Artificial Intelligence–Enabled Serial Electrocardiograms for Prediction of All-Cause Mortality in Secondary Care Settings

BACKGROUND: Prognostic assessment in secondary care settings remains challenging and may influence clinical decision-making and follow-up. Artificial intelligence-enabled electrocardiography (AI-ECG) is a promising tool, but the value of incorporating serial ECG data remains unclear. OBJECTIVES: This study aimed to develop and validate an AI-ECG model for predicting 1-year all-cause mortality and to assess the incremental value of serial ECGs. METHODS: We conducted a retrospective cohort study of adults aged over 50 years treated in emergency department or inpatient settings between 2014 and 2019. Models were developed using a deep-learning framework applied to either a single index ECG or serial ECGs, including the index and 2 prior recordings. Data were divided into training, validation, and test sets. Performance was assessed using discrimination, calibration, and decision-curve analysis. A classification threshold was selected on the validation set by maximizing the F1 score and applied to the test set. RESULTS: A total of 13,417 patients were included (median age 69 years [IQR: 57-80]; 45% women), of whom 4,372 (34%) died within 1 year. The serial AI-ECG model achieved an area under the curve of 0.84 (95% CI: 0.824-0.862). At a threshold of 0.24, sensitivity was 70%, specificity was 80%, and positive predictive value was 49%. Calibration showed good agreement between predicted and observed risks (Brier score: 0.123). Decision-curve analysis demonstrated greater net benefit than treat-all and treat-none strategies. Addition of clinical variables beyond age and sex did not improve performance. CONCLUSIONS: A serial AI-ECG model provides accurate, clinically interpretable prediction of 1-year mortality and may support risk stratification in secondary care.