Prognostic Value of SIRI in Sepsis: A Retrospective Study and Machine Learning-Based Model Development

Background: In recent years, the Systemic Inflammation Response Index (SIRI) has demonstrated unique advantages in evaluating sepsis prognosis. This study aims to investigate the predictive value of SIRI for 28-day outcomes in sepsis patients, and develop and validate a prognostic model for 28-day mortality. Methods: The demographic characteristics, disease severity, laboratory tests, treatments, and outcome measures were recorded from the adult sepsis patients. The restricted cubic splines and the ROC curve analysis were employed to evaluate the relationship and predictive capability of SIRI. Next, SIRI was categorized into tertiles, and univariate and multivariate Cox regression analyses were performed to assess its association with prognosis, supplemented by Kaplan-Meier (K-M) curves, and compare mortality differences. Patients from the First Affiliated Hospital of Soochow University were randomly allocated into training and internal validation sets at a 3:1 ratio, using the Boruta algorithm and LASSO regression and a prognostic model was constructed via logistic regression, while patients from the First People’s Hospital of Zhangjiagang City served as the external validation set. Then, the predictive performance, accuracy, and clinical utility of the model were validated using the ROC curve, Hosmer-Lemeshow test, calibration curve, and decision curve analysis (DCA). Results: The 380 patients from the First Affiliated Hospital of Soochow University and 240 patients from the First People’s Hospital of Zhangjiagang City were enrolled for the present study. The restricted cubic spline analysis revealed a nonlinear increasing trend in mortality risk with rising SIRI levels. The ROC curve analysis demonstrated that SIRI has superior predictive capability than the APACHE II and SOFA scores. When SIRI was categorized into tertiles, both the univariate and multivariate Cox regression analyses identified SIRI as significantly associated to 28-day prognosis ( p < 0.001). The K-M curves further confirmed that higher SIRI levels correlated to lower 28-day survival rates ( p < 0.001). In the training set, the Boruta algorithm combined with LASSO regression selected six independent risk factors: blood urea nitrogen (BUN), age, phosphorus (P), lactate (Lac), mechanical ventilation (MV), and SIRI. These were incorporated into the predictive model through logistic regression analysis. The ROC curve analysis revealed that the model exhibited good predictive performance across the training set (AUC: 0.851), internal validation set (AUC: 0.908), and external validation set (AUC: 0.792). The calibration of the model was verified using the Hosmer-Lemeshow test and calibration curve, while DCA was performed to confirm its clinical utility. Conclusion: SIRI is significantly correlated to the 28-day prognosis in sepsis patients, and has excellent predictive value for short-term outcomes. The prediction model that incorporated SIRI exhibited high prognostic accuracy. Keywords: sepsis, systemic inflammation response index, prognosis, prediction model, nomogram