The Use of Artificial Intelligence in Diabetes Education, Self-Management, Decision Support, Complication Prediction
In this podcast, Jennifer Smith, RD, LD, CDCES, discusses how clinicians are integrating artificial intelligence (AI) into daily clinical practice, including diabetes screening, education, and self-management for patients, decision support for clinicians and patients, and complication prediction. She also discusses possible concerns with the use of AI and more.
Additional Resources:
- Mackenzie SC, Sainsbury CAR, Wake DJ. Diabetes and artificial intelligence beyond the closed loop: a review of the landscape, promise and challenges. Diabetologia. 2024;67(2):223-235. doi:10.1007/s00125-023-06038-8
- Ponzo V, Goitre I, Favaro E, et al. Is ChatGPT an effective tool for providing dietary advice? Nutrients. 2024;16(4):469. doi:10.3390/nu16040469
For more diabetes technology content, visit the Excellence Forum.
TRANSCRIPTION:
Jennifer Smith: Of course. Thanks for having me. My name is Jennifer Smith. I'm a registered dietician and a certified diabetes care and education specialist. I work with Integrated Diabetes Services. I'm the Director of Lifestyle and Nutrition.
We work with all people with diabetes and heavily work with the people who have type one diabetes, intensively managed, people of all ages and through all walks of life, and all over the world. It's nice to be able to connect with the diabetes community and provide that education piece at length, which you know, helps so many people.
Jessica Bard: Well, welcome again. Our topic today is the use of artificial intelligence in diabetes management. To start us off, do you believe that AI can impact diabetes care, and in that same vein, if so, can you provide an overview of how it can be integrated into daily clinical practice?
JS: Absolutely. So, I think there is a lot of potential because there are so many different places that data can come from in today's world. How many apps do we have that track this and track that, right? And in terms of diabetes, systems that can bring all this information together from records, such as your medical history, medications that are in use, your lab results, and reported data from people with diabetes. Again, their blood glucose records, and their continuous glucose monitor data. If they're using a pump or smart insulin pen. As well as the tracking apps– things for nutrition and activity because care for all of those people with diabetes includes a set of targets, we're aiming for this glucose to this glucose level, we're aiming for this blood pressure, we're aiming for cholesterol levels within a certain range, and then guidelines that come essentially from some evidence-based research, with clear indications for positive outcomes vs negative, if they're not followed in terms of these targets. It's the perfect setting to be able to use AI.
In terms of integrating it into daily clinical practice, we already have some simple tools that are available in the form of, obviously glucose monitors, and again, the continuous glucose monitors, and all of the software that brings this data together, it's able to provide some insights into trends.
That in a clinical practice makes it a lot easier to direct the conversation with the person with diabetes. It allows the clinician to essentially assist in providing suggestions, along with discussion to evaluate how this person can make a change that will impact outcomes to the positive. Thankfully, AI will hopefully continue to improve with more suggestive, more data interpretation, and more directive, which hopefully will make clinical visits a lot more… with the limited time they have… more beneficial to the person or the patient, and a little bit more that the clinician can provide without having to think so much on their toes. They're getting some help from something behind that's giving some good information and putting it all together.
JB: It sounds like there are a lot of applications. To start, we'll break some out: patient education and self-management in diabetes care. How can AI be used in those categories?
JS: If we look at bringing again a lot of the data together to direct the care, then we need something that works like an “if this than that” type of process, right? Something to help with decisions based on the information pulled together for each individual person with diabetes, and many of the systems in use for glucose monitoring have clinical dashboards, essentially, or databases that can be set up in the clinician practice or office space to navigate alerts for patients directed on the specific targets that they're aiming for. So, you come in in the morning as a clinician on alert for people who have blood glucose levels above or below certain targets for a determined amount of time. They pop up, kind of as a red flag, a feed into the database, and it provides a prompt to check in with those, let's call them top priority patients.
And this follow-through, then, can assist essentially the person with diabetes who's trying to improve in a timely manner between visits, which proves in the long run to decrease the risk of complications and overall improve health, right? It can also help to direct education. Maybe the database that a clinician is looking at, they can see that more of the issue is around mealtimes or more of an issue is around when activity is logged, and so that can help to direct the education to one thing vs an overall education, where maybe the person didn't need everything at one time.
