Osteoporosis prediction in primary Sjögren's syndrome: development and external validation of a machine-learning...

Osteoporosis and fragility fractures are clinically important complications of primary Sjögren's syndrome (pSS) that may accelerate functional decline and excess mortality. In practice, osteoporosis risk is often assessed using general-population tools that do not incorporate disease activity, glucocorticoid exposure or inflammation-related bone remodelling. We aimed to develop and externally validate a prediction model for DXA-defined osteoporosis in pSS using routinely available clinical and laboratory indicators.

This retrospective cohort study included 1,000 patients with pSS from Longhua Hospital, randomly split into training and internal validation sets (7:3), and an independent external validation cohort of 266 patients from Shanghai Seventh People's Hospital. Candidate predictors were screened by univariable analysis, multivariable logistic regression and LASSO. Logistic regression was compared with seven supervised machine-learning algorithms. Performance was evaluated by area under the receiver operating characteristic curve (AUC), calibration and decision curve analysis.

The final logistic regression model retained seven predictors: sex, age, current glucocorticoid use, EULAR Sjögren's Syndrome Disease Activity Index score, 25-hydroxyvitamin D, procollagen type 1 N-terminal propeptide and β-C-terminal telopeptide of type I collagen. AUCs were 0.820, 0.807 and 0.787 in the training, internal validation and external validation cohorts, respectively, with good calibration. Machine-learning models achieved higher training AUCs but showed poorer transportability. A freely accessible web-based calculator was developed for point-of-care use.

A transparent, externally validated seven-variable model provides individualised DXA-defined osteoporosis risk estimation in pSS and may help clinicians prioritise bone density testing during routine visits.