Development and Validation of a Machine Learning-Based Early Warning System for Predicting Venous Thromboembolism Risk in Hospitalized Lymphoma Patients Undergoing Chemotherapy: A Multicentre, Retrospective Cohort Study

Background: Hospitalized lymphoma patients undergoing chemotherapy face a heightened risk of Venous thromboembolism (VTE) due to prolonged treatment and extended bed rest. However, predicting VTE risk early in these patients is challenging. The objective of this study was to develop a machine learning-based early warning system for VTE tailored to this patient group. Methods: Data were collected from 1,141 patients at four academic medical centers from January 2019 to February 2024. Twelve variables were included and six machine learning algorithms were utilized to develop a VTE early warning system. Models were evaluated based on accuracy, precision, sensitivity, specificity, and area under the curve (AUC). Subsequently, the variables importance of each model was analyzed and comprehensively evaluated by permutation importance analysis. Finally, we visualized the VTE early warning system (VTE-EWS) using a nomogram and compared its performance to the classic Khorana Score (KS) system. Results: Among the 1141 patients, 799 from Chongqing University Cancer Hospital were evaluated as a training set, while 342 from other three academic hospitals were the external validation set. In the external validation set, all six models demonstrated strong predictive performance, with accuracy ranging from 0.71 to 0.87, AUC from 0.78 to 0.84, sensitivity from 0.62 to 0.73, and specificity from 0.73 to 0.90. Subsequently, we identified the six most important variables-white blood cell (WBC), D-dimer, central venous catheter (CVC), age, chemotherapy cycles, and Eastern Cooperative Oncology Group performance status (ECOG score)-to be included in the nomogram for scoring and visually predicting the risk of VTE. Finally, compared to the KS, VTE-EWS identified more patients at VTE high risk (24/37) and demonstrated a higher probability of clinical benefit within a range of 1% to 78%. Conclusions: We developed a visual and online VTE-EWS based on machine learning algorithms. Based on simple indicators, VTE-EWS can accurately identify, visually predict, and interpret the risk of VTE in hospitalized lymphoma patients undergoing chemotherapy.