APLAR 2026: The Rheumatologist and AI

0:00 Hi, I'm Dr. Sheila Reyes from the Philippines reporting live here in Soul, South Korea for AppLer 2026. 0:11 Day two of the congress featured a session on artificial intelligence and joining me today are two of the speakers Dr. Lisa Traboko from the University of the Philippines Manila National Telealth Center and Dr. Latika Gupta from RWT and the University of Birmingham. 0:32 Good morning ladies and thank you for joining me here today. 0:35 Thanks Sheila. The pleasure is ours. Good morning. 0:40 Yes, thank you again. Um so I'll we'll head straight to the interview. 0:47 I'll ask you first Lisa. Your talk was about AI assisted literature review. 0:52 So can you walk us through your workflow for this type of literature review? 0:58 Okay, thank you. So there's two ways that I usually teach my uh teach my students and what is also taught to us in informatics. 1:09 For those who are beginners and don't know how to code, if you use research rabbit or notebook LM, make to make sure that you read the journals and not just rely on the summary that is provided by the AI powered platform. 1:30 And for those who are a little more bit more versed in using coding skills um in visual code and the like, it goes both ways that you have a planned search um pattern. 1:46 And then once the skills code has run and provided you with your literature list, always check the veracity of its output. 1:59 Okay. Okay. So, I think that's a very relevant topic of yours since um AI is already everywhere and we rheumatologists are one of those 2:14 who are also using it and um so do you would you um require or do you think there are special skill sets that we 2:23 need in order to conduct an AI assisted literature review? 2:26 M okay that's a very good question because capacity building is definitely a challenge um in our medical informatics a lot of our literature search 2:44 is on primary care so I haven't uh done a lot of um literature search for specifically for rheumatology but that is because my group 2:57 there are very experi experienced coders and um so it's very easy for them to use AI. 3:04 Um for at least for our institutions were really falling behind in terms of capacity building. 3:12 Um there are some countries where in 20% are already actively using rheumatologist actively using AI for a lot of their research activities. 3:27 But I think we just have to continue on capacity building uh until we reach the point that um a lot can teach for more. 3:35 But medical students nowadays are actually more familiar with AI and machine learning. 3:43 So that's also exciting to see how they can tip the scale. 3:49 Yeah, thank you for that. Um when you talk about capacity building is are you like referring to um formal programs or formal uh trainings or are there like um just going to the internet and then checking using AI what like how to use or conducted literature review. 4:11 Yeah, it would be nice that if we have it included in our curriculum um because currently we don't. 4:22 that's why I am also taking extra classes on it. 4:24 So definitely hopefully we can include it as part of our research curriculum. 4:31 Oh that's interesting and we'll look forward to that. 4:35 Um just another question for you Lisa before I um jump to Latica's um discussion. 4:42 uh we know that it's all it's a it's an issue probably for especially for those um for AI use that um sometimes AI may hallucinate citations or misrepresent a study's methodology. 4:56 So how do you verify the accuracy of the AI's output to ensure your review is rigorously grounded? 5:04 Okay, that's very important definitely. So um I never use the the output as the final one. 5:14 So it's always a draft. It's not what you submit. 5:21 So you always have to check every every reference. 5:26 Um not 100% oh that's okay. So that will that is a um that's a habit that must be continuously practiced and not uh what you what what we call automation bias. 5:40 So um it's another thing that cannot be simply taught but practiced daily. 5:50 Okay. Thank you very much for that. So the operative verb daily there is assisted like it's just to help assist, right? 5:58 And it still boils down to the physician to the rheatologist on what the final report would look like. 6:06 All right. Thank you, Lisa. So, I'll come back to you later. 6:08 Yeah. Hi, Latica. Thank you again for joining me here in the interview and um I'm sure a lot of our audience are also looking forward to what you um your what your talk was about. 6:24 Um you discussed the uler recommendations on big data and AI and RMDs. 6:29 So can you tell us something about it? Um what you can tell us and uh what take-home messages probably that uh your findings or your research showed. 6:43 Yeah. So um I think the talk is more about my experience serving as a coconer for the update by Ular because Ular did its first uh big data guidelines in 2019. 6:56 Um and this is an update on that and a lot has changed since as we know AI has come in there variables there is the EUI act there is the health data act of Europe and uh we also don't know how to use the tools in the clinic. 