Artificial intelligence applied to post-resuscitation ECGs for early prognostication after out-of-hospital cardiac arrest.

Background: There are limited tools for early outcome prediction after Out-of-Hospital Cardiac Arrest (OHCA). This study aimed to evaluate whether a machine learning model could help to predict neurological outcome using the 12-lead ECG acquired on scene following return of spontaneous circulation (ROSC). Methods: We conducted a retrospective analysis of prospectively collected post-ROSC ECGs from the Lombardy Cardiac Arrest Registry (January 2015-December 2023). The study included all the patients resuscitated from OHCA with a 12-lead ECG acquired on scene after ROSC. A deep neural network model was developed, validated, and tested using features extracted from these ECGs via computer vision techniques, incorporating patient age, sex, initial rhythm, and ROSC-ECG time. The dataset was split into training (80%), validation (10%), and independent testing (10%) sets. Model performance was evaluated using accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), AUROC (Area Under the Receiver Operating Characteristic curve), Matthews Correlation Coefficient (MCC), and SHAP (Shapley Additive exPlanations) values for interpretability. Results: We included 976 post-ROSC ECGs (641 poor and 335 favorable neurological outcomes). The model achieved an accuracy of 80.8%, sensitivity of 86.1%, specificity of 74.3%, PPV of 80.4%, and NPV of 81.2%, with an MCC of 0.61 and an AUROC of 0.86. Conclusions: To our knowledge, this is the first machine learning model utilizing post-ROSC 12-lead ECG data to evaluate its association with neurological outcome immediately after ROSC. Its application in the pre-hospital setting may provide additional information to support clinical decision-making regarding transport strategies and post-resuscitation care planning following OHCA.