AI in Diabetes Care

00:00:00: Diabetes Insights, Breakthroughs and Innovators.

00:00:04: The EASD TV podcast

00:00:07: from the

00:00:07: annual meeting of the European Association for the Study of Diabetes.

00:00:13: Hello again!

00:00:14: Now one other things that people really enjoy here are their debates.

00:00:18: so we've got one this year on the use of AI in diabetes.

00:00:25: For the use of AI, I have Joao Felipe Roposo from Portugal

00:00:32: and

00:00:32: I have Harold Suri from Austria.

00:00:36: So let's go to you first of all, Joao just so that we're on the same page here.

00:00:43: what uses are expected?

00:00:48: And indeed already used in treatment for people living with diabetes?

00:00:53: So I think i'll have the easiest task here because it's like discussing an existing tool.

00:00:58: It is already there, we are using so its quite easy to be in favor of use of AI Because you've already had a demonstration how can used and be favorable for people.

00:01:11: What were speaking about?

00:01:12: We're talking supporting tools that analyze large amounts data And then learn from those amount on data applied with certain individual And then support the decision taken by people living with diabetes and healthcare providers.

00:01:30: We have fantastic examples now, we use automated insulin delivery systems like the insulin pumps... ...we have analysis of data coming from glucose sensors but much more than that it identifies different phenotypes so we can personalize treatment analyze difference see the risk of hypoglycemia, hyperglycemia and cardiovascular risks.

00:01:56: And even nowadays there is a risk to develop cardiovascular disease in ten-twenty years or something like that.

00:02:02: so use of AI.

00:02:03: it's just two that we healthcare providers should master but should clearly adopt as soon as possible because they are quite useful for our job.

00:02:14: Okay!

00:02:14: So great AI enthusiasm.

00:02:16: basically its uses interrogating very large data sets in order to give clinicians and people living with diabetes useful information.

00:02:27: But you're looking a bit skeptical?

00:02:29: I do, there is some sort of misconception.

00:02:32: just measuring a lot of data gives you a lot additional information... ...I don't think this necessarily the case!

00:02:40: And i agree without that.

00:02:43: yes we have a lot CGM data or cluster data But we don't have any data how to translate this into clinical meaningful outcomes.

00:02:54: So, We can predict something but we have no idea How do deal with it and change in clinical practice?

00:03:02: In the end I mean looking at our current guidelines.

00:03:06: You get a clear recommendation for certain treatment For certain drug classes And this is irrespective of which cluster or big-data group you belong.

00:03:16: so i think Treatment is way easier currently, but we still manage not to bring people into target.

00:03:24: That should be our focus and if you ask People living with diabetes what do they expect in primary care?

00:03:33: There where most of the people are treated They want things to be explained.

00:03:38: You want somebody who's dealing with their personal problems In this specific situation empathetic with them.

00:03:50: They want to have pain relief, that's what they're looking

00:03:54: for.".

00:03:54: I've never seen in any of these papers somebody asking for give me more AI!

00:04:00: So AI is a tool for the future but isn't really going to deliver on its promise.

00:04:06: and What's needed?

00:04:06: Is action now On The Things That Really Matter To Patients.

00:04:10: May I just say the comment we said so Just go back.

00:04:14: let's travel it ten years or twenty years where we didn't were even discussing this concept and We are just seeing the reports of people Picking their fingers putting the values in a diary Looking at those values trying to extract data by themselves And then coming into the healthcare professionals doctors and nurses Trying to make sense off that huge amount of information Spending a lot of time there precious time doing that kind of analysis.

00:04:43: now In present day With the use of these kind of tools, we can analyze that in fractions of seconds.

00:04:50: And then we have more time to do what we are fantastic and I fully agree with it but this is a time for people living under conditions like change decisions they take support their decision.

00:05:04: But really AI gave us this time To spend quality times with people living with diabetes.

00:05:11: So, can you enter the argument here and have a little bit of an argument?

00:05:21: Everybody always promised us that having no paper around would free us all from all sorts of toil.

00:05:32: Has it happened?

00:05:33: No It hasn't.

00:05:34: Are we going in the same way where they are?

00:05:36: It promises much is not gonna deliver what claims.

00:05:39: I think, the ideas are great and i fully share them with them but all these?

