Derivation and validation of the Caravaggio score for the risk stratification for recurrence in patients with cancer-associated venous thromboembolism

Background: In patients with cancer associated venous thromboembolism (CAT), risk factor-based scores for recurrence could drive clinical management. The aim of this study in patients with CAT was to develop and validate a risk score for recurrent venous thromboembolism (VTE) during anticoagulation: the Caravaggio score. Methods: The Caravaggio score was developed in patients included in the Caravaggio trial and then externally validated in patients included in the TESEO registry. Potential predictors (univariate p-value ≤ 0.1) for recurrence were evaluated in a multivariable Cox regression model with death unrelated to VTE as competing event. Candidate predictors were identified and scored based on clinical relevance and β-coefficient. Patients were then categorized in three risk classes. The performance of the Caravaggio score was assessed by discrimination (c-statistics), sensitivity, specificity, positive and negative predictive value (NPV). Results: Symptomatic VTE, ovarian and/or uterine cancer, pancreatic cancer, metastatic cancer, adenocarcinoma histological subtype, and pharmacological anticancer treatment were included in the score. In the derivation cohort, the incidence of recurrent VTE in the high, intermediate and low-risk groups was 11.6, 7.7 and 2.5 %, respectively. Incidences in the validation cohort were 8.0, 3.5 and 1.7 %, respectively. c-statistics in derivation and validation cohorts were 0.641 (95 % CI 0.584–0.698) and 0.606, (95 % CI 0.557–0.653), respectively. The NPV for low vs. intermediate/high-risk group was 98 % (95 % CI 95–99) in the derivation and 98 % (95 % CI 97–99) in the validation cohort. Conclusions: The Caravaggio score is simple and able to stratify patients with CAT for the risk for VTE recurrence.