TNR 11.6.2025 - Replay of RheumNow Live 2025 - Rheumatoid Arthritis Save

This podcast is a preview of the great lectures and speakers you'll hear at RoomNow live twenty twenty six, February seventh and eighth in Dallas, Texas. You can register at roomnow.live. In this podcast, you'll hear some of the speakers and lectures from RoomNow Live twenty twenty five, specifically on RA, PSA, SPA, and vasculitis. Hope you enjoy it. Welcome to Tuesday night rheumatology.

Hi. I'm Jack Cush with RheumNow. On these next few weeks, we'll be doing Tuesday night rheumatology, 7PM eastern time every Tuesday night. Tonight, we're gonna begin with our replay of RoomNow live 2025, specifically our session or pod on rheumatoid arthritis. Thanks to UCB who has sponsored this program.

We'll be giving you excerpts of three great lectures from RheumNow Live in February 2025. You'll be hearing from first, Doctor. Una Macriess, who'll be talking about key considerations when managing older adults with RA. Our second talk is gonna be by Doctor. Brian Englund, who'll be talking about multimorbidity and who will manage multimorbidity in RA.

And our third excerpt from Doctor. Jeffrey Curtis talking about AI and its application to rheumatoid arthritis. We're almost there, although it's not yet perfect. I want to remind you that RheumNow Live is coming up twenty twenty six, February seventh and eighth in Dallas, Texas. Be there, be square.

It's going to be a great program, short lectures, lots of interaction. You can go after this session and register at roomnow.live for either the twenty twenty six meeting or the twenty twenty five meeting. And if you do register, you'll have access to review questions, board questions for 2026, 2025, and 2024. Each year, we generate about 150 to 160 questions that you might like to learn from. So I wanna remind you, go to roomnow.live to get more information about these meetings, and to register.

And now our first speaker is going to be Doctor. Una Macriess, who is an associate professor of internal medicine at UT Southwestern in Dallas. Although she's since moved on is now head of rheumatology at the VA in San Diego, UCSD. Doctor. Macri is going to talk about managing older adults with RA.

So the geriatric 5Ms were introduced by the Institute of Healthcare Improvement and then refined by the American Geriatric Society. And the idea here is, can we improve the quality of care that our older adults receive? And so this is a nice systematic way to integrate geriatrics principles within our practice. And so I ask my colleagues, fellows, residents, students, to think about the five Ms when we approach our older adult in clinic. There's multi complexity, mind, mobility, medications, what matters most.

And we will walk through this in the context of a case, Ms. L. This is a 78 year old woman with seropositive, non erosive RA. She has low disease activity on methotrexate, Plaquenil, and sulfasalazine. Her comorbidities are listed here, very common.

This is nearly all of my patients. She has many non rheumatoid arthritis meds as well. Her social history, she's retired, her husband passed last year, and she lives alone. She has two sons who live nearby, three grandchildren. They're all incredibly busy.

They don't visit too often. Tobacco, no alcohol. She uses a four wheel walker. Her hands have kind of chronic deformity with ulnar deviation, but really no active synovitis. She has evidence of osteoarthritis as well.

And her labs are unremarkable. So, even though this patient is stable, and you would think doesn't take a lot of time in your clinic you can just continue the status quo, if you would like I want us to take a step back and think about these principles of aging. And in that brief summary, I've highlighted a lot of what's in red: multimorbidity, polypharmacy, falls, social isolation, disability, depression. And so we're going to walk through this. So, the first M is multi complexity.

This is defined as two or more concurrent chronic conditions. It describes the whole person living with multiple chronic conditions, advanced illness, and or with complicated biopsychosocial needs. So it's really not just the medical comorbidities, but it's also the psychosocial. And it's important to understand your patient's rheumatoid arthritis in the context of these other comorbid conditions. This is a nice study by Doctor.

Krausen et al, with a population based cohort I clicked suggesting that the older you are, the more multi morbidity you have here at the age of 80. And then RA versus non RA also has higher comorbidity. So these are all complex relationships that Ms. L lives with. And they're all bidirectional, whether it's cardiovascular disease, function, mobility, pain, recognizing the different types of pain at play.

So I think as rheumatologists, one way to break this down is assess and prioritize. We can look at pain and function. What is her RA disease activity? She has OA concurrent pain. Think about function, both basic or essential ADLs listed here, instrumental ADLs.

We'll talk more about mobility. This is a nice concept that is emerging in the geriatric space. It's called life space mobility. This is an individual's multidimensional engagement with their environment over time. And this is a really nice measure that looks at both mobility as well as social environmental engagement.

And I would argue that that's really important the older we get. Psychosocial comorbidities. You can assess for depressive symptoms or anxiety, for example, a very brief PHQ-two or PHQ-four. Assessing pain, depressive symptoms, along with cognitive impairment is very challenging, yet important. Social isolation.

I always welcome family members, caregivers into the room. Ask them what's happening. Talk to the patient first, but you can always work with the individuals in the room. And the reason I like this is it enhances safety. If I'm trying to improve adherence to plaquenil or some medication, I really engage the caregiver in the room.

