Optimizing Cancer Associated Thrombosis (CAT) Risk Assessment Model at a Safety-Net Healthcare System

Introduction: Cancer associated thrombosis is a preventable complication that impacts the quality of life of patients with cancer. The Khorana score (KS) is the most widely used risk assessment model (RAM) to predict venous thromboembolism (VTE) in ambulatory patients undergoing chemotherapy. Potential limitations of the score include modest discrimination and small proportion of patients in the highest risk subgroup. We aimed to examine if a clinical informatics approach incorporating race/ethnicity, cancer staging, type of systemic therapy, and other known VTE risk factors from the electronic health record (EHR) can improve the RAM. Methods: We performed a retrospective cohort study at Harris Health System (HHS), a safety-net healthcare system that provides care for underserved minorities and uninsured patients in Houston. We created an integrated database that linked consecutive patients with newly diagnosed invasive cancer in the cancer registry with structured data from EPIC Caboodle database 2011-2020. Inclusion/exclusion criteria are shown in Figure 1. We followed patients from time of initial systemic therapy to time of first VTE, death, or loss of follow-up. VTE was defined as radiologically confirmed pulmonary embolism (PE), proximal or distal lower extremity deep vein thrombosis (LE-DVT), catheter-related DVT (CR-DVT), or splanchnic vein thrombosis (SVT) in inpatient or outpatient setting. We used acute, chronic or historical VTE ICD9/ICD10 facility billing codes to assess for potential events and confirmed incident and recurrent events through medical record review. We used multivariable Cox regression to assess potential risk predictors. The model was built iteratively to expand upon the KS. Kaplan Meier failures curves were used to estimate the VTE incidence. C statistic was assessed with binary outcomes at 3- and 6-month. Results: A total of 4,546 patients with newly diagnosed cancer receiving 1 st line systemic therapy met the inclusion/exclusion criteria. Relevant demographics showed a median age of 54 (IQR 46-61), 57% female, 50% Hispanic, 27% Black, and 75% uninsured. Most common cancer types included breast (17%), colorectal (13%), lung (10%), and non-Hodgkin lymphoma (8%); 32% of patients had metastatic disease. First-line systemic therapy included 89% cytotoxic chemotherapy, 9% small molecule targeted +/- endocrine therapy, and 2% PD-1/PD-L1 immunotherapy. Only 1% had remote VTE history after excluding 317 patients already on therapeu