Eligibility criteria

This review included studies with an RCT or cluster RCT design, targeted at healthy children (age 6–12 years) and adolescents (age 13–17 years).16 Specifically, participants with a diagnosed cognition disorder or with a physical condition that may interfere with them performing exercise were excluded. This review included studies that employed one or multiple types of chronic exercise modalities as an intervention (typically over 6 weeks) on at least one aspect of EF and excluded studies employing acute exercise or other combined interventions (eg, nutrition).

Literature search

The electronic databases PsycINFO, PubMed, SPORTDiscus, Academic Search Premier, Embase and Web of Science were searched for relevant studies. The combination of the three groups of search MeSH terms were used to locate studies: (1) chronic exercise, aerobic exercise, physical activity (PA), exercise, training and sports; (2) EF, working memory, inhibitory control, cognitive flexibility and planning; (3) child, adolescent, teens and youths. Reference lists of included studies were further examined as complementary sources. All included studies must be published in peer-reviewed journals with full text and conference abstracts were excluded. The language was restricted to English, and there was no restriction on publication year.

Collection of studies

Duplications of records from the searching of databases and reference lists were first excluded, then records which did not meet publication status, and language and publication year were excluded. Next, two authors (Y.X. and Y.Y.) screened the titles and abstracts independently to further exclude records according to the eligibility criteria. Following this, full-text articles were assessed independently by the two authors. Any disagreements were discussed with a third reviewer (T.H.) until a consensus was achieved.

Data extraction

The data of publication year, participant description, study design, instrument, EF variables, study region and settings of eligible studies were extracted and summarised in table 1. According to the Cochrane Collaboration Handbook,21 the mean and SD values of pre-to-post intervention difference were first extracted. More specifically, the change values, if not reported, were calculated by: ‘Mean _change = Mean _post – Mean _pre’ and ‘SD _change = SQRT[(SD _pre 2 +SD _post 2) – (2*Corr*SD _post*SD _pre)]’, where Corr (correlation coefficient) was set as 0.5.21 For those studies which only provided ‘SE’ and ‘95% CI’, the SD was calculated based on the following formula: (1) SD=SE* SQRT (N), where N is sample size; (2) SD=SQRT (N) * [(UCI − LCI)/3.92], where U=upper CI, L=lower CI. When incorporating the sample size of a cluster RCT, the formulas in ‘16.3.5 Example of incorporating a cluster-randomised trial’ of Cochrane Handbook were followed. The design effect for the trial as a whole was calculated as: 1 + (M – 1) ICC, where M is the average cluster size in the trial, and intraclass correlation coefficient (ICC) was set as 0.02.21

Assessment of risk of bias

Two authors (Y.X. and Y.Y.) assessed the study quality according to the PEDro scale.22 The summary of the quality assessment was presented in table 2. Any disagreements were discussed with a third reviewer (T.H.) until a consensus was achieved.

Statistical analysis

R studio software V.1.1.453 (RStudio, Boston, Massachusetts, USA) was used to conduct the meta-analysis. Considering the different outcomes and units of cognitive measures used in the studies, standardised mean differences (SMD) of pre-post intervention were calculated and given weight by its inverse variance. Cohen’s d values of 0.2, 0.5, 0.8 represent small, moderate and large effect size, respectively.21 A random-effects model was used based on the assumption of different true effect size.21 The heterogeneity was assessed with I2 and p-value for Q statistic. I² values of 25%, 50%, 75% indicate small, medium and large amount of heterogeneity, respectively.21 Publication bias was assessed by Egger’s test and visually presented with a funnel plot where necessary.23

Additional statistical analysis included: (1) when studies included two intervention groups, their data were combined to create a ‘single pair-wise comparison’21; (2) when studies used two or more instruments to measure the same EF domain, only the more common-used one was included19; (3) for studies that reported multiple results on one cognitive task, the result of more-executive demanding condition was included (eg, incongruent trials in Flanker task)19; (4) when studies conducted two or more measurements, only the last measurement was included.19

After conducting a meta-analysis for overall EFs, subgroup analyses were performed based on the specific EF domains: (1) three core EFs: inhibitory control, working memory and cognitive flexibility; and (2) higher-level EF, namely planning herein.

