Construction And Validation Of A Risk Prediction Model For Refeeding Syndrome In ICU Patients

Purpose:1.Explore independent risk factors for refeeding syndrome in ICU patients.2.Construct a risk prediction model for ICU patients with refeeding syndrome,verify and evaluate the model through external validation data,and provide a convenient and feasible prediction tool for clinical medical staff with refeeding syndrome.Methods:This prospective cohort study was divided into two parts.The first part is the risk factor analysis of refeeding syndrome in ICU patients: Clinical data of ICU patients from a grade-A hospital in Guangzhou from January 2022 to June 2022 was retrospectively collected as a modeling group.Patients were divided into a case group and a control group according to the occurrence of refeeding syndrome.Risk factors related to refeeding syndrome were determined by univariate analysis and binary logistic regression analysis.The second part is the construction and verification of the risk prediction model for ICU patients with refeeding syndrome: the logistic regression model and the random forest model are constructed according to the modeling group,and the internal verification is carried out.The clinical data of ICU patients from a grade-A hospital in Guangzhou from July 2022 to September 2022 was prospectively collected as the validation group,and the two models were externally validated.Sensitivity,specificity,and subject working characteristic curves were used to evaluate the clinical application effects of the logistic regression model and the random forest model.Results:1.In this study,291 ICU patients were eventually included in the modeling group,including 107 in the refeeding syndrome group and 184 in the non-refeeding syndrome group.The incidence of refeeding syndrome in patients with enteral nutrition in the ICU was 36.8%.Univariate analysis showed that age,body mass index,history of diabetes,blood transfusion therapy,insulin use,caloric intake,protein intake,APACHE II score,and NRS 2002 were different in the refeeding syndrome group and the non-refeeding syndrome group.There were 18 variables,including score,hemoglobin,glutamic oxalacetic transaminase,total protein,albumin,globulin,urea,serum calcium,fibrinogen,and procalcitonin,and the differences were statistically significant(P<0.05).The results of multivariate analysis showed that diabetes history(OR = 3.088,95% CI: 1.566~6.088),protein intake(OR = 22.667,95% CI: 10.879~47.227),albumin(OR = 0.945,95% CI: 0.916~0.975),and APACHEII score(OR = 2.471,95% CI: 1.394~4.381)were independent risk factors for refeeding syndrome in ICU patients(P<0.05).2.According to the screening of Lasso regression variables,seven variables,including age,protein intake,total protein,serum calcium,diabetes history,the APACHEII score,and the NRS 2002 score,were included in the Logistic regression model.The importance ranking of characteristic variables and the best characteristic number of the model,protein intake,diabetes history,serum calcium,insulin use,globulin,total protein,C-reactive protein,alanine aminotransferase,total bilirubin,oxygen partial pressure,creatinine,lactic acid,partial pressure of carbon dioxide,urea,procalcitonin,APACHEII score,Twenty-one characteristic variables,including the NRS 2002 score,gender,gastrointestinal decompression,diuretics,and sepsis,were included in the random forest model.The AUC of the internal verification Logistic prediction model was 0.891(95% CI: 0.849~934),the sensitivity was 69.16%,and the specificity was95.65%.The AUC of the random forest model was 0.815(95% CI: 0.714~0.893),the sensitivity was 71.47%,and the specificity was 88.74%.The AUC of the externally verified Logistic model was 0.783(95% CI: 0.682~0.883),the sensitivity was 65.52%,and the specificity was 83.33%.The AUC of the random forest model was 0.839(95% CI: 0.731~0.926),the sensitivity was 58.97%,and the specificity was 91.12%.Conclusion:1.Diabetes history(OR=3.088,95%CI: 1.566~6.088),protein intake(OR=22.667,95%CI: 10.879~47.227),albumin(OR=0.945,95%CI: 0.916~0.975)and APACHEⅡscore(OR=2.471,95%CI: 1.394~4.381)were independent risk factors for refeeding syndrome in ICU patients(P < 0.05).2.According to the external verification results,the clinical application effect of the random forest model constructed in this study is superior to that of the Logistic prediction model.