OC 63.5 - A Neural Network Approach to Predict Recurrent VTE Based on Coagulation Parameters

Tuesday, June 27, 2023

15:45 – 16:00 ET

Room: 520

Session Available As: On-Demand

Presenting Author(s)

Geke C. Poolen, MSc

PhD student
Center for Benign Haematology, Thrombosis and Haemostasis, Van Creveldkliniek, University Medical Center Utrecht, Utrecht University, The Netherlands
Utrecht, Utrecht, Netherlands

Co-Author(s)

Rolf T. Urbanus, PhD

Associate professor
Center for Benign Haematology, Thrombosis and Haemostasis, Van Creveldkliniek, University Medical Center Utrecht, Utrecht University, Utrecht, the Netherlands
Utrecht, Utrecht, Netherlands
Geert-Jan Geersing, MD, PhD

Associate Professor
Department General Practice, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands.
Utrecht, Utrecht, Netherlands
VS

Veronica Schmalschlager

Research Technician
Department of Functional Coagulation, Synapse Research Institute, Maastricht, the Netherlands Synapse Research Institute, Maastricht, The Netherlands.
Maastricht, Limburg, Netherlands
HM

Harmen Middelveld

Research Technician
Department of Functional Coagulation, Synapse Research Institute, Maastricht, the Netherlands Synapse Research Institute, Maastricht, The Netherlands.
Maastricht, Limburg, Netherlands
Roger E.G Schutgens, MD, PhD, MSc (he/him/his)

Internist-hematologist, Professor
Center for Benign Hematology, Thrombosis and Hemostasis, Van Creveldkliniek, University Medical Center Utrecht and University Utrecht, Utrecht, The Netherlands
Utrecht, Utrecht, Netherlands
Bd

Bas de Laat, MD, PhD

General Director
Department of Functional Coagulation, Synapse Research Institute, Maastricht, the Netherlands
Maastricht, Limburg, Netherlands
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

Background: After a first venous thromboembolism (VTE), the recurrence rate is high, regardless of treatment, and increases if anticoagulation is discontinued. Accurate prediction of a patient’s individual recurrence risk can aid in medication decisions after six months of standard anticoagulation. Nevertheless, it remains difficult to predict which patients will suffer from recurrent VTE with existing clinical prediction models.

Aims: Using neural networks to predict recurrent VTE based on coagulation parameters.

Methods: We developed two neural networks for the prediction of recurrent VTE with samples from the randomized controlled Vista trial which included 869 patients with a history of VTE of whom 96 patients had recurrence within two years. Both neural networks use thrombin generation, thrombin dynamics and coagulation factors (prothrombin, antithrombin, fibrinogen, FVIII, Von Willebrand Factor and active Von Willebrand Factor), in addition to age, sex and body mass index. The first and second neural network were developed with coagulation parameters obtained respectively during and after anticoagulation.

Results: With logistic regression only increased D-dimer levels and lower endogenous thrombin potential (ETP) were associated with an increased risk of recurrent VTE. Therefore, we applied neural networking methodology to integrate all acquired data into a prediction model for recurrent VTE. The first neural network, based on coagulation parameters measured during anticoagulation predicts recurrent VTE with a sensitivity of 82%, and a specificity of 44%. The second neural network, based on coagulation parameters measured after stopping anticoagulation, is more accurate. The sensitivity of this neural network is 99%, and the specificity 45%. Additionally, the positive and negative predictive values of the latter neural network were 67% and 97%, respectively.

Conclusion(s): Neural networks methodology seems suitable for the prediction of complex thrombotic events such as VTE recurrence. However, our findings need to be validated clinically, and further fine-tuning of the neural network could improve the specificity.

Learning Objectives:

Indicate which model/neural network; during coagulation or after coagulation, performed best in predicting recurrent VTE.
Recognize the limitations of the study; only one dataset was used and specificity needs to be improved.
Discuss the opportunities of neural networks in prediction of complex thrombotic events, such as VTE recurrence