Personalised cardiovascular risk model
A two-stage machine learning system that screens for peripheral vascular disease, then estimates how much it raises each individual patient's cardiovascular risk.
- Type
- Machine learning
- Status
- Manuscript in preparation
- Built with
- Python, scikit-learn, XGBoost, Flask
- Links
- View the code
The question
Peripheral vascular disease (PVD) raises cardiovascular risk, yet many cases go undiagnosed because the standard test takes time and equipment. Common risk calculators also report risk for a population, not for the patient in front of the doctor.
What I built
A dual-stage framework trained on around 70,000 patient records, following the CRISP-DM methodology:
- Stage one screens for PVD from 11 routine clinical and lifestyle features.
- Stage two estimates cardiovascular disease risk with PVD as an explicit factor.
- Counterfactual analysis answers the clinician’s real question: how much of this patient’s risk comes from PVD?
It runs as a Flask web prototype where a clinician can enter a patient’s details and see both results.
Collaboration
Joint work with Dr. Wan Noorshahida Mohd-Isa at Multimedia University, Malaysia. The manuscript is being prepared for journal submission.