What a Gastroenterologist thinks about ML in healthcare
A summary of my interview with a Gastroenterologist. This is one of my 18 interviews with clinicians for my MPH capstone (link here) for UC Berkeley’s School of Public Health.
Visit my Building Trust and Adoption in Machine Learning in Healthcare site (link here) for the abridge and full version of my MPH capstone as well as upcoming summaries of interviews and additional research.
Note that this interview is a de-identified summary for ease of reading and privacy. It has been approved for distribution by the interviewee.
“…there is a palpable sense that we are closer than we have ever been to computers being able to reason over healthcare data. That is something that is unbelievably philosophically amazing!”
Job Background
I am a Clinical Fellow and Post-Doctoral Fellow at an integrated delivery network. I specialize in gastroenterology, and I have an MD/PhD with a focus on cancer genetics. I see patients one day per week and do research the rest of the time. My research focuses on the application of data science methods to large healthcare data sets. I think I will work for roughly 40 more years.
Familiarity with ML in healthcare
To start off, what have you heard about artificial intelligence and/or machine learning in healthcare?
There is a lot to be said about ML and a lot has been said in the most recent year. Sadly, a lot is hype. We hear that there is tremendous potential to improve healthcare at all levels; but the reality from my view in the trenches is that there are many challenges to make this happen.
Past and future use
Have you used any ML tools? Would you?
Most if not all of the diagnostic algorithms are in the R&D phase now, so only few products are out there yet for clinicians to use. If you take a broader view towards population health triage, then you see that ML has been part of medicine for a very long time. These are the risk scores being used by most hospitals.
Excitement and concerns
What about ML in healthcare is concerning or exciting for you? What else is exciting or concerning for you?
There are so many things that are concerning for me. The list goes on — patient privacy, data sharing, capture of high quality and accurate data, clear definition of problems / use cases, deployment of models, clinician comfort with the models, generalizability and robustness, limited expertise at the intersection of healthcare and ML. There are many more, but these are the ones that come to mind first.
On the other hand, I am super excited that many ML methods appear to be generalizable and some of the underlying math and logic of how a lot of these methods work are quite broad. Said differently, we have the potential to answer a lot of healthcare questions with a limited set of tools, which means we can end up doing a lot more. For example, my work on natural language understanding can use the great progress that others have made in a wide array of non-healthcare industries.
The most exciting thing about all of this is that there is a palpable sense that we are closer than we have ever been to computers being able to reason over healthcare data. That is something that is unbelievably philosophically amazing!
Ethics and privacy
Where do ethics play into this? What could go wrong, or what could be done well?
Many people are concerned about data entry, ability to access care, and bias in treatments. These are challenging issues to deal with. We need to make sure that we don’t reinforce biases in the models that we train and deploy. For what it is worth, I have not had to think about it. I hope that it is not a lack of awareness, but instead that my use case is far removed from the typical kinds of biases and prejudices.
There is a call for explainable AI because we want to be able to interrogate models, understand why they came to a decision, and then confirm equity. There has been some progress in looking at the data for bias, but ignoring variables like gender and race is an antiquated approach.
