Sjögren’s is a chronic, systemic, inflammatory autoimmune disease that can affect the moisture‑producing glands (often causing dry eyes and dry mouth) and many other parts of the body. Diagnosis can be challenging, especially early on because symptoms can be non‑specific and no single test is perfect. A new study by Liu et al (2025) tested whether an artificial intelligence (AI) model could use information from routine laboratory blood tests to estimate the probability that a person has Sjögren’s. The authors’ goal was to create a low‑cost screening and decision‑support tool that could help clinicians recognize Sjögren’s earlier and reduce misdiagnosis.
Study Design
Type of study: Multicentre retrospective cohort study (researchers analyzed existing clinical/lab data).
Where the data came from: Three hospitals in China; lab records spanning 2013–2023.
Who was included: 34,958 people who were being evaluated for Sjögren’s or had Sjögren’s‑like symptoms (not simply healthy volunteers).
How the groups were organized: The dataset was split into training, testing, internal validation, and external validation cohorts.
What the AI used as inputs: 16 routine lab tests that looked at kidney and liver function, metabolic health and the presence of specific proteins were included.
What was the AI output: A probability score estimating the likelihood of Sjögren’s disease, based on lab values.
Key Findings
· Strong performance across multiple datasets: The model achieved high discrimination in the test set, the internal validation set, and the external validation set.
· High accuracy and specificity: In the external validation cohort, accuracy was 0.960 and specificity was 0.990 (meaning few false positives in that dataset).
· Outperformed single common markers: Across datasets, the model outperformed individual conventional markers such as ANA, SSA/Ro, and others when used alone.
· Potential for differential diagnosis: When tested against other conditions, the model distinguished Sjögren’s from rheumatoid arthritis, systemic lupus erythematosus, systemic sclerosis, and osteoarthritis.
Many of the selected features were not “classic” Sjögren’s antibodies, but routine indicators related to globulins/total protein, complement C4, and organ/metabolic function (e.g., creatinine, eGFR, ALT, GGT, glucose).
Clinical Significance
This study suggests that an AI model can detect patterns in everyday bloodwork that are associated with Sjögren’s disease. In practice, a tool like this could be used as a screening aid to flag people who might benefit from more Sjögren’s‑specific assessment (history, physical exam, autoantibodies, tear/saliva testing, imaging, and sometimes salivary gland biopsy). Importantly, the model is not a standalone diagnostic test; it provides a probability estimate that must be interpreted by a clinician alongside symptoms and other investigations.
Limitations and What This Means for Patients
Because the study analyzed past records (retrospective design), we do not yet know how well the tool performs when used in clinic. The data were from one country, so we cannot assume the same performance in other health systems or ethnic populations without further validation.
Future Directions
Next steps include testing the model prospectively, validating it in more diverse populations and primary‑care settings, and exploring whether adding symptoms or other clinical data improves accuracy. Work is also needed on transparency, regulation, and safe implementation so that AI tools support (rather than replace) careful clinical reasoning.
Reference:
Liu, S., Wu, G., Pan, M., Sun, Q., Gao, C., Long, X., Tang, C., Yuan, X., & Sun, L. (2025). A multi-criterion feature integration framework for accurate diagnosis of Sjögren’s disease using routine laboratory tests. *npj Digital Medicine*, 8:729. https://doi.org/10.1038/s41746-025-02110-2
