Dose- and modality-specific effects of exercise on BDNF and depression in middle-aged and older adults with mild cognitive impairment: A Bayesian network meta-analysis

Objective To evaluate the effects of exercise modality and weekly exercise dose on peripheral brain-derived neurotrophic factor (BDNF) and depressive symptoms in middle-aged and older adults with mild cognitive impairment (MCI), and to examine the robustness of the dose-specific predictions from the primary quadratic model across alternative dose–response functions. Methods PubMed, Embase, Web of Science, the Cochrane Library, and APA PsycInfo were searched from inception to 13 April 2026. Randomized controlled trials involving adults aged 50 years or older with MCI and reporting BDNF or depression-related outcomes were included. Effect sizes were expressed as Hedges’ g. Separate Bayesian random-effects network dose–response meta-analyses were conducted for BDNF, overall depressive symptoms, and modality-specific depressive symptoms. The quadratic model was used as the primary model to predict intervention effects at different exercise doses. Linear, Emax, and natural cubic spline models were fitted as functional-form sensitivity analyses. Results Twenty-eight studies involving 2,262 participants were included. In the primary quadratic analysis, the largest model-predicted effect on BDNF occurred at 440 MET-min/week (posterior mean, 0.752; 95% credible interval [CrI], 0.287 to 1.249), whereas the largest model-predicted effect on depressive symptoms occurred at 670 MET-min/week (posterior mean, 0.382; 95% CrI, 0.147 to 0.674). Dancing and multicomponent exercise showed comparatively favourable estimates at selected dose levels, whereas estimates for Tai Chi, walking, and yoga were less precise. Functional-form sensitivity analyses identified different dose locations for the maximum predicted effects. Therefore, 440 and 670 MET-min/week represent the maximum prediction points for BDNF and depressive symptoms, respectively, under the primary quadratic model, rather than established optimal exercise doses. Conclusion Exercise interventions were associated with increased peripheral BDNF levels and improved depressive symptoms in middle-aged and older adults with MCI. The primary quadratic model predicted the largest effects at 440 MET-min/week for BDNF and 670 MET-min/week for depressive symptoms, suggesting a potentially non-linear dose–response relationship. Although these prediction points are important findings of the quadratic analysis, cross-model sensitivity indicates that they should be used to inform future dose-ranging research rather than be regarded as confirmed clinical optima.