Abstract Body (Do not enter title and authors here): Background: FDA-cleared artificial intelligence algorithms applied to 12-lead ECGs (ECG-AI) offer a scalable, noninvasive approach for early detection of heart failure.

Hypothesis: We hypothesized that ECG-AI models would improve near-term identification of HF risk when compared to a previously validated clinical risk prediction tool.

Methods: We used baseline ECG and clinical data from the Multi-Ethnic Study of Atherosclerosis, Framingham Heart Study, and Cardiovascular Heart Study to estimate the risk of incident heart failure using dichotomous ECG-AI software-as-a-medical devices (SaMDs) designed to identify left ventricular diastolic dysfunction (ECG-AI DD) and low left ventricular ejection fraction (ECG-AI LEF). All subjects with adequate clinical and ECG data to estimate PREVENT-HF and ECG-AI scores were included for analysis. We assessed the utility of ECG-AI for identifying HF cases beyond those identified by the PREVENT-HF score. In this framework, a positive result on ECG-AI is treated as high risk regardless of underlying PREVENT-HF score. The C-statistic was used to assess discrimination of PREVENT-HF and PREVENT-HF plus ECG-AI and net reclassification improvement (NRI) was used to assess improvement in the identification of additional HF over PREVENT-HF for estimated risk at 3- and 10-years at the 10% and 20% thresholds.

Results: A total of 14,250 subjects from these cohorts with available data to generate both ECG-AI and PREVENT-HF scores were pooled for the analysis. The C-statistics for PREVENT-HF for identifying HF incidence in the pooled cohorts over 3- and 10-year timeframes were 0.858 (95%CI 0.841 - 0.875) and 0.881 (0.873 - 0.89). The corresponding C-statistics for PREVENT-HF with ECG-AI added were 0.898 (0.885 - 0.909) and 0.894 (0.887 - 0.902), respectively (Table 1). The corresponding NRI for ECG-AI-based up-risking relative to PREVENT-HF at 10% and 20% thresholds was 10.7% and 39.7% (3-year) and 8.3% and 32.7% (10-year), respectively (Table 2).

Conclusion: ECG-AI LEF and ECG-AI DD significantly enhance PREVENT-HF in identifying patients at high short-term risk for heart failure. These findings support ECG-AI as a scalable, actionable tool that complements standard risk scores and enables targeted HF prevention.