Aerobic exercise improves episodic memory in late adulthood: a systematic review and meta-analysis

Episodic memory (EM), supported by hippocampal and prefrontal networks, is one of the earliest cognitive systems to decline with age and in Alzheimer’s disease, contributing to functional limitations and social isolation. Given limited pharmaceutical options, identifying non-pharmacological strategies to preserve or enhance EM is critical. Aerobic exercise (AE) benefits hippocampal structure and function, yet evidence for AE-induced EM improvement in older adults has been mixed. Prior meta-analyses often used restrictive inclusion criteria (e.g., only normal cognition or only MCI), included broader memory constructs, and rarely examined exercise dose parameters, potentially obscuring true effects. This study asks whether AE improves EM in adults ≥55 without dementia and whether sample characteristics (sex distribution, cognitive status, age) and intervention characteristics (control type, intensity reporting, length, session duration and frequency, total volume) moderate effects. The authors hypothesized AE would improve EM, especially among females, younger older adults, those with normal cognition, studies with no-contact controls, and higher exercise doses.

Previous meta-analyses generally reported null or minimal effects of AE on EM in older adults with or without cognitive impairment. Limitations included restricted samples (separating normal cognition and MCI), inclusion of non-episodic memory tasks within composite scores, insufficient attention to exercise dose parameters (length, frequency, duration, volume, intensity), and smaller evidence bases. Fifteen relevant RCTs were published since the last major meta-analysis, necessitating an updated and more inclusive synthesis focusing specifically on EM and potential moderators across the cognitive aging spectrum (normal cognition, subjective cognitive decline, and MCI).

Design and registration: Systematic review and meta-analysis conducted per PRISMA guidelines and preregistered on PROSPERO (CRD42020222666). Search strategy: Databases searched on April 1, 2021: PubMed, CINAHL, EMBASE, PsycINFO, and CENTRAL, with comprehensive terms targeting episodic memory, randomized controlled trials, aerobic exercise, and older adults. English-language articles published before April 1, 2021 were included. Additional titles identified via manual journal searches and reference lists from prior meta-analyses. Screening and selection: Titles imported into Rayyan; duplicates removed. Two reviewers independently screened 699 titles/abstracts with 96% agreement; disagreements resolved by the first author. Authors were contacted for missing data (k = 9), and usable data obtained for 8. Inclusion criteria: RCTs of AE (including aerobic dance/tai chi) in adults with mean age ≥55 years with normal cognition, subjective cognitive decline, or mild cognitive impairment; acceptable controls included no contact, wait-list, stretching/toning/balance/light resistance, education, or social interaction; outcomes required an EM measure (word list, story recall, face/name recognition, object memory, paired associates); any intervention length/frequency/duration/volume/intensity except acute exercise; studies must isolate AE effects from control. Exclusion criteria: Clinical samples with dementia, stroke, depression; AE groups including strength training, exergaming, cognitive training; composite outcomes not exclusively EM; studies without a control group; inability to isolate AE effects. Data extraction: Pre/post means, SDs, and group sizes extracted where available; otherwise change scores and SD of change extracted. A pre-post correlation of 0.5 was assumed. Positive change denotes EM improvement. Quality assessment: Methodological quality assessed using the PEDro scale (0–11), categorized as poor (0–3), fair (4–5), or good–excellent (6–11). Coding of moderators: Sample characteristics coded categorically—sex distribution (low 0–64% female vs high 65–100% female), cognitive status (normal vs impaired), age (young-old 55–68 vs old-old 69–85 years). Intervention characteristics included control group type (no-contact, inactive, physically active), prescribed intensity (reported vs not), intervention length (weeks; short 6–17, medium 18–39, long 40–65), session duration (short 15–45 min vs long 50–90 min), session frequency (low 1–2/wk, medium 3/wk, high 4–7/wk), and total volume (training minutes = length × frequency × duration; low <2100, medium 2100–3900, high >3900 min). Statistical analysis: Standardized mean differences converted to Hedges’ g using Comprehensive Meta-Analysis (CMA v3). When multiple EM measures existed, effects were averaged within study. Random-effects models (inverse-variance weighting) produced pooled effects. Heterogeneity assessed with Q and I². Aim 1 estimated overall AE effect on EM; leave-one-out analyses conducted. Supplementary analyses examined memory score types (total recall, immediate, delayed, recognition) and task types (word list, story recall). Aim 2 used mixed-effects subgroup analyses and meta-regression to test moderators; Q-statistics assessed between-subgroup heterogeneity; R² from meta-regression estimated explained variance. Sensitivity analyses: Re-ran primary analysis including only high-quality studies (PEDro ≥6); alternative sex cutoff (50% female); varying pre-post correlation (0 or 0.9); excluding one study with mean age <60; excluding studies with any anaerobic sessions. Publication bias: Assessed by funnel plot, Duval and Tweedie’s Trim and Fill, Egger’s regression test, and the Copas selection model (R package metasens; significance 0.1 for residual selection bias). Power: Retrospective power for moderators and sample size estimation for future RCT (80% power, two-sided, α = 0.05).

