QOLP-30. CLINICAL PREDICTIVE MODEL FOR THE DEVELOPMENT OF VENOUS THROMBOEMBOLISM IN GLIOBLASTOMA

Abstract

INTRODUCTION: The risk of venous thromboembolism (VTE) remains high for patients with glioblastoma (GBM) throughout the disease trajectory. Our previous work demonstrated the Khorana scale lacks specificity in this population. We therefore constructed, and attempted to validate a predictive model specific for the development of VTE during adjuvant chemotherapy in glioblastoma patients. METHODS: A prior study of GBM patients treated at MD Anderson (MDACC) during the years 2005‐2011 found from a multivariate analysis that male sex, BMI ≥ 35, KPS ≤ 80, and steroid therapy were significantly associated with the development of VTE. A predictive model from the MDACC cohort was created using these risk factors, and we attempted to validate the model in an independent cohort of GBM patients treated at Henry Ford from 2010‐2015. RESULTS: To develop the model 315 patients from the MDACC cohort were randomly divided into two parts: training (75% of data) used for model building, and validation (25% of data) used for model validation. Using the predictive model, the MDACC validation cohort found 80% sensitivity and 80% specificity. We then validated the model in the Henry Ford cohort of 190 GBM patients of which 50 developed a VTE. In the external validation set, the predictive model was found to have a sensitivity = 78% and specificity = 49.3% (Fisher test p‐value = 0.0008). CONCLUSIONS: Our predictive model for the development of VTE during adjuvant chemotherapy in GBM patients retained high sensitivity in an external data set, however high specificity was lost. While the specificity in our model was higher than in previous studies examining the Khorona scale in GBM patients, further refinement to improve the models reliability to correctly identify people who will not later develop a VTE may be helpful.