Prediction of Venous Thromboembolism Utilizing Machine Learning Models - (HJF 477-17)

Background

Individuals exposed to physical trauma, including battlefield injuries and major accidents, have an elevated risk of developing venous thromboembolism (VTE). Existing diagnostic approaches can be difficult to apply in acute trauma settings because robust and invasive diagnostic methods may not be feasible in the field or at the bedside. Improved predictive tools could support clinicians in identifying high-risk patients earlier and implementing preventative treatment strategies.

Applications and Advantages

Predicts venous thromboembolism risk prior to symptom onset.
Combines clinical and biological data to improve predictive accuracy.
Supports earlier clinical intervention and preventative treatment strategies.
Utilizes machine learning approaches capable of handling large and complex trauma datasets.
May reduce complications, hospital stays, mortality, and associated healthcare costs.
Demonstrated strong predictive performance metrics, including high sensitivity and specificity.

Innovation Description

Researchers at the Uniformed Services University of the Health Sciences (USU), Walter Reed National Military Medical Center (WRNMMC), the Naval Medical Research Center (NMRC) and the Henry M. Jackson Foundation (HJF) have developed machine learning-based systems and methods to predict venous thromboembolism (VTE) in trauma patients using clinical parameters and biological data.

The platform combines data preparation, feature selection, predictive modeling, and validation approaches to identify patients at elevated risk of VTE prior to symptom onset.

The system integrates biological markers, injury severity information, and blood product administration data into predictive machine learning models. Exemplary biomarkers include IL-15, MIG, and VEGF, alongside metrics such as units of blood products transfused within the first 24 hours and indicators of soft tissue injury.

The technology may enable clinicians to make earlier and more informed intervention decisions for trauma patients, potentially reducing complications associated with VTE and improving patient outcomes.

Inventors

Seth Schobel-McHugh, PhD (HJF)
Eric Elster, MD (USU)
COL John S. Oh, MD, (WRNMMC)
Vivek Khatri, Ph.D. (HJF)
Matthew Bradley, MD (NMRC)

Innovation Status

Machine learning models demonstrated predictive performance with reported AUC values of up to 0.946, sensitivity of 0.992, and specificity of 0.838 for VTE prediction.