Chronic Exercise for Core Symptoms and Executive Functions in ADHD: A Meta-analysis

This meta-analysis was performed following the guidelines of the Cochrane Collaboration handbook,³³  and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement.³⁴  The protocol for this study was registered in PROSPERO (CRD42021229722).

Relevant studies were searched and identified using 5 databases: PubMed, Embase, Web of Science, Cochrane Library, and China National Knowledge Infrastructure from database inception to July 31, 2022. The search strategy combined the following keywords:

1. “children,” “adolescents,” “youth,” “teenager,” “attention-deficit/hyperactivity disorder,” and “ADHD”;

2. “physical activity,” “exercise therapy,” “motor activity,” “exercise,” “train,” and “sport”; and

3. “EFs,” “cognitive function,” “working memory,” “cognitive flexibility,” “inhibition,” “inattention,” “hyperactivity,” and “impulsivity.”

Additional citations were identified by searching Google Scholar using the same terms and among the reference list of the identified studies.

Studies were included if all of the following criteria were met:

1. participants: children or adolescents with ADHD aged 6 to 18 years (diagnosed according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth and Fifth editions);

2. duration of exercise interventions: ≥6 weeks;

3. outcomes: reported core symptoms (eg, inattention, hyperactivity/impulsivity) and/or EFs (eg, inhibition, cognitive flexibility, working memory);

4. study design: pre–posttest randomized control group design; and

5. published in peer-reviewed journals and written in either English or Chinese.

We excluded the studies if:

1. the studies used healthy children for comparison;

2. the intervention group was based on multimodal intervention, such as exercise combined with psychotherapy or behavioral interventions³⁵ ;

3. the study’s results focused only on the physical fitness and gross motor skills of children or adolescents with ADHD;

4. duplicate published studies were found; and

5. the studies had incorrect, missing, or unextractable data information.

Two independent researchers (H.H., and C.H.) conducted the data extraction, and disagreements were resolved by discussion with third parties (M.Q.). We extracted the following information:

1. study characteristics (eg, author[s], year of publication, the country or region where the data were collected);

2. participant characteristics (eg, age and sex of participants, sample size);

3. study protocol including design, interventions, and control conditions (eg, motor skill, exercise intensity, exercise session duration, and exercise frequency); and

4. measurement methods and outcomes (core symptoms and EFs).

Two authors (H.H., and S.G.) independently assessed the risk of bias of each included study using an 8-item checklist adapted from the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement.³⁶  Scoring discrepancies were resolved via consensus and interrater reliability was calculated using percentage agreement. A risk of bias score was awarded to each study on the basis of an 8-point scale coded as “clearly described” (√), “absent” (×), or “unclear or inadequately described” (?) for each of the following criteria:

1. eligibility criteria were specified;

2. participants were randomly allocated to groups;

3. the groups were similar at baseline regarding the primary outcome(s);

4. there was blinding of all assessors who measured the primary outcome(s);

5. data for primary outcome(s) were analyzed by “intention to treat”;

6. dropout for primary outcome(s) was described, with <20% dropout of participants;

7. conducted the sample size calculations and the study was adequately powered to detect changes in the primary outcome(s); and

8. summary results for each group with estimated effect size (ES) (difference among groups) and its precision (eg, 95% confidence interval [CI]) were reported.

Criteria were added to create an overall risk of bias score, with studies graded as low risk if scoring 7 to 8, moderate risk if scoring 4 to 6, or high risk if scoring <4.

We used a random-effects model in our meta-analysis of the included studies. For the studies that only presented data as graphs, we estimated means and SDs using the Y-axis and length of the histogram.^37,38  For the studies that reported standard errors, we converted these data to SDs.³⁹  The mean and SD values of pre- to postintervention differences between intervention and control groups were used to calculate ES for each included study. The formulas for the mean and SD pre- to postchange values were as follows^37,38 : “mean_change = mean_post – mean_pre” and ‘SD_change = square root ([SD_pre² + SD_post²] − [2 × correlation coefficient × SD_pre × SD_post]),” where the correlation coefficient was set to 0.5 based on the Cochrane Collaboration’s tool for assessing the risk of bias in randomized trials.⁴⁰  In this study, we estimated the pooled ES using the standardized mean difference (SMD) because the outcomes of interest among the included studies were measured using different methods. The ES values were quantified as large (|SMD| ≥ 0.8), moderate (0.5≤|SMD| < 0.8 SMD), small (0.2≤|SMD|< 0.5), or nonsignificant (|SMD|< 0.2).⁴¹ 