7:12 we do not know how to interpret them and then the patients are using AI um there's these frontier models um and then for research 7:20 as well there's a lot that we do not know these can be used for reviewing can these be used for drafting manuscripts so there's 7:27 so many different facets to you know AI and uh already as we know the FDA has approved more than 1,400 devices already so for 7:38 rheumatology this could be a gamecher so it is it was definitely a time for an update and uh we had set out with these 7:46 objectives. 7:46 Now the task force meeting happened two weeks ago in Amsterdam and uh so Ular has a process around it. 7:53 Um just a bit of a spoiler this is a trailer and not the final movie because uh there's an embargo on the final recommendations. 8:02 I do know what they look like but uh they can only be released after approval from the ULAR's office. 8:06 So whatever we talk today is not really Ular's official position but really my experience and also the process because the process is standard and out there for everyone to see. 8:14 These are outlined in the Ular SOPs that are published in annals and also available on the Ular website. 8:22 So typically there is um like a small group that comes together often in an update. 8:26 The process is we involve people who um were a part of the previous guidelines. 8:30 So in this case it was u professor gok who led the previous uh task force myself venerito professor g burmuster is the other convenor and then we had methodologists uh paco and um and and then a team really of 15 rheumatologists and um ethics experts, legal experts um and and data science experts. 8:59 So it was a mixed team and then we usually have patient research partners. 9:03 We have two members from young rheumatology community that is MUNET and there's a selection process behind it. 9:08 So once the group comes together then uh there's consensus on what the searches will look like and and following that there's a systematic review 9:15 which the core team takes back and works on intensely for a few months and then there is a face-to-face meeting in which the voting 9:22 of the statements uh takes place and and then within a week after the meeting there's reordering of the statement and the level of agreement 9:32 um and following that the manuscript is submitted to the ul office once they accept it's submitted to anal for acceptance. 9:39 If an doesn't take it, it goes to the ULAR second journal which is ERO. 9:43 And if doesn't accept it, then it has to be published either on the ULAR website or another journal. 9:49 And the systematic review typically is published in the ERO. 9:53 So that's the ULAR process. And in the facetoface meeting, usually when we draft the statements, uh you have to be careful, they have to be short, uh clear. 10:02 We use terms like must, would, should, should they carry a lot of weight. 10:07 For example, if you say jacks are efficacious in rheumatoid, right, is probably you mean that they should not be used, even though it doesn't sound like that in traditional English. 10:17 Um, so it's not a it's it's language, but it's also, you know, recommendations versus points to consider. 10:22 So depending on the evidence base, if it's points to consider means the evidence is not our CTS or not robust enough. 10:30 So it's more like a direction we're heading in. 10:33 And uh and then there's voting agreement. 10:34 People can abstain from voting as well. And uh usually if there's 75% agreement uh in the first round then these are approved. 10:43 If not we drop the level of agreement in round two and three to 60 and 50 respectively. 10:50 And uh so yeah go on go on. So yeah, it was an amazing experience and uh we had uh two intense days uh discussing we started off with around uh I think 30,000 articles. 11:05 We had two fellows on the task force when Venerito and Savio Labella who's also here in soul with us and uh they worked super hard for a few months and uh screened and the searches were primarily it was presented as a poster at the last u um there were three themes. 11:23 The first one was uh AI and rheumatology and uh all the work uh done including machine learning because that covered the big data aspects and then the second search string was AI and healthcare and regulation. 11:37 So this covered both big data and AI but it was broader for science overall because we needed to look at regulation in general and 11:44 there was a lot of gray literature searches as well ular lex as well uh because the sources would not always be in pub or 11:51 traditional data sets that we look at and then the third one was on ethics and publishing um and that was again a general search 11:59 not really for rheumatology. 12:01 So, as you can see, we went above and beyond rheumatology, which is a um good amount of work. 12:07 We had to eventually apply some additional filters and uh hopefully we you'll see the results of that in Berlin. 