Using Machine Learning to Understand Physician Decisions

Colloquium | November 13 | 12:40-2 p.m. | 1205 Berkeley Way West

 Ziad Obermeyer

 Public Health, School of

As the complexity of patients and the health care system grows, the human mind struggles to keep pace. Even with perfect incentives, medical decision making is maddeningly difficult, and it’s not surprising that doctors get many of these decisions wrong—with the end result of low-value care. Machine learning can help improve decision making in health care. I’ll present results suggesting that algorithms can help doctors both test less — in this case, cutting testing for acute coronary syndromes (ACS) in the emergency department (ED) by up to 40% — and also test better, by catching high-risk patients that are often missed in EDs today. Machine learning can also help us to sift through data on hundreds of thousands of testing decisions, to discover cognitive errors: patient factors linked to doctors’ tendency to over- or under-test. Overall these results suggest that machine learning, in addition to being a tool for policy makers, can also be a tool to drive new understanding of decision making and complex social systems more broadly.

 All Audiences

 All Audiences