The World Health Organization defines adolescence as the stage between 10 and 19 years of age (27). The American Academy of Pediatrics extended this stage to 21 years with the so-called late adolescence (28). This paper follows the extended definition.

Depression is one of the most common mental disorders in children and adolescents. According to a meta-analysis carried out by Costello, Erkanli (1) including studies from around the world, the estimated prevalence in young children (under 13) is 2.8%. Furthermore, 5.6% of all adolescents (13 to 18 years) suffer from depression, and the number of female patients (5.9%) is higher than the number of male patients (4.6%). In the National Comorbidity Replication study it was shown that mental health disorders' onset (like depression) peaks at the age of 14 years (29, 30). According to the WHO, depression and anxiety disorders are among the top five causes of overall disease burden among children and adolescents in Europe (31). Depressive symptoms and clinically relevant depressive disorders in young children can have a huge negative impact on the psychosocial development of the individual and can increase the risk of suicide (5, 25). Suicide is the third most common cause of death in adolescents (2). In general, depression is associated with a shorter lifespan (32).

Although psychological disorders in children and adolescents have not dramatically increased over the past decades for girls there seems to be a significant change in the prevalence of depression compared to earlier decades (33). This sex difference, however, seems not to appear before the end of Tanner stage III, suggesting hormones being involved in the pathophysiology of this disorder (30, 34). The average duration of a major depressive episode in children is 7 to 9 months (35), while 90% of children with depression recover from such an episode (36). In some cases, depression has a chronic progression that reduces the likelihood of therapy success (25). Furthermore, every single depressive episode increases the risk of recurrence (37).

In the treatment of mild and moderate depressive symptoms in children and adolescents, nonpharmacological approaches such as psychotherapy play a major role. A 2006 meta-analysis found modest benefits for psychotherapy versus control (38) which was confirmed by Eckshtain, Kuppens (39) with effects being significantly larger for interpersonal therapy (40) than for cognitive behavioral therapy (CBT). A network meta-analysis (NMA) of youth depression treatment and prevention studies, conducted by Zhou, Hetrick (41), found only CBT and IPT to be significantly more beneficial than most control conditions. A severe symptomatology may demand a combination with antidepressants. A study conducted with adolescents by Foster and Mohler-Kuo (42) found that the combination of cognitive-behavioral therapy and fluoxetine was more effective than drug therapy alone. The SSRI fluoxetine is the first-choice medication for the treatment of juvenile depression. As second-choice antidepressants the SSRIs sertraline, escitalopram, and citalopram might be used (43). Side effects of a pharmacotherapy in adolescents are comparable to those in adults including sedation, agitation, weight gain, sleep problems, vegetative symptoms, and sexual dysfunction (43). Electroconvulsive therapy use in adolescents is considered a highly efficient option for treating depression, achieving high remission rates, and presenting few and relatively benign adverse effects. Risks can be mitigated by the correct use of the technique and are considered minimal when compared to the efficiency of ECT in treating psychopathologies (44).

Preliminary results also suggest repetitive transcranial magnetic stimulation as an effective and well tolerated antidepressant treatment for adolescents with treatment resistant depressive symptomology (45).

It has been already demonstrated that physical exercise as well as physical activity can cause benefits at both, physical and mental level (46–48). Physical activity interventions have shown to be efficient not only to produce therapeutic benefits when implemented solely or as a part of a treatment for mental disorders, but also to prevent or delay the appearance of mental disorders (49). Additionally, physical exercise was found to be effective in treating symptoms and in reducing the mortality related to major depression (32). As this review discusses, meta-analyses on the depression-reducing effect of exercise in children and adolescents (50–53) so far suggest small to medium effects in this age group. Further moderating variables (e.g., dose specificity) could not be identified so far (51) although study quality and participants' characteristics (e.g. overweights as targets) seem to affect results (50). Several mechanisms have been suggested as responsible for the positive effects of exercise on depression including changes in HPA axis activity, mononamine levels, and neurotrophic growth factors as well as the adaptation of different neural structures [for an overview see Wegner, Helmich (19)].