To investigate the potential moderating effects, meta-regressions were performed based on continuous variables including age, body mass index (BMI) and intervention duration, while subgroup analyses were performed based on categorical variables including study design (RCT or cluster RCT), intervention modality, frequency and session length. The selection of these moderators was principally inspired by Álvarez-Bueno et al 5 and Wiebe and Karbach.3 For the intervention modalities, this review classified intervention modality as (1) curricular PA; (2) integrated PA, including physically active academic lessons, breaks and homework; and (3) sports and PA programmes, including after-school PA and other sports programmes.19 Sensitivity analyses were conducted by excluding studies from meta-analysis one by one.

Study characteristics

Overall, 19 studies were included in the meta-analysis and the study characteristics are summarised in table 1. Ten studies were RCTs17 24–32 and 9 studies33–41 were clustered RCTs. Regarding the study region, 11 studies were conducted in Europe, 5 in America, 2 in Asia and 1 in Oceania. The total sample included 4320 children and 718 adolescents aged from 6 to 17 years. There were 18 studies investigating the effects of chronic exercise on core EFs, of which, 16 studies assessed inhibitory control, 9 studies assessed working memory and 6 studies assessed cognitive flexibility. Compared with the studies exploring the effect of chronic exercise on core EFs, there are significantly fewer studies exploring higher-level EFs, with only two in planning and none in other higher-level EFs. Regarding the exercise intervention protocols, there were only 3 studies with sessions over 90 min, 10 studies with more than five sessions per week and 8 studies with durations over 24 weeks. The major confounding factors considered in the studies included age, BMI, duration, study design, intervention modality, frequency and session length.

The quality of the included studies is presented in table 2. In general, those studies were scored as fairly high quality, with a mean quality score of 7.1, because they were (cluster) RCTs. Moreover, over half of the studies (12 out of 19) were school-based interventions, including active class and breaks, curricular PA and extracurricular PA, whereas the remaining 9 studies were conducted in after school-time settings such as an afterschool PA programme or day camp.

Meta-analysis of effects on EFs

As illustrated by figure 2, the pooled SMD of overall EF was 0.20 (95% CI 0.09 to 0.30, p<0.05), with large heterogeneity (I²=76%, p<0.01). Regarding core EFs, the SMD was 0.26 (95% CI 0.08 to 0.45) for inhibitory control, with large heterogeneity (I²=84%, p<0.01); 0.10 (95% CI −0.05 to 0.25) for working memory, with medium heterogeneity (I²=58%, p=0.02); and 0.14 (95% CI −0.03 to 0.31) for cognitive flexibility, with medium heterogeneity (I²=58%, p=0.04). Regarding the higher-level EF, the SMD was 0.16 for planning (95% CI −0.13 to 0.45), with small heterogeneity (I²=0%, p=0.43). The high heterogeneity for core EFs reflects the importance of taking various moderator factors into consideration when analysing the effects of chronic exercise.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Forest plot for meta-analysis regarding the effect of chronic exercise on different EF domains. (1) core EFs: cognition flexibility, inhibitory control, and working memory (2) higher level EF: planning.

Moderator analysis

Table 3 summarises the results from the meta-regressions and subgroup analyses for overall EFs. In the meta-regressions, higher BMI was positively associated with greater SMD (β=0.03, 95% CI 0.0002 to 0.06, p<0.05, adjusted R²=32.09). Age and intervention duration were not related to changes in SMD.

