Machine learning-based prediction of peripherally inserted central catheter-related thrombosis in hematological
Jianhong Su1, Junlin Hu1, Wentian Wang1
1Department of Clinical Laboratory, Longyan First Affiliated Hospital of Fujian Medical University, Longyan, 364000, China.
Purpose:
Peripherally inserted central catheter-related thrombosis (PICC-RT) is a common and serious complication in patients with hematological malignancies, leading to treatment interruptions and increased morbidity. Traditional risk assessment models are often static and lack hematology-specific predictors, resulting in suboptimal performance. We aimed to develop and validate a robust machine learning (ML) model that integrates static clinical data with dynamic biomarkers to provide individualized risk prediction.
Methods:
This multicenter retrospective study screened 5420 patients, ultimately including 4015 adult patients with hematological malignancies who underwent PICC insertion across five participating centers. The cohort was randomly partitioned into a training set (n = 2810) and an independent testing set (n = 1205). Nine core predictors were selected via LASSO regression. We compared three algorithms, logistic regression (LR), random forest (RF), and XGBoost, using the area under the curve (AUC), calibration curves, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were utilized for model interpretation.
Results:
The incidence of symptomatic PICC-RT was 5.8% (n = 232). In the independent testing set, the XGBoost model significantly outperformed the LR baseline (AUC: 0.862 [95% CI: 0.825-0.899] vs. 0.774 [95% CI: 0.730-0.818], P < 0.001). Independent predictors identified included history of VTE, use of triple-lumen catheters, immunomodulatory drug (IMiD) use, and peak values of D-dimer and the neutrophil-to-lymphocyte ratio (NLR). SHAP analysis revealed a distinct threshold effect for peak D-dimer, where risk accelerated sharply once levels exceeded 1.5 mg/L. To provide a balanced predictive profile given the low baseline incidence, the model demonstrated a high negative predictive value (NPV) of 98.6% alongside a positive predictive value (PPV) of 18.8%. An open-access, web-based risk calculator was developed to facilitate clinical application (freely available at https://xxcc1114.shinyapps.io/PICC_Calculator/ ).
Conclusions:
The XGBoost model, integrating dynamic biomarkers and hematology-specific factors, provides superior personalized risk stratification for PICC-RT. This tool enables the identification of high-risk patients who may benefit from enhanced surveillance or targeted thromboprophylaxis while safely identifying low-risk individuals.