Related Experiment Videos
Development and validation of an interpretable machine learning model for venous thromboembolism risk prediction in
Ao Xia1, Jingyuan Liu1, Juanjuan Song2
1Department of Critical Care Medicine, The Affiliated Cancer Hospital of Zhengzhou University & Henan Cancer Hospital, Zhengzhou, China.
Background:
Venous thromboembolism (VTE) is a common complication in patients with lung cancer and remains difficult to predict accurately using existing risk assessment tools. This study aimed to develop and validate a machine learning model for individualized prediction of VTE risk in patients with lung cancer.
Methods:
A total of 7,959 patients with lung cancer from the Affiliated Cancer Hospital of Zhengzhou University and Henan Cancer Hospital were included, among whom 1,333 developed VTE. Patients were randomly assigned to the training and testing cohorts in a 7:3 ratio. Candidate predictors were identified from 67 clinical variables using multivariable logistic regression and least absolute shrinkage and selection operator (LASSO) regression. Seven prediction models were developed based on the selected variables, including logistic regression, Decision Tree, Random Forest, Extreme Gradient Boosting, Light Gradient Boosting Machine, Support Vector Machine, and Artificial Neural Network (ANN). Model performance was evaluated using the area under the receiver operating characteristic curve, F1 score, calibration curves, and decision curve analysis, and SHapley Additive exPlanation was used to interpret feature contributions.
Results:
Twelve variables were selected for model construction: sex, age, anticoagulant use, atherosclerosis, chemotherapy, intermittent pneumatic compression, radiotherapy, CYFRA21-1, NSE, central venous catheter placement, thrombin time, and D-dimer. Among the seven models, ANN showed the best overall performance, with the highest mean cross-validated area under the receiver operating characteristic curve in the training cohort (0.825 ± 0.016), a testing-cohort value of 0.817 (95% CI 0.794-0.840), and the highest F1 score. Calibration analysis showed good agreement between predicted and observed probabilities, and decision curve analysis supported the potential clinical utility of the model across relevant threshold probabilities. SHapley Additive exPlanation further illustrated the contribution and relative importance of each predictor.
Conclusion:
An interpretable ANN model was developed for individualized prediction of VTE risk in patients with lung cancer. The model incorporates 12 routinely available clinical variables and may support identification of patients at high risk of VTE, risk stratification, and clinical decision-making.
