Risk stratification of sepsis death based on machine learning algorithm

Abstract

<p>Sepsis, a life-threatening syndrome caused by the body's dysfunctional response to infection, carries a high mortality rate. Prompt and aggressive treatments can significantly reduce the morbidity and mortality rates. Utilizing conventional physiological and laboratory data from MIMIC Ⅲ, we applied various machine learning algorithms, including a logistic regression, XGBoost, the K-nearest neighbor, a decision tree, and a support vector machine, to predict the mortality risk. After comparing the performance of these algorithms, XGBoost emerged as the most effective, with an area under the curve (AUC) of 0.91, a specificity of 0.82, and a sensitivity of 0.84. Furthermore, we used a logistic regression to develop a scoring system for the sepsis death risk stratification, and achieved an AUC of 0.79. This scoring system identified high-risk patients upon their admission to the intensive care unit (ICU). By continuously collecting data from electronic health records and calculating the mortality risk scores, clinicians can promptly identify patients at a high risk of death and intervene early to either prevent or minimize the associated morbidity and mortality.</p>