Physical Activity and Incident Depression: A Meta-Analysis of Prospective Cohort Studies

Depressive disorders are the second leading cause of global burden of illness and account for more than 44 million years lived with disability (1). They are associated with heightened medical comorbidity (2), increased health care costs (3), and premature mortality (4). Given the breadth of depressive disorders and the individual and societal burden, strategies that may reduce the onset of depression are urgently needed (5).

One potentially modifiable risk factor for the onset of depression is low physical activity levels (6). People with major depressive disorder are known to have a 50% odds of not meeting the recommended physical activity levels (e.g., performing >150 minutes of moderate-intensity physical activity each week) compared with people without major depression (7). Moreover, structured physical activity is known to reduce depressive symptoms in those with depression (8). Systematic reviews have suggested that physical activity is a protective factor for depression onset (9, 10), with even small amounts of physical activity (e.g., walking <150 minutes per week) decreasing the incidence of future depressive episodes (9). These studies, however, have not conducted meta-analyses to quantify the magnitude of the protective role of physical activity (9). Moreover, the role of moderators such as age and sex, which may influence the relationship between physical activity and depression, have not been explored.

Given these gaps, our aims were to systematically review and meta-analyze prospective cohort studies examining the role of physical activity in reducing the risk of incident depression; to explore potential moderators, including age at baseline, geographical location, sex, length of follow-up, study quality, number of covariates used in the model, study sample size, and total person-years; and to evaluate the quality of the studies.

Method

This review adhered to the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) (11) guidelines and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (12) statement, following a protocol defined a priori (available on request).

Search Procedure

Two researchers (F.B.S., E.S.S.) searched PubMed, Embase, PsycINFO and SPORTDiscus from database inception to Oct. 18, 2017. Searches were adapted for each database, using keywords that included a combination of terms related to physical activity, depression, and longitudinal studies. (The searches are described in detail in the data supplement that accompanies the online edition of this article.) Manual searches were conducted of the reference lists from recovered articles and other systematic reviews investigating the association between physical activity, sedentary behavior or fitness, and depression (9, 10, 13, 14).

Inclusion and Exclusion Criteria

Articles were eligible if they met the following criteria:

1. The study evaluated participants, of all ages, who were free of depression or depressive symptoms at baseline.

2. Physical activity was measured with a self-report questionnaire, such as the International Physical Activity Questionnaire (15), single or multiple questions on participation in exercise, sports, or physical activity, or objective physical activity measures (e.g., accelerometers). Physical activity was defined as any bodily movement produced by skeletal muscles and requiring energy expenditure (16).

3. The study used a prospective design with at least 1 year of follow-up. Prospective studies with less than 1 year of follow-up were excluded, as this was not considered a sufficient time frame for risk and protective factors to exert a meaningful influence on depressive symptoms (17).

4. The study evaluated incident depression as the outcome, including increased depressive symptoms, through established cutoffs of depression screening instruments (e.g., the Beck Depression Inventory [18]) or based on tertiles, quartiles, or quintiles of depressive symptoms or on presence of major depressive disorder, diagnosed using structured or semistructured diagnostic interviews (e.g., instruments using DSM [19] or ICD criteria [20]) or through a self-report of physician diagnosis of depression.

5. The study reported an adjusted or unadjusted odds ratio, hazard ratio, or relative risk and 95% confidence intervals or the raw numbers of exposed and nonexposed participants who developed depression over the course of follow-up, in a way that allows calculations of odds ratios or relative risks. In instances when data were not available, we contacted corresponding authors to request the data to enable inclusion in our meta-analysis. To compare with most of the risk measures selected for the meta-analysis, the odds ratio, relative risk, or hazard ratio of studies using the lowest physical activity group as the reference group had to be inverted. Likewise, the limits of the corresponding confidence intervals were also inverted, giving rise to the limits of the confidence intervals to the reciprocal of the odds ratio, relative risk, or hazard ratio (21).

We excluded studies without primary data (reviews, commentaries, editorials); conference presentations without information about the methods or the outcomes; studies in languages other than English, Portuguese, or Spanish; and studies that evaluated physical activity as a continuous measure.

Studies of the same epidemiological cohort were included only when they reported the results in different metrics (odds ratio or relative risk/hazard ratio). For example, if one study reported odds ratio and another relative risk, each one was included in their respective analyses. This strategy allows the inclusion of the greatest number of studies without counting the same participants twice in each meta-analysis. When two or more studies reported data from the same cohort, we selected the most recently published. Studies reporting subsamples of cohorts were excluded.