So, moving on quickly to what matters most. I would argue this is the most important M. We really, you know, for older adults with rheumatic disease and multimorbidity, clinical decision making must occur in the context of their individual preferences and goals. And I don't think we always ask about it. Clinicians bring our medical expertise.

The patient brings expertise in themselves and what is important to them. And I think one of the biggest challenges that we're learning, and I'm learning from cardiology who's trying to do this first, is really how do we align our management plan with what matters most to the patient. There's no one right way to do this. The key is partnering in person and family centered care. Some ways to ask about what matters most is listed here.

And then ask yourself, is the plan we're discussing aligned with the patient's goals and preferences? We published an article reviewing the five M's in rheumatology and The Lancet Rheumatology that was published in 2024. This is one of our tables that outlines the various approaches that we can use when working with our patients to elicit what matters most. So I want to tackle barriers to complexity. I hear a lot of questions along the lines of, How will I assess and manage the multi complexity in this patient in one visit?

And I just want to remind us that I think one of the things we love most about being rheumatologists is that over time we get to know our patients. We see them every three months to six months. We can tackle this over time. I want to caution with the following questions. You know, I don't ask about it because I don't know what to do about it.

I hear that a lot with cognition. I don't know how to handle delirium or dementia, so I I won't even address it. Well, that's why we're here, and that's why we need to learn. And that's why, also, we need to rely on our interdisciplinary teams. What about, What do you expect?

She's 78 years old. Of course, she'll have pain. So, I would just also guard against that. Don't make assumptions because of someone's age, they're experiencing a certain symptom. And we're going to talk about ageism briefly towards the end.

It's time consuming. I can't fit it in today's visit. Of course, it's time consuming because it's complex. So, assess and prioritize, and you can handle this over time. You can tackle it over time.

And then, something we're going to hear from Doctor. England in the next presentation, you know, this doesn't really belong in rheumatology. This is not my issue. This is someone else's issue. It's cardiovascular disease, so it's cardiology.

This is for primary care. I think that optimizing outcomes for our patients with rheumatic disease requires managing several of these comorbidities. And then I would also say, when we go through our review of systems, we often assess a lot of these multi morbidities. And how can we reframe that in a way that can help us prioritize? And lastly, lean on interdisciplinary teams.

This is a nice schematic that Travis Welch, a rheumatologist, who was a resident at UT Southwestern, we put this together for a paper just outlining all of the different specialists who can work with us, depending on the modality. So I'm going to move towards medications, which is the third M. And a lot of these slides are borrowed or shared or modified from my colleagues who are experts in the M. So this is Jeeha Lee at University of Michigan. She is the expert, I would say, in polypharmacy.

So medications involve polypharmacy optimal prescribing AEs and medication burden. The definition of polypharmacy is concurrent use of greater than or equal to five medications. I think someone is pushing me forward, which is a sign. That's okay, too. So, I like this quote.

Numbers aren't the enemy. A necessary, ineffective, or harmful prescribing is. So I want us to keep that in mind. And keep in mind, among older adults, the prevalence of polypharmacy is very high. For RA, it's around fifty percent.

Back to the case of Ms. L, 78 year old woman. As I mentioned, she has both triple therapy for her rheumatoid arthritis and multiple non RA meds listed here. This exceeds 10 medications. I have patients who often I keep scrolling.

I see some of our fellows come to the VA for clinic. Some of our patients have 30 medications. So I want to introduce, and I'm sure everyone here already is familiar with potentially inappropriate medication use. The American Geriatric Society publishes Beers Criteria that are updated every three years. And these provide information on PIMs to be avoided by older adults.

So, let's just be aware of these. And there are drug disease and drug drug interactions. Commonly used by patients in our population include benzodiazepines, steroids along with NSAIDs, gabapentin, nifedipine, NSAIDs, opioids. And a lot of these lead to altered mental status potentially. They can lead to falls.

Some of them can lead to GI bleeding, renal issues, or liver abnormalities. So, optimizing medication use. I put the definition of deprescribing up here. And I think Doctor. Curtis brought up a great point.

You know, can we identify some patients potentially with AI who could benefit? I'm not so sure. I think this is really where shared decision making comes in. The best situations are when a patient comes to me in clinic and asks, Doctor, can I come off of some of my medications? And then we assess their disease activity.

Are they in sustained or persistent remission, which is defined as six or more months of remission? We can maybe start talking about this. And I just want to highlight that we have guidelines, clinical practice guidelines for RA, both through ACR and through EULAR, with very little mention about optimization or deprescribing. And I think the key here is we have very little data, specifically in older adults. Older adults are often excluded from our trials.

So, we have very little to work with. This is an article that I published with Doctor. Jeehali and Namrata Singh focused on, you know, deprescribing as an approach when less, maybe more. And this is a nice schematic. You know, you have medications for treatment of RA.

If they're on treatment, if they're on DMARDs, you know, and in persistent remission, can you consider tapering rather than discontinuing? If they're on glucocorticoids, can you consider tapering? I would argue that we should be tapering glucocorticoids in every single one of our patients. And then if you have an older adult who is not on DMARDs, I put a star here because I want to just highlight the fact that one third of RA patients are older. And among the patients with late onset RA, less than a third of them are on disease modifying drugs.

