Abstract

Introduction: H2FPEF score of 0-1 and ≥6 indicate low and high risks of HFpEF, while scores of 2-5 may require additional testing. We compared the prognostic abilities of unsupervised cluster modeling (USCM) and the conventional H2FPEF score. Methods: Respectively, 532 patients with suspected heart failure (HF) symptoms were included (69±6 years, 57% females, EF: 61±5%, all>50%). H2FPEF score was calculated as recommended: BMI (>30: 2 points), atrial fibrillation (3 points), and 1 point each for age>60, ≥2 hypertension drugs, E/e' ratio >9, and pulmonary systolic pressure >35 mmHg. Using these variables, a 2-step USCM was done and compared to the H2FPEF score for death and cardiac hospitalization. Results: The mean H2FPEF score was 3.2±1.5. During a median follow-up of 3.8 years, 13 patients died, 66 hospitalized (35 HF and 51 non-HF causes), and 76 had the composite outcome. H2FPEF scores of 0-1 and ≥6 had the lowest and highest frequencies of individual and combined outcomes, respectively, while scores of 2-5 was intermediate (Figure 1A). Compared to scores of 0-1, scores of 2-5 (HR: 11, 95% CI: 1.5-78) and ≥6 (HR: 29, 95% CI: 4-219) had a higher risk of outcomes. USCM (Figure 1B) identified three clusters with varying risks: (low: 145 patients, intermediate: 333 patients, and high: 54 patients) (Figure 1C). The mean H2FPEF scores differed significantly between clusters (1±0, 3.3±1.1, 6.6±1, p<0.001). Compared to the low-risk cluster, the intermediate-risk (HR: 3.4, 95% CI: 1.5-8), and high-risk clusters (HR: 7.3, 95% CI: 3-18) had a higher risk of outcomes. Clusters significantly reclassified H2FPEF scores (Figure 1D), particularly those with scores of 2-5 (23% reclassified as low risk, 7% as high risk). Among the reclassified patients with H2FPEF scores of 2-5, there was significant discrimination in outcomes (p=0.02, Figure 1A). Conclusions: Machine learning techniques, can enhance the prognostic accuracy of the H2PEF score, particularly for intermediate scores.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call