STAT+: In the battle of sepsis algorithms, performance alone doesn’t predict victory

Five years ago, the bottom fell out of sepsis prediction software. Hundreds of hospitals had adopted an algorithm from electronic health record company Epic that promised to alert physicians to predicted cases of sepsis, a life-threatening reaction to infection that kills more than 350,000 people in the United States every year.  The AI was a technical flop. Despite its results on paper, the technology failed to perform in the real world, and sent so many alerts that doctors tuned them out or ho...