|Home | About | Journals | Submit | Contact Us | Français|
Although the effects of exercise on neurocognition have been the subject of several previous reviews and meta-analyses, they have been hampered by methodological shortcomings and are now outdated as a result of the recent publication of several large-scale randomized controlled trials (RCTs).
We conducted a systematic literature review of RCTs examining the association between aerobic exercise training on neurocognitive performance conducted between January, 1966 and July, 2009. Suitable studies were selected for inclusion according to the following criteria: randomized treatment allocation, mean age ≥ 18 years of age, duration of treatment > 1 month, incorporated aerobic exercise components, exercise training was supervised, the presence of a non-aerobic-exercise control group, and sufficient information to derive effect size (ES) data.
Twenty-nine studies met inclusion criteria and were included in our analyses, representing data from 2,049 participants and 234 effect sizes. Individuals randomly assigned to receive aerobic exercise training demonstrated modest improvements in attention and processing speed (g = .158 [95% CI: .055 to .260], P = .003), executive function (g = .123 [95% CI: .021 to .225], P = .018), and memory (g = .128 [95% CI: .015 - .241], P = .026).
Aerobic exercise training is associated with modest improvements in attention and processing speed, executive function, and memory, although the effects of exercise on working memory are less consistent. Rigorous RCTs are needed with larger samples, appropriate controls, and longer follow-up periods.
Strategies to enhance neurocognitive functioning have important public health implications as subclinical neurocognitive deficits are associated with increased risk of neurocognitive impairment (1), dementia (2), and mortality (3-7), independent of traditional risk factors. One such strategy that has gained increased attention is the use of aerobic exercise to improve neurocognitive functioning (8-12). Although the value of exercise has been critically examined in review articles (13) and meta-analytic syntheses (8-11), there has been a lack of agreement as to the magnitude of improvement in neurocognitive function associated with physical activity interventions. The current lack of consensus is due to differences in the evaluation of study methodologies, studies included in the analyses, data analytic approaches, and in the classification of various neurocognitive measures.
Cross-sectional studies have shown that physically active individuals tend to exhibit better neurocognitive function relative to inactive individuals (13-22). Prospective observational studies have reported similar findings, demonstrating that individuals who maintain greater levels of physical activity show improvements in neurocognitive function relative to their sedentary counterparts (1, 23-28). However, randomized trials have provided inconsistent results, with some reporting cognitive gains (29, 30) and others equivocal findings (31). Meta-analytic reviews of randomized controlled trials (RCTs) have also reported great variation in the magnitude of improvement in neurocognition associated with aerobic exercise (10-12), with some meta-analyses reporting moderate cognitive gains (9, 10) and others more modest improvements (8, 11, 32).
In several recent meta-analyses, including a Cochrane review (11), it was concluded that current data are insufficient to show that improvements in neurocognitive function associated with physical activity are due to improved cardiovascular fitness, and that larger studies are necessary (11, 32). However, since the publication of this review there have been several large scale RCTs examining this relationship (30, 31, 33, 34). In addition, although one previous systematic review (12) examined the effects of various forms of physical activity on boosting cognitive function (primarily general orientation) among individuals with dementia, no reviews have combined data from trials attempting to prevent dementia among vulnerable populations (i.e. individuals with cognitive impairment). The Cochrane review was limited in this sense as persons with neurocognitive impairments (e.g., mild cognitive impairment (MCI) and depression) were excluded (11). Furthermore, previous meta-analyses examining this relationship may have been influenced by the inclusion of two relatively large studies reporting substantial treatment effects but that were not truly randomized (9, 35, 36), which may have overly influenced the reported effects. Therefore, we conducted a meta-analysis that included the most recent exercise intervention trials and addressed several issues including: (1) the effects of aerobic exercise training on specific domains of neurocognitive performance including attention and processing speed, executive function, working memory, and memory; (2) the influence of specific dimensions of the exercise prescription such as the mode, duration and intensity of the exercise intervention; and (3) the issue of individual differences in response to exercise training, with a focus on baseline, pre-exercise level of cognitive functioning as a potential moderator of exercise effects (i.e., we compared individuals with Mild Cognitive Impairment (MCI)) to cognitively intact samples), as well as the age of study participants.
