The algorithm was roughly 3% more accurate than the relevant diagnostic code and less biased toward certain populations, the ...