The following statistical treatments were applied:

1. if the study had 2 or more intervention groups, the data were considered independent samples. The sample size of control groups was evenly distributed to each intervention group to avoid artificially inflating the actual sample size^38,42,43 ;

2. if a study measured 2 or more core symptoms dimensions,^37,44–46  or similarly, measured 2 or more EF domains,^{37,39,42,43,45–50}  the sample size of the control group was divided by the number of domains measured in the study (Supplemental Tables 4 and 5);

3. if the study used 2 or more tasks to measure the same EF domains, the ES from the most commonly used task was selected^38,47 ;

4. if the study reported the same core symptom dimensions assessed by a different third party (eg, parents and teachers), all of the results were selected^51,52 ; and if the study reported the same EF task with multiple measurement results, outcomes that require more executive ability were selected (eg, incongruent trials in the flanker task)^43,49,53,54 ;

5. if the study reported 2 or more measurement time points, we took only baseline (ie, preintervention) and postintervention (ie, immediately after interventions) measures⁵⁰ ; and

6. of the 9 studies^{37,38,43–45,49,51,55,56}  that did not report the intensity of the exercise interventions, we estimated the intensity on the basis of the Youth Compendium of Physical Activities,⁵⁷  which coded exercise intensity as “light,” “moderate,” or “vigorous” corresponding to 1.5 − <3 metabolic equivalents (METs), 3 − <6 METs, or ≥6 METs, respectively.

We estimated the intensity in 1 study³⁹  on the basis of the same intervention protocol reported in Hattabi’s study.⁴⁷ 

The type of exercise interventions was categorized as open and closed on the basis of different conceptual characteristics identified in previous publications.⁵⁸  Open-skill exercise interventions require greater cognitive engagement to execute more complex movements and adapt to ever-changing task demands, such as tai chi or basketball. In contrast, closed-skill exercise requires more automatic behaviors or straightforward repetitive exercises with limited cognitive engagement, such as walking, running, or jumping.

To examine whether the characteristics of exercise and study modified the effects of CEIs on children and adolescents with ADHD, subgroup analyses were performed according to the categorical variables: study population (children or adolescents), core symptoms dimensions (inattention or hyperactivity/impulsivity), EF domains (inhibition, working memory, or cognitive flexibility), intervention type (exercise plus medication or exercise only), motor skills (open or closed), exercise intensity (light, moderate, moderate-to-vigorous, or vigorous intensity), exercise session duration (≤50 or >50 minutes per session, median), total exercise sessions (exercise frequency × exercise duration, <24 or ≥24 sessions, median), control group (education, medication, neurofeedback [NFB], or sedentary activity/no treatment), outcome raters (parents or teachers), and study quality (low risk or moderate risk). In addition, meta-regressions were performed on the basis of continuous variables, including age, BMI, and total exercise volume (exercise frequency × exercise duration × exercise session duration).

Heterogeneity across included studies was assessed using the I² statistic. Values of I² <25%, 25 − <50%, 50 − <75%, and ≥75% were defined as very low, low, moderate, and high degrees of heterogeneity, respectively.⁵⁹  Publication bias was assessed visually with funnel plot asymmetry and statistically by Egger’s test.⁶⁰  Furthermore, the “trim and fill” method was used to assess the stability of the pooled results if publication bias was observed.⁶¹  To test the robustness of the results, sensitivity analyses were conducted by removing 1 primary study from the pooled analysis in each instance.⁶² 

All analyses were performed using Stata version 13.0 software (Stata Corporation LP, College Station, TX). A 2-sided P value ≤ .05 was considered statistically significant.