So, what's going on there? Right? Is there something in our mind? We are not treating our older adults as aggressively potentially. So, the care team and resources are listed on the left.

If you have access to a clinical PharmD, that is certainly ideal to partner with. Moving on to the MIND, this is in collaboration with my good colleague, Doctor. Elena Miyasidova, at the Mayo Clinic. When we think of MIND, we think of delirium mentation, dementia, depression. So, this 78 year old woman with seropositive RA has had a cognitive decline over the past three years.

I want to highlight that in this population based cohort, almost forty percent there was almost a forty percent increased risk of all cause dementia in RA versus non RA, adjusting for age, sex, education, cardiovascular risk factors. This is the spectrum of cognitive impairment. So, here we have a prodromal stage. This can take decades. This is where a lot of the risk factors accumulate.

You can have a prodromal stage here where you're developing mild cognitive impairment, and then dementia. And some of the neurocognitive domains that we think about include memory, executive function, language, visual spatial skills, complex attention, and social cognition. The risk factors here are listed, and I want to say a lot of these are modifiable. And a lot of them are still unknown, so including inflammaging, gene environment interactions. So, when and how can we evaluate for cognitive impairment?

So, these are things that are observed or reported, and then we can screen. There are various tools. And then I suggest a referral. Not many of us are trained or have time to actually manage cognitive impairment directly in our clinics. Keep in mind, dementia mimics delirium, CVA, effects of medications, depression, pain.

Sometimes I question the reliability of patient reported outcomes in these situations. Onto the last is mobility. Here we think about frailty, think about body composition factors, mobility function, impaired pain, gait and balance, and then fall injury prevention. So, again, this patient's comorbidities include osteoporosis. She had a fall since the last visit.

She was coming into her house, tripped and fell. She moved from a cane to a walker for mobility. She has high fear of falling, hesitant to leave her house. She spends most of her day on the couch, and her son checks on her daily. The question is, is she frail?

We didn't talk about criteria, but there is a phenotypic freed criteria that's listed here, and she does check the box for frail because she has exhaustion, she has weak grip strength, and low physical activity when you evaluate her in clinic. This is a nice schematic from the frailty review out of Lancet Rheumatology. Frailty is really due to multifactorial contributors that include inflammation sarcopenia medications. Frailty also drives this vicious cycle where you can have falls that lead to low self efficacy, fear of falling, that leads to inactivity, that leads to reduced balance and strength, that leads to more falls. And I just want to highlight that in our patient population with rheumatoid arthritis, thirty four percent experience any falls.

Sixteen percent have recurrent falls. A prior fall is, of course, the biggest risk factor for a future fall. I want to also highlight that a lot of the risk factors in our rheumatoid arthritis patient that leads to bone density and bone quality are also contributing to bone composition changes, and all of that can contribute to falls and potentially fractures. I want to highlight the CDC falls assessment. This is the STEADI framework.

And what I like about this is it includes three very simple questions about fall risk, and we're going to talk about that at the end. This is a summary summary slide from our five Ms review in the Lancet room that basically reviews each of the five Ms and the potential collaborators or teams that you can work with to address the five Ms over time.

That was interesting. Now we're going to move on to our next lecture by Doctor. Brian Englund. Brian, as you know, from Nebraska is great on many subjects around RA, ILD, RA outcomes, but he's really interested in this topic of multimorbidity being a major determinant in RA outcomes. And he's gonna address who should be managing that.

Our first case is a 36 year old female who's been referred to you for joint pain and positive anti CCP antibody. And as a matter of fact, I actually have a picture of this patient as they come into clinic. And you see this patient and when you walk in the door you think, wow, this person looks pretty healthy. Almost looks like they're waking up on a Sunday morning getting ready to make brunch. But the truth is this isn't a clinic picture, this is from an RATV commercial.

Now let's go through another case. Now this case is a 66 year old male who's been referred for also joint pain and positive anti CCP antibody, just like the last. The difference is, well, this patient has a past medical history of COPD with chronic respiratory failure requiring supplemental oxygen, diabetes, hypertension, hyperlipidemia, CKD3. And when your office staff was trying to print their intake information, your printer ran out of toner during the medication list. And I also actually have a picture of this patient as well.

And this is what this patient looks like. Now spoiler alert, this is not a patient from an RA commercial. So these look pretty different, right? Both we suspect have rheumatoid arthritis, but they have different types of rheumatoid arthritis. The one on

the left has what I

call TB rheumatoid arthritis, whereas the one on the right has what I call real rheumatoid arthritis. And the reason that I call this real rheumatoid arthritis is when we look at large observational cohorts of people with rheumatoid arthritis, more than three quarters of the time, those people don't just have rheumatoid arthritis. They also have other chronic conditions. So the majority of the people that we see, the real rheumatoid arthritis looks like that gentleman on the right. Now this becomes problematic because if we think about how we take care of real RA, we want to go to our evidence.

