Evidence-informed exercise dose targets for rheumatoid arthritis: A systematic review and bayesian dose-response...

Exercise is strongly recommended as part of integrative management for rheumatoid arthritis (RA). However, most trials and conventional network meta-analyses compare broad exercise categories without explicitly modelling dose, which may obscure non-linear dose-response patterns and limit outcome-specific interpretation.

To estimate statistically supported dose ranges and the model-estimated doses associated with the most favourable posterior effects of different exercise modalities on disease activity, pain, physical function, and cardiorespiratory fitness in adults with RA.

Eight databases were searched from inception to 16 June 2026 for randomised controlled trials (RCTs) evaluating structured exercise interventions lasting at least 2 weeks in adults with RA. Weekly exercise exposure was standardised as metabolic equivalent minutes per week (MET-min/week) using the 2024 Adult Compendium and treated as a pragmatic proxy of exercise volume across five modalities: aerobic exercise (AE), resistance exercise (RE), combined aerobic and resistance exercise (AE+RE), traditional Chinese exercise (TCE), and yoga. Dose-response relationships for DAS28, VAS pain, HAQ-DI, and VO₂max were estimated using the MBNMAdose package with a quadratic function. Risk of bias was assessed using RoB 2, and confidence in the evidence was evaluated using CINeMA. The protocol was registered in PROSPERO (CRD420261418955).

Forty-four RCTs comprising 93 arms and 3132 participants were included. The fitted dose-response relationships were frequently non-linear or plateauing, and statistically supported effects varied by modality and outcome. TCE produced the largest reduction in DAS28 at 500 MET-min/week (MD -0.84, 95% CrI -1.16 to -0.50). RE produced the largest reductions in VAS pain at 820 MET-min/week (MD -14.03 mm, 95% CrI -19.98 to -8.68) and HAQ-DI at 1000 MET-min/week (MD -0.26, 95% CrI -0.47 to -0.06), whereas AE produced the largest improvement in VO₂max at 750 MET-min/week (MD 4.75 mL·kg⁻¹·min⁻¹, 95% CrI 3.39 to 6.11). Sensitivity analyses suggested that VO₂max findings were the most robust, whereas DAS28 estimates were more sensitive to exclusion of studies at high risk of bias. Meta-regression did not identify clear moderation by age or intervention duration. Confidence in the evidence ranged from very low to moderate across outcomes and comparisons.

Exercise management in RA may benefit from an outcome-specific and dose-informed approach rather than a uniform weekly target. Larger model-estimated exercise doses were not consistently associated with greater benefit. Given the very-low-to-moderate confidence in the evidence and inconsistent reporting of adverse events, the estimated dose values should be interpreted as provisional model-based reference targets rather than definitive prescriptions or safety thresholds.