Construction And Verification Of A Predictive Model For ICU Acquired Weakness Risk

Objectives1.To construct a ICU acquired weakness risk prediction mode.2.To verify the predictive efficiency of ICU acquired weakness risk prediction model.MethodsAccording to the modeling requirements of the clinical risk prediction model,the convenience sampling method was used to select the patients who were treated in the ICU of two tri-service general hospitals in Qingdao from May 2020 to March 2021 as the research object,and the random numbers are generated by random formula in Excel table,70%are randomly selected as the modeling group and the other 30%as the validation group.Based on literature review,the first draft of the ICU-AW risk factors collection form was developed,and the final draft of the form was determined by the method of expert meeting.Meanwhile,general data sheet,ICU-AW risk factor collection form,and Medical Research Council scoring system were used to collect relevant data of ICU patients.Univariate analysis was used to analyze the risk factors of ICU acquired weakness in the modeling group.Binary logistic regression analysis was used to screen the independent risk factors of ICU acquired weakness.Based on the results of regression analysis,the ICU acquired weakness risk prediction model was constructed,and the nomogram was drawn to visually present the constructed model.The research objects of the validation group were selected to verify the model differentiation and calibration degree.The ROC curve,and area under the ROC curve were used to evaluate the differentiation of the model.Hosmer-lemeshow chi square test,the Brier score and calibration curve were used to test the calibration degree of the model.Results1.Based on literature review and expert meeting,the ICU-AW risk factors collection form was determined,which included APACHE II scoring system and patient disease and treatment related data.2.356 subjects were included in this study,including 249 in the modeling group.Univariate analysis results of this study showed that the ICU stay(t=3.761,P<0.01),acute physiology and chronic health status score(APACHEⅡ)(t=4.237,P<0.01),type of disease(χ~2=27.585,P=0.001),mechanical ventilation(χ~2=9.315,P=0.002),high blood lactate(χ~2=9.828,P=0.002),blood lactic acid value(t=2.517,P=0.012),prolonged bed rest(χ~2=8.442,P=0.004),activity restriction(χ~2=6.308,P=0.012),use of neuromuscular blockers(χ~2=4.591,P=0.032)were risk factors for ICU-AW.According to the Binary Logistic regression analysis,the ICU stay,APACHEⅡscore,use of neuromuscular blockers,and high blood lactate were independent risk factors for ICU acquired weakness.3.Based on the results of binary logistic regression analysis,this study constructed the model equation:Y(1)=e~z/(1+e~z),where Z=0.034×the length of ICU stay+0.046×APACHEⅡscore+1.052×use of neuromuscular blockers+0.627×high blood lactate-1.351.4.In the validation group of 107 patients in this study,63 patients developed ICU-AW and 44 patients did not,while the risk prediction model predicted ICU-AW in 66patients and no ICU-AW in 41 patients.Compared with the actual results,the sensitivity,specificity and accuracy of the ICU-AW prediction model were 81.00%,65.90%and74.77%respectively.The area under ROC curve(AUC)was 0.758,95%confidence interval was[0.665,0.851],P<0.01.Hosmer-lemeshow chi-square test showed that P>0.05.The Brier score of this model was 0.205.The calibration curve showed that the predicted value was in good agreement with the measured value.Conclusions1.The hospital stay,APACHEⅡscore,using neuromuscular blockers,and high blood lactate were independent risk factors of ICU acquired weakness.2.The ICU acquired weakness risk prediction model constructed in this study has good predictive efficiency,and could be used to promote better clinical strategies for identifying patients at high risk of ICU acquired weakness.