AI for Tailored Diabetes Care: Clinician Perspectives on Patient Needs

Host 1: Welcome you. You're here because you want to understand the important stuff, to get the knowledge without being overwhelmed. That's what we aim for in every deep dive.

Host 2: And today, we're diving into something truly fascinating. The role artificial intelligence could play in making diabetes care more, um, tailored to the individual.

Host 1: Exactly. We've been poring over a really detailed study. Comes from Seoul National University Bundang Hospital. It's called Artificial Intelligence Tools in Supporting Healthcare Professionals for Tailored Patient Care.

Host 2: Right. And this research looked at an incredible amount of data. Hmm. I mean, over half a million messages.

Host 1: Half a million from more than 11,000 people managing diabetes. It's quite something.

Host 2: Our goal today is basically to unpack this research for you, to find those key insights, the real potential aha moments about how AI might, you know, change the game for supporting individuals with diabetes. Really personalizing it.

Host 1: Okay, let's unpack this then. We We all know diabetes is a huge global health issue, a massive one.

Host 2: Yeah, it really is.

Host 1: And this study reminds us that by 2050, we could be looking at what, 1.3 billion people affected, globally?

Host 2: It's a staggering number. And it's not just about blood sugar, is it? It connects to so many other health problems, even socioeconomic factors.

Host 1: Totally. And that's where this idea of patient-centered care, PCC, becomes so important.

Host 2: Right. It's all about focusing on what each specific patient needs and prefers.

Host 1: Especially for a long-term thing like diabetes, this kind of personalized approach, it's crucial for keeping people engaged, right?

Host 2: Absolutely. It helps improve their skills in managing their own care, their quality of life, and ultimately, their health outcomes.

Host 1: But here's a big challenge, and the study highlights this, clinicians are just swamped.

Host 2: Uh-huh.

Host 1: They're seeing a massive increase in workload. The study points to a 50% jump in patient messages just in recent years.

Host 2: Wow, 50%. And you have to think the pandemic probably kicked that into overdrive, with everyone shifting to digital communication.

Host 1: For sure.

Host 2: And what's really interesting about this flood of secure messages is, well, it gives a different window into patient concerns compared to, say, clinical notes from a visit.

Host 1: How so?

Host 2: These messages often reflect the day-to-day stuff, you know, the immediate questions, the practical hurdles they're facing.

Host 1: Which brings us neatly to how AI is starting to help us understand these everyday needs. The researchers used some, uh, pretty sophisticated tools. Natural language processing, NLP?

Host 2: Yeah, NLP. Basically teaching computers to understand human language.

Host 1: Yeah.

Host 2: And they didn't stop there. They used advanced AI models like BERT and, um, ChatGPT-4.

Host 1: To sift through this mountain of messages specifically about diabetes.

Host 2: Exactly. The aim was to really drill down into the main issues patients were bringing up with their doctors and nurses. What were their most common questions?

Host 1: And they managed to pin them down. They identified the top 12 clinical issues as interpreted by the AI that kept coming up, and the range is, well, it's broad.

Host 2: What was number one?

Host 1: Top of the list, dietary concerns and weight control, specifically how meals and carbs affect their levels.

Host 2: Makes perfect sense. That's constant management.

Host 1: Totally. Second was interpreting lab results, you know, blood tests, urine tests, and especially A1C levels, that long-term blood sugar picture.

Host 2: Okay, so daily management and then understanding the medical feedback. What else?

Host 1: Third was thyroid management, medication questions, TSH levels, that kind of thing. Then fourth, maybe less clinical, but super important...

Host 2: Uh-huh.

Host 1: ...administrative challenges.

Host 2: Ah, the paperwork hurdles.

Host 1: Exactly. Getting authorizations, filling out forms. Number five was bone health, questions about imaging, sometimes surgery related to their diabetes.

Host 2: Interesting, bone health.

Host 1: Yeah. Sixth was just navigating lab orders and results, the process itself. Seventh, appointment scheduling. Eighth, managing medication dosages, and they specifically mentioned metformin.

Host 2: A very common one.

Host 1: Right. Ninth was about prescriptions and supplies, getting refills, dealing with insurance, especially for things like testing strips.

Host 2: Always a challenge, the supply and insurance side.

Host 1: 10th, and this is key, I think, needing education on using devices: Dexcom, insulin pumps, understanding the data.

Host 2: That technology piece.

Host 1: Yeah. 11th was patients reporting concerns about low blood sugar, hypoglycemia. And finally, 12th was just the overall topic of blood glucose management.