In order to determine the effects of aerobic exercise interventions on neurocognitive status, an extensive literature search was conducted using the following databases between January, 1966 and July, 2009: MEDLINE, Pubmed, EMBASE, Gateway, CENTRAL, PsycINFO, Dissertation Abstracts International, Educational Research in Completion (ERIC), Sports Discus, Cochrane Register, PEDRO, Ageline, and CINAHL. The following search terms were used: cogniti*, cognitive performance, age*, elderly, mental performance, and neuropsychological in combination with fitness, aerobic, cardiovascular, VO2, and physical activity. Additional titles were identified by a manual search of relevant journals and by identifying references included in previous meta-analyses. Unpublished dissertations and conference papers were also obtained when possible.
Suitable studies were selected for inclusion according to the following criteria: (1) randomized treatment allocation; (2) mean age ≥ 18 years of age and non-demented; (3) Duration of treatment > 1 month; (4) involved aerobic exercise training (e.g. brisk walking, biking, or jogging). Age 18 was selected as a lower age limit in order to control for developmental age differences in cortical thickness and myelination, which stabilize around the second decade of life (37). Studies utilizing walking interventions that were not aerobic were not included (e.g., slow walking with frequent breaks) in order to ensure that included trials incorporated some aerobic exercise component. Combined exercise interventions with an aerobic exercise component were also included (e.g. jogging and yoga); (5) the presence of a control group that did not engage in aerobic exercise; and (6) sufficient information to derive an estimate of effect size (ES).
After initial identification and retrieval of studies, several were found to be quasi-randomized studies (36) or used case-control methodologies (15, 36, 38-44), were of insufficient duration to include (45-47), were found not to be non-randomized based on personal communication with the trial’s principal investigator (39), or did not utilize a non-aerobic exercise control group (48, 49). Another trial was conducted among adolescents and was therefore excluded (50). Several trials utilized ‘dual-task’ interventions (e.g. walking and talking) (51-53) or balance and strength-training (54, 54, 55) and were therefore not included as it could not be ascertained whether exercise was of sufficient intensity to produce aerobic changes. Several trials were not included because they utilized physical activity interventions with exclusively non-aerobic exercise components among individuals with dementia (52, 56, 57, 57, 58, 58-65). The few studies utilizing walking interventions were either explicitly non-aerobic (58) or allowed residents with limited mobility (e.g., using walkers) to rest as needed (52), thereby limiting their generalizability to more healthy samples. Accordingly, these studies were excluded from the current analyses. For two trials in which the method of randomization was unclear (39, 66) we attempted to contact the respective authors and were able to confirm that one followed a true randomization scheme (39). Results were unchanged when the remaining study was excluded and we therefore included this trial in all analyses (66).
Two raters (PJS, BMH) independently extracted information from each article using an identical review protocol, which included including study identifiers (e.g., author names, year of publication, publishing journal), duration of treatment, intensity of exercise, modality of exercise, blinding of assessment personnel to treatment status, during assessments, intention to treat analyses, and time of follow-up assessment. Effect sizes were assessed independently. Interrater reliability was assessed for the outcome domains in question (i.e., in each cognitive domain as well as for study characteristics). For all areas, interrater reliability was found to be excellent (r’s > .90; Cohen’s kappa = .75).
Neuropsychological test results were classified according to the cognitive domains described by Lezak (67). We considered neurocognitive tests that could be classified in the following categories: attention and processing speed (the sustained focus of cognitive resources with selective concentration and rapid processing of information (67, 68)), executive function (a set of cognitive skills responsible for the planning, initiation, sequencing, and monitoring of complex, goal-directed behavior), working memory (short-term storage and manipulation of information), and declarative memory (retention, recollection, and recognition of previously encountered information, hereafter referred to only as: “memory”). We considered including ‘complex processing speed’ as a measure of executive function as in previous analyses (9) but results were unchanged regardless of the classification of this test.
