One Family, Many Diagnoses: What AI Is Learning to See in Autoimmune Disease
An Ask the Expert conversation with Dr. Marjolein Klaassen, in partnership with the Autoimmune Association
“In my family, on both my maternal and paternal sides, there is extensive autoimmune disease with some of us having Hashimoto’s, arthritis, lupus, Sjogren’s, Raynaud’s, and diabetes. Why do different autoimmune diseases develop?”
This is the kind of question that reflects many autoimmune experiences — yet it rarely gets asked or answered in a doctor's appointment. It’s a scientific question. This Smart Patients member based it on her years of lived experience, a special type of expertise. This Ask the Expert session was a good example of what happens when that expertise is shared with a clinical scientist.
Smart Patients hosted Dr. Marjolein Klaassen (a gastroenterology resident and a clinical scientist at MIT's Center for Microbiome Information and Therapeutics) for a few days of member questions. Dr. Klaassen sits in an unusual seat. She treats patients, and she also works on the science of how AI reads complex biological data. That vantage point lets her answer from the perspective of a clinician who both uses these tools and studies how they are built. She offered a careful, honest account of how far science has come and what remains uncertain.
Reframing the Question
Dr. Klaassen began by describing what the family inherited. "Family members may share variants in genes that make the immune system more likely to become dysregulated: more easily activated, less well controlled, or less tolerant of the body's own tissues." Immune tendency can then show up as different autoimmune diseases in different family members, depending on which tissue or organ becomes the target. In one relative it settles in the thyroid, in another the joints, in another the gut or skin or connective tissue. Genetic studies support this. Autoimmune disorders like celiac disease, type 1 diabetes, rheumatoid arthritis, and lupus – among many others! – share overlapping mutations in genes involved in normal immune control, such as HLA, PTPN22, and CTLA4 (Harroud & Hafler, Science, 2023).
A family's pattern of autoimmune diseases reflects one or more inherited tendencies toward immune dysregulation, which is expressed in different places in different people.
Where AI Helps
For this question specifically, Dr. Klaassen explained that if the relevant information is scattered across many biological layers, including genes, antibodies, cytokines, thyroid function, and family history, then the useful signal "may not be obvious in any one layer by itself. It may be in how these layers connect."
That's the kind of problem AI is being built to work on; and it is the direction her own field is moving. She pointed to a new generation of "multimodal foundation models" for biology, designed to learn across genomics, proteomics, metabolomics, and more at once — looking for patterns that would be too complex for the human mind to pick out (Cui et al., Nature, 2025).
Used this way, she said, AI "may help move us from 'This person has several separate autoimmune diagnoses' toward 'This person has a pattern of immune dysregulation that may explain why these diagnoses cluster, and may suggest what to watch for next.'" The patient and the researcher are asking the same question from different directions.
Our visiting expert was also careful to share AI's limits. Explaining why several diseases co-occur is the most realistic near-term goal. Predicting which disease comes next remains difficult. For now, it’s only possible in narrow cases, such as islet antibodies appearing before type 1 diabetes and anti-CCP antibodies before rheumatoid arthritis. Nearly all of it stays research-stage, requiring validation and a physician's interpretation before it touches anyone's care. This is also a field that moves unusually fast. Today’s questions will look different within a few years.
Throughout the conversation, she returned to a single principle: AI helps gather, organize, and interpret, and the physician stays responsible for judgment, privacy, and the final decision.
AI in the Medical Chart
Given the increasing number of clinicians using AI to help generate notes from visits, Smart Patients asked about the risk to accuracy. Members live with their records every day and rely on them to be accurate, which is not always the case. One member summed it up for the community:
"Many of our members have had difficulty getting corrections to sometimes horrifying errors in their records."
Another described how AI had affected her personally: "This happened to me recently after my appointment was recorded using AI. I had to contact my doctor to make corrections." If it’s already hard to fix a human error, what happens when AI introduces one, or quietly copies an old one forward? Record accuracy is not a small concern. It shapes trust, and it feeds every decision a future clinician makes from the chart.
Dr. Klaassen agreed that "an error in the medical record is not harmless. Once it is in the chart it can be copied forward, summarized, pulled into referral letters, or used by a new physician who does not have time to read the entire record." She framed it as a question of design. Software built with more care can preserve the original source beside any AI summary, flag when older information conflicts with newer notes, and keep a patient's correction attached to the disputed statement so a rushed reader still sees it. Designed that way, she said, AI could help make records safer to read.
The Questions That Endure
Like many people navigating serious illness, Smart Patients members are both cautiously optimistic and skeptical: Can these tools connect what my doctors can't? Can I trust what they tell me? Do I have any control over my own information? These continuing concerns are expressed repeatedly.
Several members shared how they were already experimenting. As one put it, "Many of us are curious and exploring many of the new big-name AI platforms like Claude and Chat GPT without fully knowing where the information is coming from." Dr. Klaassen's guidance was practical: "I would use them as a medical explainer, not as a final medical source." She suggested asking a chatbot to explain a diagnosis in plain language, or to help list the questions to bring to your doctor. A question like "Do I have this disease?" sits outside what AI can answer. She also recommended asking a tool to show its sources so you can check whether the sources themselves are trustworthy. Because a model may answer from its training data without checking current evidence, she suggested asking it directly whether an answer comes from that training data or from a live search of current sources.
What to Carry With You
Dr. Klaassen's closing vision was to develop a "learning health system" where "care generates data, patients contribute their lived experience, clinicians provide clinical context and judgment, AI helps identify patterns, researchers test them, and the results come back into care faster." Patient-reported experience carries real weight in this model. When captured systematically, it has changed clinical outcomes — for example, when electronic symptom monitoring improved survival in cancer care (Basch et al., JAMA, 2017).
Over three days, our members revealed a clear unmet need for people who carry several autoimmune diagnoses: They want the connections between their conditions understood. They named barriers that will shape whether they trust AI-supported care, from unverifiable sourcing to records that are hard to correct. They showed they are already using these tools with caution and asking how to use them well.
Discussions like these are the same ones clinicians and researchers are working through right now. Because this space is changing so quickly, the questions keep evolving, and the answers rarely stay settled for long. Together, everyone asks better questions and gets better answers.
From this conversation, the Smart Patients community developed two references members can keep. The first is a "When AI is Part of Your Care" card.
The second is a "Using AI Chatbots for Health Questions" guide.
Click on each image to see and download these resources.
Thank you to the Autoimmune Association for helping to arrange this Ask the Expert Session and for all the advocacy, awareness, education and research you do. Join The Autoimmune Community Summit for free!