Risk prediction rules in individuals at risk of rheumatoid arthritis: systematic review and validation using the APIPPRA cohort.

OBJECTIVES: To review risk prediction rules (RPRs) in individuals at risk of rheumatoid arthritis (RA) and to externally validate them using an independent interception trial cohort. METHODS: We performed a systematic literature search of MEDLINE and EMBASE on November 30, 2023, updated on July 10, 2025, to identify studies reporting RA RPRs in individuals at risk of RA. Each RPR was assessed across 4 Prediction Model Risk of Bias Assessment Tool domains: participants, predictors, outcome, and analysis. Eligible RPRs were externally validated by examining risk score distributions by RA progression, and evaluating discrimination, predictive accuracy and calibration. RESULTS: We identified 25 studies describing 41 RPRs in at-risk individuals. Rules incorporating clinical and serological prognostic factors, with or without imaging, were most common. Internal validation was reported in 23 (56%) RPRs and external validation in only 4 (9%). The discrimination C-statistic ranged from 0.59 to 0.98 during development and was presented as the sole performance measure in 21 (51%) RPRs. External validation of 14 RPRs revealed 4 (29%) with moderate discrimination (C-statistic, 0.6-0.7) and 10 (71%) with poor discrimination (C-statistic, <0.6). Brier scores ranged from 0.20 to 0.24 for 6 RPRs, indicating relatively better overall accuracy, whereas 7 RPRs had poorer overall fit (Brier score ≥0.25). Calibration plots showed reasonable calibration in 2 RPRs, with most demonstrating systematic or partial miscalibration. CONCLUSIONS: Existing RA RPRs demonstrated inconsistent performance when externally validated, underscoring population heterogeneity and the need for robust model development, transparent reporting, and independent external validation before clinical implementation.