Other systems can bring together medical-record data, such as their medical history lab data, medications which can help them direct the clinician to the right type of advice, some changes maybe in the treatment strategy that they're going to put out support, including, of course, additional education, maybe some lifestyle suggestions, and maybe even directing them to additional clinical support. That may not specifically be diabetes, but it might be podiatry, or it might be the eye specialist. Right? So, in terms of putting all of this information together, AI can definitely be use useful because we can say, it directs us to say, all of this information is pointing towards this, this person would really benefit from this type of information and this type of care from this particular clinician.
JB: I know you already mentioned some of these things. If you want to dig in a little bit more, can we talk about how AI can be used for treatment, decision support, and screening advice in diabetes care? I know you already mentioned some things about helping the clinician with advice.
JS: We must consider too the potential for AI to assist with clinical time. We already know the clinical care time has a lot of constraint to it, which makes it important for AI to bring information together and direct the information that the clinician is providing to the person in a way that allows for the time allotted for that visit to be spent most appropriately. You know, diabetes is a condition that has a lot of factors to consider. It's not only blood sugar management or blood glucose management. The treatments that are needed, different medications that are in use, and there are so many of them right now, as well as the many systems and conditions that need to be managed, along with the diabetes, guidelines that need to be followed, it makes it really, really complex to manage. So, if we're only looking at data, this doesn't drive change in behavior. It doesn't alone improve a long-term outcome for the person with diabetes. And unfortunately, it doesn't really empower the clinician to help the person with diabetes make change either. If you're following within a 30-min visit, you have to check off all these boxes of things that you've discussed and talked about, it may not be hitting all the things that are really in a personal nature that are necessary for this individual. If at a clinical level, somebody is kind of lost in the next step to take for the person with diabetes sitting there sitting in their office, it means all the information that they're trying to collect, it’s I don't want to say that it's worthless, but it's not really driving any change. So, in terms of using AI again, if we've got something that's bringing enough data together to give you some direction for that visit and some ability to tell the person, “Hey, this looks like we need to get you moving in this direction. We need to get you in with this visit. I see that you haven't had an eye visit in 3 years. Well, goodness, let's do this.”
And again. in reviewing a clinical chart yourself, you might be able to pull all that data, but it might take you 30 minutes to pull it. Whereas AI could potentially bring all of that together and bring up pop-up messages that say, “Hey, this person is missing this. This this person hasn't had this done. You haven't reviewed this or check this off in a timely manner.” So, I think that's how AI can be used to some degree is probably being used with a lot of electronic medical records and messages that come through. But it's certainly, I think it's beginning to become more beneficial.
JB: I think that's well said, I think that's something that I hadn't personally thought about being able to do more within that same timeframe of a visit with the clinician. How can AI be used for complication prediction in diabetes care?
JS: We already have a lot of parameters that we're looking at and trying to keep track of. We're looking at the typical complications, being cardiovascular, neuropathy, we're looking at eye conditions and retinopathy, and all those different things. So, prediction for future health comes is beneficial, because if data can be put together to prevent typical complications, then it can have an impact at the personal level. If we can set all these parameters across the board in looking at research, collective data, from a lot of people with diabetes; if this marker is here and this marker is here, and this marker is here, you're on track to have this type of complication. But if we can have something to bring all that information together, and direct the patient visit, to be able to direct the clinician to be able to provide that information and point the person in the right direction, overall, then we're aiming for a healthcare system that can provide preventative medicine, rather than just a treatment system that says “Oh, you've got this.” Why aren't we aiming to prevent it from happening? And that starts, in terms of diabetes, that's a lot of education. That's a lot of looking at data, and then what do you do with that data in the long term? That's what can prevent a lot of the complications. If screening can prove effective based on evaluated parameters, then we're preventing loss of time in the workforce. We're preventing increased costs to the overall healthcare system, and we’d have more ability to have a wellness-focused community. That’s hopefully where this is putting us in terms of preventing complications.
JB: We discussed a lot of the benefits of AI, but with progress there can be some pain points too. So, conversely, what are some concerns or potential harms associated with using AI?
