Outcome trajectories in hospitalized heart failure with preserved ejection fraction: a machine learning cluster analysis

Background Intensive follow-up post-hospitalization for heart failure (HHF) is recommended, but difficult to pursue. Risk stratification of HHF with preserved ejection fraction (HFpEF) might help to improve resource allocation. However, it remains elusive and no study has applied cluster analysis in the acute setting. Methods Consecutive patients with HHF and left ventricular EF >40% were enrolled and evaluated at discharge. A composite endpoint of all-cause death, urgent heart transplant, HF hospitalization, or emergency department visit for decompensated HF was assessed at 12 months. Cluster analysis was performed using prespecified variables, while Cox regression and CART analysis identified predictors of adverse outcomes. Results 1052 HF patients were screened. After excluding HF with EF below 40%, 471 patients (median age 78 years, 44% women) were included. Among them, three clusters were identified. Cluster 1 comprised younger patients with de novo HF, fewer comorbidities, preserved renal function, and lower BNP (B-type natriuretic peptides) and NLR (neutrophil-lymphocyte ratio) levels. Cluster 2 consisted mainly of elderly women with hypertension and atrial fibrillation. Cluster 3 included older patients with worsening HF, higher NYHA class, renal dysfunction, anemia, and elevated BNP and NLR. Compared with Cluster 1, risk was nearly threefold higher in Cluster 2 (HR 2.9, 95% CI 1.7–5.2, p<0.001) and fivefold higher in Cluster 3 (HR 4.8, 95% CI 2.7–8.4, p<0.001). Outcome results were consistent in the vulnerable period (3 months). NYHA class and NLR emerged as key prognostic nodes. Conclusion Cluster analysis identified low, intermediate, and high-risk acute HFpEF phenotypes. These data might support personalized management strategies in hospitalized HFpEF.