2.1 Literature screening process and results
A total of 21078 studies were retrieved after systematically searching individual databases. After removing duplicate literature, 11830 studies remained. After screening the titles and abstracts, 143 studies were read in full text for refinement. After that, 33 studies met the inclusion and exclusion criteria. A manual search of references of included literature and relevant studies by senior experts in the field added two additional relevant pieces of literature. The final 35 studies were included for meta-analysis ( Fig. 1).
2.2 Literature characteristics and evaluation of literature quality
The meta-analysis included thirty-five studies published between 1982 and 2020, with a total of 5,393 participants and a mean age of 13.23 ± 3.12 years. The studies were conducted in ten countries, with fifteen in the United States, four in the United Kingdom, three in Australia, China, and Iran, two in Germany, and one each in Israel, Mexico, Korea, Brazil, and Canada. The exercise interventions were classified as follows: aerobic exercise (AE), which involved continuous or low-intensity intermittent exercises such as running; resistance training (RT), which involved resistance exercises performed by overcoming self-body mass or applying external resistance; game training (GT), which involved team sports such as soccer, basketball, and physical activity; and medium-low-intensity multi-motion training (MT), which involved martial arts, yoga, aerobics, or aerobics combined with resistance exercise. Of the thirty-five studies, twenty involved AE, sixteen involved MT, five involved GT, two involved RT, and thirty- five involved controls who did not receive the exercise intervention. The mean intervention period was 13.94 ± 7.03 weeks, with 57% of the studies having an intervention period of greater than or equal to 12 weeks. The mean number of interventions per week was 3.35 ± 1.73, and the mean duration of each intervention was 45.3 ± 18.72 minutes. The primary outcome was assessed using a validated depression rating scale to measure depressive symptoms at the end of the intervention period and the last available follow-up visit. If multiple scales were used, observer-rated scales were given priority over self-reported scales. The most commonly used scale was selected when multiple depression scales were reported to reduce heterogeneity. The specific characteristics of the included studies are described in Additional file 2.
The quality of the included literature was assessed by the Review Manager 5.3 software concerning the criteria recommended by the Cochrane 5.3 systematic appraisal manual risk of the bias assessment form, using "high risk" for random sequence generation, allocation concealment, participant blinding, rater blinding, data completeness, selective reporting, and other biases, "respectively", "unclear", and "low risk" responses, respectively, for methodological evaluation of bias in the included literature (Fig. 2). Three grades were assigned: 9 high quality (≥ 5 points), 17 medium quality (3–4 points), and 9 low quality (≤ 2 points) (Fig. 2).
2.3 Results of traditional meta-analysis
The results of the meta-analysis show a significant improvement in depression among children and adolescents who engaged in exercise (Table 1). Subgroup analyses based on different exercise modalities revealed that AE [SMD=-0.35, 95% CI (-0.59, -0.12), P < 0.01], and aerobic combined with resistance exercise (MT) had a significant effect on adolescent depression [SMD=-0.24, 95% CI (-0.45, -0.03), P < 0.01], while RT and GT had a non-significant effect (P > 0.05). Age-wise, significant improvement was observed in children and adolescents under 15 years of age [SMD=-0.41, 95% CI (-0.63, -0.19), P < 0.01]. The effect was significant for depressed patients [SMD=-0.75, 95% CI (-1.32, -0.19), P < 0.01], healthy individuals [SMD=-0.25, 95% CI (-0.41, -0.08), P < 0.01], and obese individuals [SMD=-0.15, 95% CI (-0.31, -0.00), P < 0.01], with the best effect observed in some psychiatric patients. The effect of 12 weeks of intervention was significant [SMD=-0.38, 95% CI (-0.56, -0.19), P < 0.01], while the effect of 6 consecutive weeks of intervention was not significant. The eight weeks subgroup also had better efficacy, but a high degree of heterogeneity was noted. In terms of the duration of exercise sessions, the effect of 30 minutes of exercise was significant [SMD=-0.14, 95% CI (-0.81, -0.01), P < 0.01], while 60 minutes of exercise was not significant (P > 0.05). The best effect was observed with 40–50 minutes of exercise [SMD=-0.17, 95% CI (-0.33, -0.02), P < 0.01]. Regarding exercise frequency, the effect of three times per week was significant in children and adolescents [SMD=-0.42, 95% CI (-0.66, -0.18), P < 0.01], while the effect of two times per week or greater than four times per week was not significant (P > 0.05).
