OC 36.2 - Machine-learning based risk prediction of cancer associated thrombosis: internal, external, and pre-clinical validation studies

Monday, June 26, 2023

11:00 – 11:15 ET

Room: 510

Session Available As: On-Demand

Presenting Author(s)

Rushad Patell, MD (he/him/his)

Assistant Professor of Medicine
Beth Israel Deaconess Medical Center, Boston, MA, USA
Boston, Massachusetts, United States

Co-Author(s)

DK

Dimitra Karagkouni

Post Doctoral Research Fellow
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
MB

Marisa Brake

Research Fellow
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
Anna Falanga, MD (she/her/hers)

Professor of Hematology
Department of Immunohematology and Transfusion Medicine, Papa Giovanni XXIII Hospital, Bergamo, Italy
Bergamo, Lombardia, Italy
Marina Marchetti, PhD (she/her/hers)

Hemostasis Lab Director
Department of Immunohematology and Transfusion Medicine, Papa Giovanni XXIII Hospital, Bergamo, Italy
Bergamo, Lombardia, Italy
LR

Laura Russo

Researcher
Department of Immunohematology and Transfusion Medicine, Papa Giovanni XXIII Hospital, Bergamo, Italy
Bergamo, Lombardia, Italy
Simon Mantha, MD, MPH

Attending Physician
Memorial Sloan Kettering Cancer Center
New York, New York, United States
Marc Carrier, MD

MD, MSc
Ottawa Hospital Research Institute, Ottawa, ON, Canada
Ottawa, Ontario, Canada
PW

Philip S. Wells

MD, MSc
Department of Medicine, Ottawa Hospital Research Institute, University of Ottawa, Ottawa, Canada.
Ottawa, Ontario, Canada
DN

Donna Neuberg

Senior Lecturer
Dana Farber
Boston, Massachusetts, United States
SR

Siyang Ren

Statistician
Dana Farber
Boston, Massachusetts, United States
RF

Robert Flaumenhaft

Professor
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
IV

Ioannis Vlachos

Assistant Professor
Beth Israel Deaconess Medical Center
Boston, Massachusetts, United States
Sol Schulman, MD, PhD (he/him/his)

Assistant Professor of Medicine
Division of Hemostasis and Thrombosis, Beth Israel Deaconess Medical Center, Boston, MA
Boston, Massachusetts, United States
Jeffrey I. Zwicker, MD

Chief of the Hematology Service
Department of Medicine, Hematology Service, Memorial Sloan Kettering Cancer Center, New York City, NY
New York, New York, United States

Background: Cancer increases the risk of thrombosis in cancer. There is considerable interest in developing accurate prediction models to identify patients who may benefit from primary thromboprophylaxis.

Aims: We aimed to employ machine learning to identify a panel of plasma proteins to predict cancer associated thrombosis and elucidate mechanistic insights.

Methods: We performed a proximity extension assay of 1,161 plasma proteins in samples collected from the Hypercan Study, a prospective of newly diagnosed lung and gastric cancer patients. Bayesian probabilistic machine learning approach was employed for model training and ROC defined for venous thromboembolism (VTE). External validation studies performed in the placebo-arm of the AVERT trial. Thrombin antithrombin complexes (TAT) by ELISA in ~7-week-old wild type and CD200R1-/- mice.

Results: Machine learning was applied to cohort of 77 lung cancer and 101 gastric cancer patients. The model was trained on 11-proteins and 3 clinical features (age, sex, and history of VTE) following 5-fold validation. The mean AUC was 0.84 for the prediction of VTE compared with 0.48 for the Khorana Score, and the ROC-determined prediction applied to both lung and gastric cohorts (Fig). When applied to 72 patients with mixed solid tumors of the AVERT, the cumulative incidence of thrombosis at 6 months was 23.7% in the high-risk arm compared with 5.9% in the low-risk (Gray test P=0.03, AUC=0.71). The panel included several proteins whereby reductions predicted increased VTE including CD200R1, an inhibitory receptor that limits pro-inflammatory response. We assessed the basal procoagulant state of CD200R1-/- mice and wild-type controls and observed a significant shortening of PT and increase in basal TAT complex formation.

Conclusion(s): A Bayesian probabilistic machine learning model identified proteins that are predictive of thrombosis in different cohorts of patients with solid tumor malignancies, identifying the reduction in CD200R1 as potentially novel pro-thrombotic mechanism in cancer.

Learning Objectives:

Upon completion, participant will be able to describe gaps in current prediction modeling for thromosis in patients with cancer
Upon completion, participant will be able to describe the potential role of proteomics for novel biomarker discovery in prediction of cancer associated thrombosis
Upon completion, participant will be able to describe the potential role of machine learning for novel biomarker discovery in prediction of cancer associated thrombosis