Personalized Cancer-Associated Thrombosis Risk Assessment: Integration of Plasma Proteomics, Clinical Characteristics, and Machine Learning


Jeffrey Zwicker, M.D.
Beth Israel Deaconess Medical Center
Boston, Massachusetts, U.S.

Venous thromboembolism (VTE) is a common cause of morbidity and mortality in cancer patients and leads to a significant increase in health care costs. Numerous studies have established that thrombosis is a common complication for cancer patients, contributing to the second-leading cause of mortality in cancer patients. These patients frequently suffer from multiple comorbidities, having both a greater risk of VTE recurrence and bleeding compared to non-cancer patients. Specifically, certain malignancies such as pancreatic cancer or therapies including IMiDs are associated with an increased risk of developing VTE in cancer patients. Accurate risk assessment cancer-associated thrombosis (CAT) can inform optimal patient selection for thromboprophylaxis.  Unfortunately, a uniform biomarker does not exist and the current risk scoring methods exhibit modest accuracy. Machine learning can capture rich multidimensional patient information and identify patients at increased risk for CAT. This was the foundation of the presentation delivered by Jeffrey Zwicker of Beth Israel Deaconess Medical Center in Boston. The aim of his research was to identify a novel set of variables predictive of CAT using machine learning to analyze a comprehensive feature space combining proteomic and clinical features in a well-curated prospective cancer cohort.

Machine learning predictive models were built using plasma protein concentrations and clinical/laboratory information from the accompanying patient metadata and were compared to the Khorana Score to predict VTE.  Zwicker commented that of the total cohort of 183 cancer patients, 103 had gastric cancer and 80 lung cancer, and 58% had metastasis at baseline. Approximately one-third of patients in the cohort developed a thrombotic event on follow-up. A machine learning–driven model identified 10 plasma proteins and six clinical features predictive of thrombosis. The algorithm predicted VTE with a higher accuracy than the Khorana Score. This led to the conclusion of the presentation delivered on Monday, July 19, 2021, by Zwicker that there is reason to utilize machine learning in conjunction with novel biomarker discovery for the creation of a next generation high-accuracy CAT-predictive modeling.


Previous Article Diagnosing Inherited Platelet Disorders: Immunofluorescence on the Blood Smear in Comparison With Genetic Testing
Next Article The Role of Platelet Transfusions to Manage Bleeding: Lessons From Randomized Trials (Including Neonates and Children)
Print