Patient-Similarity Function
The system found 50 comparators within preset constraints for 97% of test cases. The similarity function used a K-nearest neighbour model ranked by Manhattan distance in feature space. Feature distances were weighted by SHAP importance so…
1 sources - 4 claims
The system found 50 comparators within preset constraints for 97% of test cases. The similarity function used a K-nearest neighbour model ranked by Manhattan distance in feature space. Feature distances were weighted by SHAP importance so more important predictors required closer similarity. The study developed a patient-similarity function that retrieves outcomes from the 50 most similar prior patients.