Risk Factors Analysis And Risk Prediction Model Construction Of Hospital-acquired Pressure Injury

Objective: Describe the current situation of pressure injury in hospitalized patients and analyze related risk factors,use machine learning algorithm to build a risk prediction model of pressure injury in hospitalized patients,compare the predictive performance of each model,and select the optimal model.The prediction model with the best predictive performance is externally validated to analyze its generalization and stability.Methods:（1）Search major databases in foreign languages to obtain < Data extraction table for risk factors of hospital-acquired pressure injury patients >.（2）The data of inpatients in two tertiary A-level general hospitals in Lanzhou City were retrospectively collected,and risk factors were obtained by univariate analysis and multivariate analysis using SPSS 26.0 software.（3）The 12 risk factors were processed through the Python 3.8 programming language to construct a Logistic regression model,a decision tree model,a random forest model and a KNN model.The models were compared through indicators such as Accuracy,Precision,F1 value,and AUC value.The prediction performance is compared,and the best model is selected for external validation.Results:（1）Univariate and multivariate analysis of pressure injury in 973 hospitalized patients showed that age（Wald=4.580,P=0.032）,length of hospital stay（Wald=21.710,P<0.001）,and disturbance of consciousness（Wald=4.212,P=0.040）,surgery（Wald=4.095,P=0.043）,skin damage（Wald=4.212,P=0.02）,change in defecation status（Wald=8.235,P=0.004）,self-care（Wald=15.453,P=0.001）,Braden score（Wald=21.503,P<0.001）,indwelling nasogastric tube（Wald=10.017,P=0.002）,indwelling urinary catheter（Wald=5.371,P=0.020）,serum albumin（Wald=14.932,P<0.001）and serum calcium（Wald=7.310,P=0.007）had a total of 12 independent risk factors.（2）The AUC values of the four pressure injury risk prediction models were all greater than 0.8,showing high predictive ability.The horizontal comparison of the models found that the random forest model had the prediction Accuracy（0.96）,Recall（0.92）,Precision（0.93）,F1 value（0.94）and AUC value（0.93）were higher than other models,and were more suitable for the management of pressure injury in hospitalized patients.（3）After external validation of the optimal model showed that the prediction Accuracy（0.81）,Recall（0.79）,Precision（0.76）,F1 value（0.74）and AUC value（0.72）of the random forest model still showed good performance.Conclusions: Four different risk prediction models constructed based on 12 risk factors of pressure injury in hospitalized patients all had good predictive performance,among which random forest had the best predictive performance and better clinical value.Meanwhile,its external prediction performance is also good,it has good generalization,and it has the value of promotion and use.