00:05:45: there is plenty of papers out .I call this May shoot good woodpapers.

00:05:52: we have ideas ,but very little can apply in current routine clinical care.

00:06:00: it starts with the idea... But We

00:06:02: could

00:06:04: do!

00:06:05: I'm just being on your side.

00:06:06: Thank you for that, and we are already there also

00:06:10: Maybe in the future but it's clearly not here now for prime time.

00:06:14: so We need to do way more.

00:06:16: But i've been around a couple of years In this field And i was working On It myself.

00:06:22: But currently There is Not That Much We Can Implement In Routine.

00:06:25: But

00:06:26: Existing So much At The Moment Doesn't Mean That Its Not Useful If You Don'T Start Working Now Specifically the ambitious to do more with this kind of tool and discuss how can we integrate that in clinical practice, then it's going to happen.

00:06:42: And there is a danger because if they don't adopt it by health care system ,it would be adopted by the general population looking for counseling or introducing their data from wearables .

00:07:00: If you teach people how to use this kind of information and tools, then we'll be out-of the system because we are not stopping people to adopt these kinds of information analysis.

00:07:12: I

00:07:12: fully agree but i see a huge difference between what we do in research... ...and what we're doing clinical practice which should do where we have evidence on hard outcomes And this part is lacking.

00:07:26: so I fully agreed that We should continue doing this, but the translation into what we do in practice needs some more data backup.

00:07:36: There is a worry isn't there about AI confabulating that it comes up with stuff where it fills-in gaps and nonsense of its own?

00:07:49: The other thing In order to train AI, you need to have very high quality data.

00:07:56: And you also need to had data from diverse sources because we're talking a lot about precision medicine of Congress this year and were very aware that there are many different genotypes and phenotypes involved in diabetes.

00:08:10: so we need be sure those as many

00:08:14: just remember that those phenotypes and genotypes they were analyzed with these kind of tools.

00:08:20: Yeah, we wouldn't be there.

00:08:21: My

00:08:21: point was that you need to have quality data on which AI can work or else you have data that only works for certain parts of the population

00:08:32: and I think thats an important point your touching upon.

00:08:36: currently across the globe having standardized data very well measured there we're struggling and again i see a huge difference between research in clinical practice in the research.

00:08:48: We can work on large data set without having identifiers in routine care.

00:08:54: you need always to have that link today patient.

00:08:57: so how are going to deal with genomic data with cgm date or dated?

00:09:01: it might, be stored anywhere in the US or elsewhere else, with different data protection regulations.

00:09:10: So I see a huge problem before this is ready for primetime.

00:09:14: The issue of quality's always important in clinical practice.

00:09:19: so it's about the quality of care and we are all concerned with that.

00:09:23: but when we're analyzing these issues on artificial intelligence... Of course the quality update and source date is very important.

00:09:32: But actually, these tools nowadays make our life much easier to analyze even not so well-structured data.

00:09:39: As long as there is a quality analysis on the top of that and it's possible now to do that because we know many of us have electronic health records but most of time with unstructured information they're... Now we've got good tools for analyzing texts and then check them and learn again And so this is possible now.

00:10:03: Of course, it's not perfect.

00:10:04: we are not there again but that's the point.

00:10:07: But if you don't do that It will never get there.

00:10:10: So we'll have to tools... ...but also these issues about who owns data and I think in a region where this concept of health data space and use of information We've been trying to regulate that in using this data by the health system and even the use of this health data for research, and these kind of tools.

00:10:34: And I think if there is something that they would call European regions because we are extreme cautious – that's a view from people in the United States or other regions around the world… We're quite cautious about using access to data and anonymising it.

00:10:49: so... So i think we have a safe space here at the European region To do this kind of tools and to do these kinds of practice.

00:10:56: And

00:10:56: people use AI already in their lives without really realizing they're doing it, so if you search for cat pictures on the internet... I mean, AI is serving you up with cat pictures?

00:11:11: Yes we do but healthcare data, a very sensible data.

00:11:16: And we all know this example.

00:11:18: I mean you do three searches and all of the sudden you get adversities.

00:11:22: meant that for whatever reason just fit to your search.

00:11:27: but You want really get then additional advertisement based on your health care data?

00:11:35: We're talking here about things can be abused.

