Construction And Validation Of Risk Predictive Models For Pressure Injury In Critical Care Patients

【Objective】In order to explore risk factors related pressusre injury in patients of the Intensive care unit(ICU),construct and establish the nomogram model of critical care patients.It is hoped that the predictive model can accurately predict the risk of pressure injury,so as to facilitate the efficient screening of patients at high risk of pressure injury,and to provide basis for accurate implementation.【Method】In this study,a total of 4,540 critical care patients were selected from a multi-center survey,conducted in 28 hospitals of 13 provinces from March to December 2019.The assessment process was unified by two trained nurses,who jointly turned over the patients and examined the skin from head to toe,collected the general information,nutritional indicators and skin assessment data of all patients by Braden scale.The first 70%of the collected data were used for model construction,internal validation and evaluation,and the last 30%of data were used as the model validation data set for external validation.Taking the occurrence of ICU pressure injury as the dependent variable,and age,sex,length of stay in ICU,serum albumin level,hemoglobin level,skin tear,IAD,nutrient intake pathway,perception,humidity,mobility,activity,nutrient intake,friction and shear force were used as independent variables.Univariate analysis was used for preliminary screening of variables.Variables with significant differences(P<0.05)were included in logistic multivariate regression analysis(Forward LR),and both entry and exclusion criteria were set at0.05.The coefficientβ,Odds ratio(OR)and 95%Confidence interval(CI)were calculated.All the variables that eventually entered the Logistic model were introduced,and the software R was used to draw the nomogram model(the new model).Bootstrap was used for the internal validation of the model with the method of 1,000 repeated sampling.The prediction performance of Braden scale was taken as the standard to evaluate the new model,and the Receiver operating characteristic curve(ROC)of the new model and Braden scale was drawn.The results of Area under cure(AUC)and Hosmer-Lemeshow goodness-of-fit tests for the two assessment tools were calculated to evaluate the tools’degree of distinction and consistency.And the sensitivity,specificity,and Youden indices were calculated.The clinical practicability of the new model and the Braden scale was compared by Decision curve analysis(DCA).【Results】The incidence of pressure injury in ICU patients was 10.95%.Logistic multivariate regression analysis was showed that age(OR:1.02,95%CI:1.01～1.02),length of stay in ICU(OR:1.02,95%CI:1.01～1.02),serum albumin level(OR:0.91,95%CI:0.89～0.93),hemoglobin level(OR:0.99,95%CI:0.88～1.00),skin tears(OR:4.29,95%CI:2.75～6.67),incontinence-associated dermatitis(OR:2.94,95%CI:1.89～4.61),the ability of sensory(mild limitation[OR:3.39,95%CI:2.24～5.13],mostly limitation[OR:2.36,95%CI:1.49～3.75],completely limitation[OR:3.40,95%CI:2.05～5.63]),and the ability of mobility(mild limitation[OR:4.88,95%CI:1.47～16.24],mostly limitation[OR:2.39,95%CI:2.24～5.61).7.35,95%CI:2.25～24.08],completely limitation[OR:13.29,95%CI:3.93～44.91])were risk factors for pressure injury(all P<0.05).According to above of risk factors and based on the modeling data set to build risk predictive model,including age,length of stay in ICU,albumin,hemoglobin,skin tear,IAD,the ability of sensory and mobility,AUC of the new calculation model was 0.873(95%CI:0.86～0.89),the sensitivity,specificity and Youden index were 84.5%,74.8%and 59.3%respectively,the test of H-L likelihood ratio chi-square value forχ~2=3.86(P=0.869,df=8),via the bootstrap sampling to verify the risk predictive model of internal calibration AUC was0.868.In addition,the AUC of the prediction model verified externally based on the validation data set was 0.934(95%CI:0.91～0.96),indicating that the Logistics regression model has good differentiation and consistency.The value of AUC,sensitivity,specificity and Youden index of Braden scale were 0.845(95%CI:0.83～0.87),83.1%,72.2%and 55.3%,respectively.The prediction performance of the new model was statistically different from that of Braden’s scale(z-statistic=3.020,P<0.05).The prediction performance of the new model was better than that of Braden’s scale.When the Youden index was the largest,the cutoff value of the new model was 10.7%(the score of Nomogram model was 70 points),and the prediction performance of the new model was the best.Moreover,DCA showed that the clinical practicability of the new model was far superior to Braden scale.【Conclusion】This study screened the risk factors related pressure injury in ICU and determined the eight factors,age,the length of days in ICU,serum albumin level,hemoglobin level,skin tears,IAD,the ability of sensory and mobility for independent predictors of pressure injury,and constructed the Nomogram model for critical care patients.The new model is superior to Braden scale in terms of predictive performance and clinical practicability in improving patient outcomes.