Research on the death risk prediction model for patients with sepsis based on machine learning

This is the authors' abstract. We don't add key points for this paper.

Objectives: Sepsis is a critical public health emergency that poses a severe threat to human life and health. However, there is currently no reliable prognostic assessment tool for such patients. This study utilized machine learning methods to develop a 28-day mortality risk prediction model for patients with sepsis, aimed to provide a basis for early identification of high-risk patients and timely clinical decision-making. Methods: The clinical data sets of patients with sepsis treated in the intensive care unit (ICU) of a tertiary hospital from January 2017 to June 2023 were retrospectively collected. Logical regression model (LR), Gaussian naive Bayesian model (GaussianNB), support vector machine (SVM), nearest neighbor algorithm (kNN), random forest (RF), and extreme gradient boosting model (XGBoost) were used to construct a 28-day mortality risk prediction model for patients with sepsis, and multiple evaluation indicators were used to analyze the prediction performance of the model. Considering the particularity of the medical industry, we applied the Shapley additive interpretation (SHAP) method to visualize the optimal model to meet the needs of clinicians for model understanding and personalized prediction. Results: A total of 631 patients with sepsis were included in this study, and the 28-day mortality rate was 44.06%. A total of 16 features were included in the screening of clinical features for model construction. According to a number of evaluation indicators, the XGBoost model performed well and stably. The area under the curve (AUC) of the receiver operating characteristic (ROC) of the training set and the validation set was higher than that of other models (0.996 and 0.759, respectively). The SHAP summary map provides relevant clinical characteristics of the risk of death in patients with sepsis within 28 days, and is highly dependent on SOFA score, age, and neutrophils. Conclusions: The XGBoost model has a good predictive ability for the short-term prognosis of ICU patients with sepsis, which can assist clinicians to identify high-risk patients early and provide clinical decision-making guidance.