Metabolomic profile for understanding heart failure classifications
Maria V. Kozhevnikova,Yuri N. Belenkov,Ksenia M. Shestakova,Anton A. Ageev,Pavel A. Markin,Anastasiia V. Kakotkina,Ekaterina O. Korobkova,Natalia E. Moskaleva,Ivan V. Kuznetsov,Natalia V. Khabarova,Alexey V. Kukharenko,Svetlana A. Appolonova
DOI: https://doi.org/10.1101/2024.11.13.623517
2024-11-14
Abstract:Background: The 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. Methods: MP 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. Results: The 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, model's 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 <50% showed good precision. The HFrEF vs. 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. Conclusions: Our 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.
Biochemistry