A new study published in Scientific Reports suggests that machine learning algorithms can accurately predict high-risk days for out-of-hospital cardiac arrest (OHCA). The researchers used meteorological and chronological data from a large pool of OHCA patients in Japan to create and test these algorithms. The study drew on data from the nationwide, population-based OHCA registry, which collected data on OHCA events across Japan between 2005 and 2015.
The study’s primary goal was to find ways to identify high-risk days for OHCA incidence accurately, which can help clinicians better prepare for a possible surge in OHCA patients. The team used explainable features such as meteorological data, chronological information, and the percentage of elderly populations per region to train the machine learning algorithms.
The study found that the average and diurnal temperature from the previous day were the best predictive meteorological features for OHCA incidence in the Japanese population. The team used various machine learning algorithms such as XGBoost and SVM to build eight classification models, which were trained using the Tokyo cohort data between 2005 and 2012. They then selected the best machine learning algorithm, XGBoost, to use in the test cohort, which included patients from six highly populated Japanese prefectures, except Tokyo.
The performance of the XGBoost model was evaluated based on the AUROC values, which were estimated from six-fold cross-validation. The results showed that the XGBoost algorithm had an AUROC value of 0.738, indicating its high accuracy in predicting high-risk days for OHCA incidence in Japan.
The team also used the SHAP algorithm of the XGBoost model to interpret the contribution of each feature to the predictive model. The analysis revealed that other features such as the percentage of the elderly population, onset day, and month were also important predictors for OHCA incidence in Japan.
The study suggests that machine learning algorithms can be a useful tool to predict high-risk days for OHCA incidence. The researchers believe that this information can help clinicians prepare better, allocate resources efficiently, and reduce the risk of death from cardiac arrest. However, they acknowledge that more research is needed to generalize these findings to other populations and regions.