Construction and validation of a predictive model for septic shock based on propensity score matching

Objective To construct a predictive model for septic shock based on the propensity score matching (PSM) method and validate its effectiveness. Methods A total of 114 patients with sepsis were enrolled as study objects, and were divided into septic shock group (40 patients) and sepsis group (74 patients) according to whether they developed septic shock. PSM was performed with a ratio of septic shock to sepsis of 1∶2, resulting in the inclusion of 30 patients in the septic shock group and 60 patients in the sepsis group after matching. The levels of C-reactive protein (CRP), procalcitonin (PCT), interleukin-6 (IL-6), serum amyloid A (SAA), soluble endothelial protein C receptor (sEPCR), endothelial cell-specific molecule 1 (ESM-1), clusterin (CLU), and the Acute Physiology and Chronic Health Evaluation Ⅱ (APACHE Ⅱ) score and Sequential Organ Failure Assessment (SOFA) score at admission were compared between the two groups. Cox proportional hazards regression analysis was used to identify the factors influencing septic shock, and a predictive model for septic shock was constructed and internally validated using the receiver operating characteristic (ROC) curve. Kaplan-Meier survival curves were plotted to analyze the differences in survival prognosis among patients with different expression levels of the indicators. Results After matching, there were no statistically significant differences in general information between the two groups (P>0.05). At admission, the septic shock group had higher levels of serum PCT, CRP, SAA, IL-6, sEPCR, ESM-1, and higher APACHE Ⅱ and SOFA scores, as well as a lower level of serum CLU compared with the sepsis group (P<0.05). Cox regression analysis showed that PCT, CRP, SAA, IL-6, sEPCR, ESM-1, APACHE Ⅱ score, and SOFA score were independent risk factors for septic shock (P<0.05), while CLU was an independent protective factor (P<0.05). The predictive model for septic shock, constructed based on these factors, showed an internal validation accuracy of 94.44%, an area under the curve of 0.950, a sensitivity of 93.33%, and a specificity of 96.67%. Dead patients had higher levels of PCT, CRP, SAA, IL-6, sEPCR, ESM-1, and higher APACHE Ⅱ and SOFA scores, as well as a lower level of CLU at admission compared with survivors (P<0.05). Compared with patients with low expression levels or low scores, patients with high expression levels of PCT, CRP, SAA, IL-6, sEPCR, ESM-1, and high APACHE Ⅱ and SOFA scores had higher fatality rates, while patients with high CLU expression levels had a lower fatality rate (P<0.05). Conclusion The serum biomarkers including PCT, CRP, SAA, IL-6, sEPCR, ESM-1, CLU, and the APACHE Ⅱ and SOFA scores in sepsis patients are closely related to the occurrence of septic shock and survival prognosis. The predictive model constructed by combining these indicators can accurately predict the occurrence of septic shock.