Predicting sepsis within 24 hours at triage stage in emergency department using machine learning

Abstract

Abstract Background Sepsis is one of the main causes of mortality in intensive care units. To reduce its damage, prediction should be made the earlier the better. As around 36% of sepsis onset took place within 24 hours after ED admission in MIMIC-IV (v2.2), a prediction system at ED triage stage would certainly be of help. Previous methods, such as qSOFA, are more suitable for screening instead of prediction in ED. And we aimed to find a light-weight, convenient prediction method through machine learning. Methods We utilized the open medical database MIMIC-IV (v2.2), to obtain sepsis patients’ corresponding data in the emergency department. We built our dataset with demographic data, vital signs and synthesized features. We used XGBoost as the classifier, to predict if the patient would develop sepsis within 24 hours after ED admission, and used SHAP to interpret the model’s outcome. Results For 10 fold cross validation on the 14,957 samples, we reached an accuracy of 84.1 ± 0.3%, and an area under the ROC curve of 0.92 ± 0.02. The model achieved similar performance on the testing set of 1,662 patients. SHAP values showed that the five most important features were acuity, arrival transportation, age, shock index, and respiratory rate. Conclusions Machine learning models like XGBoost can be used for sepsis forecast, using just a small amount of data conveniently collected in the ED triage stage. This can help reduce the workload in the ED, and warn medical workers against the risk of sepsis in advance.