Optimal exercise prescription for balance after stroke: a Bayesian dose–response network meta-analysis

Background: Post-stroke balance impairment is common and clinically consequential, contributing to increased fall risk, reduced functional independence, and long-term disability. Exercise is widely prescribed to improve balance after stroke, yet the dose required for meaningful benefit-and whether higher doses yield additional gains-remains uncertain due to heterogeneous exercise prescriptions across trials. Objective: To quantify the dose-response relationship between exercise dose (standardized as METs-min/week) and balance improvement after stroke, and to estimate the minimum effective dose and the dose range associated with maximal benefit using Bayesian model-based dose-response meta-analysis. Methods: We searched PubMed/MEDLINE, Embase, Web of Science Core Collection, Scopus, CENTRAL, and major Chinese databases, from inception to December 31, 2025, and conducted backward citation tracking. We included randomized controlled trials enrolling stroke survivors. The primary outcome was the Berg Balance Scale (BBS). We used Bayesian random-effects network meta-analysis to compare exercise modalities and Bayesian model-based methods to estimate dose-response relationships. Risk of bias was assessed using RoB 2, and certainty of evidence was evaluated with CINeMA. Results: We included 42 randomized controlled trials. Overall, the dose-response relationship was non-linear: the predicted effect suggested an increase with dose up to an apparent peak region around ~1,200 MET-min/week, followed by attenuation/decline at higher doses, with greater uncertainty where high-dose data were sparse. In the ascending portion of the curve, the model-implied average local change was approximately +0.12 Hedges' g per 100 MET-min/week, whereas beyond the peak region it was approximately -0.05 per 100 MET-min/week. The minimum dose associated with a credible improvement was ~270 MET-min/week, within the observed dose range of included trials. Across modalities, resistance training showed the largest pooled benefit versus control, while aerobic, resistance, and water-based exercise exhibited non-linear patterns with modality-specific peak regions; Chinese exercise and balance training showed positive associations within the evidence-supported dose range. Estimates for HIIT were small and generally imprecise, with credible intervals frequently including no effect. Conclusion: Exercise significantly improves balance in patients with stroke, with ≥270 METs-min/week representing the credible minimum effective dose. Based on our dose-informed recommendations, clinical programs may prioritize moderate-intensity training performed 3-5 times per week, with resistance training (RT) and water-based exercise (WBE) most likely to yield larger gains (predicted peak doses: RT ~ 666 METs-min/week; WBE ~ 1,616 METs-min/week), while aerobic exercise (AE) is also effective at moderate doses (predicted peak ~554 METs-min/week). In addition, Chinese exercise (CE) and balance training (BT) show a stable positive dose-effect association within the evidence-supported range. By contrast, HIIT shows small and uncertain effects and is therefore not recommended as a routine first-line option at present. Systematic review registration: PROSPERO, CRD420261297953.