Analyses were conducted using Comprehensive Meta-analysis software (Englewood, NJ). Data were analyzed using both fixed and random effects models and Cohen’s G for between-group differences (69). Briefly, fixed effects analysis assumes that all studies are drawn from the same population, such that differences in treatment effects across studies are attributed to sampling and methodological variability (i.e. error variance). In contrast, random effects analysis allows for the possibility that studies are drawn from different populations, such that differences across studies may be due to unidentified sources of variation and provides a more conservative estimate of treatment effects (70). However, because results did not differ between fixed and random effects analyses and because random effects are generally recommended for examining treatment effects in meta-analytic studies (70), we have presented the random effects findings only. In trials reporting multiple effect sizes within the same neurocognitive domain, data were collapsed by averaging all effect sizes within each neurocognitive domain for each study, such that each study produced no more than one effect size per domain. For the purposes of the study quality analyses, treatment effects were collapsed for each study for all neurocognitive domains. In addition, two trials in our literature search produced multiple publications in either peer-reviewed journals (71-73) or book chapters (74, 75) that were combined for the purposes of analysis. Homogeneity of treatment effects was assessed using the Q statistic. Three trials collected neurocognitive data at multiple time points in which participants continued to receive treatment (30, 73, 76). However, in only one study were the effects of treatment un-contaminated by crossover between groups (30). For this study only (30), we chose data from the longest follow-up assessment for inclusion in our analyses, although results were unchanged when other time points were examined.
Exploratory sensitivity analyses (77, 78) were conducted in order to investigate sample characteristics that may have moderated the effects of treatment on neurocognitive outcomes. Specifically, three trial characteristics were examined: Duration, Intensity and Mode of exercise intervention. We also examined two important methodological characteristics associated with methodological quality: blinding of assessors of neurocognitive outcomes and use of intention-to-treat (ITT) analyses. As an additional analysis, we examined whether treatment effects varied by cognitive status of participants at baseline (i.e., ‘non-impaired’ or mild cognitive impairment [MCI]; patients with dementia [Alzheimer’s disease] were excluded) and age of study participants.
Our initial literature search yielded 5,538 potentially relevant studies, 68 of which were retrieved for full-text review. Twenty nine studies (N = 29) incorporating data from 2,049 participants met inclusion criteria and were included in the present analyses (Table 1), including data for 1,024 experimental participants and 997 controls. Two hundred thirty four (n = 234) effect sizes were available for analysis. Trials ranged in duration from six weeks (79) to 18 months (30). As shown in Table 1, the primary exercise modality was brisk walking and/or jogging and control groups were typically assigned to a wait-list control, although stretching and toning, health education, and relaxation exercises were also used. Rates of attrition varied widely (range 0 – 41%; mean attrition = 12.2%). Only 13 studies (44.8%) utilized blinded assessments and only seven studies (24.1%) utilized intention-to-treat (ITT) analyses. The effects of exercise on individual neurocognitive measures are presented in Table 3. Due to the substantial number and heterogeneity of neurocognitive tests, only those tests used in more than one study are presented.
Twenty-four studies examined the effects of aerobic exercise on attention and processing speed. Exercise training was associated with modest improvements in attention and processing speed (g = .158 [95% CI: .055 to .260], P = .003) (Figure 1) and this effect was consistent across studies (Q23 = 26.249, P = .289). Moderator analyses demonstrated that trials of greater duration did not improve attention and processing speed to a greater extent than briefer interventions (r = .17, Q1 = 3.555, P = .399). Similarly, intensity was not associated with variations in attention and processing speed outcomes (r = −.375, Q1 = 1.41, P = .235). Results did not differ between individuals with MCI (g = .028, P < .001) and other samples (g = .181, P = .825) (Q1 = 1.228, P = .268). Combined interventions improved attention and processing speed to a greater extent (g = .250 [95% CI: .042 to .658], P = .026) than aerobic only interventions (g = .098 [95% CI: −.012 to .208], P = .152) (Q1 = 4.373, P = .037). There was no observed association between the mean age of study participants and improvements in attention and processing speed (r = −.047, P = .817).
