Document resource
Background Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous and frequently under recognised syndrome, complicating timely diagnosis and intervention. Previous clustering studies have been predominantly trial-based, limiting their real-world applicability. This study aimed to identify and validate HFpEF phenogroups through cluster analysis of routinely collected electronic health record (EHR) data and investigate their clinical trajectories and outcomes.Methods A retrospective cohort study was conducted using EHR data from two UK centres. HFpEF patients were identified per European Society of Cardiology (ESC) criteria (left ventricular ejection fraction ≥ 50% and evidence of structural or functional cardiac abnormalities) using a validated natural language processing (NLP) pipeline. Unsupervised clustering was performed with latent class analysis (LCA) on ten clinical features, and the resulting phenogroups were validated in an external cohort. All-cause mortality was analysed, and a Random Forest model trained on baseline clustering variables mapped longitudinal transitions between phenogroups by incorporating newly identified comorbidities.Results Among 2,223 patients (median age 75, 60% female), 89.6% met ESC criteria but lacked a clinician-documented diagnosis. LCA identified four phenogroups: (1) Elderly-Atrial Dysfunction [N=703 (32%)]; (2) Cardio-Renal-Metabolic [N=530 (24%)]; (3) Obesity-Predominant [N=530 (32%)]; and (4) Young-Low Comorbidity [N=487 (22%)] ( table 1). The Young-Low Comorbidity group had the lowest clinician-documented diagnosis rate (3%) and lowest estimated five-year mortality (26%) (figure 2). In contrast, the Cardio-Renal-Metabolic and Elderly-Atrial Dysfunction phenogroups had significantly higher adjusted mortality risks (HR: 1.49; 95% CI: 1.18–1.87; P < 0.03 and HR: 1.30; 95% CI: 1.02–1.67; P < 0.001, respectively). Over a median 4-year follow-up, 53% of the Young-Low Comorbidity group progressed to higher-risk phenogroups, while the Elderly-Atrial Dysfunction and Cardio-Renal-Metabolic groups demonstrated high stability (figure 2). External validation (N=3,349) confirmed phenogroup reproducibility.Conclusions NLP-driven phenotyping identified four clinically distinct and prognostically relevant HFpEF phenogroups. Our findings highlight the lack of clinician recognition and rapid progression in early-stage HFpEF, reinforcing the urgent need for improved recognition and early intervention. Targeted therapies may offer significant benefit for this group, warranting further investigation.Abstract 5-007 Table 1Baseline characteristics Elderly-Atrial Dysfunction, 32% (n = 703) Cardio-Renal-Metabolic, 23% (n = 503) Obesity Predominant, 24% (n = 530) Young Low Comorbidity, 22% (n= 487) Age, years 84 (81, 88) 74 (65, 81) 70 (61, 76) 60 (48, 71) Female 482 (69%) 253 (50%) 339 (64%) 257 (53%) Obesity 58 (8%) 321 (62%) 391 (74%) 219 (45%) Hypertension 487 (69%) 480 (95%) 498 (94%) 14 (3%) Atrial Fibrillation 264 (38%) 138 (27%) 101 (19%) 17 (4%) Kidney Disease 190 (27%) 471 (92%) 27 (5%) 52 (11%) Diabetes 62 (9%) 333 (66%) 176 (33%) 11 (2%) NT-proBNP, pg/ml 3,562 (2,128, 5,349) 3,824 (1,506, 7,298) 987 (404, 1,853) 1,775 (971, 3,134) LVEF (%) 59.6 (58.0, 62.0) 60.9 (58.9, 63.0) 60.4 (58.1, 62.7) 59.7 (57.9, 62.3) Median (IQR) or n (%)Abbreviations: LVEF, left ventricular ejection fraction; LA, left atrium; TR, tricuspid regurgitation; LVEDD, left ventricular end-diastolic diameter.Abstract 5-007 Figure 1Kaplan-Meier Survival Curves for hfpef Phenogroups. (A) Kaplan-Meier survival curves illustrating five-year survival estimates stratified by hfpef phenogroup. Log-rank tests were used to assess significant differences in survival across phenogroups. Phenogroups are as follows: Elderly-atrial dysfunction (ad) (blue), cardio-renal-metabolic (orange), obesity-predominant (green), and young-low comorbidity (red)Abstract 5-007 Figure 2Phenogroup trajectories. Alluvial diagram demonstrating the movement of patients between phenogroups at baseline and follow-up, including ‘Death’ as a separate endpoint category. The width of the flows (arrows) is proportional to the number of patients transitioning between clusters