Host 2: So quite a mix of clinical, practical, and administrative issues.

Host 1: Exactly. And what really struck me was that these top concerns, they were pretty much the same before and during the COVID-19 pandemic.

Host 2: That's significant. It suggests these aren't just temporary issues, they're fundamental, ongoing areas where people with diabetes consistently need support and information.

Host 1: Right. So once the researchers had this clearer picture of patient needs, they moved to the next logical step.

Host 2: Which was?

Host 1: Thinking about, okay, what kind of AI tools could actually help clinicians address these specific issues? And critically, they asked endocrinologists, the specialists, what they thought. How useful would these tools be and, importantly, how risky?

Host 2: What was the general feeling? Were they open to AI?

Host 1: Generally, yes. Quite positive, actually. The average usefulness rating was 4.3 out of 5.

Host 2: That's pretty high. Suggests real potential there.

Host 1: Definitely. And some specific AI tools really stood out as highly useful in the clinicians' view. For instance, AI that could provide evidence-based answers to those common, frequently asked patient questions.

Host 2: Well, I can see that immediately. Imagine the time saved. Clinicians answer the same basic questions over and over.

Host 1: Exactly, freeing them up for more complex stuff. Another big winner was AI that could summarize insurance policy changes and give real-time updates on covered meds.

Host 2: Yes, anyone who's navigated insurance knows that pain point.

Host 1: Uh-huh. And also automating patient education, specifically on managing and preventing low blood sugar, hypoglycemia, that was seen as really valuable, too.

Host 2: Which makes sense, given how serious hypoglycemia can be. Having clear, automated guidance could be a real safety benefit.

Host 1: Totally. Other highly rated tools included things like, um, generating templated responses for common lab result questions, but importantly, ones that could be customized.

Host 2: Ah, so efficiency, but still personalize.

Host 1: Right, finding that balance. Also, creating templates for authorization letters, another time sink, and developing educational content about how meal timing and composition impact diabetes.

Host 2: You see a pattern, don't you? AI for streamlining communication, providing quick, accurate info, and easing that administrative load.

Host 1: Absolutely. And good nutrition education's just so fundamental.

Host 2: It really is.

Host 1: The list of highly useful tools kept going. Automating responses for common admin questions, helping patients schedule appointments via the portal, um, crafting personalized, templated answers for dosage inquiries...

Host 2: ...like for that metformin example earlier.

Host 1: Exactly. And generating patient education on medication titration, you know, adjusting doses, and self-monitoring techniques.

Host 2: These tools really sound like they could empower clinicians to support more patients more effectively, like a force multiplier.

Host 1: Yeah, that's a good way to put it. And finally, also rated highly useful, AI providing updated medication guidelines or supply chain info...

Host 2: ...like drug shortages, maybe.

Host 1: Could be. Offering AI-driven insights to personalize education on glucose monitors and pumps, and generating really clear, step-by-step instructions for using those devices and interpreting the data.

Host 2: All things that empower the patient, giving them the info they need when they need it.

Host 1: Definitely. Now, it wasn't all just enthusiasm. While the overall vibe was positive, clinicians did flag some areas where they felt more caution was needed, where the perceived risk was higher.

Host 2: Okay, that's crucial. Where did they draw the line or at least feel a bit hesitant?

Host 1: They were less comfortable with AI, say, synthesizing all of a patient's data to try and prioritize urgent requests.

Host 2: Hmm, I can see that. Triaging based purely on AI.

Host 1: Yeah. And similarly, an AI system that automatically triaged incoming messages by urgency and topic also raised some flags.

Host 2: Okay.

Host 1: And probably the most understandable point of caution was around AI offering real-time interpretations of glucose data and suggesting immediate treatment adjustments.

Host 2: Right, that feels like it steps much more into direct clinical decision-making. Requires nuance, context, the human judgment element.

Host 1: Precisely. They clearly felt human oversight is essential there. The study summed it up nicely. The common theme for the useful AI tools was automating feedback, creating education, and crafting those templates.

Host 2: So AI as a helper, an assistant, providing information, reducing workload...

Host 1: ...and not taking over the critical clinical judgment calls independently.

Host 2: Makes sense. That seems like a reasonable balance, at least for now.

Host 1: Now, beyond the tools, the study also dug into who was asking what. They found some really interesting links between patient characteristics and the topics of their messages.

Host 2: Ah, so looking at different demographic groups, what did they find there?

Host 1: Well, for example, patients identifying as white tended to send more messages across a really wide range of topics compared to patients from other racial groups.

Host 2: More messages overall or on specific things?

