Abstract
BACKGROUND: Thromboembolism (TE) is a serious complication in lymphoma, driving excess morbidity and mortality. Existing prediction tools perform suboptimally in lymphoma-specific settings. METHODS: We retrospectively analysed 790 newly diagnosed lymphoma patients (January 2019-December 2021). Patients were randomly split 7:3 into development and internal-validation cohorts. Forty-eight candidate predictors were screened with LASSO, followed by multivariable Cox modelling to construct a nomogram. Discrimination and calibration were assessed at 6, 12 and 24 months using time-dependent ROC analysis and bootstrap calibration. RESULTS: TE occurred in 77/790 patients (9.8%). Independent predictors were ECOG performance status, prior venous thromboembolism (VTE), coronary artery disease, central venous catheterisation, and APTT category. The nomogram showed good discrimination: AUCs were 0.813, 0.818 and 0.733 at 0.5, 1.0 and 2.0 years in the development cohort, and 0.724, 0.731 and 0.659 in the validation cohort. Conventional scores performed poorly in this population (e.g., at 1 year ThroLy 0.587 vs. Khorana 0.527). Calibration plots indicated close agreement between predicted and observed risks. Patients who experienced TE had poorer overall survival, with the greatest divergence in survival curves occurring within the first six months after diagnosis. CONCLUSIONS: This lymphoma-specific model improves TE risk stratification and can inform individualised prophylaxis and early monitoring. External, multi-centre validation is warranted to confirm generalisability.