Abstract 4345167: A Contemporary Machine Learning-Based Risk Stratification for Mortality and Hospitalization in Heart Failure with Preserved Ejection Fraction Using Multimodal Real-World Data

Background: Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous condition with high morbidity and mortality. Risk stratification of patients with HFpEF is important for advancing therapeutic development and improving clinical care. Research Question: Predicting overall mortality and heart failure (HF) Hospitalization in real world HFpEF population Aims : to leverage machine learning model to develop prognostic models based on real-world data, with the potential to support risk stratification in routine clinical practice Methods: CONFIDENT is an observational, multi-cohort study across three centers in Europe and the US. Patients with HFpEF, according to the HFA-PEFF criteria with ≥ 2 years of follow-up, were included from 2013 to 2022. The dataset included multimodal data from electronic health records, lab tests, echocardiography, and electrocardiography, with 82 baseline candidate variables. We developed machine learning-based prognostic models to predict all-cause mortality and HF hospitalization. Model performance was compared to conventional risk score in an external validation cohort. Results: A total of 1208 patients were included in the study. The mean age was 72±12. The 2-year risk of HF hospitalization and all-cause mortality ranged from 13 to 44% and 9 to 19% respectively. The all-cause mortality prognostic model achieved good discrimination with a C-index of 0.67 [95%CI, 0.66-0.68], and 0.68 [95%CI 0.66-0.69] in the training cohorts, and 0.72 [95%CI 0.65-0.78] in the validation cohort, and performed better than the PREDICT-HFpEF score (C-index: 0.66 [95%CI 0.65-0.68], p-value =0.012; 0.60, [95%CI 0.59-0.62],p-value < 0.01 and 0.67 [95%CI 0.58-0.73], p-value =0.013 respectively. Similar results were observed when compared to the Meta-Analysis Global Group In Chronic Heart Failure Risk Score (MAGGIC). Similarly, the model derived for HF hospitalization outperformed PREDICT-HFpEF and MAGGIC + natriuretic peptide. Conclusion: CONFIDENT prognostic models for all-cause mortality and HF hospitalization using routinely collected variables can reliably predict outcomes and facilitate personalized care and trial recruitment strategies in HFpEF.