Work

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

Year: 2026

Type: preprint

Abstract: Background Rheumatoid arthritis (RA) has a preclinical period characterized 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 modeled longitudinal autoantibody trajectories from two Department of Defense Serum Repository cohorts (Cohort A: 209 matched case-control pairs, 1566 samples, six autoantibody biomarkers; Cohort B: 309 cases with two matched controls each, 2758 samples, eight autoantibody biomarkers). Change-points and magnitudes of change were estimated simultaneously under a multivariate normal likelihood on log-transformed concentrations, treating values at each assay’s upper limit of detection as right-censored through data augmentation and modeling an unstructured residual correlation among biomarkers; a diagonal-covariance sensitivity analysis is reported in the Supplement. Results In Cohort A, 5/6 biomarkers shifted before diagnosis, with 95% highest posterior density intervals (HPDIs) excluding zero. RF-IgA changed earliest, 9.2 years before diagnosis (HPDI: −13.8, −4.2), followed by RF-IgM at 8.2 years (−11.7, −5.4) and ACPA-IgG at 7.8 years (−9.9, −6.1). In Cohort B, all eight biomarkers changed before diagnosis, anti-CCP3 (IgG) earliest at 7.2 years (−8.1, −6.4); IgG isotypes changed earlier and more steeply than IgA. Ranked by a composite time integrating timing and magnitude, ACPA-IgG and anti-CCP3 (IgG) came first. Conclusions This Bayesian model estimates change-points and magnitudes simultaneously while accounting for the upper limits of detection and for correlation among biomarkers, and it characterizes the uncertainty in both. It complements prior divergence-based methods for understanding preclinical RA autoimmunity. Key Messages Treating autoantibody concentrations at the upper limit of detection as right-censored and modeling the residual correlation among biomarkers allow joint estimation of when and how steeply each autoantibody begins to rise before RA diagnosis. RF-IgA, RF-IgM, and ACPA-IgG exhibited the earliest pre-diagnostic trajectory acceleration (7.8 to 9.2 years before diagnosis), and all eight markers in a second panel of anti-citrullinated protein antibodies shifted before diagnosis, with IgG isotypes changing earlier than IgA. A composite time integrating the timing and magnitude of autoantibody change provides a different ranking of biomarkers than change-point estimation alone, which may inform future screening and prevention strategies.

Source: medRxiv

DOI: 10.64898/2026.05.22.26353892