In the subgroup analyses, the effect of chronic exercises on EFs was significantly moderated by intervention modality and session length, but not by study design and frequency. As illustrated by figure 3, the SMD was 0.02 (95% CI −0.05 to 0.09) for integrated PA (ie, physically active academic lessons, breaks and homework), with small heterogeneity (I²=23%, p=0.22); and 0.21 (95% CI 0.12 to 0.31) for sports and PA programme, with small heterogeneity (I²=0%, p=0.46). At the same time, the SMD for curricular PA was 0.39 (95% CI 0.08 to 0.69), with large heterogeneity (I²=85%, p<0.01). With regard to the exercise session length, the SMD for interventions with a session length < 90 minutes was 0.24 (95%CI 0.10 to 0.39, p=0.02) with large heterogeneity (I²=79.5%, p<0.01), whereas the SMD for interventions with a session length ≥ 90 minutes was 0.05 (95%CI -0.03 to 0.14).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Forest plot for subgroup analysis of intervention modality (curricular PA, integrated PA, and sports and PA program) as a moderator.

Sensitivity analysis

Two studies36 39 were found to be major contributors to the high heterogeneity by the leave-one-out sensitivity analysis. After excluding these two studies, heterogeneity was medium (I²=41.1%, p=0.01), and the pooled SMD of overall cognitive measures was 0.12 (95% CI 0.05 to 0.19).

Publication bias

The funnel plot is presented in the online supplementary figure 1. Egger’s test was used to assess publication bias (t=2.44, df=31, p=0.02). The results indicated the possibility of publication bias for all included studies.23

Main study findings

This review included only RCT studies with consideration of various moderators to examine the effects of various types of chronic exercise interventions (typically over 6 weeks) on EF among children and adolescents. We found a small but significant effect size of chronic exercise interventions on overall EFs and inhibitory control. In addition, the effect of chronic exercise on overall EFs was moderated by BMI and intervention modality.

Comparisons with previous studies

Our findings of a small effect size on overall EFs and inhibitory control extends a 2014 systematic review that included 19 articles.16 In that study, chronic exercise has no significant effect on overall EFs, without further investigation on specific EF domains. Importantly, there were only five chronic exercise studies in that review and the authors did not limit their review to RCTs. Our findings are strongly supported by another systematic review, which included 36 studies.19 Álvarez-Bueno et al 19 reported a small effect size of chronic exercise on overall EFs and inhibitory control with large heterogeneity. They also revealed a small effect size on higher-level EFs, including planning, with small heterogeneity. Despite the similar results to our findings, their review also included non-RCT studies, similar to Verburgh et al.16 Additionally, they deemed two cohorts or intervention groups as independent samples in their meta-analysis, and this may inflate the true sample size and distort the true weight of each study.

Although aerobic exercise has relatively broad positive effects across a variety of cognitive functions, the benefits of exercise intervention seem to be more significant for EFs.42 However, there is no consensus on which cognitive task is more sensitive to exercise interventions. Among the three core EFs, previous reviews have suggested that inhibitory control is especially sensitive to chronic exercise interventions among children and adolescents,3 19 and our findings also support this point. Among higher-level EFs, our results on the planning domain indicated no effect size with low heterogeneity. However, since there were only two included studies with outcome of this domain, it is premature to draw a clear conclusion.

BMI as a moderator

Between-study heterogeneity is common in a meta-analysis,43 and, in our findings, the effect of chronic exercise interventions on overall EFs and inhibitory control has high heterogeneity, whereas the one on working memory, cognitive flexibility and planning has small or medium heterogeneity. It is worth noting that the meta-regression revealed that the BMI is a moderator on the effect of chronic exercise on EFs, which means that the effects of exercise interventions were comparatively larger on the population with higher BMI. Previous studies have shown that children with excessive body weight demonstrated poorer EF.44 45 Therefore, the potential change of EFs determined by the proportion of overweight participants in the sample may explain the inconsistent findings among the general population. Crucially, it is possible that the children with obesity may benefit more from chronic exercise than their normal-weight peers.

