Construction And Validation Of A Risk Prediction Model For Post-ICU Syndrome

Objective:To investigate the factors influencing Post-Intensive Care Syndrome(PICS)and to construct and validate a risk prediction model for PICS.The aim was to accurately predict and screen out people at risk of PICS,to help clinical identification of people at risk,and to provide a reference screening tool for clinical anticipatory care.Methods:(1)Establishment of predictors for the post-ICU syndrome risk prediction model:The main method used was a combination of literature research and Delphi expert correspondence to identify predictor entries.(2)Construction of the post-ICU syndrome risk prediction model: mainly through a prospective cohort study method,917 patients who were hospitalized and successfully transferred out of comprehensive ICUs in two tertiary care hospitals in Bengbu from September 2021 to November 2022 were conveniently selected as the investigation population,and the final development cohort entering the model was 705,with September 2021 to 494(70%)from June 2022 were in the training set and 211(30%)from July to November 2022 were in the validation set.Data were collected during the patients’ ICU stay,followed up at the time of transfer and at one month of transfer,and divided into PICS and non-PICS groups based on the results of the follow-up.The variables were screened using univariate analysis and binary logistic regression analysis respectively,and a line graph model for PICS risk prediction was drawn using R language software based on the screened independent risk factors for PICS.(3)Internal and external validation of the post-ICU syndrome risk prediction column line graph model: The enhanced Bootstrap method was used to resample the training set data 1000 times for internal validation and the validation set data for external validation,and the subject Operating Characteristic Curve(ROC)and Calibration calibration curve analysis was used to evaluate the discrimination and calibration of the PICS risk prediction line graph model,and clinical decision curve analysis(DCA)was used to evaluate the clinical validity of the PICS risk prediction line graph model.Results:(1)Establishment of predictors for the post-ICU syndrome risk prediction model: A total of 41 risk factors for PICS were screened in the literature part of the study;later,a fter Delphi expert correspondence and incorporating real-life clinical scenarios,28 risk f actors were finally obtained as predictors and included as observed variables during the cohort study.(2)Multi-factor logistic regression analysis showed that infection in the ICU(OR=19.562),tracheotomy(OR=10.240),sedation(OR=4.420),vasoactive drugs(OR=3.276),APACHEII score(OR=1.238),and length of ICU stay(OR=1.105)were(p<0.05)we re independent influencing factors of post-ICU syndrome.(3)Internal and external validation of the model: Hosmer-Lemeshow test results showed χ2 = 8.974 P = 0.345 for the training set and χ2 = 5.366 P = 0.718 for the validati on set.ROC curve analysis showed that the Area Under the ROC Curve(AUC)value fo r the training set was 0.935(95%CI: 0.912～0.959),with a best cut-off value of 0.312,sensitivity of 0.867 and specificity of 0.888,and the AUC value of 0.875(95%CI: 0.822～0.927),with a best cut-off value of 0.661,sensitivity of 0.763 and specificity of 0.862 f or the validation set;the Calibration calibration curve The results showed that the predic ti-on model had high prediction calibration;the results of the decision curve analysis sh o-wed that the threshold range in the training set was 0.30～0.80,and the threshold range i-n the validation set was 0.30～0.98,which had good clinical validity.Conclusion:(1)Infection in ICU,tracheotomy,sedation,vasoactive drugs,APACHEII score an d length of stay in ICU were independent influencing factors for post-ICU syndrome.(2)The PICS risk prediction line graph model constructed with the six independent risk factors showed good discrimination,calibration and clinical validity;forming a vis ual graph that makes clinical application easier and faster.(3)The constructed PICS risk prediction line graph model has been validated internally and externally,showing good calibration,differentiation and clinical validity.It can be used as a prediction tool for the risk of PICS and provide a reference tool for clinical staff to identify high-risk groups.