Our electronic search yielded a total of 1333 relevant records. Of these, 1311 were excluded according to the exclusion criteria by reading the title, abstract, and full text. Twenty-two studies were included in our meta-analysis, of which 5 focused on the effects of core symptoms, 11 focused on the effects of EFs, and 6 reported both core symptoms and EFs. The literature screening process is summarized in Fig 1.

Of the total of 22 studies, 4 studies were conducted in North America,^39,46,52,54  11 in Asia,^{37,43,44,48–50,55,63–66}  3 in Europe,^38,42,45  3 in Africa,^47,56,67  and 1 in Oceania.⁵¹  The total number of participants was 1309, with 667 in intervention groups and 642 in control groups. Fifteen of 22 studies included participants of both sexes,^{38,39,42–48,50,52,56,65–67}  whereas the remaining 7 studies included only male participants.^{37,49,51,54,55,63,64}  The shortest intervention period was 6 weeks,^37,63  and the longest was 72 weeks.⁵⁶  The exercise frequency in the intervention group was at least 1 session per week,^38,51  with a maximum of 7 sessions per week.⁵²  The duration of each session ranged from 15 to 180 minutes. Ten studies did not report exercise intensity,^{37–39,43–45,49,51,55,56}  5 reported moderate-intensity exercise, ^{46,47,63,67,68}  6 reported moderate-to-vigorous–intensity exercise,^{42,46,48,50,52,64}  and 1 reported vigorous-intensity exercise.⁶⁶  The intervention involved various types of exercises, including running,^42,66,67  combined aerobic exercise,^{37,38,46,50,52,54,63,64}  exergaming,⁴⁵ orienteering exercise,⁶⁵  swimming,^39,47  yoga,⁵¹  hippotherapy,⁴⁴  taekwondo,⁵⁶  cycling,⁴⁸  soccer,⁴⁹  and table tennis.^43,55  In contrast, the control group included education,^37,63  medication,^42,44  NFB,⁴²  and sedentary activity/no treatment.^{38,39,43,45–52,54–56,64–67}  In addition, 19 studies were written in English^{37–39,42–47,50–52,54–56,63,64,66,67}  and 3 studies were written in Chinese^48,49,65  (Table 1).

Details of the risk of bias in the 22 included studies are provided in Supplemental Table 4. Three studies were considered to have a low risk of bias,^45,46,52  whereas 19 had moderate risk.^{37–39,42–44,47–51,54–56,63–67} 

Our meta-analysis showed that CEIs improved core symptoms (SMD = −0.39, 95% CI: −0.64 to −0.41), as well as EFs (SMD = −0.68, 95% CI: −0.91 to −0.45) (Figs 2 and 3). Low heterogeneity was found in the results of studies on the effect of CEIs on core symptoms (I² = 41.6%, P = .03) and moderate heterogeneity in the results of studies on the effect on EFs (I² = 54.4%, P < .01). The Egger test showed no significant publication bias for the included studies (P = .11 for core symptoms and P = .85 for EFs) (Supplemental Fig 4 and 5).

The effects of CEIs on core symptoms and EFs in children and adolescents with ADHD may be moderated by different characteristics (eg, age, motor skills, exercise intensity, and type of control) (Figs 4 and 5). The subgroup analyses showed that the CEIs had a small beneficial effect on inattention dimension (SMD = −0.32, 95% CI: −0.63 to −0.004) but no significant effect on hyperactivity/impulsivity dimension (SMD = −0.41, 95% CI: −0.92 to 0.09). Meanwhile, in exercise plus medication (SMD = −0.56, 95% CI: −0.89 to −0.24), closed-skill exercise (SMD = −0.83, 95% CI: −1.30 to −0.35), moderate-intensity exercise (SMD = −0.43, 95% CI: −0.69 to −0.18), parent-raters (SMD = −0.44, 95% CI: −0.76 to −0.13), and moderate-risk studies (SMD = −0.56, 95% CI: −0.93 to −0.19) subgroups, CEIs had a small-to-large beneficial effect on overall core symptoms compared with controls. Additionally, CEIs achieved significantly better improvement in core symptoms than education (SMD = −0.80, 95% CI: −1.35 to −0.26) and sedentary activity/no treatment (SMD = −0.39, 95% CI: −0.68 to −0.10). Lastly, the beneficial effect of CEIs on overall core symptoms was not affected by the study population (children or adolescents), exercise session duration (≤50 or >50 minutes per session, median), and the total exercise sessions (<24 or ≥24 sessions, median).

