Q&A: AI in diabetes care poses many challenges, benefits to patients and clinicians
Click Here to Manage Email Alerts
Key takeaways:
- AI in diabetes care may help with diagnosis prediction and facilitate administration for clinicians.
- It is important for clinicians to be involved with and supervise AI to avoid worsening existing disparities.
AI in diabetes care is becoming more popular with various potential benefits for patients and clinicians as well as important challenges to keep in mind, according to a conversation between two diabetes experts.
“There’s a lot of opportunity and potential [with AI]. I think we still have to approach this with caution just like any new technology, device and any new medication,” Edward C. Chao, DO, clinical professor of medicine at the University of California, San Diego, said during an interview with James R. Gavin III, MD, PhD, chief medical officer at Healing Our Village, clinical professor at Emory University and Healio | Endocrine Today Co-editor.
Gavin spoke with Chao regarding AI in medicine and its use in endocrinology, with a focus on diabetes in particular. “I know that this will test our capabilities to meet those challenges, but I’m confident and optimistic that we can meet these challenges to ultimately help our patients who we have the privilege to serve, and also to potentially address issues and be one of the tools that can help with addressing burnout and reclaiming the joy and passion.”
Read the interview transcript, lightly edited for clarity, below and/or watch the Healio Video Exclusive series linked here:
- Video 1: AI poses potential benefits, challenges in diabetes care
- Video 2: The role of clinicians, stakeholders for AI in diabetes
- Video 3: Important to not ‘exacerbate’ existing disparities with AI
Gavin: The AI theme has been a strong focus of this year’s meeting, and it has been greeted with enormous energy. Why do you think that we have that kind of reaction at this point in time to this theme?
Chao: I would say that it’s the topic I feel like we hear so much about and it’s top of mind. It’s like a trending topic that we don’t know so much about, at least patients like myself, in a lot of ways. The technical aspects are capturing interest because technology has been at the forefront of diabetes, which has so much data collected by patients and collected by health care professionals. And yet, we still have a divide as far as how much technology can be harnessed to help our patients and ourselves as we serve our patients.
Gavin: Now, along those same lines, there’s been a sharp increase in the volume of published work on AI in medicine in the clinical setting over the last several years. Why do you think that that has happened in that way?
Chao: I know that there was a quote from the chat session yesterday about AI and how it can change clinical practice. One of the speakers said 20,000 papers in the last 6 years have been published on AI topics in diabetes care. I think this is because it’s a burgeoning technology, and the fact that so many folks are involved and interested, whether it’s data scientists or other stakeholders involved. So, it’s a very important field of inquiry.
Gavin: When we think about diabetes, and the juncture of AI with diabetes, in this meeting and in other settings, people have called diabetes a highly quantified disease, making it perhaps more amenable to problem-solving and prediction by AI. Do you agree with that?
Chao: Yes, I definitely agree. There are studies that have tried to harness predicting diagnosis and very early studies predicting hyperglycemia with assistant commands, for example.
Gavin: Now, I’ve been doing this for a long time in this diabetes space and we’ve come a long way in our understanding of diabetes as a complex cardiometabolic disease involving multiple hormones and multiple organ systems. Now, I wonder, as you think about it through the lens of somebody that is really familiar with AI, and where it has come, does this pose special challenges to how much we can expect from AI in the clinical universe of diabetes care?
Chao: That’s a great question. I know that there’s a lot of challenges. How do you reconcile the technology with clinicians’ workload that’s already overwhelming, if you will? Are you adding more to the workload? Are you trying to streamline it? That can be unintentional, but adding to the workload? One of the speakers yesterday, Mudassir Rashid, PhD, pointed out that we don’t want to have a situation where if folks see that patients are being seen more efficiently, we can add more patients to somebody’s schedule, a health care professional’s schedule, as a result. I think the other thing is that we can have some unintended consequences, certainly with AI that we don’t even know about yet.
Gavin: We always talk about the challenges, but maybe there are some special opportunities that are also presented. Maybe we should talk about the upside a little bit.
Chao: Yes, there are plenty of upsides. We do know that AI is used for diabetic retinopathy with high accuracy vs. human interpreters and we do know that it can harness and process six times faster than conventional methods. There’s a lot of exciting work, some of which was talked about yesterday, about how the ideal is taking multimodal datasets. A lot of AI right now is unimodal, which is just one text or modality-like images. But trying to combine multiple can potentially give you much more information than where we are now. And I think part of the excitement, and what I’m interested in, is human-centered design, in terms of how that can apply to research in AI. How do we harness individual stakeholders like patients and health care professionals? How can we take their insights and make AI work? There’s a brief example about how there’s a small set of 11 commercially available apps, and of 26 patients, I believe, they said 43% needed help in just using those apps. I think there’s a lot of opportunity for intervening upstream and having patients involved saying, how do you think this could be more user-friendly, accessible and usable?
Gavin: What should we be doing more of to make sure that diabetes care professionals are ready to leverage everything that AI is going to bring to the table with patient management?
