Risk Assessment Models for Predicting Venous Thromboembolism in Patients with Pancreatic Cancer

Data on the performance of the Khorana, PROTECHT, and ONKOTEV risk assessment models (RAMs) to predict venous thromboembolism (VTE) in patients with pancreatic cancer (PC) receiving outpatient chemotherapy remain limited. We performed a head-to-head comparison of these RAMs in patients with newly diagnosed PC enrolled in the nationwide, multicenter, and prospective BACAP cohort.

The Khorana, PROTECHT, and ONKOTEV scores were calculated at enrollment prior to chemotherapy. Patients were stratified into intermediate- and high-VTE-risk groups according to each RAM. The primary study outcome was VTE at a 6-month follow-up. The accuracy and discriminatory performance of the scores were assessed by calculating time-dependent Brier scores and c-indexes. Sub-distribution hazard ratios (SHRs) between high- and intermediate-risk patients were estimated.

Of 762 PC patients, 73 developed VTE within 6 months. In the competing risk analysis, the cumulative incidence of VTE at 6 months was 16.4% (95% CI, 13.8-19.1). The time-dependent Brier score was 0.14 (95% CI, 0.12-0.15) for all scores, indicating well-calibrated predictions. The respective time-dependent c-index of the Khorana, the PROTECHT, and the ONKOTEV scores was 0.50 (95% CI, 0.46-0.55), 0.50 (95% CI, 0.49-0.51), and 0.53 (95% CI, 0.48-0.58), indicating poor discrimination. The SHRs between high- and intermediate-risk patients ranged from 1.05 (95% CI, 0.76-1.44) for the ONKOTEV score to 1.06 (95% CI, 0.77-1.45) for the Khorana score.

In newly diagnosed PC patients receiving outpatient chemotherapy, the Khorana, PROTECHT, and ONKOTEV scores demonstrated a poor performance in predicting VTE at 6 months, highlighting the need for new tools to guide thromboprophylaxis decisions.