What an Endocrinologist thinks about ML in healthcare
A summary of my interview with an Endocrinologist. 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.
“I am excited by the amount of information that I will get access to.”
Job Background
I am an Endocrinologist at an integrated delivery network. I see patients about 80% of the time and then the remainder is administrative, because I am a Chief. Most of my clinical work revolves around thyroid diseases, diabetes, osteoporosis as well as adrenal and calcium-related diseases. Administratively, I am in charge of making sure we have enough appointments available, learning and training my clinicians in new technologies, setting clinical standards for our medical group, as well as the other basic administrative work. I anticipate another five to 10 years until I retire.
Familiarity with ML in healthcare
To start off, what have you heard about artificial intelligence and/or machine learning in healthcare?
I have mostly heard about ML in healthcare as it relates to diagnosing patients. You can tell a computer system all of the symptoms and labs, and then perhaps it could give you a good differential diagnosis. But I am not really worried about it replacing me, since there is such a human component to care delivery.
Past and future use
Have you used any ML tools? Would you?
Not that I know of, but I certainly would if I could trust it.
Excitement and concerns
What about ML in healthcare is concerning or exciting for you? What else is exciting or concerning for you?
My big concern is that I have no idea how these ML tools could even work because there aren’t good data for them. So much of medicine is an intangible art that these tools wouldn’t have access to. Also, great clinicians are great because they have enough experience to see the nuance in all of the information. Sometimes less experienced clinicians may miss something that someone like myself won’t because I can more easily know what to look for. I just don’t think ML will replace clinical experience; but hey, you never know. And that is the exciting part, right? There is a lot of human error in medicine and many things get overlooked, so maybe ML could help with some of the basic issues and misses.
I am excited by the amount of information that I will get access to. Nowadays, there are these continuous glucose monitors and semi-closed loop systems. When I entered this field, we relied on finger sticks and hoped for patient compliance. Now with these tools, we don’t have to worry about it as much and maybe ML will make that even easier. Because so much of healthcare is about patient adherence, and smarter ways of doing that would help a lot, and that is exciting.
Ethics and privacy
Where do ethics play into this? What could go wrong, or what could be done well?
I think ethically challenging spaces come when you are making assumptions. Everyone makes assumptions all of the time; it makes decision making possible. However, these assumptions can have negative consequences. If you make an incorrect assumption about a patient using ML and there is prejudice baked in, then we really are doing ethically wrong things. This becomes worse when ML is automating things because there is no one to identify the issue or stop the train.
How does privacy fit into all of this?
It’s a big problem and we talk about it a lot. All of these new technologies that make our jobs as clinicians easier are cloud-based. So, we are always asking the questions about who owns the data and who controls the data. We need to make sure to strike a balance, so that research and innovation can happen.
ML knowledge and model explainability
At what level do you need to understand how the model makes its prediction?
I don’t need to know how the model is built, instead I just need to know that it is safe and efficacious. My personality and interests are not connected with the exact details of the algorithms.
External validation needs
For you to be willing to use an ML tool, what external validation would you need to see? What types of government and/or non-government institutions would play a role?
Well of course, I would want to see that the FDA approved the ML tool for the specific indication. Considering other technology, like CGMs, those were approved for diagnostics purposes only first. But later, they were approved for therapeutic purposes. I was only ready to use it for therapeutics after it was approved for that.
I would also need to see clinical trial results. In our semi-closed loop pump system, we are trusting algorithms. For that, we had to see that the clinical trials found the technology to be both safe and efficacious.
Desired use cases
Where are there opportunities to assist clinicians with ML? Imagine this: A world-class technology company, developed an ML tool that suggests possible diagnoses or triages a patient population. What is the best thing for them to build now and why?
I think there is a big need around screening at-risk populations to make sure that diseases are detected and then appropriately treated. I also think epigenetics and genetics are going to be great new frontiers of medicine. There will be a day when patients get an assessment of their genomics and learn actionable ways to improve the quality of their lives.
Buying process
Where you practice medicine, who all would be involved in making the decision to purchase and use an ML tool?
I would be involved all the way — from initiation, to vendor reviews, to training, to deployment. As a Chair, I help decide whether or not some technology would be useful for our specialty.