In terms of EFs, CEIs yielded the greatest improvement in inhibition (SMD = −0.88, 95% CI: −1.43 to −0.34), followed by cognitive flexibility (SMD = −0.71, 95% CI: −0.99 to −0.42) and working memory (SMD = −0.50, 95% CI: −0.74 to −0.26). Meanwhile, moderate-risk studies (SMD = −0.77, 95% CI: −1.03 to −0.51) subgroups had moderate beneficial effects on EFs, and CEIs had a significantly better effect on EFs than education (SMD = −0.70, 95% CI: −1.23 to −0.16) and sedentary activity/no treatment (SMD = −0.84, 95% CI: −1.07 to −0.60). Additionally, the effects of CEIs on EFs were not influenced by study population (children or adolescents), intervention type (exercise plus medication or exercise only), motor skills (open or closed), exercise intensity (moderate or moderate-to-vigorous intensity), exercise session duration (≤50 or >50 minutes per session, median), and total exercise sessions (<24 or ≥24 sessions, median).

Furthermore, meta-regression analyses showed that age, BMI, and total exercise volume were not related to changes in overall effect of CEIs on core symptoms and EFs (Supplemental Tables 5 and 6).

After removing the study comparing the effect of exercise with medication,⁴⁴  the effects of exercise only on core symptoms changed from being nonsignificant (SMD = −0.25, 95% CI: −0.68 to 0.19) to significant (SMD = −0.45, 95% CI: −0.73 to −0.16).

In this meta-analysis study, we examined the effects of CEIs on core symptoms and EFs among children and adolescents with ADHD. We found that CEIs had a small beneficial effect on overall core symptoms, as well as the inattention dimension. In addition, there was a moderately beneficial effect of CEIs on overall EFs and a moderate-to-large effect on the specific domains of EFs. We also observed that different types of exercise had varying effects on core symptoms, showing that closed-skill exercise was more beneficial for core symptoms.

We found a beneficial effect of CEIs in overall core symptoms of children and adolescents with ADHD, which is consistent with most previous meta-analyses.^21,25,27  In contrast, another meta-analysis²³  showed no significant effect of exercise on core symptoms, possibly because of the lack of relevant studies.^{37,38,42,44,46,64,67}  With regard to the different dimensions of core symptoms, findings reported by Xie²¹  and Sun³²  were consistent with our study in that the exercise interventions improved inattention, but not hyperactivity/impulsivity. However, Cerrillo-Urbina’s study²⁵  showed a moderate effect of aerobic exercise on hyperactivity/impulsivity symptoms, possibly because of the inclusion of studies with acute exercise,^69–71  where the intervention effects were more easily observed when tested immediately after exercise. Additionally, ADHD core symptoms are evaluated chiefly on the basis of self-reporting information provided by parents or teachers, which may lead to variability in the results across studies.⁷² 

Regarding the EFs, different studies, including our own,^{22,24,28,29,31}  consistently show that exercise interventions have beneficial effects on the inhibition and working memory domains. However, the current study revealed a moderately significant effect of exercise on cognitive flexibility in ADHD (SMD = −0.81, 95% CI: −1.17 to −0.45). In contrast, Tan found that this effect was not significant,²⁴  possibly because this study included older participants (18–25 years)^73,74  who tend to have more stable and mature cognitive flexibility⁷⁵  that is less likely to change.

Preliminary results in this study suggest that exercise only may not have a significant effect on the improvement of core symptoms. However, when the study with medication as the control group was removed in the sensitivity analyses,⁴⁴  the effects of exercise only on core symptoms changed from nonsignificant (SMD = −0.25, 95% CI: −0.68 to 0.19) to significant (SMD = −0.45, 95% CI: −0.73 to −0.16). Since medication is more effective at improving core symptoms than exercise only,⁴²  using medication as a control group may obscure the beneficial effect of exercise. Our analysis of the results suggested that both exercise only and exercise plus medication interventions improved core symptoms and EFs compared with the effect of nonpharmacological controls, including education and sedentary activity/no treatment.

