Construction of a decision tree model for predicting the 28-day prognosis of sepsis patients based on serum SLPI and S100P

Sang Zhenzhen; Pang Xiuyan; Cui Jie; Wang Weifeng; Rao Xin

doi:10.3969/j.issn.1002-1949.2025.10.010

Abstract：Objective To construct a prediction model for the 28-day prognosis of sepsis patients by using machine learning algorithms and to explore the predictive value of serum secretory leukocyte peptidase inhibitor(SLPI)and S100 calcium-binding protein P(S100P).Methods This study adopted a prospective observational research method,enrolling sepsis patients admitted to the emergency department of Cangzhou Central Hospital from March 2024 to March 2025 as the research subjects.They were divided into the sepsis without shock group(58 cases)and the sepsis with shock group(95 cases)based on the severity of the condition.According to the 28-day prognosis after admission,they were further divided into the death group(44 cases)and the survival group(109 cases).Forty healthy individuals who underwent physical examinations during the same period were included as the control group.The levels of SLPI and S100P in the serum of sepsis patients were detected on the day of admission by using enzyme-linked immunosorbent assay(ELISA).Multivariate Logistic regression was used to analyze the risk factors for 28-day mortality in sepsis patients,and Spearman correlation analysis was conducted.The R software was used to construct a decision tree model based on the risk factors affecting 28-day mortality in sepsis patients,and the predictive performance of the model was evaluated by using the receiver operating characteristic(ROC)curve and the area under curve(AUC).Results The levels of SLPI[492.40(424.45,556.9)pg/mL,vs.389.85(332.3,446.72)pg/mL]and S100P[60.5(52.05,77.3)ng/mL,vs.(42.62,55.27)ng/mL]in the serum of sepsis with shock patients were significantly lower than those in the sepsis without shock patients(P<0.001).The levels of SLPI[422.1(345.7,496.2)pg/mL,vs.547.6(456.22,607.85)pg/mL]and S100P[50.5(44.5,55.6)ng/mL,vs.78.5(69.45,87.65)ng/mL]in the serum of the survival group were significantly lower than those in the death group(P<0.001).There were statistically significant differences in the SOFA score,APACHEⅡ score,blood lactate(Lac),albumin(ALB),and procalcitonin(PCT)in sepsis patients between the survival group and the death group(P<0.001).The results of multivariate Logistic regression analysis identified that SLPI[OR=1.01,95%CI(1.00,1.02)],S100P[OR=1.23,95%CI(1.14,1.37)],SOFA score[OR=1.41,95%CI(1.15,1.83)],APACHEⅡ score[OR=1.16,95%CI(1.03,1.34)],and Lac[OR=1.15,95%CI(1.03,1.32)]were risk factors for 28-day mortality in sepsis patients(P<0.05).Spearman correlation analysis showed that S100P and SLPI were significantly positively correlated with APACHEⅡscore,SOFA score and Lac(r=0.768、0.676、0.699,r=0.691、0.803、0.702,P<0.001).A decision tree was constructed based on the risk factors affecting 28-day mortality in sepsis patients,and the three variables most associated with to patient mortality(S100P,SOFA score and SLPI)were identified ultimately.The accuracy rate of the decision tree model was 89.54%,the sensitivity was 84.09%,the specificity was 91.74%,and the area under ROC curve was 0.935[95%CI(0.877,0.968)].The AUC of APACHEⅡ score and SOFA score alone for predicting 28-day mortality in sepsis patients was 0.748[95%CI(0.712,0.773)]and 0.798[95%CI(0.765,0.827)],respectively.The decision tree model showed significantly better predictive performance than APACHEⅡscore and SOFA score alone.Conclusions The levels of serum SLPI and S100P in sepsis patients are significantly elevated.The decision tree model constructed based on SLPI and S100P can more accurately assess the adverse prognosis of sepsis patients.