Jeff McCullough, Associate Professor of Health Management and Policy at University of Michigan talks about the ability to use machine learning methods for causal inference in conjunction with techniques from other disciplines.
Transcript of video is provided below.
JM: Hi my name is Jeff McCullough, I’m an Associate Professor of University of Michigan.
Interviewer: Can you tell us the most exciting thing that you are working on?
JM: The most exciting thing I’m working on is probably trying to understand the relationship or use of machine learning methods for causal inference. The reason it is interesting is because there is a dearth of clinical trials and the standard for evidence based for medical care as well as practice of medicine businesses. I want things to be based on modulus population under extremely controlled settings.
What machine learning can give us is the ability to do a better job at understanding the variations that exists in the real world and measuring outcomes that are relevant to actual people, long run things that are captured in clinical trials. At the same time, machine learning by itself is actually going to give us the wrong answer, it’s meant to solve other problems. So we can combine it with techniques from other disciplines like economics to try and interpret actual causal counterfactual predictions. It gives us the tool kit we need to get the real causal effects of medicine, practice, and policies that matches real world rather than laboratory settings.