A Machine Learning Model Integrating Systemic Immune-Inflammation Index for Predicting 28-Day Mortality in Geriatric Sepsis Secondary to Community-Acquired Bacterial Pneumonia.

Purpose: Sepsis is a life-threatening clinical syndrome characterized by a dysregulated host response to infection. The systemic immune-inflammation index (SII) is a novel prognostic biomarker, However, its predictive value for early prognosis in elderly individuals suffering from sepsis secondary to community-acquired bacterial pneumonia (CABP) remains unclear. This study intends to apply machine learning techniques to develop an interpretable model for predicting prognosis. Patients and Methods: This medical records review included elderly patients with sepsis secondary to CABP admitted to ICUs at Beijing Shijitan Hospital. Clinical outcomes based on 28-day survival status served as the basis for dividing participants into survivor and non-survivor groups. For prognostic prediction, five machine learning algorithms were developed, including Gradient Boosting Machine (GBM), Logistic Regression, Random Forest (RF), Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting (XGBoost). Model efficacy was quantified using the area under the receiver operating characteristic curve (AUROC) metric and assessed clinically through decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) visualization provided insights into the models' decision-making processes. Results: In this investigation, clinical information from 364 geriatric participants suffering from sepsis secondary to CABP was examined. For developing prediction models, twelve predictors were chosen. During an extensive evaluation of multiple computational frameworks, the XGBoost algorithm exhibited superior prognostic capability regarding 28-day mortality (AUROC = 0.901, 95% CI: 0.853-0.949,). The SHAP summary plot generated from the optimal XGBoost model ranked the importance of predictive features. The SII, APACHE II score, age, Pneumonia Severity Index (PSI), gender was identified as the top five most influential factors. Conclusion: Elevated SII concentrations correlated significantly with this mortality rate. The interpretable XGBoost model accurately detected the precise link between admission SII measurements and subsequent 28-day all-cause mortality in the target patient population.