Ó³»­´«Ã½

BACKGROUND: Out-of-hospital cardiac arrest (OHCA) is a public health burden with the majority occurring in the general population for whom there is no firm strategy to predict risk.OBJECTIVES: The authors evaluated whether artificial intelligence enhanced electrocardiography (ECG) and clinical information from electronic health records (EHRs) can stratify risk of OHCA in the general population.METHODS: We use a case-control study design (matching on age and sex), to derive and temporally validate models to predict OHCA. To evaluate the potential use case of models in a real-world context, we evaluated the 2-year cumulative incidence of OHCA in individuals undergoing ECG in a health care system, while accounting for the competing risk of non-OHCA mortality.RESULTS: In the temporal validation cohort, discrimination of OHCA was highest for the multimodal ECG + EHR model (area under the receiver operating characteristic curve: 0.83; area under the precision recall curve: 0.44) followed by the EHR-only model and the ECG-only model (Bonferroni adjusted P for all pairwise comparisons <0.05). In the real-world cohort of individuals undergoing ECG, the EHR + ECG model flagged two-thirds (153 of 228) of those with incident OHCA over a 2-year period as high-risk. Using the ECG + EHR model, the 2-year cumulative incidence of OHCA was 2.4% (95% CI: 2.0%-2.8%) in individuals identified as high-risk compared with 0.5% (95% CI: 0.3%-0.8%) in individuals designated as low risk.CONCLUSIONS: In a large U.S. health care system, artificial intelligence-enhanced ECG and EHR data effectively discriminated individuals at risk of OHCA and identified those at clinically relevant risk of incident OHCA over a 2-year period.