Early outcome prediction for out-of-hospital cardiac arrest with initial shockable rhythm using machine learning models.

Abstract

Early outcome prediction for out-of-hospital cardiac arrest with initial shockable rhythm is useful in selecting the choice of resuscitative treatment by clinicians. This study aimed to develop and validate a machine learning-based outcome prediction model for out-of-hospital cardiac arrest with initial shockable rhythm, which can be used on patient's arrival at the hospital. Data were obtained from a nationwide out-of-hospital cardiac arrest registry in Japan. Of 43,350 out-of-hospital cardiac arrest patients with initial shockable rhythm registered between 2013 and 2017, patients aged <18 years and those with cardiac arrest caused by external factors were excluded. Subjects were classified into training (n = 23,668, 2013-2016 data) and test (n = 6381, data from 2017) sets for validation. Only 19 prehospital variables were used for the outcome prediction. The primary outcome was death at 1 month or survival with poor neurological function (cerebral performance category 3-5; "poor" outcome). Several machine learning models, including those based on logistic regression, support vector machine, random forest, and multilayer perceptron classifiers were compared. In validation analyses, all machine learning models performed satisfactorily with area under the receiver operating characteristic curve values of 0.882 [95% confidence interval [CI]: 0.869-0.894] for logistic regression, 0.866 [95% CI: 0.853-0.879] for support vector machine, 0.877 [95% CI: 0.865-0.890] for random forest, and 0.888 [95% CI: 0.876-0.900] for multilayer perceptron classifiers. A favourable machine learning-based prognostic model available to use on patient arrival at the hospital was developed for out-of-hospital cardiac arrest with initial shockable rhythm.