Development of a machine learning–based prediction model for peripherally inserted central catheter (PICC)-related thrombosis in older adult cancer patients

Peripherally inserted central catheter (PICC)-related thrombosis is a common complication among patients with cancer, and older adults may be particularly vulnerable due to age-related physiological changes, and treatment-related factors. This study aimed to develop and evaluate machine-learning-based prediction models for PICC-related thrombosis in older adults with cancer and to identify important predictors associated with thrombosis risk. This retrospective study included 532 older adult patients with cancer (aged ≥ 65 years) who underwent PICC placement. PICC-related thrombosis was defined as the primary outcome. Least absolute shrinkage and selection operator (LASSO) regression was applied for predictor selection, and seven variables were retained, including chemotherapy, fibrinogen, D-dimer, thrombin time, albumin, white blood cell count, and platelet count. Eight prediction models, including logistic regression, XGBoost, support vector machines (SVM), Random Forest, NeuralNetwork, LightGBM(GBM), K-nearest neighbours (KNN), and Adaboost, were developed and compared. Model’s performance was assessed using area under the receiver operating characteristic (AUC), accuracy, sensitivity, specificity and F1 score. Calibration curve and Brier score were used to evaluate the degree of calibration of the model, and decision curve analysis was generated to assess the net clinical benefit. This study also employed explainable machine learning techniques to provide insights into the model’s predictions and outcomes using the SHAP (SHapley Additive exPlanations) approach. Among the 532 patients, 132 (24.8%) developed PICC-related thrombosis. In the training set, several complex machine-learning models achieved high discrimination; however, some exhibited substantial performance deterioration in the testing set, suggesting potential overfitting. Logistic regression demonstrated the best overall performance in the testing set, achieving an AUC of 0.724 (95% CI: 0.620–0.828), with the lowest Brier score (0.149), favourable calibration, and clinical net benefit. SHAP analysis showed that fibrinogen, thrombin time, chemotherapy, and white blood cell count were the major contributors to thrombosis risk prediction. A machine-learning-based prediction model was developed to estimate PICC-related thrombosis risk in older adults with cancer. The model may assist clinicians in early risk stratification and individualised catheter management for older patients undergoing PICC placement. Not Applicable.