# Comparative Evaluation of Risk Assessment Models for Predicting Venous Thromboembolic Events in Cancer Patients with Implanted Central Venous Access Devices

## TL;DR

Cancer patients with implanted devices for chemotherapy face high early risk of blood clots, and current risk models need improvement for better prevention strategies.

## Contribution

The study evaluates and compares the performance of three VTE risk models in cancer patients with implanted devices, identifying ONKOTEV as the most accurate.

## Key findings

- ONKOTEV had the best predictive performance with 74.4% accuracy and 85.7% specificity.
- Nearly half of VTE events were device-related and occurred within 68 days of insertion.
- VTE was independently associated with worse overall survival (hazard ratio of 1.39).

## Abstract

Patients with solid tumors receiving chemotherapy through implanted venous access devices are at increased risk of venous thromboembolism (VTE), but existing risk prediction models are not well validated for this group. We conducted a retrospective study of 446 patients with solid tumors to assess the incidence, timing, and predictors of VTE, and to evaluate the accuracy of the Khorana, COMPASS, and ONKOTEV risk assessment models. We observed an overall VTE rate of 18.4%, with nearly half of events being device-related and occurring early after insertion (median 68 days). Among the models assessed, the ONKOTEV score showed the best predictive performance, while all models showed limited value for device-related events. Importantly, VTE was independently associated with worse overall survival. These findings highlight the need for improved risk stratification and targeted, potentially time-limited prophylaxis strategies in cancer patients receiving chemotherapy through implanted devices.

Background/Objectives: Cancer patients using implanted venous access devices (ICVADs) for chemotherapy are at increased risk of venous thromboembolism (VTE), but the performance of risk assessment models (RAMs) in this setting is understudied. This study evaluated VTE incidence, risk factors, and the predictive performance of the Khorana, COMPASS-CAT, and ONKOTEV models. Methods: We retrospectively reviewed records of adult cancer patients treated with chemotherapy via ICVADs. The cumulative incidence (CI) of VTEs was estimated using the Fine–Gray method, and RAM performance was assessed by sensitivity, specificity, predictive values, accuracy, and AUC. Overall survival (OS) was analyzed using Kaplan–Meier and log-rank tests. Results: A total of 446 patients were included. The most common cancers were colorectal (29.6%), gastric (26%), pancreatic (18.4%), and breast (13.9%). During a median follow-up of 16.5 months, VTEs occurred in 82 patients (18.4%), including 43 (9.6%) that were ICVAD-related. Median time to VTE was 117 days and 68 days for ICVAD-related events. The CI of VTEs was 9% at 1 year and 18.4% at 2 years. ONKOTEV showed the best performance (accuracy of 74.4%, specificity of 85.7%, and AUC of 0.607), with 1-year incidence higher in the high-risk group (28.5% vs. 12.4%, p < 0.001). In contrast, all RAMs showed limited ability for ICVAD-related VTEs. VTE was independently associated with inferior OS (HR 1.39, p = 0.037). Conclusions: Cancer patients with ICVADs face a substantial risk of early VTEs. Among evaluated RAMs, ONKOTEV performed best for overall but not ICVAD-related events. Prospective studies are needed to guide prophylaxis strategies using validated RAMs.