Interpretive machine learning predicts short-term mortality risk in elderly sepsis patients.

Backgrounds: Sepsis is a leading cause of in-hospital mortality. However, its prevalence is increasing among the elderly population. Therefore, early identification and prediction of the risk of death in elderly patients with sepsis is crucial. The objective of this study was to create a machine learning model that can predict short-term mortality risk in elderly patients with severe sepsis in a clear and concise manner. Methods: Data was collected from the MIMIC-IV (2.2). It was randomly divided into a training set and a validation set using a 7:3 ratio. Mortality predictors were determined through Recursive Feature Elimination (RFE). A prediction model for 28 days of ICU stay was built using six machine-learning algorithms. To create a comprehensive and nuanced model resolution, Shapley Additive Explanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were used to systematically interpret the models at both a global and detailed level. Results: The study involved the analysis of 4,056 elderly patients with sepsis. A feature recursive elimination algorithm was utilized to select eight variables out of 49 for model development. Six machine learning models were assessed, and the Extreme Gradient Boosting (XGBoost) model was found to perform the best. The validation set achieved an AUC of 0.88 (95% CI: 0.86-0.90) and an accuracy of 0.84 (95% CI: 0.81-0.86) for this model. To examine the roles of the eight key variables in the model, SHAP analysis was employed. The global ranking order was made evident, and through the use of LIME analysis, the weights of each feature range in the prediction model were determined. Conclusion: The study's machine learning prediction model is a dependable tool for forecasting the prognosis of elderly patients with severe sepsis.