Enhancing venous thromboembolism risk prediction after immunotherapy for lung cancer

Assessing venous thromboembolism (VTE) risk after immunotherapy remains important for lung cancer management. We analyzed 2,300 patients receiving first-line immunotherapy from two centers, randomly assigned to training (70%), validation (15%), and internal test (15%) sets, and included 491 patients from an independent external center for external validation. Five feature-selection methods and five machine-learning algorithms were compared to develop a 6-month VTE prediction model. The Lasso-logistic model showed the best performance, with areas under the curve of 0.692 and 0.728 in the internal and external test sets, respectively, outperforming Khorana, Padua, PROTECHT, ONKOTEV, and COMPASS-CAT scores (all p < 0.05). The high-risk group had a higher cumulative VTE incidence than the low-risk group (12.3% vs. 4.8%, p = 0.006). Shapley additive explanations (SHAP) analysis identified D-dimer and Eastern Cooperative Oncology Group (ECOG) performance status as the most influential predictors, supporting individualized thromboprophylaxis decisions.