Optimal exercise dose and modality-specific effects on lower-limb balance and functional motor performance in middle-aged and older adults with stroke: a Bayesian network meta-analysis

To compare the effects of different exercise modalities on lower-limb balance and functional motor performance after stroke and to examine their nonlinear dose–response relationships to identify potentially optimal exercise doses using a Bayesian network meta-analysis. A systematic search of PubMed, Embase, Web of Science, the Cochrane Library, and SPORTDiscus identified 54 randomized controlled trials (RCTs). A Bayesian random-effects dose–response network meta-analysis examined associations between total exercise dose, exercise modalities, and lower-limb balance and functional mobility after stroke. Both lower-limb balance and functional movement outcomes showed clear nonlinear improvements with increasing total exercise doses, peaking at approximately 1,800–2,200 MET×min/week. Beyond this range, marginal benefits plateaued or declined, with greater inter-individual variability. For lower-limb balance, gains stabilized or slightly decreased past 2,200 MET×min/week. Functional movement improvements formed an early plateau at around 1,800 MET×min/week, sustained up to 2,200, then gradually diminished. Exercise modalities displayed distinct dose sensitivities: integrated aerobic training (IAT) rapidly achieved peak benefits at moderate loads, supporting early rehabilitation; motor control training (MCT) maintained stable effects over a broader range, facilitating long-term neuromuscular reinforcement; virtual reality training (VRT) showed localized peaks at low to moderate doses but exhibited increased uncertainty at higher levels. This study suggests that moderate weekly exercise doses may improve lower-limb balance and functional mobility after stroke, with dose-response patterns differing across exercise modalities. These findings support dose regulation and individualized exercise prescription, although further high-quality trials are needed to confirm the observed dose ranges and modality-specific effects. This study is the first to apply a Bayesian non-linear network meta-analysis to model the joint effects of exercise dose and training modality in stroke rehabilitation, addressing the critical gap in prior research that lacked integrated, quantitative guidance for optimizing intervention strategies. It identifies an optimal total exercise dose of approximately 1800–2200 MET×min/week for improving lower-limb balance and functional motor performance in stroke patients. Beyond this threshold, benefits plateau or slightly decline, accompanied by increased interindividual variability. It demonstrates distinct dose-response profiles across modalities: integrated aerobic training (IAT) peaks at moderate doses, motor control training (MCT) sustains effects over a broader range, and virtual reality training (VRT) shows localized benefits at low-to-moderate doses with greater variability in outcomes.