Study Selection

In the first stage of study selection, two authors (F.B.S., E.S.S.) independently screened titles and abstracts of all articles retrieved from the search. Afterward, the full text of potentially eligible references was reviewed in detail by the same investigators. Disagreements were resolved through discussion until consensus was achieved. A third reviewer (B.S.) was available for mediation.

Outcomes

The primary outcome measure was the adjusted odds ratio (and 95% confidence interval) for incident diagnosed depression or depressive symptoms.

Data Extraction

Five authors (F.B.S., E.S.S., M.H., J.F., and S.R.) independently extracted data, including geographical location, name of cohort, number of participants included at baseline, age at baseline, physical activity assessment (instrument or questions used, what aspects of physical activity were considered by the measure to define physical activity levels [e.g., frequency, intensity, time, type, amount of energy expended, steps, or other]), depression assessment (e.g., instrument and cutoff used, diagnostic criteria, medical records), follow-up period, odds ratio, relative risk, or hazard ratio and 95% confidence interval, and the number of covariates. The data utilized for the adjusted meta-analysis were those of the most adjusted model presented in each of the respective reports.

Study Quality

The methodological quality of studies was assessed with the Newcastle-Ottawa Scale by two authors (F.B.S. and S.R.). The Newcastle-Ottawa Scale uses three elements to evaluate the risk of bias of prospective studies: 1) selection of participants (four items: representativeness of the exposed cohort, equal derivation between source of exposed and nonexposed participants, ascertainment of the exposure, and demonstration that the outcome of interest was not present at the start of the study); 2) comparability (one item: comparability of cohorts on the basis of the design of the analysis) (studies where the odds ratio or relative risk were calculated on the basis of the raw number of participants provided from the original reports received zero points for comparability); and 3) outcomes (three items: adequate assessment of outcome, adequate follow-up time, and adequacy of follow-up). A study can be awarded a maximum of 1 point for each numbered item within the selection and outcome categories, and a maximum of two points can be given for comparability. The maximum score on the Newcastle-Ottawa Scale is 9 (highest quality), and we assigned scores of 0–3, 4–6, and 7–9 for low, moderate, and high quality of studies, respectively (22). In case of disagreement, a consensus was reached through discussion.

Meta-Analysis

A random-effects meta-analysis was conducted investigating the relationship between baseline physical activity level and incident depression. Procedures included first pooling data across all studies comparing the incident depression in the highest physical activity level group (the group with greater frequency, intensity, volume, energetic expenditure, or other, from each study, as defined by the study authors) and the lowest physical activity level group (reference group). Analysis for adjusted odds ratio, crude odds ratio, adjusted relative risk/hazard ratio, and crude relative risk/hazard ratio were conducted separately. Specifically, adjusted odds ratio, odds ratio, relative risk/hazard ratio, adjusted relative risk/hazard ratio, and 95% confidence intervals were calculated for incident depression. For the adjusted odds ratio and adjusted relative risk/hazard ratio, we pooled the estimates using the model with the greatest number of covariates presented by the authors. Second, subgroup analyses were performed investigating the relationship between 1) different geographical regions (different continents), 2) how physical activity levels were assessed (e.g., asking about intensity, frequency, volume [time spent in physical activity], or composite variables including two or more variables, and studies using metabolic equivalents as units were classified together with the metabolic equivalents category); 3) the mean age of the sample at baseline (e.g., children and adolescents [<18 years of age], adults [ages 18–65], or elderly persons [over age 65]; 4) the use of self-report questionnaires or objective measures to assess physical activity; 5) depression assessment method, including screening instruments, major depression diagnosis, assessed by structured or semistructured diagnostic instruments, or self-report of physician diagnosis of major depression; and 6) the adjustment for potential confounders (age and sex, body mass index, smoking, and baseline depressive symptoms; age and sex and one more of the other three; and age and sex and two of the other three). Third, we evaluated potential moderators: percentage of males (only for crude odds ratio and relative risk/hazard ratio), length of follow-up, year of publication, person-years, total number of participants at baseline, study quality according to the Newcastle-Ottawa Scale score, and the score for the selection of participants, outcome, and comparability (only for adjusted analyses), and the number of covariates included in the model (only for adjusted odds ratio and adjusted relative risk/hazard ratio, to evaluate whether studies using more covariates are more likely to find significant or stronger effects) (23) through meta-regression analysis. Lastly, we evaluated publication bias using the Begg and Mazumdar (24) and Egger tests (25) and corrected for this using the Duval and Tweedie trim and fill method (26). To maximize statistical power, studies pooling participants with incident depressive disorders along with incident anxiety disorders were included in the main analysis. However, a sensitivity analysis excluding those studies was performed to assess whether they had an impact on the results obtained. Sensitivity analyses were also performed excluding studies of the same cohorts that have any potential sample overlapping. Heterogeneity was quantified using the Q and I² statistics, with scores of <25%, 25%−50%, and >50% indicating low, moderate, and high heterogeneity, respectively (27). Finally, the fail-safe number of negative studies that would be required to nullify (i.e., make p>0.05) the effect size was calculated (28). All analyses were performed using Comprehensive Meta-Analysis, version 3.