Host 1: On specific things like scheduling, lab orders, pharmacy refills, thyroid questions, medication, devices, insurance, symptoms, imaging, even mental health and sleep. A lot of categories.

Host 2: That's quite a broad difference. What about ethnicity? Say, Hispanic versus non-Hispanic patients?

Host 1: The non-Hispanic group sent more messages about meals and diet, lab orders, insurance, symptoms, and imaging compared to Hispanic patients.

Host 2: Okay. And gender differences?

Host 1: Yes. Female patients sent more messages than male patients regarding scheduling, lab orders, thyroid issues, medication, general symptoms, imaging, and mental health concerns.

Host 2: Interesting. And did male patients send more messages about anything in particular?

Host 1: They did, actually. More messages about meals and diet and about devices and sensors compared to female patients.

Host 2: Right, so different focus areas potentially. Any other groups?

Host 1: One more: marital status. Unmarried patients apparently sought more advice on insurance or coverage issues and also on devices and sensors compared to married patients.

Host 2: These are fascinating insights. It really drives home that managing diabetes isn't a one-size-fits-all experience, does it?

Host 1: Absolutely. Different groups clearly have different information needs, different points of friction. It underlines why that tailored, personalized support is so vital.

Host 2: And it suggests where AI interventions might need to be targeted differently, too.

Host 1: Exactly. Which leads us to think about the key areas where AI could make the biggest difference based on all this. Diet, for instance. Huge volume of messages, right?

Host 2: Top of the list.

Host 1: So, the potential for AI to offer personalized nutrition advice, maybe linking CGM data with what patients report eating, that seems really promising.

Host 2: And clinicians rated AI help with things like carb counting and personalized advice as highly useful, didn't they?

Host 1: They did. It lines up. Then there's the administrative burden. We heard how much clinicians liked the idea of AI automating responses, generating templates, helping with scheduling...

Host 2: Yeah, freeing up their time significantly.

Host 1: The study even suggests this kind of admin support could be particularly beneficial for older adults or maybe patients who aren't native English speakers, potentially making things more accessible.

Host 2: That's a great point about accessibility. What about the thyroid issues?

Host 1: Well, given the high volume and distinct nature of those messages, especially from white and female patients, the researchers floated the idea of maybe a dedicated communication channel for thyroid topics.

Host 2: Hmm, interesting. With AI potentially helping with the initial sorting, basic info, or education.

Host 1: Something like that. And bone health, again, we saw white and female patients expressed more concern there. AI could perhaps handle common queries or even help streamline referrals if needed.

Host 2: Okay.

Host 1: And finally, mental health. Seeing higher rates reported by white and female patients suggests AI might even have a role in, say, initial symptom screening or helping connect people to support resources.

Host 2: It's really powerful when you start connecting the patient needs identified in the messages with potential, targeted AI solutions like this.

Host 1: It is. But we do need to keep the study's context in mind, of course.

Host 2: Right, the limitations. It was primarily one academic medical center.

Host 1: Yes, although it did include 22 affiliated centers and the patient population was diverse, which helps, still it might not perfectly represent every health care setting.

Host 2: And they mentioned not having detailed socioeconomic status data.

Host 1: Correct. That's a missing piece that could definitely influence patient needs and questions.

Host 2: And I suppose it only reflects the needs of people already using the secure messaging portal.

Host 1: True. It misses those who don't or can't use it. Although the study does note that diabetes patients generally have pretty high engagement with these portals.

Host 2: Okay. So limitations acknowledged, but the scale of the data, that half a million messages, still gives us some really valuable takeaways.

Host 1: Absolutely. I think the big picture, the main takeaway, is that AI genuinely has significant potential to support both patients and clinicians in managing diabetes better.

Host 2: By focusing on those specific needs that patients themselves are raising through their communications.

Host 1: Right. And it highlights that balance, doesn't it? High perceived usefulness for AI in areas like education, admin support, information provision...

Host 2: ...versus that necessary caution when it comes to AI handling raw patient data directly or making complex clinical judgment.

Host 1: Exactly. It's about smart integration, not blind adoption.

Host 2: Finding the right role for the technology alongside the essential human touch.

Host 1: So, for you, our listener, thinking about all this, what parts of managing your own health information or maybe understanding health topics do you think AI could most help with? And maybe, what bigger questions does this deep dive raise for you about the future of personalized healthcare, not just for diabetes, but overall?

Host 2: Yeah, it's definitely something to ponder. How do we leverage these incredible tools ethically and effectively, always keeping the individual patient at the center? That's the ongoing challenge, isn't it?