PB0942 - Prediction of Venous Thromboembolism in patients with cancer using Neural Network

Monday, June 26, 2023

18:30 – 19:30 ET

Room: Exhibition

Presenting Author(s)

Romy De Laat-Kremers, PhD

Post-doctoral Research Project Leader
Department of Data Analysis and Artificial Intelligence, Synapse Research Institute, Maastricht, the Netherlands
Maastricht, Limburg, Netherlands

Co-Author(s)

ML

Marina Loizou

.
Division of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine/New York Presbyterian Hospital, New York, United States
New York, New York, United States
ST

Scott Tagawa

.
Division of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine/New York Presbyterian Hospital, New York, United States
New York, New York, United States
MY

Madeline Young

.
Division of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine/New York Presbyterian Hospital, New York, United States
New York, New York, United States
CG

Christine Garcia

.
Division of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine/New York Presbyterian Hospital, New York, United States
New York, New York, United States
MS

Manish Shah

.
Division of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine/New York Presbyterian Hospital, New York, United States
New York, New York, United States
PK

Pashtoon Murtaza Kasi

.
Division of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine/New York Presbyterian Hospital, New York, United States
New York, New York, United States
Jacob H. Rand, MD

.
Department of Pathology and Laboratory Medicine, Weill Cornell Medicine/New York Presbyterian Hospital, New York, United States
New York, New York, United States
Bd

Bas de Laat, MD, PhD

General Director
Department of Functional Coagulation, Synapse Research Institute, Maastricht, the Netherlands
Maastricht, Limburg, Netherlands
Maria T. DeSancho, Master in Clinical Science

Professor of Clinical Medicine
Division of Hematology and Medical Oncology, Department of Medicine, Weill Cornell Medicine/New York Presbyterian Hospital, New York, United States
New York, New York, United States

Background: Venous thromboembolism (VTE) is a frequent complication in patients with cancer and can be provoked by the tumor or treatment. High fibrinogen levels are known to be associated with a poorer outcome in patients with cancer. Antithrombin (AT) inhibits key coagulation enzyme thrombin, and AT deficiency is associated with an increased VTE risk.

Aims: To investigate whether neural networks based on coagulation factors and patient characteristics can predict VTE in patients with cancer.

Methods: Coagulation factors were determined in 100 patients of the ongoing Cornell Blood Coagulation in Cancer (COBLOCA) study. Demographics, cancer type and treatment, medications, and VTE events before and during the study were documented. The artificial intelligence method neural networks was used to develop prediction models for VTE in patients with cancer. The predictive capacity of the available variables was determined individually, and the top performing variables for VTE prediction were selected for the neural network. Final input parameters were age, sex, body mass index (BMI), prior VTE, use of anticoagulants and statins, fibrinogen, and AT.

Results: Prostate (36%), lung (16%) and colon (12%) cancer were most common in the COBLOCA cohort, and a total of 21 different cancer types were registered. Thirty-six patients experienced VTE prior to study enrollment (36%). Seven patients experienced VTE during follow-up (median (IQR): 13 (3-19) months). The accuracy of the neural networks developed to predict VTE events was 99.4%. The sensitivity and specificity were 100% and 99%. The positive and negative predictive values were 99% and 100%, respectively. Age was the most important determinant of the neural network (39%), followed by sex (20%), fibrinogen (12%), statin use (10%), AT (9%), anticoagulant use (6%), prior VTE (4%) and BMI (1%).

Conclusion(s): A neural network based on patient characteristics, coagulation parameters and medication predicts VTE in the COBLOCA cohort, and should be further clinically validated.