Prediction of the risk to PICC associated bloodstream infection in cancer patients:a comparative study of two prediction models
Objective To compare the effect of extreme learning machine(ELM)vs logistic regression analysis on prediction of the risk to PICC-related central line associated bloodstream infections(PICC-CLABSI)in cancer patients.Methods Clinical data of 1,146 patients who received PICC,from January 2019 to March 2023,in the Department of Oncology of a ⅢA hospital were analysed.A total of 786 patients who received PICC between January 2019 and December 2021 were assigned to the modelling group,and the rest of 360 patients who received PICC between January 2022 and March 2023 were assigned to the validation group.The risk prediction model was established based on the data of modelling group analysed by Chi-square test,and then by the binary logistic regression to determine the statistically significant variables.Based on the analyses of the two models,a nomogram was plotted to evaluate the fitting and predictive effectiveness.Performance of the two models were evaluated using Hosmer-Lemeshow test as well as the area under the curve(AUC)of receiver operating characteristic(ROC).Risk factors identified by the logistic regression and the PICC-CLABSI risks were used as input and output parameters respectively,to establish an ELM prediction model.The two models were compared in terms of predictive effectiveness using the data of the validation group.Results History of diabetes mellitus,frequency of chemotherapy(≥3 times),maintenance cycle(>7 days),maintenance site(out of hospital),white blood cell count(<3.5×109/L),and albumin(<40g/L)were risk factors for PICC-CLABSI in cancer patients.The logistic regression model demonstrated a good predictability by Hosmer-Lemeshow test(χ 2=5.201,P=0.736),with an AUC-ROC of 0.860(95%CI:0.799～0.922),sensitivity at 0.893,specificity at 0.704 and accuracy at 72.8%.The ELM prediction model exhibited a determination coefficient of 0.823 and mean squared error of 0.051,with a fitting rate at 74.5%,hence it indicated a good predictive power.The ELM model showed a superior predictive power than the logistic regression model.Conclusion The ELM model and logistic regression model,based on logistic regression analysis,offers higher prediction accuracy.It provides valuable guidance to healthcare providers in identification of high risks of PICC-CLABSI for cancer patients.
cancer patientsperipherally inserted central cathetercentral line associated bloodstream infectionlogistic regressionextreme learning machine
万方数据
维普