• Included studies: 36 RCTs; 2750 participants; publication years 1985–2021; mean age range 59–85 years; weighted mean age 70.82; 66.4% female; 20 studies normal cognition, 16 impaired (1 SCD, 15 MCI). - Overall effect on episodic memory: Hedges’ g = 0.28; 95% CI [0.10, 0.46]; p = 0.002 (random-effects). Leave-one-out: all g > 0.17; all p < 0.005. Sensitivity to pre-post correlation: r = 0 (p = 0.005); r = 0.9 (p < 0.001). - Heterogeneity: Q(35) = 170.00; p < 0.001; I² = 79.41% indicating substantial true heterogeneity. - Publication bias: Egger’s test t(34) = 2.07; p = 0.023 (suggesting asymmetry). Trim and Fill imputed studies increased pooled effect to g = 0.38 [0.20, 0.56]. Copas selection model adjusted effect g = 0.10 [−0.04, 0.24]; p = 0.16; estimated probability of publishing the least precise study = 0.71; ~14 unpublished studies estimated. - Power: 404 participants total (202 per group) required for 80% power to detect g ≈ 0.28 at α = 0.05. - Subgroup and moderator results: • Age (significant moderator): Q = 4.92; p = 0.027; R² = 0.09. Young-old (55–68): g = 0.49 [0.24, 0.75]; p < 0.001. Old-old (69–85): g = 0.10 [−0.14, 0.34]; p = 0.418. • Sex distribution: High female (65–100%): g = 0.34 [0.09, 0.60]; p = 0.009. Low female (0–64%): g = 0.23 [−0.03, 0.50]; p = 0.086. • Cognitive status: Normal cognition: g = 0.42 [0.17, 0.66]; p = 0.001. Impaired (SCD/MCI): g = 0.14 [−0.12, 0.39]; p = 0.300. • Control group: No-contact: g = 0.38 [0.07, 0.70]; p = 0.018. Physically active (stretching/toning): g = 0.35 [0.04, 0.67]; p = 0.028. Inactive (education/social): g = 0.12 [−0.20, 0.44]; p = 0.456. • Prescribed intensity reporting: Reported: g = 0.36 [0.15, 0.57]; p = 0.001. Not reported: g = 0.10 [−0.25, 0.45]; p = 0.587. • Intervention length: Medium (18–39 weeks): g = 0.46 [0.16, 0.76]; p = 0.002. Short (6–17): g = 0.25 [−0.08, 0.57]; p = 0.140. Long (40–65): g = 0.13 [−0.18, 0.44]; p = 0.405. • Session duration: Short (15–45 min): g = 0.32 [0.05, 0.59]; p = 0.021. Long (50–90 min): g = 0.26 [0.01, 0.51]; p = 0.040. • Session frequency: Medium (3/week): g = 0.39 [0.14, 0.64]; p = 0.002. Low (1–2/week): g = 0.14 [−0.18, 0.45]; p = 0.387. High (4–7/week): g = 0.23 [−0.27, 0.74]; p = 0.369. • Total volume: High (>3900 min): g = 0.33 [0.004, 0.65]; p = 0.047. Medium (2100–3900): g = 0.31 [−0.01, 0.63]; p = 0.060. Low (<2100): g = 0.23 [−0.11, 0.57]; p = 0.188. - Outcome-type analyses: Significant positive effects for immediate and delayed recall; not for learning or recognition. Task analyses showed significant effects for word-list and story-recall tasks. Results robust to excluding studies with anaerobic sessions and to removing one study with mean age <60.