Results

Search Results

The initial search yielded 13,474 results. After the removal of duplicates and exclusion at the level of titles and abstracts, 10,099 abstracts were considered. At the full-text review stage, 430 studies were considered; 383 studies were subsequently excluded, and two additional studies were identified in the references of other included articles. Therefore, 49 unique studies were included in the review (29–77). (A flowchart [Figure S1] and a list of excluded articles are provided in the online data supplement.)

Studies and Participant Characteristics

Across the 49 unique prospective studies, 266,939 individuals were included, with nearly equal sex distribution (47% males), followed up for an average of 7.4 years. The total person-years was 1,837,794. Of these, 36 cohorts from 34 unique studies provided data for adjusted odds ratio, 19 cohorts from 18 studies provided for odds ratio, 18 cohorts from 12 studies provided for adjusted relative risk, and 17 cohorts from 15 studies provided for relative risk. Table 1 lists the studies included in each analysis. Only one study used objective measures to evaluate physical activity. Fifteen studies evaluated major depression using structured or semistructured diagnostic instruments or self-reported physician diagnosis of major depression only. The details of the included studies are summarized in Table 1. The list of included studies is provided in the online data supplement.

Study Quality

The mean study quality score of the studies was 6.34 (SD=0.8) out of 9 on the Newcastle-Ottawa Scale, representing moderate to high methodological quality. A detailed quality assessment is presented in Table S1 in the data supplement.

Physical Activity and Incident Depression

Highest versus lowest physical activity.

People with higher physical activity levels were at reduced odds of incident depression when compared with people with lower physical activity levels in adjusted (adjusted odds ratio=0.83, 95% CI=0.79, 0.88, p<0.001; I²=0.00, Q=25.93, N=36) (Figure 1) and crude odds ratio analyses (odds ratio=0.59, 95% CI=0.51, 0.68, p<0.001; I²=52.38, Q=37.80, N=19) and had decreased risks on adjusted and crude relative risk analyses (adjusted relative risk=0.83, 95% CI=0.76, 0.90, p<0.001; I²=0.00, Q=14.86, N=18; relative risk=0.68, 95% CI=0.60, 0.78, p<0.001; I²=33.40, Q=24.02, N=17). The plots for odds ratio, adjusted relative risk, and relative risk are provided in Figures S2, S3, and S4 in the data supplement, and the incidence rates are listed in Table S3 in the data supplement. Publication bias was evidenced for adjusted odds ratio (Egger’s intercept=−0.65, p=0.002), adjusted relative risk (Egger’s intercept=−1.25, p<0.001; Begg and Mazumdar tau=−0.43, p=0.01). The Duval and Tweedie trim and fill technique adjusted the effects to an adjusted odds ratio of 0.85 (95% CI=0.81, 0.89), an odds ratio of 0.63 (95% CI=0.54, 0.74), an adjusted relative risk of 0.86 (95% CI=0.78, 0.96), and a relative risk of 0.80 (95% CI=0.69, 0.94). The classic fail-safe N test revealed that 380, 519, 102, and 210 studies with negative results would be required to nullify the protective effect of physical activity on incident depression for adjusted odds ratio, odds ratio, adjusted relative risk, and relative risk analyses, respectively.

Subgroup and sensitivity analysis.

