Development and validation of a 10-metabolite prognostic model for 4-year all-cause mortality in heart failure with preserved ejection fraction in the Chinese Han population

For patients with heart failure with preserved ejection fraction (HFpEF), accurate predicting the risk of death remains an unmet clinical need. We sought to develop a metabolite-based prognostic model for risk stratification and use transcriptomic analysis to provide biological insights into HFpEF prognosis. We characterized the plasma metabolomic profile of 500 hospitalized Chinese Han patients with HFpEF from a national multicenter prospective cohort. The cohort was randomly divided into non-overlapping training and validation sets at a 1:1 ratio. In the training set, candidate metabolites were first identified by logistic regression, then further refined using the LASSO regression followed by the Boruta algorithm and correlation analysis. The model was validated in a separate, non-overlapping validation cohort together with bootstrap analysis. Meanwhile, weighted gene co-expression network analysis (WGCNA) was performed using mRNA microarray data from 291 HFpEF patients from the same cohort to identify key gene modules and their associated biological processes. Integrative metabolomic and transcriptomic enrichment analyses were performed to explore biological pathways associated with HFpEF prognosis. Ten metabolites were selected to establish a risk stratification model (10-metabolite model). In the validation set, the 10-metabolite model demonstrated good discrimination for predicting 4-year all-cause mortality, with an area under the curve (AUC) of 0.878 (95% confidence interval (CI) 0.835–0.921), and showed better discrimination than the clinical risk score (AUC 0.714, 95% CI 0.649–0.778). The 10-metabolite model significantly improved risk classification, with net reclassification improvement (NRI) of 0.585 (95% CI 0.469–0.701; P < 0.001) and integrated discrimination improvement (IDI) of 0.413 (95% CI 0.344–0.482; P < 0.001). Compared with individuals in the lowest tertile, those in the highest and medium tertiles of the metabolite score had a 23-fold (hazard ratio (HR) = 22.95, 95% CI 11.07–57.54) and 2.8-fold (HR = 2.78, 95% CI 1.32–5.86) increased risk of mortality respectively. Multi-omics enrichment analysis revealed the metabolic patterns, such as glyceropholipid metabolism, tyrosine metabolism, the AMPK signaling pathway and TCA cycle, which may contribute to HFpEF prognosis. We developed a 10-metabolite model that demonstrated supportive discrimination for predicting 4-year all-cause death in hospitalized Chinese Han patients with HFpEF. Transcriptomic analyses provided complementary biological insights into the metabolic pathways associated with HFpEF prognosis. The model is promising but requires further validation in other populations before being used in routine clinical practice.