediction was evaluated using Harrell’s C‐statistic. Results We included 3942 MESA and 1515 RS participants, with a median of 15 years follow‐up. In MESA, 15 metabolites were significantly associated with incident HF (false discovery rate–adjusted P <0.01), with 5 replicating in RS (false discovery rate–adjusted P <0.05). However, after further adjusting for diabetes or hypertension, observed associations lost statistical significance. Methanol, glucose, proline, and acetoacetate showed significant correlations with fasting glucose (Pearson coefficients, 0.41–0.89). Incorporating the 5 replicated metabolites or all 15 metabolites identified from MESA into PREVENT‐HF (Predicting Risk of Cardiovascular Disease Events–Heart Failure) did not significantly improve C‐statistics in either cohort. Conclusions Fifteen proton nuclear magnetic resonance–measured metabolites were associated with incident HF, but these associations were not independent of diabetes or hypertension. These metabolites provided minimal predictive utility beyond PREVENT‐HF equations.