Significant protective associations of physical activity on incident depression were found across the four continents (Asia, Europe, North America, and Oceania) with available data for adjusted odds ratio and relative risk analyses. Protective effects were found for Asia, North America, and Oceania for odds ratio analysis, and for Europe, North America, and Oceania in adjusted relative risk analysis. Significant associations of high physical activity were found in all analyses for studies that assessed physical activity levels in terms of different volumes and composite/metabolic equivalents. Higher frequency of physical activity provided protective effects in adjusted odds ratio and odds ratio analyses, but not in adjusted relative risk or relative risk analyses. Higher intensity was significantly associated with lesser incident depression in all but adjusted odds ratio analysis. Protective effects were found for adults and older persons in all analyses and for children in adjusted odds ratio and relative risk analyses. Significant associations were found for studies assessing depressive symptoms across the four analyses. Physical activity was protective for major depression diagnosis in adjusted odds ratio, odds ratio, and relative risk analyses. Completing 150 minutes per week of moderate to vigorous physical activity was protective for incident depression in adjusted odds ratio and adjusted relative risk analyses. Lastly, subgroup analyses of studies that adjusted for age and sex, body mass index, smoking, baseline depressive symptoms, or age and sex and one more confounder, or age and sex and two more confounders were all significant in adjusted odds ratio analyses. Adjusted relative risk analyses adjusting for age and sex, body mass index, smoking, or age and sex and one more confounder were all significant. The details of the subgroup analyses are summarized in Table 2.

We performed sensitivity analyses (available on request) removing the study that pooled participants with anxiety disorders together with those with depression both in the overall analysis and in major depression only (78), excluding the study that used objectively measured physical activity (49). The results remained significant for all analyses.

Meta-Regressions

Sample size at baseline, year of publication, length of follow-up, individual study person-years, percentage of males, number of covariates used in each study for adjusted analyses (the list of the covariates used is provided in Table S2 in the data supplement), and the study quality according to the Newcastle-Ottawa Scale were investigated as potential moderators through meta-regression analysis. None of the investigated moderators significantly explained the variance of the effects of physical activity on depression onset in any of the analyses. The detailed results of the meta-regressions are summarized in Table 3 (plots are available on request).

Discussion

To our knowledge, this is the first study to meta-analyze the relationship between physical activity levels and incident depression. Study findings indicate that across 49 studies, higher physical activity levels are associated with a decreased odds of developing future depression. The results remained robust after adjustment for potential publication bias. Moreover, our results indicate that higher levels of physical activity offer a protective effect on future development of depression for people of all ages (youths, working-age adults, elderly persons), and this finding is robust across geographical regions around the world.

Previous narrative systematic reviews have suggested that physical activity can be protective against the development of depression (9, 10). Our study advances the field by conducting the first pooled meta-analysis investigating this relationship, which allows a clearer understanding of a true association between an exposure and outcome, rather than when studies are considered separately, as in previous reviews (79). Recently, a meta-analysis including 11 prospective studies found that sedentary behavior is associated with an increased incident depression at follow-up (relative risk=1.14, 95% CI=1.06, 1.21) (14). While sedentary behavior and physical activity are related constructs—with the former existing at the low end of the physical activity spectrum—it is of clinical relevance to quantify the pooled relationships of physical activity with subsequent depression onset independently of sedentary behavior.

Mammen and Faulkner (9) reported that gender may modify the effect of physical activity on incident depression. This assumption was not supported in our meta-regression analysis, however, suggesting that the potential protective association of physical activity is similar for men and women. Also, we demonstrated that physical activity has protective effects on depression across different geographical regions, and for people of all ages. Notably, physical activity was assessed by different parameters, such as frequency, intensity, volume, and type, that can be captured to discriminate different physical activity levels. Our subgroup analyses demonstrated that the protective effects of physical activity are found in studies in which the different aspects of physical activity (intensity, frequency, volume) were measured individually or when two or more aspects (metabolic equivalents/composite) were considered.

Our meta-analysis suggests that physical activity is associated with a decrease in the risk of developing depression, which raises an inevitable question: How might physical activity offer protection against depression onset? It is likely that no single mechanism can explain this relationship. A range of biochemical and psychosocial factors are likely responsible, including biological mechanisms through which exercise increases neurogenesis and reduces inflammatory and oxidant markers (80) and activates the endocannabinoid system (81). People with depression have decreased hippocampal volumes and levels of markers of neurogenesis, and increased levels of inflammatory (e.g., interleukin-6) (82) and oxidant markers (82). Physical activity may regulate these abnormalities, increasing hippocampal volume (83) and neurogenesis levels (84), as well as adjusting the imbalance between anti- and proinflammatory (85) and oxidant markers (86, 87). Also, physical activity may directly increase psychological factors such as self-esteem or perceptions of physical competence. Finally, an improved level of fitness leads to both subjective and objective improvements in physical health status (88). Productive areas of future research include physical activity interventions to prevent symptoms of depression and the underlying biological and psychological mechanisms.

