Optimal exercise modalities for enhancing motor function recovery after stroke: a Bayesian systematic review with pairwise and network meta-analyses

Summary: Background: This study aimed to compare the effectiveness of different exercise interventions in improving motor function recovery after stroke and to identify optimal rehabilitation strategies. Methods: A comprehensive search of five databases was conducted for randomized controlled trials (RCTs) up to March 13, 2025. Primary outcomes included the 6-min walk test (6MWT), 10-m walk test (10MWT), Berg Balance Scale (BBS), and FMA (total, UE, LE). Pairwise and Bayesian network meta-analyses were performed to compare the relative effects of different exercise modalities. Risk of bias was assessed using the RoB 2 tool, and evidence certainty was rated using CINeMA. This systematic review was prospectively registered in PROSPERO (CRD420251091242). Findings: A total of 317 RCTs involving 14,464 stroke patients were included. Network meta-analysis showed that, compared with routine care, electrical stimulation plus exercise (ESX) was associated with the largest improvement in walking endurance (MD = 53.4 m, 95% CI 14.27–92.58, p < 0.05, I2 = 0%), while aerobic exercise (AE), functional training (FT), gait training (GT), and mind-body exercise (MBE) showed smaller effects. Lower limb training (LLT) demonstrated the greatest improvement in gait speed (MD = 0.18 m/s, 95% CI 0.14–0.23, p < 0.05, I2 = 0%), with TCMEX, task-oriented training (TOT), and MBE also showing benefits. For balance (BBS), core stability training (CST) ranked highest (SMD = 0.77, 95% CI 0.30–1.24, p < 0.05, I2 = 93.24%), followed by MBE and FT. FT showed the largest pooled effect for total FMA (SMD = 2.15, 95% CI 0.48–3.81, p < 0.05, I2 = 93.86%), whereas MBE showed larger pooled effects for FMA-LE (SMD = 1.94, 95% CI 0.04–3.85, p < 0.05, I2 = 97.54%) and FMA-UE (SMD = 1.85, 95% CI 1.08–2.63, p < 0.05, I2 = 95.57%). Substantial heterogeneity was observed in several pooled comparisons. Meta-regression analyses identified multiple outcome-specific clinical and contextual factors (e.g., intervention duration, stroke phase, stroke severity, routine rehabilitation, age, sex distribution, and socioeconomic setting) that partially explained between-study variability, although no single factor accounted for heterogeneity across all outcomes. Importantly, global network diagnostics indicated acceptable overall model fit and consistency, supporting cautious interpretation of the comparative estimates at the network level. Interpretation: Several structured exercise interventions were associated with improved motor outcomes after stroke. High heterogeneity was observed in some comparisons; however, meta-regression analyses suggested several potential sources of variability, and global network assessments indicated generally acceptable heterogeneity. These findings should therefore be interpreted cautiously, particularly for outcomes with substantial between-study variation. Funding: This study received no specific funding.