Nineteen studies assessed the effects of aerobic exercise on executive function. Aerobic exercise was associated with modest improvements in executive function (g = .123 [95% CI: .021 to .225], P = .018) (Figure 2), and effects were of similar magnitude across studies (Q18 = 13.418, P = .766). Neither duration (r = −.436, Q1 = 3.627, P = .057) nor intensity (r = −.203, Q1 = 0.413, P = .520) were related to improved executive function. Improvements in executive function were somewhat smaller among individuals with MCI (g = −.004, P = .973) relative to other samples (g = .153, P = .008) (Q1 = 1.377, P = .241), and findings did not differ between studies that included only aerobic exercise (g = .109, P = .074) or combined aerobic exercise with other exercises (e.g., strength training) (g = .163, P = .106) (Q1 = 0.214, P = .644). Finally, there was no observed association between the mean age of study participants and improvements in executive function (r = −.348, P = .130).
Twelve studies examined the effects of aerobic exercise on working memory. Exercise did not appear to improve working memory performance (g = .032 [95% CI: −.103 to .166], P = .642) (Figure 3) and this effect was relatively consistent across trials (Q11 = 12.241, P = .346). Similar to other cognitive domains, neither the duration of the intervention (r = .346, Q1 = 1.438, P = .230) nor the intensity of exercise (r = .109, Q1 = 0.123, P = .725) appeared to moderate the effects of treatment. Only one study examined the effects of working memory among individuals with MCI and test for moderation was therefore not examined. Combined interventions (n=2) appeared to improve working memory (Q1 = 4.817, P = .028) (g = .288 [95% CI: .030 to .546], P = .028) relative to aerobic only interventions (g = −.042 [95% CI: −.184 to .101], P = .567). In addition, a significant association was observed between mean age of study participants and improvements in working memory, with older samples demonstrating greater improvements relative to younger samples (r = .564, P = .051).
Sixteen studies assess the effects of aerobic exercise on memory function. Aerobic exercise was associated with modest improvements in memory relative to controls (g = .128 [95% CI: .015 - .241], P = .026) (Figure 4) and effects were of similar magnitude across studies (Q15 = 9.030, P = .876). Neither intensity (r = −.051, Q1 = 0.026, P = .871) nor duration (r = .373, Q1 = 1.381, P = .240) appeared to moderate the observed effects on memory. Sensitivity analyses demonstrated that the effects of exercise were stronger among individuals with MCI (g = .237 [95% CI: .000 to .474], P = .050) relative to non-cognitively compromised individuals (g = .096 [95% CI: −.032 to .224], P = .143), although the statistical test for moderation did not achieve significance (Q1 = 1.055, P = .304). Only one study assessing memory utilized a combined intervention, so this was not examined as a potential moderator. In addition, there was no observed association between the mean age of study participants and improvements in memory (r = −.222, P = .175).
Analyses of whether methodological quality moderated the observed pattern of results, we examined whether treatment effects varied by 1) blinding of assessors and 2) use of ITT analyses. Studies did not differ in neurocognitive treatment effects whether they did (g = .143, P = .013) or did not use (g = .185, P = .012) blinded assessments (Q2 = .204, P = .651). Similarly, the effects of treatment on neurocognitive performance did not differ between those studies that did (g = .161, P = .004) or did not (g = .166, P = .087) utilize ITT analyses (Q2 = .002, P = .964).
Results indicate that aerobic exercise training confers modest improvements in neurocognitive function among healthy older adults, including improvements in attention and processing speed, executive function, and memory. Aerobic exercise did not appear to benefit working memory, however. Moderator analyses demonstrated that studies utilizing combined aerobic exercise and strength training interventions improved attention and processing speed and working memory to a greater extent than aerobic exercise alone. In addition, we found preliminary evidence that trials among individuals with MCI may be associated with greater improvements in memory relative to those among non-cognitively compromised samples. In contrast, neither training characteristics, such as study duration and intensity, nor methodological quality were associated with differential improvements in neurocognition.
