Predictive Modeling of Sjögren's Disease Using US Healthcare Claims Data.

Area of Science:

• Rheumatology
• Autoimmune Diseases
• Data Science in Healthcare

Background:

• Sjögren's disease (SD) is an autoimmune disorder often associated with delayed diagnosis.
• Early identification of SD is crucial for effective patient management and treatment.

Purpose of the Study:

• To develop predictive models for Sjögren's disease using electronic health records.
• To identify key factors associated with early SD diagnosis.
• To characterize distinct patient subgroups within Sjögren's disease.

Main Methods:

• Utilized de-identified claims data (Clinformatics® Data Mart Database).
• Applied machine learning models: LASSO, random forest, and XGBoost.
• Employed Latent Class Analysis (LCA) for subgroup identification.

Main Results:

• Identified 5,632 Sjögren's disease cases and 56,320 controls.
• Key predictors included joint pain, autoimmune comorbidities, abnormal serology, and immunosuppressants.
• LASSO model achieved an AUC of 0.80; other models showed similar performance.
• LCA revealed four SD patient classes, primarily defined by serological abnormalities and disease burden.

Conclusions:

• Machine learning models show promise for predicting Sjögren's disease from routine healthcare data.
• Further refinement is needed to address data imbalance and model limitations for clinical application.