Why Clinicians Don't Trust Your AI - And What Engineers Can Do About It
Let me tell you about the most expensive math problem in healthcare: A hospital system enters into a contract with a provider to utilize a sepsis prediction model. The vendor has a cool demo. The AUC is 0.94. The clinical informatics team spends six months integrating it into the EMR. Nurses and physicians sit through training sessions. A year and a half later, someone shuts it off quietly.
Why? Because it went so far as to raise false alarms so frequently, the clinical staff stopped paying any attention to the alerts. Every patient with a fever and a slightly elevated lactate triggered a sepsis alarm. ICU nurses received 40-50 alerts per shift. They rationally and defensibly learned to ignore them.
I’m describing the Epic Sepsis Modelthat was rolled out to hundreds of hospitals throughout the United States and was externally validated to miss 67% of actual sepsis cases while...
Copyright of this story solely belongs to hackernoon.com. To see the full text click HERE
