Lecture | October 31 | 1:10-2:30 p.m. | 202 South Hall
Machine learning is earnestly being applied to automate tasks in medicine â in effect as a medical technology. The same tools, though, can also be used to improve our understanding of the health system itself. Rather than simply being a medical technology, they can also contribute to empirical science and better grounded policy. I describe results from two projects where the predictive approach proves particularly illuminating, both on 'wasted' spending: one on over-testing and the other on the high concentration of spending at the end of life. I will also describe methodological issues that arise that are relatively neglected in the machine learning literature, such as measurement error and the impact of unobserved variables.