The Khorana, PROTECHT, and CONKO risk scores demonstrated limited discriminative performance (AUC ~0.57) and low sensitivity (~30%) for predicting venous thromboembolism in Brazilian cancer outpatients.
Cohort (n=803)
None
Yes
Do established VTE risk assessment models (Khorana, PROTECHT, CONKO, Vienna CATS) accurately predict symptomatic VTE in adult ambulatory cancer patients?
Current VTE risk assessment models have limited sensitivity and discrimination for predicting VTE in Brazilian ambulatory cancer patients, suggesting the need for locally validated models.
Effect estimate: AUC 0.567 (95% CI 0.462-0.672)
Cancer-associated thrombosis (CAT) is a relevant cause of morbidity and mortality in oncology patients. Although several venous thromboembolism (VTE) risk assessment models (RAMs) are recommended to guide thromboprophylaxis in ambulatory cancer patients, their performance varies across populations. Data from Brazilian cohorts are limited, and the real-world performance of these models in this setting remains unclear. We conducted a prospective external validation study of four established VTE RAMs-Khorana, PROTECHT, CONKO, and Vienna CATS-in ambulatory cancer patients from a Brazilian tertiary-care registry. A total of 803 adult patients with complete data were followed for 12 months for objectively confirmed symptomatic VTE. Predictors included baseline variables from each RAM, applied according to their original definitions. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC), with 95% confidence intervals (CI) obtained by bootstrap resampling. Vienna CATS was evaluated in a biomarker-defined subcohort of 470 patients. During follow-up, 36 of 803 patients (4.5%) developed VTE. The Khorana (AUC 0.567; 95% CI 0.462-0.672), PROTECHT (AUC 0.575; 95% CI 0.470-0.680), and CONKO (AUC 0.567; 95% CI 0.464-0.671) scores showed limited and comparable discrimination, with sensitivities of approximately 30% at the conventional high-risk threshold (≥3 points). In the Vienna CATS subcohort, 24 of 470 patients (5.1%) developed VTE. Vienna CATS demonstrated modest but statistically significant discrimination (AUC 0.672; 95% CI 0.559-0.779) over chance. However, pairwise comparisons using DeLong's test showed no significant differences in AUC between the clinical-only scores and the same subcohort. Additional analyses suggested heterogeneity in risk across component combinations despite equal point weighting. Among patients with very-high-risk tumors plus one additional component, VTE incidence varied substantially depending on the specific factor present, ranging from 0% for BMI to 38.5% for low hemoglobin. Overall, current RAMs demonstrated limited sensitivity for identifying patients who developed VTE in this Brazilian cohort. These findings indicate that RAMs alone may be insufficient to guide thromboprophylaxis decisions in this setting, particularly given their limited sensitivity and the occurrence of VTE events among patients classified as low risk. Furthermore, the results support the need to develop locally validated models for Brazilian patients, as well as to explore novel biomarkers, alternative predictor thresholds, and potentially nonlinear approaches to variable weighting.
Abreu et al. (2026) conducted a cohort in Cancer-associated venous thromboembolism (n=803). Khorana score was evaluated on Discriminative performance (AUC) for 12-month VTE outcome (AUC 0.567, 95% CI 0.462-0.672). The Khorana, PROTECHT, and CONKO risk scores demonstrated limited discriminative performance (AUC ~0.57) and low sensitivity (~30%) for predicting venous thromboembolism in Brazilian cancer outpatients.