Along with antidepressant drugs and psychotherapy, physical exercise is a promising option to treat depression. However, in reports exercise is often used synonymously with physical activity (54), which can be misleading. Physical activity is an umbrella term that includes sub-categories such as sports, leisure activities, dance, and physical exercise (55). The American College of Sports Medicine (56) defines physical activity as any bodily movement produced by the contraction of skeletal muscles that results in a substantial increase in caloric requirements over resting energy expenditure. Physical exercise, the concept of interest in this article, is characterised as a training exercise intervention that is planned and structured, repetitive and purposeful, leading to a change in fitness (54). We think that interventions can only run by exercise not by physical activity. Therefore, it can be said that physical exercise is always physical activity, but physical activity is not necessarily physical exercise. Nevertheless, we also included physical activity to our search because the use of exercise as an intervention for the treatment of mental health diseases is still used vaguely (57).

This article aims to evaluate whether children and adolescents with depression benefit in the same way from physical exercise training as adults do. The relief of depressive symptoms in both clinical and nonclinical samples was analysed.

The protocol of this systematic review was registered on July 20, 2018 in PROSPERO (International prospective register of systematic reviews) at www.crd.york.ac.uk under the PROSPERO-ID CRD42018100357.

To determine if a meta-analysis was appropriate for this article the single studies included had to fulfill the eligibility criteria displayed in Table 1 . In order to structure the eligibility criteria, the PICOS approach (58) was used by implementing five categories: population, intervention, comparator, outcome, and study design. Only meta-analyses, including longitudinal studies with control groups, were considered for inclusion. Their results are usually put in relation to a baseline collected at the beginning of the study, which allows a comprehensive understanding of the degree and direction of change over time (59).

Furthermore, the meta-analyses had to be published in a peer-reviewed journal and in the English language. All limitations were set before the literature search was conducted.

A wide literature search strategy was developed using keywords and Medical Subject Headings from four categories: population, outcome, intervention type, and study design ( Supplementary Material Table S1 ). The search terms from each category were combined in order to locate all relevant literature using the following databases: PsycINFO (EBSCO Interface), PsycARTICLES (EBSCO Interface), MEDLINE (via PubMed) and PubMed. The search was last conducted December 12, 2019.

The selection of meta-analyses was carried out independently by two researchers. Any disagreement between them was solved through discussion with a third reviewer. After deleting duplicates, the relevant articles were selected by screening the titles and examining the abstracts. Full-text articles were retrieved and scanned when abstracts did not provide sufficient data.

For each of the included meta-analyses the following data was extracted independently by two researchers: Information about the included single studies (design, sample size, sample characteristics, depression assessment, intervention characteristics, control group characteristics) and information about the meta-analyses (risk of bias analysis, effect sizes, methodological characteristics). A data extraction form was used.

All contained meta-analyses performed a risk of bias analysis to assess the quality of the included single studies. Additionally, they focused on publication biases. In order to compare the data, the quality analyses were extracted and examined. To assess the methodological quality of the included systematic reviews with meta-analyses, the AMSTAR 2 checklist (a measurement tool to assess the methodological quality of systematic reviews) (60) was filled out independently by three researchers.

In all meta-analyses included, the standardized mean difference (SMD = M₁ – M₂ /SD_pooled ) was used as the measure for effect size. The reported effect sizes of each one of the included meta-analyses were combined and a general effect size was calculated and discussed.

A subgroup analysis regarding the effect size for only clinical samples was carried out, understanding clinical samples as those including participants in treatment for a depressive-related disorder or with a formal diagnosis of a depressive disorder. Therefore, the single studies within the selected meta-analyses which examined clinical participants were extracted, the overlap studied, and the effect size calculated. Regarding the assessment of heterogeneity, a visual inspection of the forest plot and the I² value was made. According to the interpretation guide provided by Higgins and Green (61), while I² test results ranging from 0% to 40% might not report relevant heterogeneity levels, results from 30% to 60% may indicate moderate heterogeneity and between 50% and 90% substantial heterogeneity.

