Exercise modalities for overall and domain-specific balance performance in older adults: A systematic review and Bayesian network meta-analysis

Background Ageing impairs balance in older adults, yet the exercise modality suited to balance domains remains uncertain. The study compared overall and domain-specific effects by fall risk, setting and dose. Methods Five databases were searched to 22 December 2025 for randomised controlled trials of adults aged ≥ 60 years without disease-specific recruitment. Bayesian network meta-analysis estimated Hedges’ g and 95% credible intervals (CrIs). Results 423 trials (33 901 participants; 2546 effect sizes) were included. Compared with control, 18 of 19 modalities improved overall balance; stretching showed no clear benefit. The largest effects were for perturbation training (0.66, 95% CrI 0.49–0.82), sensory training (0.55, 0.42–0.68), dance (0.54, 0.45–0.64) and agility stepping (0.53, 0.39–0.67). Tai Chi, perturbation training, Pilates, and agility stepping or aquatic exercise had the largest estimates for global, static, dynamic and proactive balance, respectively; balance-specific training, aquatic exercise and conventional resistance training showed tentative reactive-balance signals. Perturbation training showed larger estimates in increased-fall-risk and supervised groups, whereas sensory training produced the largest home-based estimate. Dose-response associations were nonlinear and exploratory. Conclusions Exercise improves balance, but modality choice should consider balance domain. Current estimates were most favourable for perturbation training in overall and static balance under supervision, sensory training at home, Tai Chi in global balance, Pilates in dynamic balance, and agility stepping or aquatic exercise in proactive balance. These estimates do not establish superiority. Dose should reflect task-specific challenge, safety and feasibility, because METs capture metabolic cost rather than balance-specific challenge. Copyright © 2026. Published by Elsevier B.V.