Establishment And Validation Of A Predictive Model For Sepsis-related Coagulopathy In Diabetic Patients Based On MIMIC-Ⅳ Database

Objective The purpose of this study is to establish and validate a predictive model for sepsis-associated coagulation dysfunction in diabetic patients based on the MIMIC-Ⅳdatabase.By analyzing the clinical data of diabetic patients in the MIMIC-Ⅳ database,relevant factors affecting sepsis-associated coagulation dysfunction are identified,a predictive model is established,and the model is validated to evaluate its predictive performance and clinical value.The aim of this study is to provide clinicians with a tool to assist in the diagnosis of sepsis-associated coagulation dysfunction in diabetic patients,improve clinical treatment efficacy and patient survival rate,and provide data support and reference for subsequent related research.Methods Obtained permission to access the MIMIC-Ⅳ database,extracted and installed the data using PostgreSQL and 7-Zip software,extracted and filtered the required data using PostgreSQL and Navicat software,checked for missing values,and used R software to perform multiple imputations on indicators with missing values less than 40%.The data was randomly divided into a modeling group and a validation group in a ratio of 7:3.The LASSO logistic regression analysis was used to screen out candidate predictive factors with larger regression coefficients,and a multivariate logistic regression analysis was performed based on the training data to determine independent risk factors associated with the development of coagulation dysfunction.A risk model was established,in which the score of each predictive factor was calculated based on the coefficients of the logistic regression variables,and a column chart model for predicting sepsis-related coagulation dysfunction in diabetic patients was constructed using R software,making the model visualized.The model discrimination was evaluated by ROC curve analysis.R version 3.4.1 was used for data analysis and plotting.For normally distributed continuous data,Mean±SD was used,and t-tests were used for intergroup comparisons.For count data,x2 tests or Fisher’s exact probability tests were used,and the results were presented in the form of rates(%).A multivariate analysis was performed using the logistic regression model,and factors with statistically significant differences in univariate analysis were included in the model to calculate the relative risk(OR)and 95%confidence interval.P<0.05 was considered statistically significant.Results According to the inclusion and exclusion criteria,a total of 2745 diabetic patients with sepsis were selected for the study,including 1152 patients who experienced coagulation dysfunction during sepsis(designated as the "occurrence group")and 1593 patients who did not experience coagulation dysfunction(designated as the "non-occurrence group").70%of patients were randomly selected from the occurrence and non-occurrence groups for model building,while the remaining 30%were used for internal validation.Basic characteristics of the occurrence and non-occurrence groups were compared in the training set,revealing significant differences in age,SAPS Ⅱ score,respiration,mean arterial pressure,temperature,international normalized ratio,prothrombin time,platelets,partial pressure of carbon dioxide,bicarbonate,total carbon dioxide,hematocrit,hemoglobin,chloride,calcium,lactate,blood glucose,urine output,SOFA score,gender,congestive heart failure,kidney disease,cancer,use of vancomycin,and use of CRRT.Using LASSO logistic regression,12 predictors with high regression coefficients were selected.Based on the training data set,a multiple logistic regression analysis was performed,and age,SOFA score,international normalized ratio,platelets,hemoglobin,lactate,blood glucose,and mild liver disease were identified as independent risk factors.Using R software,a column line graph model was constructed to visualize the prediction of coagulation dysfunction in sepsis patients with diabetes,and the model’s discriminative ability was evaluated through ROC curve analysis.In this study,the ROC curve AUC was 0.931 for the model building group and 0.922 for the validation group,indicating that the established prediction model had good performance,high accuracy,and reliability.Conclusion This study developed a new model to predict sepsis-related coagulation dysfunction in diabetic patients using clinical variables.The model can provide more accurate diagnostic and therapeutic decision-making support for physicians.