Comparative Predictive Modeling for PICC Line Complications in Oncology: A Retrospective Study

Abstract

Aims/Background Peripherally inserted central catheter (PICC) are increasingly used in cancer treatment, offering significant therapeutic benefits while also posing risks for complications such as infection, thrombosis, and catheter migration. Effective prediction and management of these complications are crucial to optimizing patient outcomes and reducing healthcare costs. Methods This retrospective study analyzed PICC line insertion in 266 cancer patients implemented from January 2019 to December 2023 at a regional healthcare facility in China. Using least absolute shrinkage and selection operator (LASSO) logistic regression, we identified key factors influencing PICC line complications and developed a tailored nomogram for individual risk assessment. The efficacy of the model was compared with support vector machine (SVM), random forest, and gradient boosting machine (GBM) using receiver operating characteristic (ROC) and decision curve analysis (DCA) metrics. Results Factors such as body mass index (BMI), diabetic status, and age were found to be significant predictors of PICC line complications. The LASSO model demonstrated superior predictive capability (area under the curve, AUC = 0.79) over SVM (AUC = 0.40), random forest (AUC = 0.70), and GBM (AUC = 0.64). A tailored nomogram was developed for clinical use, enabling personalized risk evaluation. Conclusion The study underscores the utility of LASSO logistic regression in the personalized risk evaluation of PICC line complications, recommending its integration into clinical practice. The tailored nomogram provides a practical tool for clinicians to enhance care customization for patients requiring PICC lines, thereby improving treatment outcomes and patient safety.