But our evidence starts usually with clinical trials. And which patient makes it to the clinical trial? Well, it's usually the TBRA patient that makes it into the clinical trial that informs our clinical practice guidelines that get read by us as providers, but then when us providers knock on the door to go see a patient, we see the real RA patient. So it's a bit of a conundrum. Now I want to get everybody on the same page in terms of what multimorbidity really is.

We're gonna start with this concept of comorbidity, which most of you are familiar with. And in comorbidity, we have RA at the center of our focus, And we have these other chronic conditions that we're interested in how they relate to rheumatoid arthritis. Now, multimorbidity is different because at the middle of all of this is the patient themself. And rheumatoid arthritis, instead of being at the middle, is one of these other chronic conditions relating and impacting the patient. So this makes it more patient centric.

But the benefit really goes beyond that. Because what we realize is we know that rheumatoid arthritis is associated with these other chronic conditions. And many of these other chronic conditions interact with each other. So if we add those connections to our multi morbidity model, we really move to what I call this multi morbidity web. You can see that this patient is now trapped in all these interconnected chronic conditions, including rheumatoid arthritis.

This is helpful to think of it this way, because when a patient gets trapped in the web, they have poor outcomes, not unlike a fly that gets trapped in a spider's web. These poor outcomes are general health related, such as reduced quality of life, impaired physical function, shortened lifespan, but they also directly impact RA related outcomes. We perhaps use advanced RA therapies less frequently in people who are trapped in the multimorbidity web. There tend to be poor treatment responses and less likelihood of ever achieving target disease activity thresholds. And we also worry about that these patients trapped in the multimorbidity web having more complications after we start them on different RA immunomodulatory therapies.

So if this web is problematic, it's really important for us as clinicians to understand when this web is starting to get spun. So we did this study using a large US commercial claims database. And we identified when people were diagnosed with rheumatoid arthritis and matched them to people without rheumatoid arthritis. You can see on this figure that the trajectory of gaining other chronic conditions. The RA patients in blue gained these conditions at a faster rate than the non patients in orange.

But importantly, even at the time of RA diagnosis, as indicated on this slide, there's a difference in multimorbidity burden between RA and non RA. And if we look a couple years before their RA was diagnosed, we see that is really when these lines start to separate. So it's really right at the time of RA, or even before RA, where this multimorbidity trajectory diverges between RA and non RA patients. Now another related concept is that of difficult to treat RA. And this has been a hot topic.

We've seen EULAR's definition. We've now seen a number of studies that have come out that are characterizing what those people look like who have difficult to treat RA. And it was struck me as I looked at one of these articles, a nice review by this group that's shown on this slide, and they characterized what are these major causes of difficult to treat RA. And I said, you know, most of those overlap with multi morbidity. And so indeed if you go through, I went through and starred all of these causes that really can be related to multimorbidity.

So in fact, multimorbidity and difficult to treat RA are not necessarily completely separate entities, it's that our people that are trapped in that multimorbidity web are indeed the high risk, most likely to end up with difficult to treat rheumatoid arthritis. So we've gotten all on the same page in terms of what multimorbidity and the multimorbidity web is of rheumatoid arthritis, But that's not enough just knowing about the problem. We actually have to intervene if we wanna make our patients' lives better. So how do we act? Let's start with what we do as rheumatologists, where we treat rheumatoid arthritis.

And the question comes up is should we treat rheumatoid arthritis differently in terms of multimorbidity? And this can be confusing. And one of the ways I think that makes more sense is if we separate out our goals. We can have the goal of multimorbidity prevention, or we can have this goal of trying to reduce multimorbidity progression. The difference is with multimorbidity prevention, we're starting with a patient who's not already multimorbid.

And in this patient population, it's probably better to aim for a target of remission than low disease activity, for a couple of reasons. The first reason is that we know that people who are able to achieve remission have the lowest risk of developing other chronic diseases, such as heart disease, or lung disease, or infections. The other reason is that when we use our advanced therapies or combination therapies or how much we're willing to suppress the immune system in these patients, our concern about adverse events such as serious infection is not as big of a concern. These patients are not at higher risk of those complications. Now we can contrast that with multimorbidity progression, where the patient in front of us is already multimorbid.

And while we love to be able to reverse that, that's not typically the case. So more often our goal is to try and prevent this from progressing. And in this patient we may favor targeting low disease activity than remission. And a big reason why is because the adverse effects that can happen with immunosuppression are increased in this population. So if a patient with more multimorbidity is a higher risk of having serious infections or other complications with our therapies.

The other piece is that we aren't sure that our disease activity measures are telling us the perfect answer in this patient population. So we have a number of different RA disease activity measures we can use in practice. And these pull from patient information, provider information, laboratory data, or imaging data. And with these we can construct all sorts of different definitions of what is remission. And these perform differently.

We know we categorize more people in remission by a DAS28 than we do by a C dye or an S dye. And so, as we look at what pieces inform this that might be impacted by multimorbidity, the patient global is one that really jumps out. And you've probably seen this in your clinic where a patient reports a patient global that's quite high, you assess them and you think the rheumatoid arthritis is doing really well. And that's because it's very hard for patients to sometimes discern why they're not feeling well, if it's from the rheumatoid or other chronic conditions. So recognizing this, the new ACRU law remission criteria have actually loosened how stringent they are in terms of the patient global.