Although previous meta-analytic reviews have reported that exercise may improve neurocognitive performance (8-12, 32), ours is one of the largest reviews to date demonstrating that aerobic exercise improves neurocognition among non-demented adults and the first to show that physical activity may enhance memory performance among individuals with MCI, a group at elevated risk for Alzheimer’s Disease (24). Several previous meta-analytic studies have examined the relationship between physical activity and cognitive function (8-12, 32). Colcombe and Kramer (9) reported that randomized controlled trials of exercise are associated with clinically meaningful improvements in executive function, processing speed, memory, and motor function. Our findings showed markedly weaker effects relative to this review, most likely as a result of excluding two decidedly positive studies trials included in Colcombe and Kramer meta-analysis (35, 36) which, upon closer examination were not truly RCTs. In a Cochrane review, Angevaren and colleagues (11) concluded that, although RCTs of aerobic exercise among individuals without cognitive impairment were associated with modest improvements in attentional processes, cognitive speed, and motor function, the existing data were insufficient to show that improvements in cognition were attributable to changes in cardiovascular fitness. Similarly, Etnier and colleagues (32) have demonstrated that, although higher levels of fitness were associated with better neurocognitive performance among cross-sectional study designs, studies examining pre-post comparisons found that larger gains in aerobic fitness were associated with lesser improvements in cognitive performance (32). Etnier and colleagues (8) have also noted that methodological limitations contributed to significant variability in treatment effects, with higher quality studies tending to show smaller effects, and studies with the highest quality rating demonstrating no effect of exercise on neurocognition. Most recently, van Uffelen and colleagues (12) reported that physical activity interventions among individuals without cognitive decline, on average, tended to report improved neurocognitive function. However, van Uffelen and colleagues (12) did not attempt to statistically combine treatment effect sizes across studies, reported that the majority of existing trials examining this question have failed to demonstrate a treatment benefit, and found that the extant literature is marked by a lack of high-quality studies. The present analyses address many of the issues raised by this previous review by including several large, high-quality RCTs not previously incorporated in systematic literature syntheses (30, 31, 33, 34).
The finding that exercise may produce larger improvements in memory for individuals with MCI than other patient groups is novel and warrants further investigation, although this must be viewed as preliminary. Although Heyn and colleagues (10) demonstrated that physical activity is associated with improvements in mental status among individuals with dementia, the majority of these trials were conducted among institutionalized adults with dementia and utilized balance and isometric exercises and did not examine the effects aerobic exercise, specifically, on neurocognition. The finding that aerobic exercise improves memory is consistent with several animal studies, which have indicated that physical activity increases brain-derived neurotrophic factor (BDNF) expression (80) in the hippocampus and peri-hippocampal structures (81, 82). In an important examination of mediators, Periera and colleagues (83) demonstrated that increased BDNF in the dentate gyrus, an area of the brain proximal to the hippocampus, was associated with dose-response improvements in memory performance among younger adults participating in an exercise intervention. In addition to plausible neurotrophic mediators, it is also possible that individual differences influenced the present findings. For example, it is possible that the MCI samples in our study were composed of a greater number of individuals with the apolipoprotein E type 4 allelic genotype (APOE-4), which has been associated with an increased risk for incident MCI and Alzheimer’s disease (84). In addition, recent evidence suggests that individuals with this genotype may exhibit relatively greater neurocognitive improvements with physical activity compared with healthy, older adults (85-87).