Interestingly, closed-skill exercise showed a beneficial effect on core symptoms in children and adolescents with ADHD, whereas open-skill exercise did not. It may be because of the simplicity and repetition of closed-skill exercises that make it easier for children and adolescents with ADHD to master and automate, thereby maintaining the established exercise intensity (eg, moderate intensity) that romotes the secretion of neurotransmitters, such as catecholamines.⁷⁶  Conversely, children with ADHD generally have poorer fundamental motor skills⁷⁷  and EFs,⁷⁸  and they may have difficulty maintaining established exercise intensity when participating in open-skill exercises requiring higher motor skills and greater cognitive engagement.⁵⁸  However, light-intensity exercises are less effective than moderate-intensity exercise because of their inability to improve low arousal in individuals with ADHD.⁷⁹  Thus, closed-skill exercises may be more beneficial than open-skill exercises in improving core symptoms in children and adolescents with ADHD.

The study findings varied because of differences in the control groups. The CEIs resulted in more significant improvements in overall core symptoms and EFs compared with the education and sedentary activity/no-treatment groups. Additionally, a small number of studies directly compared CEIs with medication^42,44  and NFB,⁴²  with the results failing to show that CEIs are better than medication and NFB in improving the observation outcomes. Nevertheless, we should note that CEIs have the advantages of having no medication-related side effects (eg, sleep problems and loss of appetite) and do not require as complex a device as NFB. What’s more, among the many treatment options, parents appear to prefer more holistic approaches such as exercise and nutrition.²⁰  Thus, exercise intervention should be considered as a component of a multimodal treatment approach to ADHD.

The exercise intensity usually moderates the effectiveness of exercise interventions.⁸⁰  In this meta-analysis, CEIs with moderate intensity had a beneficial effect on core symptoms, whereas other intensities did not (eg, light and vigorous intensity). The findings are consistent with the previous meta-analyses,^22,27  as well. This observation may be because children and adolescents with ADHD generally have a poor arousal state because of delayed prefrontal cortical development,⁸¹  and moderate-intensity exercises can produce an optimal arousal state in the cerebral cortex.⁸²  This explanation is consistent with the inverted “U”-shaped relationship between exercise intensity and cognitive improvement,⁸⁰  whereby moderate-intensity exercise promotes cognitive development in healthy populations more effectively than light- and vigorous-intensity exercise. These findings require further validation, especially the high-quality research evidence of direct comparison among different exercise intensities.

The findings of this study showed that CEIs had a positive effect on core symptoms when raters were parents, but not when raters were teachers. In other nonpharmacological treatments, such as cognitive training, NFB, and behavioral interventions, teachers were also found to have significantly lower ratings of intervention effectiveness than parents.⁷² 

This difference is probably caused by inflation of ES estimates by parents, most of whom were unblinded and had expectations of intervention. It is also possible that parents’ assessment was correct, but the improvements did not generalize from parent-observed to teacher-observed settings. Additionally, teachers are more likely to blindness or may not know the children with ADHD well, thus reducing their sensitivity to observing changes before and after the intervention. In general, it is necessary to evaluate the effectiveness of an intervention with different raters, especially when benefits can be found in more conservative results.

Current mainstream studies agree that ADHD is mainly caused by lower dopamine levels,⁸³  and the mechanisms by which exercise interventions benefit inattention symptoms may be related to increased dopamine levels.⁸⁴  However, exercise interventions are not effective for hyperactivity/impulsivity symptoms, and this may be because of the brain’s dopamine reward pathway being associated with inattention symptoms, but not hyperactivity/impulsivity symptoms.⁸⁵ 

Several explanations may help explain the mechanism by which exercise interventions can improve EFs of ADHD. First, exercise interventions improve EFs by increasing cerebral blood flow and the supply of oxyhemoglobin to the prefrontal cortex,⁸⁶  which has been linked to positive changes in EFs. Second, exercise interventions can promote connectivity and activation between ADHD-related brain functions, thereby improving EFs.⁸⁷  Third, evidence from typically developing children⁸⁸  suggests that exercise interventions may improve brain function by improving brain structure, positively affecting core symptoms and EFs in children and adolescents with ADHD.

