[Study on the diagnosis model of sepsis based on Medical Information Mart for Intensive Care-IV database]

To explore the diagnostic model of sepsis based on Medical Information Mart for Intensive Care-IV (MIMIC-IV).

The clinical data of septic patients admitted to intensive care unit (ICU) with MIMIC-IV database, including demographic characteristics, disease severity scores [Simplified Acute Physiology Score II (SAPS II), Logistic Organ Failure Score (LODS), Oxford Acute Disease Severity Score (OASIS), Sequential Organ Failure Assessment (SOFA)], vital signs and laboratory indicators (blood routine, serum electrolytes, blood gas analysis, blood glucose, liver and kidney function, coagulation function, blood culture results) at ICU admission, underlying disease, site of infection and whether shock was combined were retrospectively analyzed. The primary outcome indicator was 28-day death. The patients were divided into survival group and death group according to 28-day prognosis. The differences in various indicators between the two groups were compared, and the indicators with differences between the two groups were incorporated into the classification tree model to screen out the top 5 indexes of importance for predicting the 28-day death in septic patients, which were enrolled in the diagnostic model of sepsis. Receiver Operator characteristic curve (ROC curve) was used to screened out the scoring system with the greatest predictive value for 28-day death in septic patients from four disease severity scores, which was enrolled in the diagnosis model of sepsis.

A total of 12 743 patients were enrolled in the study finally, including 7 411 males (58.157%) and 5 332 females (41.843%), aged from 18 to 100 years old, age (65.38±16.16) years. There were 9 097 patients (71.388%) in the 28-day survival group and 3 646 patients (28.612%) in the death group. Age, four disease severity scores, vital signs and most laboratory indicators at ICU admission [body temperature, heart rate (HR), respiration rate (RR), systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse oxygen saturation (SpO₂), white blood cell count (WBC), platelets count (PLT), hemoglobin (Hb), K⁺, Na⁺, Cl^-, HCO₃^-, alanine aminotransferase (ALT), aspartate aminotransferase (AST), prothrombin time (PT), activated partial thromboplastin time (APTT), blood urea nitrogen (BUN), serum creatinine (SCr)] and ratio of positive blood culture, infection site and combined with shock were statistically significant differences between the two groups (all P<0.05). The 28-day death risk factors classification tree model for septic patients was consisted of 4 layers and 25 nodes, among which there were 10 terminal nodes. The top 5 variables in terms of importance were combined with shock or not, age, combined with metastatic tumor or not, infection site, and blood culture results. Except for age, which had an insignificant warning effect, all other variables were included in the diagnosis model of sepsis. ROC curve analysis showed that all four scores had predictive value for the 28-day death in septic patients. Among them, the area under the ROC curve (AUC) of the SAPS II score was the largest, which was 0.726 [95% confidence interval (95%CI) was 0.716-0.736]. And it was higher than the LODS score [0.693 (95%CI was 0.683-0.703); Z=7.949, P<0.001], OASIS score [0.665 (95%CI was 0.654-0.675); Z=13.041, P<0.001], and SOFA score [0.656 (95%CI was 0.646-0.667); Z=14.299, P<0.001], which was enrolled in the diagnostic model of sepsis. The final diagnostic model of sepsis was "sepsis [shock (+/-), site of infection, blood culture (+/-)/metastatic tumor (+/-), SAPS II score]".

Based on the analysis results of the MIMIC-IV database, it was recommended that the diagnostic model of sepsis was "sepsis [shock (+/-), site of infection, blood culture (+/-)/metastatic tumor (+/-), SAPS II score]".