The meta-analysis demonstrates that aerobic exercise confers small but reliable improvements in episodic memory among adults aged ≥55 without dementia. The effect is strongest in younger older adults (55–68 years), suggesting a potential window of greater plasticity or fewer comorbid barriers to adherence and efficacy. Benefits are more evident in samples with higher proportions of females and in participants with normal cognition, implying that initiating AE before clinically detectable impairment may yield greater gains. Control condition characteristics influenced detectability: benefits were observed versus no-contact and physically active (non-aerobic) controls but not versus inactive education/social controls, which may themselves modestly benefit EM. Reporting of prescribed intensity was associated with significant effects, potentially reflecting better intervention fidelity and individualized dosing. Subgroup patterns suggest that interventions of 18–39 weeks, with three sessions per week lasting 15–90 minutes and achieving >3900 total minutes, are most likely to show EM benefits. Mechanistically, AE may enhance EM via improvements in sleep and mood, cardiovascular and metabolic health, and direct neurobiological effects on hippocampal structure and function (e.g., increased BDNF, synaptic plasticity, angiogenesis, and anti-inflammatory pathways). Despite overall positive findings, substantial heterogeneity and indications of publication bias temper certainty about the precise magnitude of the effect. Nonetheless, pooling across underpowered trials reveals consistent benefits, informing clinical and public health strategies to prescribe AE for cognitive aging.

Aerobic exercise improves episodic memory in adults aged ≥55 without dementia, with the largest benefits in younger older adults, samples with more females, individuals with normal cognition, and interventions with reported intensity, medium duration (18–39 weeks), thrice-weekly sessions of 15–90 minutes, and total volumes exceeding 3900 minutes. These findings support AE as an accessible, non-pharmacological approach to promote episodic memory in late adulthood and underscore the value of early implementation along the aging trajectory. Future research should: (1) include more diverse populations and larger samples to ensure adequate power; (2) standardize and report intervention fidelity (intensity prescription, adherence, supervision, fitness changes); (3) test optimal dose–response parameters (length, frequency, duration, intensity) and modality combinations; (4) conduct long-term follow-up to assess persistence and progression to MCI/dementia; (5) examine mechanisms at behavioral and biological levels; and (6) extend investigation into middle age to determine preventive effects earlier in the lifespan.

• Substantial heterogeneity across studies (I² ≈ 79%), limiting precision of pooled effects. - Evidence of publication bias (Egger’s test significant); divergent bias adjustments (Trim and Fill vs Copas) yield larger vs attenuated effects. - Limited ethnic/racial diversity reduces generalizability. - Inconsistent or absent reporting of critical intervention fidelity variables (intensity prescription, adherence, supervision, attrition, changes in fitness) precluded finer-grained moderator analyses. - Variable and study-defined diagnostic criteria for normal vs impaired cognition complicate subgroup interpretations. - Mix of intent-to-treat and per-protocol analyses across studies may bias effect estimates. - Insufficient numbers to analyze certain EM subdomains (e.g., face/name recognition, object memory, paired associates) and EM components (what/where/when). - Lack of individual participant data restricted moderator analyses to study-level aggregates. - Potential reduced power to detect moderation effects beyond age; small moderator effect sizes warrant cautious interpretation.