“I think with any technology, we also have to make sure there are safeguards for equity in patients and that there isn’t bias that’s creeping in there,” Dr. Bhana said.
He gave the example of cutaneous lupus flares, which present differently in patients who are Black when compared with those who are white. An AI model would need to have recognition models across skin tones.
Disease Prediction
Better disease prediction is an area in which the use of big data may come into play.
“Imaging is a piece of it, but then you can integrate all this data that’s floating around in data warehouses, like clinical records, lab markers, genomics, and then incorporate that along with population-based data,” Dr. Bhana said.
By using information, such as multiomics and symptoms, and placing that in large language models, AI may help the rheumatologist predict disease outset, flares and management.
“That’s coming, but I think a lot of this comes into the battle between who has the data? Where’s the data coming from? What’s the purity of the data? What’s the cost of it going to be to leverage these different warehouses and synthesize and integrate them?” he said.
Rheumatologists and other clinicians may already use AI to help review signs and symptoms or develop a differential diagnosis, Dr. Hausmann noted. On the flip side, some patients look up their symptoms online and get AI-generated feedback—some of which may be accurate and some of which may be wrong or even harmful.
The challenge going forward will be to preserve the human role in shaping this information vs. a techno-autocracy, Dr. Bhana said.
Remote Patient Monitoring
Remote patient monitoring is more common in cardiology, with the use of defibrillators and pacemakers that can hook into telemedicine clinics for monitoring, Dr. Bhana said. Although it’s not yet common in rheumatology, the future may bring the use of smartwatch accelerometer data to detect if a patient’s gait is changing or if they’re having falls. Depending on what’s happening, this type of monitoring may trigger an earlier return to the provider’s office for a check-up.
Treatment Optimization
The rheumatology world would likely welcome a more personalized approach to personalized medicine, Dr. Bhana noted.
“Can you use large data models of proteomics and try to figure out therapeutic targets that were not seen before?” he said.
Dr. Hausmann gives the example of deciding if an interleukin 6, Janus kinase or tumor necrosis factor (TNF) inhibitor may be better for a patient vs. assuming methotrexate or a TNF inhibitor would work for a patient. AI with such data sets may be more efficient than the current trial-and-error approach to medications that is used now to find the right mix for patients.