12:15 Wow, that sounds really interesting. And I and while I was listening to you, I mean, I could just imagine the intensity, the extent of of um your work. 12:26 So hopefully um in Berlin we get to um we get to put you here again on room now so that we could uh you could share more data um and the updates the specific updates at that. 12:43 So you've mentioned about um just just because it came to mind um the ethics and I think um just a question for any of 12:54 you can answer um how important do you think is um the disclosure on AI use is like just thinking of the practicality for or 13:03 for example for the rheumatologists listening out there like how important is disclosing the use of AI in practically anything that you used AI for? 13:16 Yeah, I think this is a matter of you know time and uh situation uh really like if of course we're using it for diagnostic 13:26 decision making then the question is can you even use it and and that is pretty much why we needed this update as well that 13:35 how far can we go and what can we do with these tools um uh because there's so many tools out there the average rheumatologist 13:41 they're being asked to choose from their hospitals which tool can be used and which not and they do not know how to take that 13:47 decision. 13:48 So how can we judge a good tool and does it only depend on the approvals because the life cycle of a device pretty much starts after approval. 13:55 So you may use you know um a scribe uh to um to uh do letters quickly but the patients get discharged quicker and then you have new patients come in who are sicker and they occupy beds for longer. 14:08 So maybe the system actually the efficiency does not improve. 14:12 some of the invisible work becomes visible now and the bottleneck is moving and and then you know this reminds me of the industrial revolution 14:21 where there were factories designed around uh steam engines and then they had these uh electric motors come in and then they designed the factories 14:30 around that but that didn't really change anything you know the productivity didn't improve because the whole system had to be rewired around uh the 14:38 new development and AI is one such thing it's not going to change you For one small aspect of care, we need like redesigning of 14:47 the entire system. 14:47 So we are at swap the engine stage and for rheumatologists it's very important to know what they can do, what they can't and how to interpret these tools and where to disclose their use. 14:57 Okay, well said. I think the the intentional use of AI should also include the responsible um disclosures of of these things and not just 15:12 like applying this to a certain um situation or a certain tool but the importance of really looking into um how does this help uh 15:20 particularly in in clinical practice. 15:22 Lisa, do you want to add any? Um, I want to add like um in the terms of how we teach our fellows because I'm sure you must have noticed that in some conferences, some fellows actually use generative AI to design their posters. 15:44 But in reality, we don't have rules for can they use it or not use it. 15:53 So it's very hard to fault them for um doing that uh when there's there's no guidelines. 15:58 At the same time as Latica mentioned the additional work. 16:04 So my co our oncologist colleague um uh we met earlier said that her fellows were using um generative AI for journal clubs. 16:19 And then what she did was make them journal club 10 articles so that they don't know how to wing it anymore. 16:28 So in the end they got more toxic more work rather than making it easier. 16:36 But again because we don't have a curriculum, we don't have a guide to follow on how um to teach our younger generation use of this. 16:48 So I think it's another unmet need that and a challenge that probably will be addressed in future meetings. 16:56 Um, so we're almost out of time, but before I let you let you girls go, um, can you give us just one key message from your talks? 17:06 Um, that is, uh, like what is the most critical thing, sorry, the most critical thing you want physicians to start doing differently or, you know, adapting or um, yeah, doing differently? 17:20 Uh, Latika. Yeah, I think the pace of technology is you know evolving very rapidly and uh you know uh we're just talking we might need another update soon. 17:37 So I think uh we as physicians clearly this technology talks about you know taking over. 17:44 We don't know if it will take over really but what we do know is people who are in sync with technology have an advantage over physicians who don't. 17:52 So uh I think the key message here is to keep pace with times and upscale ourselves. 17:58 So AI literacy really is the solution and that is something probably should be a priority for all organizations moving forward. 18:06 Thank you. Thank you Latica Lisa. Uh same for me. 18:13 um literacy capacity building because it's hard to make rules for something that you don't know how to use. 18:22 Correct. Yeah, I agree. And so I think this has been a very insightful um discussion. 18:30 I know it's been short but you know there will be other um I think other opportunities and um with your talks um lots of people will definitely have more questions and um become interested more in the use of AI particularly in rheumatology. 18:50 So there you have it guys, follow I am Sheila Reyes once again and you may follow me on X at roomarampa and please tune in to room now for more information and updates on the applar 2026. 19:04 Thank you ladies once again. Thank you. Thank you.