Metabolomic profile for understanding heart failure classifications

BackgroundThe existing classifications of heart failure (HF) remain a topic of debate within the cardiology community. Metabolomic profiling (MP) offers a means to define HF phenotypes based on pathophysiological changes, allowing for more precise characterization of patient groups with similar clinical profiles. MP may thus aid in refining HF classifications and offer a novel approach to phenotyping. MethodsMP was performed to 408 patients with different stages of HF. Patients with symptomatic HF were divided into phenotypes by left ventricle ejection fraction (LVEF). Liquid chromatography combined with mass-spectrometry were used for the MP. Data were analyzed using machine learning. The relationship between the incidence of all-cause death and LVEF trajectory changes and metabolomic clusters was evaluated. Follow-up period was 542 days [16;1271] in average. ResultsThe classification model achieved an AUC ROC - 0.91 for distinguishing of Stage A from Stage B and an AUC ROC - 0,97 for Stage B vs. Stage C using metabolomic analysis, models performance for differentiating Stages C and D was lower (AUC ROC 0.81). For HF phenotypes, the HFrEF, HFmrEF, and HFpEF model demonstrated moderate accuracy (AUC ROC 0.74), whereas the model distinguishing HFpEF from HF with EF 40% model, however, displayed low accuracy. Biostatistical processing of MP identified four metabolomic clusters, and 26 metabolites demonstrated the greatest significance (metabolites of the kynurenine and serotonin pathways of tryptophan catabolism, glutamine, riboflavin, norepinephrine, serine, long- and medium-chain acylcarnitines). Patients with reduced LVEF had the poorest prognosis (HR 1,896; 0,711-5,059), with an LVEF decrease linked to a threefold rise in all-cause mortality risk. Cluster 3 was associated with a 2,880-fold increase in all-cause mortality. ConclusionsOur findings suggest that MP provides an effective alternative approach for stratifying HF patients by stage. The observed metabolic similarities between HFpEF and HFrEF phenotypes highlight limitations in the current classification, underscoring the need to refine HF phenotyping into two primary categories. Hierarchical clustering by metabolomic profile produced a high-accuracy model, supporting MP as a valuable tool for HF classification. Novelty and SignificanceO_ST_ABSWhat Is Known?C_ST_ABSO_LIClassifying HF by stages offers the advantage of incorporating preventive aspects, however, diagnosing Stage B can be challenging due to the necessity of analyzing numerous parameters for verification. C_LIO_LIThe classification by LVEF, particularly distinguishing HFmrEF, remains controversial because of limited evidence for specific treatment strategies. C_LIO_LIMetabolomic profiling (MP) offers a means to identify unique pathophysiological changes across phenotypes, presenting a promising approach for diagnosing HF and developing targeted therapies. C_LI What New Information Does This Article Contribute?O_LIThis study demonstrates high accuracy in HF stage classification by integrating chromatography-mass spectrometry data with bioinformatic analysis through multiparametric machine learning (ML) models. C_LIO_LISimilarities between metabolomic profiles of patients with LVEF View larger version (65K): org.highwire.dtl.DTLVardef@173f141org.highwire.dtl.DTLVardef@e41c97org.highwire.dtl.DTLVardef@11386eorg.highwire.dtl.DTLVardef@8fee01_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOGraphical abstractC_FLOATNO EF - ejection fraction; HF - heart failure; HFmrEF - heart failure with mid-range ejection fraction; HFpEF - heart failure with preserved ejection fraction; HFrEF - heart failure with reduced ejection fraction. C_FIG