A machine learning model for predicting out-of-hospital cardiac arrest incidence using meteorological, chronological, and geographical data from the United States

Abstract

AbstractBackgroundDespite advances in pre- and post-resuscitation care, percentage of survival to hospital discharge after out-of-hospital cardiac arrest (OHCA) was extremely low. Development of an accurate system to predict the daily incidence of OHCA might provide a significant public health benefit. Here, we developed and validated a machine learning (ML) predictive model for daily OHCA incidence using high-resolution meteorological, chronological, and geographical data.MethodsWe analyzed a dataset from the United States that combined an OHCA nationwide registry, high-resolution meteorological data, chronological data, and geographical data. We developed a model to predict daily OHCA incidence with a training dataset for 2013–2017 using the eXtreme Gradient Boosting algorithm. A dataset for 2018–2019 was used to test the predictive model. The main outcome was the predictive accuracy for the number of daily OHCA events, based on root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). In general, a model with MAPE less than 10% is considered highly accurate.ResultsAmong the 446,830 OHCAs of non-traumatic cause where resuscitative efforts were initiated by a 911 responder, 264,916 in the training dataset and 181,914 in the testing dataset were included in the analysis. The ML model with combined meteorological, chronological, and geographical data had high predictive accuracy in relation to nationwide incidence rate per 100,000 at the nationwide level) in the training dataset (RMSE, 0.016; MAE, 0.013; and MAPE, 7.61%) and in the testing dataset (RMSE, 0.018; MAE, 0.014; and MAPE, 6.52%).ConclusionsA ML predictive model using comprehensive daily meteorological, chronological, and geographical data allows for highly precise estimates of OHCA incidence in the United States.Clinical PerspectiveWhat is new?A machine learning predictive model developed with a high-resolution meteorological dataset and chronological and geographical variables predicted the daily incidence of out-of-hospital cardiac arrest (OHCA) in the U.S. population with high precision. The predictive accuracy at the state level was greater in medium and high-temperature areas than in the low-temperature area.What are the clinical implications?This predictive model revealed complex associations between meteorological, chronological, and geographic variables in relation to predicting daily incidence of OHCA. It might be useful for public health strategies in temperate regions, for example, by providing a warning system for citizens and emergency medical services agencies on high-risk days.