So now you can be in remission using the Boolean two point zero criteria, with a patient global up to two on a zero to 10 centimeter scale. So if we understand our target, the other question then from a rheumatoid arthritis treatment perspective is if we should use our therapies differently in people who are multimorbid. The challenge is, at the very beginning I told you, the evidence is generally not on people who are multimorbid. It's the TBRA patients that make it to the trials. But I told you a little bit of a fable, because actually we do have a multimorbidity trial.

We just don't think of it that way. So oral surveillance was really a multimorbidity trial. As we know from this trial, there's over four thousand active RA patients despite methotrexate who are at least 50 years old with a cardiovascular risk factor. But when we look at the characteristics of these patients, we see it's a multi morbid population, with twothree of people having hypertension, almost twenty percent diabetes, almost forty percent with extra articular RA, and and over ten percent with coronary artery disease. When we look at the outcomes in this multi morbidity trial, we see that the efficacy of TNF inhibition versus JAK inhibition was very similar.

The change in disease activity was very similar throughout this trial. But we know the story for the safety. The safety, we did not see the similarity. This trial did not meet non inferiority with higher risk of MACE and cancers in those who were treated with JAK inhibition compared with TNF. So really importantly, this is telling us that in a multi morbid population, the safety of our therapies may indeed differ.

This is really striking when we look at some of the post hoc analyses that came out of this trial. So when we look at MACE risk for example, and people who had a prior cardiovascular event, which is identified here in red, compared with those who only had risk factors, the risk associated with JAK inhibition was very different. Those with prior cardiovascular events had a number needed to harm of only 16, whereas that number really skyrocketed up to almost nine hundred in people who only had risk factors. Again, really importantly telling us that in multi morbid populations, we very well may see differential comparative safety of our disease modifying therapies. And I really hope that we can see more trials in these types of populations of real RA instead of just TV RA going forward.

Now let's move on beyond what we are very comfortable with as rheumatologists, treating their RA. What about the things beyond RA treatment? And particularly, who's going to manage these? We don't wanna end up in this situation that Spider Man is in right here where the rheumatologist and the PCP are standing there just pointing at each other. So who's going to run the show?

And some of our best evidence on what's happening out there already comes from Christy Bartle's group at the University of Wisconsin, where they did these interviews of rheumatologists and primary care providers on the preventative care services they were providing. And so on this bar graph what you see is the percent of primary care or rheumatologists who are doing these different preventative screenings. And what we see from this study is that rheumatologists and PCPs have complementary roles. So PCPs were more likely to do things like checking lipids or screening for diabetes or ordering cancer screenings, Whereas the rheumatologists were more likely to do things such as administering vaccines, talking about bone health or lifestyle modifications. So if we think back to that picture, it's not a finger pointing exercise, but rather it's both, it's all providers as part of the care team, really looking at what do I offer, what can I provide for our patients to make sure they're getting the care they need as part of this team?

In terms of kind of broad multi morbidity interventions, there's not a lot of data specifically in rheumatoid arthritis. So some of the strategies that have been proposed about how we can help multi morbidity patients really come from the general population and primary care practice settings. These have included things such as having structured multi morbidity screening and assessments to make sure that we're capturing these risk factors and these other chronic conditions. There are things such as enhanced care delivery models, where we can try and build multidisciplinary clinics or multidisciplinary care teams to really meet the needs of a multimorbid patient. It's also empowering these patients who may be dealing with multiple chronic conditions, balancing taking care of all of them, monitoring them.

And so things like goal setting, self management support, or digital health can be really helpful in these settings. Where we do have a little bit of data specifically in rheumatoid arthritis comes from the COMEDRA trial. So this was a six month open label randomized control trial of almost 1,000 RA patients. And these patients were randomized to either a structured comorbidity assessment versus self assessment of disease activity as a control. And in this comorbidity assessment, what they received was a nurse led screening, where they reported existing other chronic conditions, they went through and identified different risk factors, And then they were provided some management recommendations to the patient, to the rheumatologist, and to the primary care provider.

Over the next six months they tracked how many actions were taken for multi morbidity. And what you can see from this slide is that the group that was randomized to this nurse led comorbidity assessment intervention had more actions taken to try and prevent the progression of multi morbidity. These included things in cardiovascular disease, infections, cancers, and osteoporosis. So this is pretty impressive data showing that even a fairly light touch that potentially could be integrated to our clinical settings has the ability to really improve how much we intervene on multimorbidity. But beyond these general strategies, these broad multimorbidity intervention strategies, more often what we're probably finding ourselves within clinic is trying to address some of the key conditions that make up multimorbidity and rheumatoid arthritis.

So for the rest of the talk I wanna focus on this slide. Cardiovascular disease, cancer, lung disease, infection, osteoporosis, and mental health. And I think you understand why we chose these. These are all really closely connected with rheumatoid arthritis. Now as we go through these and talk about how we might address them, I wanna acknowledge that there are gonna be many factors that influence how you specifically do these interventions compared with me or compared with who you're sitting next to in this room.

