.@BrookingsEcon-@SchaefferCenter: When @UPMC evaluated the risk of death from pneumonia of patients in their ER, the #AI model predicted that mortality dropped when patients were over 100 years of age or had a diagnosis of asthma! UPMC did indeed have low mortality for these two groups, but the assessed risk of pneumonia for them was so high that staff had given them lifesaving antibiotics before they were even registered into the #ElectronicMedicalRecord: thus antibiotic administration time stamp would have been wrong! #DataBias. AI-based medical protocols like this could harm the most vulnerable and high-risk of patients. Therefore, one must identify #Bias in data sets, & also, evaluate AI-based medicine on whether it mitigates or perpetuates healthcare disparities, i.e., with a social impact metric.
— Satyen Baindur (@Satyen_Baindur) March 9, 2019
