Deep-learning model for screening sepsis using electrocardiography

Abstract Background Sepsis is a life-threatening organ dysfunction and major healthcare burden worldwide. Although sepsis medical emergency that requires immediate management, screening for the occurrence of difficult. Herein, we propose deep learning-based model (DLM) using electrocardiography (ECG). Methods This retrospective cohort study included 46,017 patients who were admitted to two hospitals. A total 1,548 639 had septic shock, respectively. The DLM was developed 73,727 ECGs from 18,142 patients, internal validation conducted 7774 7,774 patients. Furthermore, an external with 20,101 another hospital verify applicability across centers. Results During validations, area under receiver operating characteristic curve (AUC) 12-lead ECG 0.901 (95% confidence interval, 0.882–0.920) 0.863 (0.846–0.879), respectively, 0.906 interval (CI), 0.877–0.936) 0.899 CI, 0.872–0.925), detecting shock. AUC 6-lead single-lead 0.845–0.882. sensitivity map revealed QRS complex T waves associated sepsis. Subgroup analysis 4,609 infectious disease, predicting in-hospital mortality 0.817 (0.793–0.840). There significant difference in prediction score according presence infection dataset (0.277 vs. 0.574, p < 0.001), including severe acute respiratory syndrome coronavirus 2 (0.260 0.725, = 0.018). Conclusions delivered reasonable performance 12-, 6-, ECGs. results suggest can be screened not only conventional devices but also diverse life-type machines employing DLM, thereby preventing irreversible disease progression mortality.