The finding that aerobic exercise alone did not improve working memory performance is an interesting and unpredicted finding. Although it is unclear why aerobic exercise improved other cognitive functions but did not appear to benefit working memory, this finding is somewhat consistent with previous brain imaging studies of aerobic exercise. Previous studies have demonstrated that cerebral alterations associated with exercise are preferentially in the peri-hippocampal region (83), anterior white matter tracks (88), and anterior cingulate (89). Although there is substantial overlap in the brain circuitry for carrying out complex cognitive processes, such as working memory, no imaging studies have demonstrated volumetric changes in the dorsolateral prefrontal cortex, which is primarily subserved by white matter projections from the corpus callosum and most consistently associated with working memory performance (90, 91). The finding that combined aerobic exercise and strength training interventions improved attention and working memory to a greater extent than aerobic exercise alone is consistent with previous reviews (9), as well as mechanistic studies demonstrating that strength training may improve neurocognitive function by increasing insulin growth factor, which has been implicated as a mediator of the exercise and neurocognition relationship (92-94). It is also possible that combined interventions more effective in reducing cerebrovascular risk factors (e.g. high blood pressure) (95) and improving aerobic fitness relative to aerobic training alone (96). These improvements in cardiovascular function may reduce the white matter degradation and cerebral ischemia that often results from these conditions (97-99). Alternatively, it is also possible that combined interventions may result in greater improvements in vascular health (100) and basal levels of inflammation (101, 102), although these relationships have yet to be investigated.
The present meta-analysis has several limitations. First, the literature is marked by a lack of high-quality trials examining the effects of aerobic exercise on cognitive endpoints. Trials included in our analyses differed substantially in their use of blinded evaluations, intention-to-treat analyses, and clinically validated cognitive assessment tools. Second, randomized controlled trials are limited by logistical constraints in their ability to sustain interventions over prolonged periods of time. Accordingly, the majority of studies examining cognitive end-points have done so after several months of aerobic training (79, 103) or, in some instances, incorporated follow-ups several years later (33, 104). There are limited data regarding how physical activity sustained over the course of several years may affect cognitive endpoints (2, 105), despite observational data indicating that physical activity and cardiovascular health may take years to affect brain health (106, 107). In addition, RCTs that have examined the neurocognitive effects of aerobic exercise over an extended time period have demonstrated greater improvements in memory over longer follow-up periods (30, 73). Third, the majority of extant studies have utilized interventions with frequency and intensity prescribed in accordance with the American Heart Association recommendations for cardiac rehabilitation (i.e., heart rates at 70% peak VO2 3 times per week). It is therefore possible that there was not enough of a range in exercise prescriptions to observe an effect on neurocognition. Finally, there is a lack of consensus as to which neurocognitive measures are most appropriate to examine changes in neurocognitive function associated with exercise. As shown in Table 3, there is substantial heterogeneity in treatment effects among neurocognitive measures. Accordingly, future studies would benefit from the identification of a standardized neurocognitive battery with the appropriate psychometric characteristics to examine neurocognitive measures associated with aerobic exercise.
In conclusion, aerobic exercise training results in modest improvements in cognitive performance among non-demented adults. Trials utilizing longer interventions were associated with greater gains in attention and processing speed, whereas trials conducted among individuals with MCI tended to demonstrate greater improvements in memory relative to non-MCI samples. Additional randomized trials are needed with larger samples, more extensive follow-up periods, appropriate controls, and more extensive measurement of potential mediators of cognitive change. Accordingly, future studies would benefit from the assessment of subclinical vascular health as a potential mediator of the exercise and neurocognition relationship, as this has been associated with improvements in aerobic capacity (100) and neurocognitive performance in other samples (108, 109). Future studies should also collect functional magnetic resonance imaging (fMRI) or diffusion tensor imaging (DTI) measures to track cerebral alterations following exercise, as several previous studies have demonstrated that exercise and improved fitness may increase cerebral blood flow (16) and alter blood oxygen level dependent (BOLD) response patterns to cognitive tasks (89), as well as improving structural brain health, such as by increasing white (88) and gray matter integrity (22) and brain volume (88). Finally, more rigorous studies should examine the effects of aerobic exercise training among individuals with MCI in order to determine whether this is a plausible strategy to delay or prevent incident dementia (101).
The research was supported by grants MH 49679 and HL080664-01A1 from the National Institutes of Health and M01-RR-30 from the General Clinical Research Center Program, National Center for Research Resources, National Institutes of Health, awarded to James A. Blumenthal.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.