An Exploratory Study On The Efficacy Of Time-dependent Variables In Predicting Adverse Outcomes Of Sepsis Based On Dynamic Bayesian Networks

Objective: In view of the rapid changes in the condition of patients with sepsis in the intensive care unit,the high risk of organ dysfunction and death,relying on the high correlation between time-dependent variables and time,the dynamic Bayesian network(DBN)was used to build a dynamic change model of time-dependent variables of patients with sepsis in the intensive care unit.It is expected to provide methodological support or reference for the change of patients with sepsis in ICU and the prediction of adverse outcomes.Methods: Extract the data of patients with sepsis from the Medical Information Mart for Intensive Care(MIMIC-IV)and build a dynamic Bayesian network model.Then the time dependent variable changes,organ dysfunction and death risk ability were tested at the patient’s ICU for 24 hours and 48 hours respectively.Through the evaluation of the area under receiver operating curve(AUROC)of subjects in 24 hours and 48 hours,as well as the evaluation of sensitivity and specificity,we can better understand the predictive ability of different network learning algorithms on the adverse outcome of sepsis patients in the intensive care unit.Results: When the DBN model is used to predict the change of time-dependent variables,the difference ratio between the actual value of most time-dependent variables and the predicted value of the 24 th and 48 th hours is less than 5%,and the error of the predicted value of the 48 th hour is slightly greater than the error of the predicted value of the 24 th hour.The accuracy of DBN model in predicting kidney,liver,cardiovascular and blood dysfunction was more than 0.8,and the accuracy of DBN model was less than 0.8 for neural dysfunction at the 48 th hour.When using the DBN model to predict the death risk,the 24-hour and 48-hour death risk predicted from AUROC and 95% confidence interval were 0.975(0.965-0.986)and 0.943(0.930-0.956)in MIMIC-Ⅳ,respectively;According to the data in the first 24 hours of the results,the AUC scores of the Simplified acute physical score(SAPS-Ⅱ)and the Acute Physiology and Chronic Health Disease Classification System Ⅱ(APACHE-Ⅱ)were0.948(95% CI,0.936-0.960)and 0.943(95% CI,0.927-0.959).The AUROC predicted by DBN is better than SAPS-Ⅱ and APACHE-ⅡConclusion: DBN can help us better evaluate the physiological changes,organ dysfunction,and mortality risk of ICU sepsis patients by learning and inferring the causal relationships of time-dependent variables in the MIMIC IV database.It can also be used as a real-time tool to predict the physiological changes and adverse outcomes of sepsis patients during ICU admission,providing reliable reference for the treatment of sepsis patients.