Potential publication bias was evaluated using a funnel plot. All statistical analysis and calculations were performed using the Review Manager (RevMan) software (62).

A total of 1,941 studies were identified in the literature search process to seek out systematic reviews with meta-analysis focused in this field. After removing duplicates, two independent researchers reviewed 1,152 titles and abstracts. Any discrepancy between researchers was discussed with a third reviewer. A consensus was reached and ended in a total of 23 potentially relevant studies. Those 23 studies were reviewed in full text. Four studies met the eligibility criteria and were included in this review (50–53) (for more information see the Flow chart of the selection process, Figure 1 ). A table with the excluded studies can be found as Supplementary Material Table S2 .

The general characteristics information of the four meta-analyses included were extracted by two reviewers using a data extraction form and were summarized in Table 2 .

Of the four meta-analyses included, only two included RCTs (52, 53). The other two meta-analyses included RCTs and also other study designs of lower quality. Radovic, Gordon (51) included five RCTs and three controlled trials (CTs). The meta-analysis of Brown, Pearson (50) covered five RCTs, two CTs, one cluster randomized CT (CRCT) and one quasi-experimental study (QES).

Several questionnaires for reporting depression outcome measures were included in the single studies. The Beck Depression Inventory (BDI) (63) was by far the most frequently used one. The existence of control groups is reported in the four meta-analyses (50–53). Their nature varied between single studies: without intervention, on a waiting list, low intensity exercise, or the usual exercise routine, among others ( Table 2 ).

The type of intervention differed between single studies. In three of the four included meta-analyses (50–52), aerobic exercise was the most used intervention. Carter, Morres (53) also included articles using some form of aerobic and/or resistance/strength training in the intervention. Within the single studies, the duration of the interventions varied from 4 to 40 weeks. In the meta-analysis of Radovic, Gordon (51), the maximum duration was only 20 weeks, whereas the minimum in Brown, Pearson (50) was at least 9 weeks. The approximated medium duration of all of the interventions was 11.5 weeks.

Regarding the frequency of the interventions, Brown, Pearson (50) and Radovic, Gordon (51) included single studies with a frequency of 2 to 5 days per week. In Carter, Morres (53) and Larun, Nordheim (52), the majority of studies included three sessions per week. In general, the mostly adopted frequency of implementation was three times per week. The duration of the exercise interventions varied between 5 and 90 min per session. The minimum of min was included by Larun, Nordheim (52). Carter, Morres (53) also included single studies with a minimum duration of 15 min only, whereas the other meta-analyses included minimums of 20 or 25 min. The maximum of a 90-min intervention was shared by all meta-analyses. The approximated medium duration of sessions was 41 min.

The number of participants included in each meta-analysis varied according to the sample size of the single studies. Larun, Nordheim (52) had the largest sample. They included 16 studies and reported 1,191 subjects. The smallest sample was found in Radovic, Gordon (51), with eight studies and 297 subjects. Brown, Pearson (50) integrated nine single studies and a total of 581 subjects. Carter, Morres (53) used eleven single studies, although only eight of them (including 445 subjects) were eligible for their meta-analysis. Those numbers led to a total of 41 single studies and 2,514 subjects as a base for this analysis.

The overlap of single studies within the four analyzed meta-analyses ( Table 3 ) caused a reduction to a final number of 30 single studies and 2,110 subjects for this review. The number of participants reported from each meta-analysis regarding the single studies did not always match. Some took the starting sample while others selected the final sample after drop outs. In this article, the latter was reported. This can be seen with the example of a single study (64) which was included in the four selected meta-analyses for this review (50–53). Two of the meta-analyses (51, 52) reported a sample size of 43 participants whereas Carter, Morres (53) reported 60 participants and Brown, Pearson (50) reported 30 participants. In the original article (64) 60 subjects started the program (30 control group; 30 experimental group). However, only 23 subjects of the experimental group and 20 of the control group completed the whole study, giving a final sample of 43 participants.

