Construction and validation of secondary ectopic risk prediction model for PICC catheters in preterm infants based on random forest algorithm

ObjectiveTo investigate the key risk factors for secondary ectopics of catheter after central venous catheter insertion(PICC) through peripheral veins,and to construct and validate a risk prediction model based on the random forest algorithm.MethodsThe data of 590 preterm infants hospitalized in the department of neonatology of a tertiary specialty hospital in Shanxi province with PICC catheterization were retrospectively collected,and the data were randomly divided into training set(<italic>n</italic>=413) and validation set(<italic>n</italic>=177) in a ratio of 7∶3.Eighteen clinical indicators were selected as predictors for whether secondary ectopic of the catheter occurred at the tip after PICC catheterization in preterm infants.Based on the random forest algorithm,a risk prediction model for secondary ectopic ectopic of PICC catheters in preterm infants was constructed,the importance of risk factors was ranked,and a SHAP plot was plotted to explain the contribution of each variable to the prediction of model output.The confusion matrix analysis of the model was carried out using the validation set data,and the prediction effect of the model was evaluated by using the accuracy,sensitivity,specificity and receiver operating characteristic(ROC) curves.ResultsThe incidence of secondary ectopic of PICC catheter in 413 preterm infants in the training set was 39.0%,and the key factors screened by the random forest algorithm were the first X⁃ray localization result,secondary catheter fixation,mechanical ventilation,and weight gain rate,respectively.In the validation set,the accuracy of the risk prediction model was 81.92%,the sensitivity was 71.21%,the specificity was 88.29%,the positive predictive value was 78.33%,the negative predictive value was 83.76%,and the AUC value was 0.870(95%CI:0.817⁃0.923).ConclusionsThe risk prediction model constructed based on the random forest algorithm demonstrates good predictive performance.It could potentially provide a scientific basis for the safe and efficient clinical use of PICC in premature infants.