Nine trials reported outcomes using the Freezing of Gait Questionnaire (FOGQ). Pooled analysis using a random-effects REML model found that exercise interventions were associated with significantly lower FOGQ scores compared with control conditions (SMD = −0.62, 95% CI −1.06 to −0.18, P = 0.01). Heterogeneity was moderate (I² = 64.52%). Leave-one-out sensitivity analyses produced pooled estimates that remained statistically significant across omissions, indicating the result was not driven by a single study. The GRADE assessment rated certainty for FOGQ as low. Details regarding specific exercise types, dose parameters, or patient subgroups driving the effect were not identified in subgroup analyses.
Thirteen comparisons reported motor outcomes using the Unified Parkinson’s Disease Rating Scale Part III (UPDRS-III). Exercise interventions were associated with significantly lower UPDRS-III scores compared with control (SMD = −0.80, 95% CI −1.02 to −0.59, P < 0.001). Heterogeneity for this outcome was low (I² = 0.00%), and leave-one-out analyses produced consistently significant estimates, supporting robustness of the pooled effect. The certainty of evidence for UPDRS-III was rated as moderate. The included trials used a range of exercise modalities; exploratory subgroup analyses did not demonstrate differential effects by modality, intervention dose, age, or disease duration.
Eight trials reported the Berg Balance Scale (BBS). Pooled results favored exercise, with higher BBS scores in intervention groups (SMD = 0.85, 95% CI 0.27 to 1.42, P = 0.004). However, heterogeneity was high (I² = 77.51%), indicating substantial variability among study results. Leave-one-out analyses showed the pooled effect remained statistically significant regardless of which study was omitted, suggesting the positive finding was not driven by a single trial. Despite the statistically significant pooled effect, the certainty of evidence for BBS was rated as low, reflecting risk-of-bias concerns and inconsistency across studies.
Four trials reported Mini-BESTest outcomes. The pooled effect favored exercise but did not reach statistical significance (SMD = 0.23, 95% CI −0.24 to 0.70, P = 0.34). Heterogeneity was moderate (I² = 53.59%). Leave-one-out analyses showed variability in statistical significance across iterations, with only one omitted-study model remaining significant. The GRADE certainty for Mini-BESTest was rated low, and the authors concluded evidence for Mini-BESTest outcomes is inconclusive.
Risk-of-bias assessment using the Cochrane RoB 2 tool found no trials rated as low overall risk. Sixteen of 20 trials had some concerns and four had high overall risk. Specific domains with issues included deviations from intended interventions (usually related to blinding challenges inherent to exercise trials), incomplete reporting of randomization procedures, missing outcome data, and outcome measurement. Publication bias and small-study effects were assessed where feasible; formal assessments were limited to outcomes with sufficient comparisons. Certainty of evidence using GRADE was moderate for UPDRS-III and low for FOGQ, BBS, and Mini-BESTest.
The review included 20 RCTs published between 2009 and 2024, enrolling a total of 714 patients with Parkinson’s disease (372 randomized to exercise interventions, 342 to control). Exercise interventions encompassed diverse modalities: dance-based training, Tai Chi, Qigong, yoga, treadmill training, balance/gait training, resistance training, aquatic exercise, walking training, and multimodal programs. Intervention dose parameters (duration, frequency, single-session length) varied across trials. The authors conducted exploratory subgroup analyses for age, disease duration, intervention duration, frequency, session duration, and modality but did not identify differential effects according to these factors.
The systematic review followed PRISMA 2020 guidance and was prospectively registered in PROSPERO (CRD420261281718). Multiple databases (PubMed, Embase, Web of Science, CENTRAL, Scopus) were searched through January 10, 2026, with backward citation searches and manual supplementation. Inclusion required adults (≥18 years) with idiopathic PD in randomized controlled trials comparing exercise-only interventions to usual care or no structured exercise, and reporting at least one of the targeted outcomes. Continuous outcomes were pooled as standardized mean differences (Hedges’ g) with 95% CIs using random-effects REML models. Heterogeneity was quantified with I² and τ²; sensitivity analyses included leave-one-out tests.
Across multiple outcome domains, exercise interventions were associated with improvements in freezing-of-gait symptoms (FOGQ), overall motor function (UPDRS-III), and performance on the BBS. Evidence for Mini-BESTest outcomes was inconclusive. Despite statistically significant pooled effects for several outcomes, methodological limitations in the included trials and generally low-to-moderate certainty ratings warrant cautious interpretation. The available evidence supports inclusion of exercise as part of PD rehabilitation strategies, but specifics of optimal modality, dose, and patient selection remain uncertain based on current trial data.