Radiographic progression in psoriatic arthritis: head-to-head comparison of the Ratingen score and the Sharp-van der Heijde method.
TL;DR
SVDH and PARS showed broadly comparable psychometric performance for PsA damage assessment and were robust for cross-sectional scoring but showed important limitations in reliably detecting cumulative change in this low-progression cohort.
Abstract
Objectives
To compare the psychometric properties (inter-rater reliability, agreement, responsiveness, and measurement error) of the Psoriatic Arthritis Ratingen Score (PARS) and modified Sharp-van der Heijde (SVDH) methods using the Belgian Psoriatic Arthritis (BePas) cohort.
Methods
Radiographs of hands and feet (baseline T0, follow-ups T1, T2) were independently scored by two blinded readers using SVDH (max 528) and PARS (max 360). Statistical analyses included Intraclass Correlation Coefficient (ICC), Bland-Altman analysis, Standardized Response Mean (SRM), and Smallest Detectable Change (SDC).
Results
Both methods detected positive mean change, primarily between T0 and T1. Inter-rater reliability for absolute scores was high for both (ICC >0.865; baseline ICC: PARS 0.906, SVDH 0.888). For cumulative change scores (ΔT2-T0), inter-rater reliability was low and very similar for both methods (ICC=0.483 for SVDH and 0.485 for PARS). Both methods showed moderate responsiveness over the cumulative interval (two-reader average SRM: PARS 0.571, SVDH 0.531). For both methods and both readers, mean cumulative change scores were smaller than the corresponding SDC values (PARS: 2.8-3.7% of maximum score; SVDH: 2.8-4.1%), indicating that average progression fell below the threshold of reliable individual-level detection. Bland-Altman analysis showed slightly narrower absolute limits of agreement for PARS in selected comparisons, whereas SVDH showed narrower relative limits after normalization to the maximum possible score.
Conclusions
SVDH and PARS showed broadly comparable psychometric performance for PsA damage assessment. Both methods were robust for cross-sectional scoring but showed important limitations in reliably detecting cumulative change in this low-progression cohort. Neither method demonstrated clear superiority.