Young talents SPIDeRR project: Tjardo Daniel Maarseveen
“Our results support the project’s aims by showing that it’s possible to disentangle disease heterogeneity and identify clinically relevant patient subgroups early on. This opens the door to more personalized treatment decisions, helping to match the right therapy to the right patient during this crucial timeframe, and ultimately reducing delays in effective care”
We recently spoke with Tjardo Daniel Maarseveen, a PhD candidate at Leiden University Medical Center (LUMC). His research explores how artificial intelligence (AI) can help improve care for people with rheumatic diseases. In the interview, Maarseveen discussed about two promising AI projects the Knevel lab is working on. Both aim to support earlier and more accurate identification of musculoskeletal conditions, helping to improve patient care pathways.
Early identification and classification of rheumatic risk
The first project that Maarseveen is involved in focuses on the early identification of patients at increased risk of developing conditions such as rheumatoid arthritis, fibromyalgia, and osteoarthritis. By analyzing referral letters using AI (for more on this, listen to Maarseveen’s pitch video), he and his team aim to better stratify patients and guide them toward the most appropriate care pathway — such as rheumatological treatment or physiotherapy [1].
In the second project, which was discussed in more detail during the interview, the team is investigating the diversity among rheumatology patients [2]. By examining factors such as autoantibody profiles and symptom patterns, they aim to determine which patients are likely to respond well to treatments like methotrexate or alternative medications. The goal is to move toward more personalized and effective care pathways.
Data-driven clustering in early arthritis
In their study, Maarseveen and his colleagues explored whether it is possible to identify patterns among patients with rheumatoid arthritis based on routinely collected data from their first rheumatology visit, such as blood tests and joint involvement. By applying data-driven cluster analysis, they were able to define patient groups with distinct patterns of joint involvement. These included a cluster with typical hand arthritis, one with predominantly foot involvement, another with a polyarticular form (affecting many joints), and an oligoarticular form mainly involving large joints such as the knees. Importantly, these clusters were not only consistent over time but also predictive of treatment outcomes. For instance, patients in the hand arthritis cluster responded particularly well to methotrexate, showing higher remission-and retention rates. In contrast, those with foot-dominant involvement had less favorable responses. Maarseveen explains: “This suggests that early pattern recognition is possible and may help guide more personalized treatment strategies. For example, we might consider alternative medications for certain patient subgroups from the outset”.
Improving outcomes by targeting the therapeutic window
Discussing how their findings support the broader goals of the SPIDeRR project, Maarseveen notes that the current care pathway for rheumatoid arthritis is often lengthy and fragmented, which is an issue because there is a critical ‘window of opportunity’ early in the disease when treatment is most effective. At present, most patients begin with the same first-line treatment, typically methotrexate. However, some do not respond well and only switch to a more suitable medication later — often after missing this optimal treatment window. Maarseveen: “Our results support the project’s aims by showing that it’s possible to disentangle disease heterogeneity and identify clinically relevant patient subgroups early on. This opens the door to more personalized treatment decisions, helping to match the right therapy to the right patient during this crucial time frame, and ultimately reducing delays in effective care”.
Future ambitions
Looking ahead, Maarseveen aims to focus over the next five years on closing the gap between research and everyday clinical practice, ensuring that innovations, such as the referral letter tool, are effectively implemented within healthcare systems to truly improve patient care. “Although we’re still in the early stages of data sharing, it’s crucial that we learn from one another. For example, we can adopt successful approaches from countries like Sweden and integrate them into our own systems. I see a lot of unrealized potential here”, he says.
Currently, Maarseveen is researching how to implement the referral letter AI model in an actual hospital (Reumazorg Zuid West Nederland) as part of a pilot study to determine its efficacy. When it comes to employing the power of AI, it is always important to keep stakeholders in mind, he notes, adding that he works in close collaboration with patients and healthcare professionals — particularly rheumatologists and experts familiar with existing clinical systems and software. “Together, we can explore where AI can genuinely support clinical practice and improve care”, he says, emphasizing that responsible implementation requires ongoing dialogue. “It’s vital that we ensure these technologies truly contribute to care and don’t unintentionally exclude certain patient groups. Right now, there’s still some confusion about what AI actually means for healthcare and what its broader implications are. That’s why a collaborative, thoughtful approach is essential for creating sustainable and effective solutions.”
Real-world data, real patients
At the end of the interview, Maarseveen explains that their research uses routinely collected data from several hospitals and a wide range of patients, which helps make the findings more relevant to everyday healthcare. This means the approach is easier to apply in routine clinical practice. Using real-life data is especially important in areas like rheumatology. “In this field, the patients doctors see every day often don’t match the typical profiles included in clinical trials or the official classification criteria, such as the 1987 or 2010 guidelines from the ACR/EULAR (American College of Rheumatology/European League Against Rheumatism). By focusing on real patients, rather than the carefully selected — and sometimes healthier people — in trials, this research aims to reflect the true variety of patients seen in regular outpatient care.”
References
[1] Maarseveen, T.D., Glas, H.K., Veris-van Dieren, J. et al. Improving musculoskeletal care with AI enhanced triage through data driven screening of referral letters. npj Digit. Med. 8, 98 (2025). https://doi.org/10.1038/s41746-025-01495-4
[2] Maarseveen, T.D., Maurits, M.P., Agra Coletto, L., et al. npj Digit. Med; under revision. Link to preprint: https://www.researchsquare.com/article/rs-6256181/v1.