Researchers applied unsupervised machine learning to deep patient data to classify heart failure with preserved ejection fraction into two biologically distinct subgroups, according to findings published in Nature Communications. The cardiac condition accounts for more than half of all heart failure cases, though clinical heterogeneity has historically limited clear biological definitions.
The study evaluated 155 participants enrolled in the prospective PACIFIC-Preserved cohort, cataloged under clinical identifier NCT04189029. Investigators combined clinical metrics, imaging data, and high-throughput proteomics through multiview clustering algorithms. The analysis identified two primary phenogroups: PEF1, which presented with inflammatory signaling, TNF receptor activation, cardio-skeletal injury, and increased renal dysfunction, and PEF2, marked by endothelial stress and less severe biological perturbations.
Clinicians could not distinguish the two patient categories through standard diagnostic assessments alone. To resolve the subgroups, the researchers built a multimodal classifier integrating clinical measurements, imaging markers, and proteomic variables, recording an F1 accuracy score of 0.87. External validation conducted across 456 patients in the MEDIA-DHF cohort linked the PEF1 subgroup to adverse clinical outcomes, recording an adjusted hazard ratio of 2.27 with a 95 percent confidence interval between 1.04 and 4.94.
Assistance Publique - Hôpitaux de Paris served as the official sponsor for the PACIFIC-Preserved clinical trial through its clinical research division. Public funding came from the French state investment entity BPI under grant number BPI2018-PSPC-07, with institutional grant-in-aid support contributed by drugmakers Sanofi and Servier. Study co-authors included Jean-Sébastien Hulot, Pierre-Yves Hervé, and Nicolas Girerd, alongside industry researchers from Sanofi R&D, Servier R&D, and technology company Fealinx.
