Construction And Validation Of Death Risk Factors And Prediction Models For Patients With Severe Autoimmune Diseases

Objectives(1)Understand the mortality rate of patients with severe ADs;(2)To explore the risk factors affecting the death of patients with severe ADs;(3)Construct and verify the death risk prediction model for patients with severe ADs.Methods Using a cross-sectional study,using structured query language to carry out data mining on the MIMIC-Ⅲ database,according to the ICD-9 code,the ADs population was screened out from the database.Using STATA 16.0,R 4.2.1 and SPSS 23.0 software to explore the risk factors affecting the death of patients with severe ADs,after screening out the predictive model variables,using Logistic regression,random forest and decision tree to construct a severe ADs death risk prediction model,for three The model is internally validated,and finally the optimal model is selected from the three models according to the evaluation index of the model.Results(1)The fatality rate of severe ADs patients was 28.49%.(2)Univariate analysis of prognostic factors in patients with severe ADs:Patient-related factors: age,body temperature,respiration,blood oxygen saturation,urine output,length of hospital stay and length of ICU stay,chloride ions,sodium ions,bicarbonate,platelets,serum urea concentration,and serum creatinine were statistically different between the two groups The disease-related factors: congestive heart failure,hematological malignancy,organ dysfunction,sepsis,organ failure,depression,and infection had statistically significant differences between the two groups(all P<0.05).<0.05);related factors of medical care: OASIS,SAPSII,MELD,APSⅢ,Elixhauser,SAPS,SIRS,GCS,SOFA,LODS,mechanical ventilation,endotracheal intubation,surgery,liquid electrolyte therapy were statistically different between the two groups Significance(both P<0.05).(3)Prediction model construction and verification:In Logistic regression,mechanical ventilation,depression,chloride ion >106,surgery,and Elixhauser score >12.5 were predictors;in decision tree,SAPSII,Exlixhauser,endotracheal intubation,mechanical ventilation,urine output,APSⅢ,MELD score were Predictors;in random forest,urine volume,body temperature,chloride ions,SAPSII,Exlixhauser score,length of ICU stay,infection,and age were predictors with relatively large influence in the model.Logistic regression,decision tree,random forest: AUC values were 0.699,0.675,0.770;sensitivities were 0.28,0.32,0.38;specificities were 0.94,0.90,0.92;positive predictive values were 0.65,0.54,0.64;classification The correct rates are 0.76,0.73,0.77 respectively.Whether the difference between the AUC of the three models of Logistic regression,decision tree,and random forest is statistically significant(P = 0.0167): there is no statistically significant difference between the AUC of Logistic and decision tree(P=0.547);the AUC of logistic and random forest model The difference was not statistically significant(P=0.044);the AUC of the decision tree(0.675)was smaller than that of the random forest(0.770),and the difference was statistically significant(P=0.002),so between the two models,the random forest Better predictive power.Conclusion(1)The prognosis of patients with severe ADs is still a long-term concern for medical workers.In this study,the fatality rate of patients with severe ADs was 28.49%.(2)The prognosis of patients with severe ADs is affected by many factors: mechanical ventilation,chloride ions,and Exlixhauser entered the model many times,while body temperature,age,APSⅢ,SAPSII,tracheal intubation,urine output,surgery,depression,infection,Variables such as MELD score and ICU length of stay are sometimes included in the prediction model,and sometimes they are not included in the model,indicating that although these factors have an impact on the prognosis of patients,their influence is not as significant as the previous factors.(3)Fatality prediction model for severe ADs patients: Combined with the statistical analysis results of sensitivity,classification accuracy,AUC value and the predictive efficacy of the three models,the random forest model performed best.Finally,this study used the random forest model to predict severe ADs patients prognosis.