These include things like our comfort level with managing this condition specifically. It could be due to our healthcare system, how connected we are in that healthcare system, our practice environment, the relationships we have with the patient's other treating providers, and even how closely the patients are following with those other providers. Things like insurance status and social determinants of health, as well as the patient's capacity to take on these other medical actions, all influence specifically what we can do. So the important thing is it's not a one size fits all approach with multimorbidity. There's not one perfect way of doing it.

We'll all have to adapt to the patient in our own environments. But we don't have to fret, we don't have to worry about that, because as long as we are doing something, that's much better than doing nothing. So doing something to address multimorbidity in our RA patients is really the best thing for.

Sage advice from Doctor. Englund. Our last but not least wrap up talk is gonna be about artificial intelligence. Doctor. Jeff Curtis from the Rheumatology Immunology Division at University of Alabama Birmingham.

Jeff has a long interest in teaching education research, but he's really interested in AI. He's gonna talk about some practical applications and maybe answer the question, is AI gonna put all of us out of business? Listen up.

So I'm going to zip through this quickly. 24 year old pregnant woman, she's twenty eight weeks along, she has fever, chills, pain in her ankles and her knee, headaches, she has an erythematous rash, pregnancy is otherwise going fine, baby's okay on an ultrasound. What is the best thing that you would do to treat somebody for presumed Lyme disease? This isn't a polling question. Probably you're not going to give ibuprofen or only ibuprofen.

So if you ask GPT this, GPT says you should give this patient tetracycline. And the physician who answered this said you should give them amoxicillin. So tetracycline, of course, might be a reasonable consideration except that the patient is pregnant. And what maybe those in the back can't see, so the statement coming out of GPT, this is a direct quote that tetracycline is a commonly used antibiotic for Lyme disease and is safe to use in pregnancy. That sounds really definitive, like do you see any qualifiers like might be safe, or know, pharmaspeak with rabbit droppings of safety?

No, no, no. This is just go do this, and yet this is wrong. Like this would not be a good idea. I think most people would say, you don't give tetracycline to a pregnant patient. So the point is that just because it's coming out of the computer or a large language model doesn't mean it's right.

Those are called hallucinations. That is literally the technical term. I could give you lots of other examples, this is the best rheumatology example I've got. It makes up stuff. It really makes up stuff.

I found it and corrected it on addition errors because these models are quite good with text, they're terrible with math. The companies that make them are trying to do things about that, it's definitely gotten better, but this is a big concern when your model hallucinates in a very authoritative way. Let me now switch gears. Other uses of artificial intelligence, not so much generative AI. So this is thermography, this is a thermal imaging camera, the one we have at UAB is about $1,500.

Patients could buy these, connect them to their iPhone for a couple $100. It uses the heat signature in the hands looking for joint inflammation. It cleans it up, converts it to grayscale. In this study, the group in gray basically is RA in remission, the group in dark is active discrimination to separate is this active versus not active RA is pretty good. The receiver operator curve or accuracy is point eight, almost point nine.

So the use of this AI technology would be your patient has a camera at home or attached to their iPhone, she takes a picture of her hands, sends it to you, and now you don't need to see her in clinic because you know her RA is totally quiet, she's in remission, and you've essentially substituted a thermal imaging camera coupled with AI for the joint exam that maybe now you don't have to do, now you have a telehealth visit. So another mechanism if you don't own a thermal camera, because most people don't own a thermal camera, so you have your iPhone or other smartphone take a picture of the finger folds on the dorsum of the hand. So basically AI is reading the length, width, shape, and the diameter of those finger folds, essentially the wrinkles on the back of your hand. And if your joints are swollen, it is able to detect using this measure they created and validated called the finger fold index, does this patient have synovitis. So now she doesn't need a thermal camera at home, now she just needs her iPhone, and you could, assuming this, you know, is really born out and becomes ready for prime time, it's not yet.

But this is AI looking at images to triage how many swollen joints does the patient have. Again, these are ideas, this is not, oh, I'm going to go download that app and start using this routinely. It's an application of AI. A third one, if you have a ton of data in your EHR, as we now all do, could you cluster patients into unique and distinct phenotypes to predict the likelihood that they persist on drug or they do well, that's definitely a possibility, and harnessing large scale data from the EHR to be able to cluster RA patients and phenotypes, you could give her information about prognosis, provide better education, management, etcetera. That's the concept of harnessing AI to cluster patients based upon their trajectory of improvement or their prognosis.

Last one, and I think Doctor. Giles may talk a little bit more about this topic. So one group out of Europe took the tofacitinib clinical trials, extracted a bunch of variables, and had the goal to predict serious infections using a machine learning based model. So the idea is Mrs. Smith is in front of you, she's getting ready to start a JAK inhibitor, could you predict the likelihood using AI or machine learning, just a branch of AI, bad things that might happen to her to help you pick a drug based on safety, or at least talk to her about the safety?

