A systematically derived risk model for 3-year all-cause mortality after acute venous thromboembolism

Existing risk scores for venous thromboembolism (VTE) mainly support acute triage, whereas medium-term prognosis after acute VTE has been less systematically evaluated.

In COMMAND-VTE Registry 1 (2010-2014; n = 3027), 77 harmonized baseline variables were screened using least absolute shrinkage and selection operator regression with 10-fold cross-validation after single imputation. Variables retained at the best lambda were entered into a Cox proportional hazards model for 3-year all-cause mortality. Temporal validation was performed in COMMAND-VTE Registry 2 (2015-2020; n = 5197), primarily using the imputed cohort.

Six predictors were retained: active malignancy, cardiopulmonary arrest on arrival, age, alkaline phosphatase, hemoglobin, and systolic blood pressure. In derivation, the area under the receiver-operating-characteristic curve (AUC) was 0.862 and the concordance index (C-index) was 0.832. In temporal validation, the AUC was 0.835 and the C-index was 0.806. Calibration showed close agreement in the first and second quartiles, a trend toward overestimation in the third quartile, and underestimation in the fourth quartile. In the complete-case sensitivity cohort, discrimination remained similar (AUC 0.843; C-index 0.817), and Kaplan-Meier curves stratified by quartiles of model-estimated 3-year mortality showed graded separation.

A systematically derived model based on routinely available variables provided a practical estimate of 3-year all-cause mortality after acute VTE across two Japanese registry cohorts.