The strengths of this study are as follows:

1. only RCTs were included, which increases the reliability of our findings;

2. we simultaneously explored the effects of exercise on 2 clinically important and relevant impairments (core symptoms and EFs) in children and adolescents with ADHD, thereby contributing to a more comprehensive overall understanding of the effects of CEIs on ADHD; and

3. we comprehensively analyzed the moderators of exercise interventions affecting core symptoms and EFs in ADHD to provide theoretical references for subsequent targeted exercise interventions.

Several limitations must also be addressed. First, core symptoms, EF measurements, and exercise intervention characteristics varied across studies, possibly leading to biased estimates of intervention effects. However, we attempted to enhance the accuracy of our estimates of the effects of exercise interventions through subgroup and sensitivity analyses. Second, some of the subgroup analyses could not be completed because of the limitations of the included studies (eg, sex and subtypes of ADHD). In addition, the subgroup analysis results must be further confirmed because of the small number of included studies (eg, studies of exercise versus drugs and exercise versus NFB). However, we have combined the findings of all the studies available to date. Third, potential publication bias because of the exclusion of studies published in other languages cannot be completely excluded, although the statistical results in our study suggest no publication bias.

Findings from this meta-analysis have implications for future research. First, exercise interventions need to be further refined and optimized to enhance their impact on core symptoms and cognitive functions. Currently, most of the comparative studies on the effects of different exercise prescription are derived from indirect evidence from reviews or meta-analyses. Direct comparisons of different exercise prescription parameters, such as motor skills (open versus closed) and exercise intensity (moderate versus vigorous intensity) on ADHD core symptoms and EFs, are required to confirm the optimal exercise therapy program. Second, there needs to be a paradigm shift in intervention delivery, moving from in-person interventions to eHealth delivery, including Web-based exercise interventions. Web-based exercise interventions can break the constraints of time and space, allowing more individuals with ADHD to have the opportunity to participate. Especially in the current period of the coronavirus disease 2019 epidemic, it is of great practical significance to explore the effect of Web-based exercise interventions on ADHD. Third, identify the potential sports injury risks. It is unclear how often sports injuries occur during exercise interventions for children and adolescents with ADHD; future research should report the number of injuries in exercise interventions, estimate the probability, and investigate how to reduce the occurrence of sports injuries. Fourth, the effects of exercise were compared with other treatment methods. In most previous studies, no treatment was included as a control condition. Future studies should compare the differences in the effects of exercise interventions and other nonpharmacological treatments on ADHD to clarify the effect of exercise on ADHD further. Fifth, identify the effects of exercise on preschool children with ADHD. Few studies have focused on the effects of exercise in preschoolers with ADHD, those who recommend nonpharmacological treatment alone first.⁸⁹  Therefore, high-quality exercise intervention studies for preschoolers with ADHD are needed. Sixth, the effects of exercise on different subtypes of ADHD and the underlying mechanisms should be explored. There is a lack of empirical studies comparing the effects of exercise on different ADHD subtypes (inattention, hyperactivity/impulsivity, or combined). Additionally, future studies should also explore the potential physiologic mechanisms of exercise interventions to improve ADHD by measuring actual physiologic changes (eg, neurotransmitter levels in the brain cortex⁶⁸ ). Longer follow-up is also recommended to track the long-term effects of exercise interventions.

This meta-analysis suggests that CEIs have small-to-moderate beneficial effects on overall core symptoms and EFs in children and adolescents with ADHD. In addition, closed-skill exercise may be more beneficial than open-skill exercise for improving overall core symptoms in ADHD. Given their effectiveness and popularity with parents, CEIs should be considered as a complementary treatment of children and adolescents with ADHD. High-quality RCTs are necessary in the future to determine the optimal exercise prescription for the treatment of ADHD with different ages, symptoms, and subtypes.

ADHD

attention-deficit/hyperactivity disorder

CEIs

chronic exercise interventions

CI

confidence interval

EFs

executive functions

ES

effect size

METs

metabolic equivalents

NFB

neurofeedback

RCTs

randomized controlled trials

SMD

standardized mean difference