So for example, if this AI model told you, hypothetically, that a patient's one year risk of a serious infection, she's 63, she's diabetic, she smokes, and a TNF inhibitor is eight percent, and with abatacept it was four, with a JAK it was 11, useful? Would you use this? If this was built into your EHR, would this be helpful for counseling or drug selection based on safety? How big of a difference would it have to be for you to say, Yeah, this is actually a helpful tool in my practice? This is, I think, where we're headed as a field.

I think we're not too far away from this application, where we are predicting risk or even predicting the likelihood of benefit in RA, probably as one of the first diseases we could do this with in rheumatology. Lastly, you could extract RA related disease activity data from the EHR notes for those who even bother to do a CDI or any sort of a joint count and document it. Most EHRs don't have structured data fields to type numbers into, even assuming you measure things. In fact, when this was done within the ACR's RISE Registry, the accuracy as measured by this metric called the F1 score, it was only about 50%. So the caution here is that you can use AI to essentially suck all the data out of your notes because you don't want to type numbers in because that's too bothersome.

But if you're going to use what the AI says is coming out of your note for doing a prior auth or knowing if the drug is working, the accuracy at present out of any of these models is not great. I don't want the insurance company to reject the patient's prior auth next year saying, Oh, this drug isn't working because AI screwed up how active this patient's disease is on a biologic or a JAK inhibitor. Switching gears to another application, could we predict use of, or the effectiveness of a drug you get ready to start? So this is a paper that Stan Cohen led. This is the Prism RA test by Cypher Medicine.

So this is looking at gene expression features, and although I don't want to get into the merits, whether you love or hate this or indifferent, basically this is an application of AI. Some of you, if you use this lab test, may already be working with. Essentially, you create this single decision tree with branches, you know, is the CRP elevated, is rheumatoid factor positive, other biomarkers. You create a tree, and then you create a forest of trees to make a decision, is this drug likely to work? You're only using the data, you don't have to prespecify any of the biologic predictors, and then it got very fancy.

And again, this is the predictive AI element of what's happening. So for those using this test or things like it, this is already in clinical practice, you're already using AI if you're using this diagnostic test. So the study that was done basically said if you are a predicted non responder to TNF, and in fact if you, and it was about fiftyfifty in this study about whether you're a predicted non responder, if you go ahead and use a TNF anyway, and you're a predicted non responder, the outcome is not very good. If in fact you do what the test suggests, don't use a TNF, the outcome is better. But in fact, if the signature was not detected, the AI says do whatever you want, the result is roughly about the same.

So this is an application that some are already harnessing to be able to use AI. Another mechanism that isn't RNA sequencing data, so some of the newer mechanisms in RA that we have to grab biomarkers and then apply AI would look at cell free DNA. When I first heard about this I thought, wait, isn't that for cancer detection and infections? But in fact, no, we have this in rheumatology. The idea is that when live cells die, they release cell free DNA, and a lot of that regulatory grade chromatin gets released, you can recover it in the bloodstream, you don't have to do a synovial biopsy here.

And in fact, you can measure the cell free DNA and the regulatory active chromatin fragments, you can map them back to the epigenome and the transcriptome, and what that ends up getting you is using some mapping tables, you can reconstruct gene expression, but you don't have to do RNA sequencing. RNA sequencing is expensive, it takes a bunch of time, those samples degrade, but using some of the databases now, you can essentially extract what you would otherwise get out of RNA sequencing, analyze it with RA, and this has been looked at in a number of different RA cohorts. And so far as a predictor of whether a TNF or a JAK or something else is going to work. So far, at least in the hands of this group, sensitivity and specificity is in the 90 ish percent range. And the interesting thing compared to some of the endeavors to predict treatment response in RA, it's not enough to predict response or non response to a TNF.

You need to know that response to this drug class is good, but to this other drug class is bad. It's not enough to predict response or nonresponse to everything. I can already tell you who's not going to respond. You know, your sick, depressed, anxious fibro patient that has chronic back pain on opioids and is depressed, like that's not useful. I need to know that you're going to respond to something different, and then mapping that to the biology, and AI is being used to do that both synovial biopsies and liquid biopsies like this.

However, one note of caution: what can you realistically expect if you're going to predict treatment response out of RA or any other disease? So what we care about is the positive predictive value of that classifier or the diagnostic test. You want to know if the test says I should use a JAK inhibitor, that it's accurate, that I can trust it to make decisions. But the important part is that if you're looking for a needle in a haystack, it's really hard for AI or any other algorithm to do a good job. And the problem is that that accuracy or that positive predictive value decreases as the prevalence of what you're looking for decreases.

So what does that mean? So it means that if you have a biologic naive person, her likelihood of having an ACR fifty response on average is about forty percent. So if you're predicting non response, you know, if you've got a test, you can do pretty well. Your accuracy on the Y axis here could be better than 90%. If you're trying to predict response, it could be almost that good.

But what if you're not looking for a needle for something common, 40 to sixty percent? What if you're looking for somebody who is bio experienced? Now the ACR fifty is only 30%, and now the best that that test can do. Same test, same sensitivity, same specificity, it's the same algorithm. Now the accuracy is only seventy five percent.