Chao: I think that’s a question that’s evolving and we’re going to continue to find out. I think getting involved, asking the questions and getting up to speed on the advances as it continues rolling. I feel like, in some ways, AI is like a train that has already left the station that we’re struggling to run on foot to catch up. And I think that what we need to do is to engage all stakeholders, all folks involved, whether it’s patients, clinicians or the technological companies that provide the electronic health record.
Gavin: Well, I like that notion of the train having left the station, but it also brings to mind the fact that if we are already sort of behind, what should we be doing with our trainees, and with those who are earlier in the pipeline?
Chao: My trainees are definitely very tech-savvy. I think the question is, how do we prepare them for this? And I think in some ways, they have the tools, they have the digital-native mentality that that folks like me may not have had as a younger trainee.
Gavin: Now, there’s this thorny issue that comes up with every new set of interventions in clinical care and that’s the whole issue of regulatory oversight. Where’s that most likely to come from as we start thinking about clinical applications?
Chao: As far as data security, privacy, could this be vulnerable to hacking and risky information that could have potentially very serious adverse consequences for individuals? I think the regulation has to be integrated. They have to be at the table as far as the conversations. David C. Klonoff, MD, and others have done a lot of great work as far as addressing EHR and trying to integrate, for example, CGM data in the EHR is a massive undertaking that still is not centralized. So, I think that’s something that has to be continued to have a conversation about.
Gavin: Do you think the FDA or other such agencies are going to get more deeply involved as we see more applications?
Chao: I think that it will be the companies, but also, I think this is where the human-centered design aspect comes in where folks should be thinking about this ahead of time and hopefully try to anticipate, as much as possible, the challenges.
Gavin: What do you think is the most likely framework for how liability is going to be attributed in case there’s some untoward consequence of applying AI in clinical management scenarios?
Chao: That is a question that was brought up during yesterday’s symposium and I think there’s no clear answer as far as if it’s the company’s responsibility rather than the practitioner, the clinician. There are applications and early studies of clinicians talking to people about AI and getting recommendations and getting input from AI. So, I think that brings a whole great new world, if you will, of some questions and challenges that we still have to tackle.
Gavin: Do you think that there are people who are serious about it and who are doing the deep thinking on those issues?
Chao: Yes, yes, ethical AI and all those considerations that are so important.
Gavin: Now, when we think about AI playing out and finding its place in clinical practice session settings, what are the chances that the big systems, the elite players, are going to get access to all of the features and all of the nuances of AI, while the small players, the less advantaged ones, will be left behind?
Chao: I think that’s a potential possibility and a real concern. I would hope that there’s more cooperation in the spirit of sharing and not having folks left behind.
Gavin: So, we really have to be intentional about equity and creating a level playing field when it comes to who gets the advantages of AI? Because I worry about the likelihood of disparities emerging, even in a situation like the application of AI.
Chao: I share that concern, as far as exacerbating the already existing disparities among individuals and communities of color or folks who may not have access to technology and I think that the real issue is using datasets to train the AI that are not necessarily representative, and that’s a problem, globally, for patients of diverse communities. AI is only as good as the data you connect it to. So, I think that’s a real concern and it has to be top of mind.
Gavin: Can you imagine a scenario that you, as a person who is more familiar than most with this area, would consider as your worst nightmare in the application of AI in the clinical setting? And, by contrast, when you sit back and think about it, what do you view as your fondest dream? What is your ideal scenario, in the way AI will play out in the clinical setting?
Chao: I think one of the worst things would be exacerbating those disparities that already exist, contributing to a further divide amongst patients in terms of their care. I think that it could potentially, I don’t want to sound alarmist, but it could potentially be used for purposes that are not consistent and that can erode trust at the bedrock of our patient-clinician relationships. There are surveys of attitudes of patients about AI and one of the top points was that we want to have the clinician heavily involved in supervising or being aware of what AI is doing to our patients, and then that plays into the theme of trust. I think, if anything kind of gets so far out of where we should be, would that trust be eroded and undermine the patient-clinician relationship?
I think the best-case scenario would be, would this be to the point where it could almost be like a translator? My dream was always imagine having a scribe, which we know can be helpful, but having it automatically translate the conversations that we have with our patients into well thought out and well put together notes. And then also be able to generate an after-visit summary of a very cogent, well thought out, organized way. Not replacing us but helping us to be more what we came into as a clinician, which is to spend time with our patients to get to know our patient. If that could serve us in that way, I think that would be the most ideal use. And, in a lot of cases, being a partner in terms of gathering and processing data a lot faster than we humans can and being able to interpret that data accordingly so that we can best help our patients.
Gavin: Now we are here at the American Diabetes Association meetings, and of course there are mixed medical specialties represented here. But as we think about where AI is going and how it’s likely to be applied in clinical settings, is it your view that at every major medical care or medical specialty type of meeting that we have, there should be some presence of AI so that we don’t have that train too far down the track for everybody who’s taking care of patients going forward?
Chao: I agree that we need to have the conversations, we need to have folks understand the issues, the potential opportunities, clinical challenges and then talk openly about them. And as far as one of the speakers yesterday, Dr. Rashid, was talking about how it has to be a diverse group of folks coming together and working to collaborate.
Collapse
Healio Interviews
Click Here to Manage Email Alerts