What a Plastic & Reconstructive Surgeon thinks about ML in healthcare
A summary of my interview with a Plastic and Reconstructive Surgeon. 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 know how to use the outputs of X-ray machines, CT scanners, and MRI machines to help my patients. And as I think of it, I only have a very rudimentary understanding of how those things work. So maybe for an ML tool, I don’t need to know as much.”
Job Background
I am a Plastic and Reconstructive Surgeon at a purely clinical office of an academic medical center. The majority of reconstructive surgery nationwide is on breasts given that they need repair from cancer or trauma. In America, insurance plans are mandated to cover reconstructive surgery for breast cancer patients and breast cancer is sadly common. Since I really like what I do and more recently finished my fellowship, I think I have about 30 more years to practice.
Familiarity with ML in healthcare
To start off, what have you heard about artificial intelligence and machine learning as it related to medical care?
During my fellowship, I saw a grand rounds talk from the Chief Resident about how AI will impact plastic surgery. I don’t know much beyond that but it was an interesting talk.
But when you get into AI completely taking over what I do, I just don’t buy it. AI is a fluffy word that isn’t true. My gut says that we aren’t even close to AI. However, I do acknowledge that ML is real and see that as advanced data analytics. If you give an ML algorithm the right data, then you will get a good prediction.
Past and future use
Have you used any ML tools? Would you?
No, but I absolutely would. During that grand rounds presentation, the Chief Resident talked about how you could use an ML tool to diagnose fractures and dislocations. It was early stages then, so I am not sure how far that has come today, but I do think it would be relevant in some clinical practices.
Excitement and concerns
What about ML in healthcare is concerning or exciting for you? What else is exciting or concerning for you?
ML in medicine is exciting to me. I am a huge advocate for integrating new technologies into medicine and especially for my specialty of plastic and reconstructive surgery. For example, I have used 3D printing to help with surgery planning. So, if ML could contribute to something like this, then that would be a great use of ML.
Of course, I also have some concerns. Nothing is perfect, so I am concerned about the errors that get built in. Humans with natural intelligence are prone to errors, so ML tools that learn from them would produce the same types of errors. So, I have a concern about how sensitive and specific these ML technologies could be.
Ethics and privacy
Where do ethics play into this? What could go wrong, or what can be done well?
I don’t know if I have really thought about it all that much. With regard to assisting diagnoses, I am not sure. I think it would be unethical to solely rely on ML to make diagnosis, but I am not sure. The whole idea is to replace human decisions and our errors with something better. So if something is better, then I think it would not be ethical to use the worse option. However, I would have to think about this more.
How does privacy fit into all of this?
I don’t have a lot of concerns around privacy. If a machine can do something as good as or better than a human, then that is a big win. Also, an ML tool would be more HIPAA compliant than a human. Machines would be programmed to be HIPAA compliant, while you can’t actually force a human to stay quiet.
How should the data be used? Who should or should not have access to it?
All providers taking care of a specific patient within a specific encounter should have access to the data.
Who else should help inform you or decide for you if an ML tool is ethical and sufficiently private?
When it comes to me learning about new technologies and the potential ethics, I rely on three sources: (1) the journal Plastic and Reconstructive Surgery, (2) medical conferences, and (3) my own academic medical center’s internal presentations. I read the journal religiously cover to cover, and that information affects my clinical practice the most. At conferences and my health system, I focus on the well-known lecturers. So, if someone has a good presentation in ML in healthcare, I will listen to that.
Do you trust these ML tool developers to have access to these data? Why or why not?
Sure.
ML knowledge and model explainability
At what level do you need to understand how the model makes its prediction?
The only things that I need to know are the ML tool’s sensitivity and specificity and maybe its reproducibility and reliability. That is probably it. Can I trust it with that? Sure. I don’t need to know the math behind the models.
I know how to use the outputs of X-ray machines, CT scanners, and MRI machines to help my patients. And as I think of it, I only have a very rudimentary understanding of how those machines actually work; more of an intuition versus expertise. So maybe for an ML tool, I don’t need to know as much.
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?
I am so convinced that ML is just advanced computing. So, it is just another tool and another device. I am not convinced that it is much different than a CT scanner. So, I think it would follow the same certification process. I would like to see some outcomes studies, depending on what it is used for. And of course, it would have to go through the standard clinical trials.
Clinical education
How would clinical education be impacted?
If you have a hospital full of ML tools that can diagnose and triage patients for you, what is the point of having an ER clinician? You don’t need that significant of a qualification nor medical education. Could ML eventually deliver treatment? Maybe. Generally speaking, I think surgeons are more protected than others. But even for us, a lot of what we do — patient history, vitals, physical exam, pathology, radiology, and lastly diagnosis — is repetitious and could be automated with ML. I might be biased, but I think plastic and reconstructive surgeons are protected the most since so much of our surgery work is an art. But then again, one day, who knows. So to answer your question, I could see medical students becoming basically technicians.
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 lot of value in helping to manage the flow of patients to the ER. It would be great if there were a patient tool that could better direct them to where they need to go. It could tell them if they should go to the ER, urgent care, or wait to see how things are tomorrow. The ER is always swamped, so that would be helpful to allocate the patients better.
Implementation
When an ML tool gets implemented, how should that be done? Who should have access first; who should not?
I think an ML tool would first be piloted with physicians, who can use the suggestions and their advanced clinical judgement. As the tool gets better and we can trust it more, I would then roll it out to others, like nurses.
Buying process
Where you practice medicine, who all would be involved in making the decision to purchase and use an ML tool?
Each service line has a committee that considers new products. If I find something good, I need to present it to the surgery department’s new products committee. They will primarily want to know if it is cost effective and/or bring in more money for the health system.
What data, references, and promises would they need to learn about to ultimately say yes or no?
My current hospital is less research oriented compared to our larger academic medical center. So, we are less interested in experimentation and instead want to see that something is tried and true. If I were to present something completely new, I would also need to come to them with institutional review board (IRB) buy-in and some grant funding.
