A Meta-Analysis of Exercise Intervention for Depression Symptoms

Summary

Exercise significantly increased depression remission rates (OR = 2.40) and reduced symptom severity (SMD = –0.96) in 40 RCTs (2,667 participants). These findings support exercise as a complementary approach, though heterogeneity and bias limit certainty.

Abstract

This meta‑analysis estimated the effect of exercise on depression symptoms. Unlike network meta‑analyses that compare exercise modalities, this pairwise meta‑analysis specifically quantifies the effect of structured exercise versus no exercise, using rigorous control-group definitions and updated methods. A literature search up to November 2025 identified 10,123 records. Forty randomized controlled trials met the inclusion criteria. Pooled effect sizes were calculated using random‑effects models; odds ratios (OR) for remission and standardized mean differences (SMD) for depression scores. Heterogeneity was assessed with I². Exercise significantly increased remission rates (OR = 2.40; 95% CI, 1.97–2.93; p < 0.001; I² = 49%) and reduced depression scores (SMD = –0.96; 95% CI, –1.27 to –0.65; p < 0.001; I² = 87%) compared to control groups. The number needed to treat for one additional remission was 2.1. The pooled SMD corresponded to clinically meaningful reductions. Subgroup analyses showed larger effects for supervised exercise and passive controls (e.g., waitlist) versus active comparators (e.g., pharmacotherapy). However, findings have major limitations; only 12/40 studies (30%) had low-risk of bias (RoB 2); most had short follow‑up (≤16 weeks); control conditions varied widely (waitlist to pharmacotherapy); and very high heterogeneity for depression scores (I² = 87%) persisted despite using SMD. The 95% prediction interval for SMD crossed zero, indicating exercise may not benefit all future populations. Grading of Recommendations Assessment, Development and Evaluation (GRADE) was moderate for remission but low for depression scores. Despite these limitations, exercise was associated with significantly higher remission and lower depression scores. These results support exercise as a complementary approach for depression, particularly for achieving remission. However, high‑quality, large‑scale RCTs with standardized outcomes and longer follow‑up are needed before clinical recommendations can be made.

Introduction

Depression is a common and incapacitating mental illness linked to higher mortality, medical comorbidity, and lower quality of life¹. More than 300 million people, or around 4.4% of the global population, suffer from depression². Psychotherapy and antidepressant medication (or a mix of both) are currently advised therapies³. However, psychotherapy is usually expensive and only produces remission rates of 50%⁴. Antidepressant medication frequently causes withdrawal symptoms, relapses, and side effects⁵. Notably, around two-thirds of adults who suffer ....

Protocol

Eligibility criteria

Studies were included if they met all of the following criteria, organized according to the PICOS framework (Population, Intervention, Comparator, Outcomes, Study design)¹⁷. All platforms and materials used in this study are listed in the Table of Materials. All database searches were performed in November 2025. No additional proprietary software (e.g., SAS, SPSS, R) was used beyond RevMan 5.3 and EndNote 20. The GRADE assessment was performed manually using the GRADEpro GDT online tool; the version is recorded as of the access date.

<....

Results

Study selection

The PRISMA 2020 flow diagram (Figure 1) summarizes the study selection process. Database searches identified 10,123 records (Cochrane Library: n = 1,845; Embase: n = 2,567; Google Scholar: n = 3,210; MEDLINE via OVID: n = 1,892; PubMed: n = 609). After removing 1,018 duplicate records, 9,105 unique records remained for screening. Discrepancy between stated and applied inclusion criteria for control groups. The innovation section in.......

Discussion

This meta‑analysis of 40 randomized controlled trials (2,667 participants with depression)^{24,25,26,27,28,29,30,31,32,33,34,}

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Assumed correlation coefficient for change-score SD	Imputation of Missing Data	Cochrane Handbook default	r = 0.5 (sensitivity analysis performed)
ClinicalTrials.gov	Clinical Trial Registries	U.S. National Library of Medicine	Database as of November 10, 2025 (no NCT-specific filter)
Cochrane Central Register of Controlled Trials (CENTRAL)	Databases	Cochrane Library	Issue 11 of 12, November 2025
Cochrane Handbook formulas	Formulas & Calculations	Cochrane Collaboration	Handbook version 6.4 (updated September 2024)
Egger’s linear regression test	Funnel Plot Asymmetry	RevMan 5.3 / custom macro	N/A
Embase (OVID platform)	Databases	Elsevier / OVID Technologies	Embase Classic+Embase 1947 to 2025 Week 47
EndNote	Reference Management	Clarivate Analytics (London, UK)	Version 20 (build 20.1.0.16478)
Google Scholar	Databases	Google LLC	Search conducted November 15–16, 2025 (first 500 records by relevance)
GRADE (Grading of Recommendations Assessment, Development and Evaluation)	Certainty of Evidence	GRADE Working Group	GRADEpro GDT (online version, accessed November 2025)
MEDLINE (OVID platform)	Databases	OVID Technologies	Ovid MEDLINE® 1946 to November 14, 2025
NNT from SMD (Furukawa & Leucht method)	Effect Size Conversion	Based on area under normal curve	N/A
Number needed to treat (NNT) from OR	Effect Size Conversion	Formula: NNT = 1 / [CER × (1 – OR-¹)]	N/A
PubMed	Databases	U.S. National Library of Medicine	NLM ID: 101775772 (database version 2025)
Review Manager (RevMan)	Statistical & Meta-Analysis Software	Cochrane Collaboration	Version 5.3 (Copenhagen: The Nordic Cochrane Centre)
RoB 2 tool	Risk of Bias Assessment	Cochrane Collaboration	Version: August 2019 (latest update)
Standardized mean difference (Hedges’ g)	Effect Size Conversion	Custom calculations in RevMan 5.3	N/A
WHO International Clinical Trials Registry Platform (ICTRP)	Databases	World Health Organization	Search portal version 1.6 (accessed November 2025)