Journal: American journal of hematology
This study addresses the challenge of diagnosing venous thromboembolism (VTE), a common and serious condition affecting up to 900,000 people annually in the US.
Using structured and unstructured electronic health record data from nearly 4,700 adults presenting with VTE-related symptoms in primary care, researchers developed and evaluated seven machine learning models to predict VTE incidence.
The logistic regression model performed best, with an AUC of 0.88.
Key risk factors identified included:
- Cancer history
- Smoking
- Spinal cord trauma
The study also distinguished risk factors associated with timely versus delayed VTE diagnosis. This underscores the potential of data-driven models to improve early detection and reduce diagnostic delays in clinical practice.