Examine how machine learning applied to UK primary care records can identify a likely-undiagnosed population of GEP-NET patients.
Key Takeaways
- Model achieved a test ROC-AUC of 0.756 and PR-AUC of 0.427
- Applied to 6.8 million records the model estimated roughly 1,200 undiagnosed patients
- Estimate had a precision of 0.85 when extrapolated across the database