Detecting change-points in preclinical rheumatoid arthritis biomarkers using Bayesian multivariate segmented regression.

Background: Rheumatoid arthritis (RA) has a preclinical period characterised by elevations in serum autoantibodies. Identifying the timing and magnitude of autoantibody trajectory changes may inform screening strategies and preventative interventions. Methods: Using a Bayesian multivariate segmented regression, we jointly modelled longitudinal autoantibody trajectories from two Department of Defense Serum Repository cohorts (Sample A: 209 matched case-control pairs, 1566 samples, six biomarkers; Sample B: 309 cases with two matched controls each, 2758 samples, eight biomarkers). Change-points and magnitudes of change were estimated simultaneously under a multivariate likelihood with an unstructured residual correlation matrix. Results: In Sample A, five of six biomarkers exhibited pre-diagnostic trajectory shifts with 95% highest posterior density intervals excluding zero. RF-IgM demonstrated the earliest change-point at 8.10 years before diagnosis (95% HPDI: -10.47, -5.73), followed by ACPA-IgG at 7.43 years (95% HPDI: -9.33, -5.76). In Sample B, only the four IgG isotypes showed pre-diagnostic shifts, with anti-CCP3 (IgG) earliest at 7.00 years (95% HPDI: -8.48, -5.29). A composite metric integrating timing and magnitude reordered rankings. Conclusions: This Bayesian framework enables simultaneous estimation of change-points and magnitudes across correlated autoantibodies while fully characterising uncertainty, offering a complementary approach to prior divergence-based methods for understanding preclinical RA autoimmunity.