The age of the participants ranged between 5 and 20 years. The characteristics of the included populations in each meta-analysis varied from normal population: healthy samples, to at-risk groups: juvenile delinquents, pregnant adolescents, obese children, or clinically depressed populations: with major depressive disorder (MDD), with primary diagnostic of childhood depression and dysthymia or with moderate depressive symptoms.

Three of the included meta-analyses consisted of both clinical and nonclinical samples. Brown, Pearson (50) did not include clinical samples with regards to understanding a clinical sample as the sample with a formal diagnosed depression using clinical recognized diagnostic criteria. Larun, Nordheim (52) and Radovic, Gordon (51) both included three single studies with clinical samples. Carter, Morres (53) included five studies with clinical samples. Due to the overlap of those 11 studies, six single clinical studies were finally identified.

Two of those six final single studies integrated adolescents with diagnosed major depressive disorder (67, 68). One study included adolescents with dysthymia and primary diagnosis of conduct disorder (66). Cohen-Kahn (69) and Kanner (65) included psychiatric inpatients. In the second one, the patients had moderate or severe levels of depression. In Carter, Guo (70), the participants were also receiving a health or social care professional treatment for depression.

The aim of this article is to systematically review the meta-analyses that focus on the effects of physical exercise interventions on clinical and nonclinical depression in children and adolescents. Therefore, a first calculation of the effect size reported in the four meta-analyses included was carried out. The overlap was not taken into account at this stage.

Following the interpretation guideline according to Cohen's criteria (small d = 0.20; medium d = 0.50; large d = 0.80) (71), the calculated overall effect size is medium (d = −0.50).

An additional analysis with all of the single studies including clinical samples was carried out. The effect size of the physical exercise intervention in clinical depressive subjects was calculated.

After analyzing the overlap between single studies with clinical samples ( Table 4 ), the reported effect sizes were studied. The study of Hughes, Barnes (68) was included in Carter, Morres (53) and Radovic, Gordon (51) reporting different effect sizes. After analyzing the original study, the data of Carter, Morres (53) were used because the same results (d = −0.69) were reached. Radovic, Gordon (51) reported an effect size of d = −0.54. However, Carter, Morres (53) and Radovic, Gordon (51) showed the same results (d = −1.39) regarding the single study of Roshan, Pourasghar (67).

The single study of Brown, Welsh (66) was included in three meta-analyses (51–53). The original article was checked in order to understand the reported data of each of the meta-analyses. Radovic, Gordon (51) reported an effect size of d = 0.15. Larun, Nordheim (52) reported an effect size of d = 0.78. Carter, Morres (53) explained the impossibility of estimating the effect size due to the insufficient data reported in the original study (standard deviation missing) (53). Due to the missing data, it was agreed upon the impossibility of calculating the effect size.

The study by Cohen-Kahn (69) was included in Larun, Nordheim (52). The effect size reported was d = −0.14.

Kanner (65) single study was included in two meta-analyses (52, 53). The selected data was extracted from Larun, Nordheim (52) reporting an effect size of d = −0.46. In Carter, Morres (53) two different effect sizes were reported for the single study. The explanation to this fact relapse in the two separate intervention arms—low intensity/high intensity—of the exercise intervention. The reported results were the following: d = 0.01 for the first condition and d = −0.31 for the second condition.

Another single study (70) was included in the meta-analysis of Carter, Morres (53). The effect size of the exercise intervention with severe depressive participants was d = −0.19 in favor of the intervention group.

The I² statistic test was performed to assess heterogeneity. The results (I² = 36%; p = 0.18) showed a moderate level of heterogeneity. Which means that the level of variation across studies is due to the moderate differences between them rather than to chance. The results reported a small to medium effect size taking the base of Cohen's criteria (d = −0.48) (71). Effect size calculations for clinical samples can be found in the clinical sample analysis forest plot ( Figure 2 ). The publication bias analysis based on the visual inspection of the funnel plot indicated minor to no publication bias due to the asymmetric inverted funnel shape. The funnel plot can be found as Supplementary Material Figure S1 .