C3.ai Digital Transformation Institute Colloquium on Digital Transformation Science Webinar

Event Type

Lecture

Sponsor

C3.ai Digital Transformation Institute

Virtual

Date

Oct 8, 2020   3:00 pm  

Speaker

Sendhil Mullainathan, Roman University Professor of Computation and Behavioral Science, University of Chicago Booth School of Business; Ziad Obermeyer, Associate Professor of Health Policy and Management, University of California, Berkeley

Registration

Registration

Views

7

Originating Calendar

NCSA-related events

The next C3.ai Digital Transformation Institute Colloquium on Digital Transformation Science webinar will be Thursday, October 8 at 3:00 p.m. U.S. Central. Presenting "Solving 'Prediction Problems' in Health, from Heart Attacks to COVID-19" will be Sendhil Mullainathan, University of Chicago, and Ziad Obermeyer, University of California, Berkeley.

Registration is required for this webinar.

Abstract: In order to treat a disease, doctors must first know whether it is present. Thus, a key task in medicine is to judge the likelihood of a disease given rich, observable patient data. Because this resembles the prediction tasks where machine learning algorithms shine, we use them to study two important clinical problems. The first is the decision to test for heart attack. Because a test is only useful if it yields new information, efficient testing is grounded in accurate prediction of test outcomes. By comparing doctors’ testing decisions to tailored algorithmic predictions, we show that doctors both over-test (52.6% of high-cost tests for heart attack are wasted) and also under-test (many patients with predictably high risk go untested, then go on to experience frequent adverse cardiac events including death in the next 30 days). The second is the study of triage decisions in COVID-19. In emergency rooms across the world, doctors must decide if patients with suspected or confirmed disease are safe to go home, or if they need hospital-level monitoring. Clinicians have noted that many patients are tragically sent home, only to deteriorate rapidly. In ongoing work, we use machine vision to find subtle predictors of pulmonary collapse that human doctors miss. Together, these examples suggest that training algorithms to solve clinical “prediction problems” can yield both improvements in clinical care, and new insights into physician behavior and human health.

See details of upcoming talks

Register for any upcoming talks