Development And Validation Of A Short-term Prognostic Model Of In-hospital Cardiac Arrest

BackgroundIn recent years,although the care for in-hospital cardiac arrest(IHCA)has been greatly improved,the prognosis of patients with IHCA is still very poor.According to the Get With the Guidelines-Resuscitation(GWTG-R)registry study,the return of spontaneous circulation(ROSC)of patients with IHCA after cardiopulmonary resuscitation(CPR)had increased from 44%in 2002 to 72.4%in 2019,while the discharge survival rate only increased from 17%to 26.7%.The trend of increase in discharge survival rate was less than the rate of ROSC,which indicates that the follow-up treatment and care of IHCA patients with ROSC still face great challenges.2020 American heart association guidelines for cardiopulmonary resuscitation and emergency cardiovascular care emphasized once again that intensive post-resuscitation treatment should be given to IHCA patients with ROSC.The outcomes of IHCA patients should be accurately predicted and appropriate treatment decisions should be made,which is of great significance for improving the prognosis of IHCA patients.Serval prognostic models of IHCA have been established in many countries and regions.However,the existing prognostic models of IHCA only include pre-arrest or intra-arrest variables,which are mostly used to assist decision-making on whether to do not attempt resuscitation(DNAR)or to terminate resuscitation.They are not completely applicable to IHCA patients with ROSC.Aims1.Explore the prognostic factors affecting IHCA patients with ROSC.2.Develop and verify the prognostic model,select the best model for clinical application,and provide the basis for individualized treatment and nursing plans for the realization in IHCA patients with ROSC.Materials and MethodsThe adult IHCA patients with ROSC≥24h from medical information mart for intensive care(MIMIC)-III were included as the study subjects.The demography,location of arrest,initial rhythm,complications,vital signs and lab tests after ROSC,and hospital outcomes were collected.Patients were divided into death group and survival group according to whether hospital death occurred.Patients were randomly divided into training set and internal validation set in a ratio of 7:3.Based on R language,the factors of discharge outcome in IHCA patients were screened out by logistic regression,random forest,and neural network algorithms in the training set.And prognostic model 1,model 2 and model 3 were developed respectively.In order to simplify the models and facilitate use,model 2 and model 3 are included in the top 10 variables in the random forest and neural network algorithms respectively.The area under the receiver operating characteristic curve(AUROC)and area under the precision-recall curve(AUPRC)was compared to select the best prediction model in the training set and internal validation set respectively.Then the data of a 3a grade hospital from July 2019 to July 2021 were included for external validation of the best model.Results1.A total of 1,119 IHCA patients with ROSC≥24h in MIMIC-Ⅲ were included in this study,66.8%(748 patients)survived to discharge,and 33.2%(371 patients)died in hospital.There were 653 patients in the training set,281 patients in the internal validation set,and 185 patients in the external validation set.2.In multivariate logistic regression,9 variables including age,temperature,heart rate,first diagnose,tumor,red cell volume distribution width(RDW),the potential of hydrogen(PH),blood glucose and blood urea nitrogen(BUN)were prognostic factors in IHCA patients with ROSC.Model 1 was developed based on the above 9 variables.In the development set,the AUROC and AUPRC of model 1 were 0.840[95%confidence interval(CI):0.808-0.871]and 0.687,respectively.In the internal validation set,the AUROC and AUPRC were 0.782(95%CI:0.724-0.840)and 0.620,respectively.3.Variables included in model 2 based on the random forest algorithm were BUN,serum creatinine,heart rate,blood glucose,partial pressure of carbon dioxide(PCO2),PH,RDW,respiratory rate,phosphate and respiratory failure.The AUROC and AUPRC were 0.780(95%CI:0.744-0.817)and 0.606,respectively.In the internal validation set,AUROC and AUPRC were 0.752(95%CI:0.689-0.816)and 0.600,respectively.4.The variables included in model 3 based on the neural network algorithm were international normalized ratio(INR),serum creatinine,pneumonia,PCO2,blood glucose,white blood cell count,WBC,platelet count(PLT),respiratory failure,sepsis,and RDW.The AUROC and AUPRC of model 3 were 0.791(95%CI:0.756-0.826)and 0.612,respectively.In the internal validation set,the AUROC and AUPRC were 0.791(95%CI:0.756-0.826)and 0.616,respectively.5.Combining the test results of the development set and the internal validation set,model 1 performed best.The results of external validation showed that the score still had a good degree of differentiation[AUROC=0.702(95%CI 0.628-0.779)].They were visualized using a line map and named the short-term cardiac arrest prognosis(STCAP)score.Conclusions1.Age,body temperature,heart rate,first diagnose,malignant tumor,RDW,PH,blood glucose and BUN were independent factors affecting the prognosis of IHCA patients with ROSC.2.The STCAP score has better ability to predict short-term prognosis of IHCA patients with ROSC.