Intervention modality and session length as moderators

In addition to BMI, intervention modality (ie, curricular PA, sports and PA programme, and integrated PA) and session length were also recognised as moderators in our review. The beneficial effects of chronic exercise on overall EFs were observed in the modalities of curricular PA as well as the sports and PA programme rather than integrated PA. Although integrated PA in classrooms is posited to be a promising intervention strategy, current studies have identified that this kind of interventions had no significant effects on EFs. In addition, interventions with a session length < 90 minutes had a positive effect on overall EFs, whereas the exercise session length ≥ 90 minutes had no significant effects on EFs. The reasons for the findings are not clear. As opposed to sports programme and curricular PA, integrated PA possesses less amount of social component and group dynamics. Additionally, the exercise intensity of classroom-integrated PA may be lower than that of sports programme and curricular PA. The results may therefore be partly explained. Also, since these studies had no long-term follow-ups, it remains unclear whether a potential benefit will emerge after a longer period postintervention of integrated PA. Nevertheless, given that no adverse effects were observed, integrated PA in classroom might be a potential way to intervene sedentary behaviour among children and adolescents, without impairment in cognitive development.

As well as integrated PA, cognitively engaging exercise, which is characterised with increased cognitive demands and challenges inherited in the exercise, is also attracting significant attention in this field. Recent studies40 46 47 have hypothesised that cognitive engagement and efforts, instead of physical fitness or exercise dose, may bring about functional changes and contribute to improvements in EFs. Therefore, if an intervention modality involves cognitive engagement, it might enhance the effectiveness of exercise intervention on EFs.

Other potential reasons for the heterogeneity

While the high heterogeneity of pooled effects on overall EFs was partly explained by BMI, intervention modality and session length, our study found that the effect of chronic exercise on EFs was not moderated by other variables such as study design, age, duration, or frequency. This may be due to four potential reasons. First, the high variability among adherence to the exercise protocol varied from study to study and it might lead to the inconsistencies.48 Second, it is especially hard for researchers to control the potential exercise and sports participation of the control group when they are out of the study context, and this may further influence the effect sizes.48 Third, although many studies have taken moderating factors such as age, gender and pubertal stage into consideration, other factors with high inter-individual variability, such as genotype, may also have influenced the intervention effects.48 Lastly, there is no gold standard measurement for EFs thus far.29 Therefore, the cognitive measurements in the included studies may partly explain the incongruous results within the same domain.

Over half of the included studies (12 out of 19) were enacted under real-life settings (eg, school) rather than laboratory settings, which may increase the generalisability of their results. However, while the settings might increase the generalisability, the sample size is comparatively small. It is important to note that most studies conducted post hoc analyses rather than calculating the sample size in advance based on the power requirement and expected effect size. Thus, it may impact the reliability of their results.

Strengths and limitations

One strength of this systematic review is the strict inclusion of RCT studies and exclusion of observational or longitudinal studies. This strict inclusion increases the reliability of the causal relationship inferences. Another strength is the full consideration of various key moderators such as intervention modality, study design, age, duration, frequency and BMI. This provides a much clearer picture than previous reviews of the effects of chronic exercise interventions on EFs. The limitation of this review is the inclusion of various cognitive tasks, which may lead to the high heterogeneity in this review. Furthermore, although we made a comprehensive literature search and an objective study selection, the Egger’s test indicated a possibility of publication bias. Therefore, the findings of the meta-analysis need to be interpreted with caution. Another limitation is the inadequate number of studies in adolescents included. Because of this, the effects of chronic exercise on EFs among different age groups remain to be determined. The third limitation is the insufficient number of studies on the higher-level EFs, which need to be further studied. Lastly, some other potential moderators may not be identified and addressed in our review.

Future studies need to pay particular attention to exploring the following aspects. First, higher-level EFs should be given more attention due to the extremely limited number of studies. Second, more studies in adolescents are needed, considering that different age groups have different speeds and stages of EF development. Third, whether cognitive engagement is a key moderator and whether it substitutes the moderation effect of other potential variables should be further validated. Lastly, longer follow-ups are recommended to track the long-term effects of exercise interventions over time.