00:11:39: people are really threatened by.

00:11:42: This could go to health insurances, this goes for employers.

00:11:46: So I fully agree that data protection is a major issue and we're not really talking here about social media but just putting out the photo of healthcare data.

00:11:59: The problem also is perception from public because most people wouldn't have that feeling they would post their health information in social media.

00:12:12: Actually, there are quite good papers for us and my country to know when the influenza peak is coming because it's just analyzing data from the social media doing a kind of text search with using AI And we know before people come into the health system With the peak off the flu It's already there Because people just comment... ...and its easy now To get information about Health coming from social media.

00:12:38: So

00:12:39: I mean, certainly in the UK.

00:12:42: it said that the major supermarkets will know you're pregnant before just

00:12:47: pause

00:12:48: simply because of your choice of foods?

00:12:50: Yeah

00:12:51: and is this really something we want to have?

00:12:55: The question if not If You Want To Have It Because It Already Exists And i think its better if We Can Discuss It.

00:13:01: Take The Advantage Help To Capacitate People to do this kind of use-of data, because we are not in the position to say don't do that.

00:13:12: It's already there so it cannot fight... The forces

00:13:16: left the stables and they're all gone!

00:13:18: So have you given up then?

00:13:20: If nothing can be done about them let us go with the flow.

00:13:26: That is where I agree on your side when saying well ...we need time for people.

00:13:33: But I can say we can use this to save our time and best empower people where they find reliable information.

00:13:42: Reliable information?

00:13:44: Yeah,

00:13:44: that's our job!

00:13:47: So for example one of the people living with diabetes was telling me there were something like a hundred sixty different CGM type devices available from China.

00:14:03: All of those devices will pull the patient's data and pool it.

00:14:10: There are a lot very high skills in AI, In China And you really have absolutely no control over the data at all.

00:14:21: what is being used for how its been use?

00:14:24: where has be used The purpose how data is being used, it's always the biggest worry for patients.

00:14:34: Absolutely right on that.

00:14:35: but again... It doesn't mean we don't have to control the source data and be able to ask for a demonstration of evidence.

00:14:46: so we agree with the existence.

00:14:48: We should take decisions based upon evidence.

00:14:50: Of course use technology has to be adopted according to evidence And we need strict rules.

00:14:58: Unfortunately, and that's not good for the European region.

00:15:01: We don't have such good rules to assess technologies especially in our area.

00:15:07: I don't know about others.

00:15:08: imagine it would be the same but we have an easy adoption of technology And quality coming from evidence might be not so good But doesn't mean that we should implement just regulations.

00:15:23: And that's one more

00:15:24: final point from you.

00:15:25: I think You made a very important point, but this is CGM data only This is glucose trends.

00:15:32: if you then start to put these together with health record data was medication data with lab data With genomics data Then it gets really tricky.

00:15:44: and i think there There Is Really?

00:15:47: Justifiable threat That the will be data break breaches that the data might be abused.

00:15:54: So I think

00:15:54: it should be

00:15:55: very cautious what we're doing.

00:15:57: Okay, so let me have your predictions of how you think a debate is going to go?

00:16:02: Is it gonna go your way or yours?

00:16:05: What clearly goes my way!

00:16:08: Well i have no doubts.

00:16:09: also because i don't think we can deny reality.

00:16:13: and so AI's already there being used... We just have to adopt it.

00:16:18: So it's going to be a very interesting debate.

00:16:21: I'm not sure where I fall down on this, but iIm not convinced.

00:16:27: either way lots of caution needed But I absolutely agree with you Xiao It's already here.

00:16:33: so on that when is it?

00:16:35: Tomorrow morning kids.

00:16:37: Friday

00:16:37: friday eight thirty

00:16:39: so friday Eight thirty.

00:16:41: if you want a good old argument first thing in the morning In Sophia Hall, then that's for you.

00:16:47: And if you want to watch it at your leisure later on You'll find us available through the EASD website.

00:16:53: So thank you both very much made The Best Man win and I'll be back with again shortly.

00:16:58: Thank you for joining us On the podcast of the European Association For the Study Of Diabetes.

00:17:04: We hope you enjoyed todays insights.

00:17:07: If you did Go ahead and hit subscribe, And you'll always stay up to date with the latest in diabetes research & discoveries.