Pathophysiological insights into machine learning-based subphenotypes of acute heart failure with preserved ejection fraction

Abstract

ObjectiveThe heterogeneous pathophysiology of the diverse heart failure with preserved ejection fraction (HFpEF) phenotypes needs to be examined. We aim to assess differences in the biomarkers among the phenotypes of HFpEF and investigate its multifactorial pathophysiology.MethodsThis study is a retrospective analysis of the PURSUIT-HFpEF Study (N=1231), an ongoing, prospective, multicentre observational study of acute decompensated HFpEF. In this registry, there is a predefined subcohort in which we perform multibiomarker tests (N=212). We applied the previously established machine learning-based clustering model to the subcohort with biomarker measurements to classify them into four phenotypes: phenotype 1 (n=69), phenotype 2 (n=49), phenotype 3 (n=41) and phenotype 4 (n=53). Biomarker characteristics in each phenotype were evaluated.ResultsPhenotype 1 presented the lowest value of N-terminal pro-brain natriuretic peptide (NT-proBNP), high-sensitive C reactive protein, tumour necrosis factor-α, growth differentiation factor (GDF)-15, troponin T and cystatin C, whereas phenotype 2, which is characterised by hypertension and cardiac hypertrophy, showed the highest value of these markers. Phenotype 3 showed the second highest value of GDF-15 and cystatin C. Phenotype 4 presented a low NT-proBNP value and a relatively high GDF-15.ConclusionsDistinctive characteristics of biomarkers in HFpEF phenotypes would indicate differential underlying mechanisms to be elucidated. The contribution of inflammation to the pathogenesis varied considerably among different HFpEF phenotypes. Systemic inflammation substantially contributes to the pathophysiology of the classic HFpEF phenotype with cardiac hypertrophy.Trial registration numberUMIN-CTR ID: UMIN000021831.