Limitations and Future Research

Our meta-analysis has some limitations. First, most of the studies analyzed used self-report questionnaires to measure the exposure factor and the outcome. While common in the physical activity literature, self-report questionnaires are associated with recall bias. However, only one of the included studies used an objective measure (pedometer) (49) to evaluate physical activity, thus precluding exploration of whether results were different with self-report questionnaires compared with objective measures. Also, subgroup analyses showed that physical activity decreased the risk of developing depression, regardless of whether this was based on self-report measures or major depression diagnosis from structured clinical diagnostic interviews. Second, we found some evidence of publication bias in adjusted odds ratio and adjusted relative risk analyses. Nonetheless, adjusting for publication bias, after trimming 10 studies for adjusted odds ratio and eight studies for adjusted relative risk, resulted in smaller but still significant associations (adjusted odds ratio=0.85; 95% CI=0.81, 0.89; adjusted relative risk=0.86; 95% CI=0.78, 0.96). Therefore, the primary results of our analyses were not altered by considering the potential number of unpublished studies. Third, it should be noted that we included only studies in which there were no depressed participants at baseline, which minimizes the risk of selection bias. Despite this, the risk of selection bias was not entirely excluded, since depression is a recurrent disorder and previous depressive episodes were not well documented in the studies we investigated. Fourth, we were able to perform subgroup analyses including studies that evaluated the protective effect of 150 minutes of moderate to vigorous physical activity per week. However, these analyses included a small number of studies. Also, in all the other studies, the definition of low or high physical activity, as well as the aspects of physical activity that were captured by each instrument (intensity, frequency, volume, or two or more) varied widely. These limitations prevent us from establishing a “minimum” or an “optimal” dosage of physical activity necessary to decrease the odds of incident depression. However, we can conclude that people with higher levels of physical activity have a lower risk of developing depression than those with lower levels of physical activity. Fifth, seven of our subgroup analyses were nonsignificant. It should be considered that those analyses included a small number of studies and potentially are underpowered. Lastly, the included studies assessed physical activity participation using questionnaires referring to the preceding days or weeks. Thus, it is not possible to evaluate whether being engaged in higher levels of physical activity for longer periods confers greater protection in comparison to shorter periods.

Despite the robustness of our findings across age ranges, geographical regions, and the different aspects of physical activity (frequency, intensity, time, type), some caution is required, given that there may be a number of covariates that were not assessed. For example, some evidence suggests that the protective effect of physical activity seems to be greater in the noncarriers of the E type 4 allele of the apolipoprotein E (APOE) gene (89), and that carriers of the Met allele of the brain-derived neurotrophic factor (BDNF) gene are more likely to experience greater benefits for somatic symptoms from exercise interventions (90). Also, the effects of physical activity in people with increased risk for depression, such as people with a familial history of depression, has not yet been examined.

Differences in the assessment of depressive symptoms at baseline across studies are also a limitation. It is possible that the inclusion of participants who exhibited subthreshold depressive symptoms at baseline influenced the likelihood of developing depression at follow-up not only because of a lower engagement in physical activity but also because of an inherently higher risk of developing full-blown depression. Nonetheless, significant associations between high physical activity levels and lower development of depression was reported by included studies that controlled for baseline depressive symptom severity in a subgroup analysis for adjusted odds ratio, thus showing the protective effect of physical activity also in people with subthreshold depressive symptoms. Only one study adjusted for depressive symptoms at baseline for adjusted relative risk and found no significant associations, but it should be noted that this analysis was based on a single study. Also, people with lower physical activity levels may have other risk factors for depression, as such as obesity, poor diet, use of tobacco, and other clinical comorbidities. Therefore, given the observational nature of the included studies, it is possible that these other correlated factors contributed to increased risk of incident depression among people with low physical activity.

Further studies are warranted to evaluate the minimum physical activity levels required as well as the effects of different physical activity types and “dosages” on subsequent risk for depression. Also, further studies accounting for genetic variations and assessing people with increased risk for depression are required. Lastly, considering the burden of disease and the global impact of mental illness, further studies should evaluate the cost-effectiveness of physical activity in the prevention of depression.

Conclusions

Higher levels of physical activity are consistently associated with a lower odds of developing future depression. The protective effects of physical activity were observed regardless of age and sex and were significant across all geographical regions. Our data further emphasize the importance of policies targeting increased physical activity levels. Randomized controlled trials are required to address whether or not physical activity can prevent the development of depression in those at high risk.