And now if we have a train wreck patient, failed multiple things, she's on opioids, fibro, a bunch of medical problems, you know, if the likelihood she's going to get an ACR fifty without your AI model is 10 or fifteen percent, even with a great predictive model, you're only going to get about a fiftyfifty coin flip and then Roy Fleishman is on you and is mad about that. So the point is that how much of a needle in a haystack you're looking for is very related to how well your AI prediction model is going to perform. So that's just Epi 101. But if we're all expecting, oh, my AI accuracy better be 90% or better, think again, because if you're predicting something really rare and hard, you're never going to get accuracy of 90% or better. So switching gears, if we're going to look at AI helping with value based reimbursement, something that we're talking about a lot, probably more than we're doing.

So this is a laundry list about what AI could help us with. Could it help use biomarkers and AI algorithms to select the right drug? Probably. I've given you a couple examples, some commercially available tests that some but others may not use, some in development that maybe some of you are part of. Yeah, I think it can do that.

Could it predict the likelihood that you could dose reduce or stop a drug? Yeah, I think it could do that. Again, it's an algorithm. You're basically curve fitting. So could you predict pre identify outlier patients?

Sure, you could probably do that. Could you use data patients contribute from a smartphone app, or things they wear, like the ring that I wear that has biosensor data in it? Yeah. So I think there's a lot of applications that AI could harness data and help us be smarter. Could you figure out who's going to do great on triple therapy with combo CSD marts?

Sure, if you really want to do that and you live in the Midwest. So anyway, there's some other things that AI can help us with. Could it help with patient education? Well, yeah, but I've introduced some cautionary notes here. Could this help with extracting your clinical data to help with prior auths?

Sure, but again, you really want it to be right, because you don't want a rejection because your AI algorithm screwed something up. So again, we could talk about any of these at great length, we're not going to, but AI can help with a lot of the things that you and I want to do or our staff are doing on a pretty regular basis. So I think there's a lot of opportunities either at hand or might be soon available if we bend our attention to helping with these things. It does make the assumption that if we get patients into better disease control, we pick the right drug out of the gate, patients are cheaper. In fact, I and other people, there's a handful of papers showing a year in remission costs a lot less than a year in high disease activity for somebody with RA.

So we know that that's true. So if you get the right drug more quickly, you will save payers money, they care about that. Let me tell you though, when AI goes wrong, you can really do a great job wrecking any AI based, value based reimbursement program. So this was a program that I and others were at least nominally part of. So Blue Cross Blue Shield said, hey, we're going to give you the opportunity to have a machine learning, read AI tech platform by this company, and we're going to help you figure out who to dose taper or take a medication holiday.

So this is an AI based program that's going to figure out who you shouldn't use a biologic or JAK and how you should take away drugs. What this really was interpreted is AI or machine learning is going to quote help you figure out who you can't treat or aren't going to be allowed to treat with a biologic or a JAK, and who you're going to have to stop. Some of the guardrails that I and others suggested based on data, like a data safety monitoring board, but this isn't research, this is just clinical practice in North Carolina. The payer, BCBS, Blue Cross rejected all of those. Everyone expected great, you're going to use my data, you're going to build an AI algorithm and then you're going to use it against me, and this is what the new world and the disutopia is going to end up being for us and our patients.

When you force me to abide by an AI algorithm making treatment decisions for me, yeah, no thanks. CSRO, the ACR, the North Carolina Rheumatology Association said, We hate this idea, many stakeholders including myself, Maddie Feldman and others withdrew. It was modified into irrelevancy, but this is one of the first and worst examples of AI being harnessed within rheumatology by a payer, and I think you couldn't have done much worse in messaging this to our field. Importantly, no rheumatologists designed this. The idea might not be fatally flawed.

We know from the CMRA trial and a number of other discontinuation studies, you can take away biologics, you know, maybe twenty eight, thirty percent of people can do well. You recapture them back quickly. If AI could help us find the one in three likely to do well, great. You could save a lot of money, and for the sake of time and to not get gonged, I'm gonna I'm gonna zip through that. AI will not put us out of a job.

AI can do the things on the the on the left, but you and I build relationships, we do exam, we provide judgment, wisdom, and empathy, and we're there to provide patient communication and to do all the things that is more than just answering multiple choice questions. In your handout I have my own 10 commandments of AI, particularly generative AI. You can read them. We don't need to go through all of them, but I think that they are are useful. And so with that, I want to acknowledge the many talented people that works in in my data science lab that I get to work with at UAB and elsewhere.

Thank you.

That's it for Tuesday night rheumatology this week. Be sure to tune in next week, November 11, when we will be discussing psoriatic arthritis a week after spondyloarthritis. The final week, right before Thanksgiving, we'll be talking about vasculitis. All of these are Tuesday night, seven p. M.

Eastern. Be sure to sign up. You can watch it on Zoom, on the RheumNow channel, on YouTube, Facebook, LinkedIn, and Twitter. Thanks to UCB for sponsoring this program and supporting RheumNow live twenty twenty five and 2026 upcoming. Be sure to go to rheumnow.live to register for Room Now 2026.

We'll see you there.