Prognostic Factor Analysis And Development And Validation Of Mortality Risk Prediction Model For Patients With Sepsis Based On Machine Learning

Objective:In order to explore the risk factors affecting the prognosis of sepsis patients in the intensive care unit(ICUs),we used a machine learning approach to develop a risk model for predicting death in sepsis patients within 28 days based on Medical Information Mart for Intensive Care Ⅲ(MIMIC Ⅲ).Methods:This study was a retrospective design.Sepsis patients from the Medical Information Mart for Intensive Care-Ⅲ(MIMIC-Ⅲ)database were enrolled.All patients were≥ 16 years old and admitted to ICU who are diagnosed with sepsis within 24 hours.The criteria for diagnosing patients with sepsis were based on the new Sepsis 3.0 criteria.Patients’ demographic characteristics,vital signs,and laboratory parameters were collected 24 hours after ICU admission,and mortality at 28 days was the primary outcome.Patients were randomly allocated to modelling and validation groups in a 7:3 ratio,A predictive model for predicting 28-day morality in sepsis was performed based on the logistic regression and machine learning approach and.We built a nomogram to present the predictive model.The sensitivity,specificity and the area under the curve of predictive model and clinical scores were applied to appraise the function of the model on the validation dataset.The decision curve was plotted to assess the clinical practical application value of the model.Results:A total of 4,209 patients with sepsis were included and 28-day mortality was 25.5%.The study population was categorized into the modeling set(n=2949)and validation set(n=1260)based on the ratio of 7:3.12 variables which showed significant differences between survivor group and non-survivor group in modeling set were selected for constructing the logistic model.In modeling set,the predictive performance based on the area under the receiver-operating characteristic curve(AUC)were 0.789 for logistic model,0.641 for SOFA score,0.591 for qSOFA score,0.552 for SIRS score,0.661 for SAPS score and 0,670 for OASIS score.In validation set,the AUCs of logistic regression,random forest model,SOFA score,qSOFA score,SIRS score,SAPS score and OASIS score 0.781,0.778,0.647,0.536,0.546,0.654 and 0.670,respectively.The multivariate logistic regression indicated that age,cardiac failure,stroke,vasopressin use,OASIS score,respiratory rate,body temperature,lactate,lymphocyte,red blood cell distribution width,total bilirubin and prothrombin time were independent mortality risk factors for sepsis patients.The prediction model was presented by constructing nomogram.After internal validation,the calibration curve of the model suggested a basic fit with the ideal curve,and the predicted and actual values agreed well(H-L test,p=0.212).The decision curves showed that the column line graph model had some clinical utility on the high risk threshold range(0.2 to 0.8).Conclusion:Age,cardiac failure,stroke,vasopressin use,OASIS score,respiratory rate,body temperature,lactate,lymphocyte,red blood cell distribution width,total bilirubin and prothrombin time are the independent risk factors affecting prognosis in patients with sepsis.A model for predicting 28-day mortality in sepsis is performed by machine learning approach and logistic regression,which is superior to the sepsis related score(SOFA,qSOFA or SIRS)or ICU severity score(SAPS,OASIS).