Examining radiographic outcomes over time.

Area of Science:

• Rheumatology
• Biostatistics
• Longitudinal Data Analysis

Background:

• Statistical analysis is crucial for clinical data interpretation and treatment decisions.
• Rheumatoid arthritis (RA) progression is often assessed using radiographic scores.
• Longitudinal analysis of RA radiographic scores requires robust statistical methods.

Purpose of the Study:

• To examine patterns and radiographic scores in an early rheumatoid arthritis cohort over a 3-year follow-up.
• To apply and compare advanced statistical models, including novel two-part models, for longitudinal radiographic score analysis.
• To identify the best statistical model for predicting radiographic progression in RA.

Main Methods:

• Analysis of total Sharp radiographic scores interpolated over 3 years.
• Application of generalized estimating equations (GEE) and two-part models (zero-inflated Poisson, zero-inflated negative binomial).
• Modeling of scores assuming Poisson or negative binomial distributions with AR(1) or exchangeable correlation structures.
• Adjustment for clinical, demographic, and therapeutic characteristics and pattern outcomes.

Main Results:

• Two-part models were utilized to address excessive zero counts in radiographic data, representing an innovation in rheumatology.
• Comparison of different statistical models using goodness-of-fit criteria was performed.
• The study aimed to determine the optimal model for predicting radiographic progression in RA patients.

Conclusions:

• Two-part models are highly appropriate for analyzing longitudinal radiographic data in rheumatic diseases, despite not being previously used in rheumatology.
• The application of these advanced statistical techniques enhances the interpretation of RA progression.
• This research provides a foundation for improved predictive modeling in RA management.