Presentation (CTI)

OBJECTIVES: Risk scores derived from the Framingham Heart Study identify patients at elevated four-year risk for congestive heart failure (CHF), but do not explain why many high-risk individuals do not progress to CHF. This study aimed to (1) identify demographic, clinical, and care-delivery factors that differentiate high-risk patients who did versus did not develop CHF within four years, and (2) develop and validate a machine learning (ML) model to further stratify CHF risk within this high-risk population.
METHODS: A retrospective cohort study used real-world data from individuals aged 45+ with coronary artery disease, hypertension, or valvular disease. Inclusion required ≥1 year baseline data (heart rate, blood pressure, height, weight, BMI) and ≥4 years follow-up. High-risk status, defined by Framingham CHF risk factors, was used to stratify patients by CHF diagnosis within four years. Comparative analyses assessed demographics, comorbidities, procedures, medication, and specialist use. Supervised ML models, trained with cross-validation, predicted CHF onset in high-risk patients. Feature attribution methods identified key risk drivers.
RESULTS: Among the highest-risk patients, those who developed CHF had higher cardiometabolic and renal comorbidity burden, greater baseline physiologic instability, and increased acute care use. Patients who did not progress to CHF were more likely to receive sustained antihypertensive and cardioprotective drugs, show slower comorbidity progression, and engage earlier with cardiology specialists. Machine learning (ML) models significantly improved the discrimination of CHF onset within the high-risk cohort. Feature analysis consistently identified comorbidity burden, long-term treatment exposure, and specialist care as key differentiators of CHF progression.
CONCLUSIONS: Within populations already classified as high risk, machine learning applied to real-world data can identify clinically meaningful heterogeneity in the progression of CHF. These findings suggest that modifiable treatment and care-delivery factors may mitigate CHF onset despite elevated baseline risk, supporting more targeted prevention strategies.