JS: I can think of one. I think we've certainly covered a lot of the benefits of it. Potential harms, in terms of the adoption in a clinical setting, there is a lot of hesitancy. I think most of the people in the world with diabetes have type 2 diabetes, which means that the majority of those navigating diabetes are doing so with the assistance of primary care or general medicine practitioners. And so, when we start to get complicated with all the things that need to be considered, to adapt the use of these AI systems, education to the general medical practice needs to be thorough in being able to use them well and to use them safely. So, I think that hesitancy for most of the clinicians to use it for pattern management, and to direct care, it lies in their abiltity to receive the right information to use this type of beneficia computer algorithm to help their patients. So, I think that's not necessarily a harm, but it's certainly one of the things that is hard to get it adopted at a comfort level for most clinicians who could be doing an awesome job beyond what they're already doing, if they just had something that was bringing the information together.
In terms of the harm, I read a great nutrition-based article about AI, and it went into some detail about what AI could do in terms of nutrition, education, and what it really couldn't do. And the bottom line really seemed to be, and I truly agree with it, the bottom line tends to be in very simple cases of education.
AI can do a really good job because it's almost like AI is filtering out and really navigating one type of education need. Let's say it's a low sodium diet, a low cholesterol diet, or a low sugar diet, it can do a pretty good job of navigating that, even on a personal level, you could feed it some information–I follow a Vegan diet– so then it can direct it that way. But when we talk about diabetes within the realm of nutrition, there are many conditions that a person with diabetes might be navigating beyond the diabetes. There might be heart health issues. There might be kidney issues. They may have some type of digestive component that they need to consider. And so that's were, at least at this point, AI seems to falter a bit in terms of directing the information the right way. It doesn't get as specific as needed, and unfortunately, some of the information is almost counter. If you have somebody who needs to follow a diet that's considering diabetes management, but then they also have kidney health issues, sometimes that information is completely opposite ends.
So, you might have some confused people getting information that it looks like they’re being told to do this, but then the other aspect is of being told to do this, and I think without the human piece of education, they should really talk with an educator about something like that. AI can't quite get there yet.
Maybe AI is not considering on a personal level that this person runs 20 miles a day. Maybe someday it will get there. I mean, that's the hope, right? But at this point I think those are some of the, not necessarily a harm, but just a caution in use.
JB: I imagine there might be a couple concerns in health inequity as well and potentially regulation and security concerns. Can you speak to any of those?
JS: Yeah, in the regulation and security concerns… because a lot of the data is being collected in a healthcare system, there are good rules and regulations that a brick and mortar, hospital or clinic, they must have very, very tight constraints on what types of systems they're using, where they're allowing data to come in from, and how they're receiving data. But there are a lot of other apps that can bring data together. And unfortunately, some of the security concerns for those are that they're not necessarily 100% approved by a healthcare system.
And so, when we're looking at regulation and security concerns, it's often on the person with diabetes to really check out: “Is this safe? If I put my data in here… if I put my information from my insulin pump or my blood glucose meter, is it really going to the person that I want it to go to? Is it only going to my clinician? Am I only sharing it in one direction?
Those are some of the regulatory things to be concerned about again. All the major brands that are on the market have checkoffs in terms of those regulations. Some of the other apps that can kind of feed into that data, whether you're using an IOS system or an android-based system or whatnot. Some of them have health data, collective systems that grab from other apps, and if you're not careful about turning things on or off, then, again, that could be a security concern.
JB: How would you summarize our conversation today if you just had to say a couple quick sentences?
JS: I think that AI is a definite benefit in the realm of diabetes management and in many other things, as well, but obviously in the world of diabetes to bring together a lot of information to direct visits in a way that can help to personalize and improve long term outcomes for people. I think there's certainly a way to go with AI. But I really, I look forward to AI getting better. In the type one world, it's certainly proving to be beyond even that of which we use clinically with all the smart pump systems and all the algorithms and everything kind of in use. Technology is quickly moving these days in the world of diabetes, which it's great to see.
JB: I can imagine this is a topic that we'll need to update often. Well, Jennifer, thank you so much for joining us today, we appreciate it.
JS: Of course. Thank you.