Table 1
Results of traditional meta-analysis
Variable | Number of comparisons | Heterogeneity test results | Meta-analysis results |
Q value | I2 value | SMD effect size (95% CI) | P value |
Type | AE | 20 | 78.14 | 77.0% | -0.35(-0.59, -0.12) | <0.01 |
MT | 14 | 68.12 | 80.9% | -0.24(-0.45, -0.03) | <0.01 |
RT | 2 | 0.19 | 0.0% | -0.08(-0.36, 0.20) | = 0.557 |
GT | 5 | 51.80 | 92.3% | -0.26(-0.71, 0.18) | = 0.240 |
Age, y | <15 | 17 | 147.76 | 89.2% | -0.41(-0.63, -0.19) | P<0.01 |
≥ 15 | 11 | 29.66 | 66.3% | -0.24(-0.49, 0.01) | P = 0.062 |
Healthy status | Healthy | 19 | 85.89 | 77.9% | -0.25(-0.41, -0.08) | P<0.01 |
Obesity | 6 | 2.94 | 0.0% | -0.15(-0.31, -0.00) | P<0.05 |
Mental illness | 9 | 89.84 | 91.1% | -0.75(-1.32, -0.19) | P<0.01 |
Session duration, min | 30 | 7 | 30.92 | 77.4% | -0.14(-0.81, -0.01) | P<0.05 |
40–50 | 13 | 35.24 | 63.1% | -0.17(-0,33, -0.02) | P<0.05 |
60 | 7 | 15.19 | 60.5% | -0.13(-0.33, 0.07) | P = 0.200 |
Program duration, wk | 6 | 5 | 23.32 | 82.8% | -0.37(-0.86, 0.12) | P = 0.137 |
8 | 5 | 38.76 | 89.7% | -0.65(-1.49, 0.18) | P = 0.124 |
12 | 6 | 6.23 | 19.7% | -0.38(-0.56, -0.19) | P<0.01 |
>12 | 14 | 110.62 | 88.2% | -0.23(-0.46, -0.00) | P<0.05 |
Frequency, d/wk | 2 | 6 | 39.77 | 87.4% | -0.41(-0.85, 0.03) | P = 0.06 |
3 | 16 | 51.45 | 70.8% | -0.42(-0.66, -0.18) | P<0.01 |
≥ 4 | 8 | 20.85 | 66.4% | -0.19(-0.40, 0.02) | P = 0.08 |
| overall | | | | -0.28(-0.41, -0.14) | P<0.01 |
2.4 Network meta-analysis
Based on traditional meta-analysis, this study included 35 studies with 5293 participants in a network meta-analysis. The dots in Fig. 3 represent different exercise interventions, with the dot's size indicating the sample size, and the lines between the dots indicating a direct comparison between the two exercises. The thickness of the line represents the number of studies that made the comparison. If there is no line between two interventions, they cannot be directly compared and require reticulated meta-analysis. The overall inconsistency test showed that the P-values were all greater than 0.05 for all outcome indicators, indicating good overall consistency. Further examination of the consistency of each closure showed that the inconsistency factor (IF) values ranged from 0.06 to 1.06, and the lower limits of 95% CI all contained 0, indicating good consistency for each closure. Thus, the consistency model was used for analysis.
2.4.1 Network meta-analysis and probability ranking
The results of the network meta-analysis demonstrated that AE, MT, RT and GT significantly reduced depressed mood in children and adolescents compared to the control group (P < 0.01) (Fig. 4). According to the effect size and probability ranking scale, the SUCRA values for the intervention effects of the four sports were ranked as follows, in descending order: AE (66.2%), GT (62.5%), RT (59.0%), and MT (57.9%). (Table 2)
Table 2
Ranking table of sport’s type
Treatment | SUCRA | PrBest | Mean Rank |
CT | 4.4 | 0.0% | 4.8 |
AE | 66.2 | 23.8% | 2.4 |
RT | 59.0 | 33.9% | 2.6 |
GT | 62.5 | 30.0% | 2.5 |
MT | 57.9 | 12.3% | 2.7 |
2.4.2 Sensitivity analysis
In order to explore the sources of heterogeneity, sensitivity analyses were conducted by excluding individual studies in turn, and the results are presented in Fig. 5. Consistent with the initial analysis results, the exclusion of single studies had little effect on the combined results, indicating that the results of the combined effect values of this study were more stable.
2.4.3 Publication bias test
In addition, a publication bias test was performed on the included literature, and the results are shown in Fig. 6. An asymmetry at the bottom right of the funnel plot suggests the possibility of publication bias, but the method is subject to subjective judgment and may be inaccurate. Therefore, Begg's and Egger's tests were also applied, and the results indicated that P > 0.05, indicating no significant publication bias in the literature.