Interventions to Promote Physical Activity and Healthy Digital Media Use in Children and Adolescents: A Systematic Review

The protocol for this review was registered within the International Prospective Register of Systematic Reviews (PROSPERO #: CRD42020213361). This review was originally designed to evaluate the effectiveness of both (1) nondigital interventions to reduce screen use and sedentary behavior, and (2) digital-based interventions for universal health promotion in school-aged children and adolescents. One search strategy was used (Supplemental Information), and eligible studies were screened together until the abstraction phase, at which time included studies were abstracted and analyzed separately between studies reporting nondigital interventions and those studies assessing digital-based interventions. Given the large number of studies included, the review authors decided to report the evidence synthesis separately.²⁴  As guidance, we propose a socio-ecological conceptual framework for digital and nondigital health interventions (Fig 1).

Searches were conducted using a specified search strategy (Supplemental Information) in the following databases: PubMed, Embase, Medline, PsycINFO, Cumulative Index to Nursing and Allied Health Literature, The Cochrane Library, Cochrane Central Register of Controlled Trials (CENTRAL), Cochrane Methodology Register, and the World Health Organization regional databases. The terms were combined with the Cochrane Medline filter for controlled trials of interventions. There were no limitations on geographical settings, publication language, or duration of intervention follow-up. The final search was completed March 16, 2021. Additional details about the search strategy development and other information sources are included in Vaivada et al.²⁵ 

Although all screening was conducted by a single reviewer, full-text review and data abstraction were conducted in duplicate. Expanded details of the screening and selection process for this review can be found in Vaivada et al.²⁵  Specific eligibility criteria were used to screen and select studies for inclusion (Table 1).

Eligible study designs included randomized controlled trials (RCTs), quasi-experimental studies (QES), and nonrandomized trials that already assessed the feasibility of the intervention to evaluate the research question.²⁶  As such, small pilot or feasibility trials without any follow-up larger trials were excluded. Studies were eligible if published in 2000 or after. Classification of high-income countries (HIC) and low- and middle-income countries (LMIC) was conducted according to the World Bank’s 2019 fiscal year country income classification. Studies that included both children and adolescent participants without disaggregating the age groups were included, where the majority of the study’s sample age fell within the selected age range, or the average mean age reported was between 5 and 19.9 years.

Interventions were defined as any planned action, program, or policy that was implemented to promote healthy digital media use and to reduce sedentary behaviors, screen use, or screen time (Table 1). Eligible comparisons were no intervention (placebo), standard arm of care (eg, existing school programs, activities, or initiatives), or other intervention arms in the case of a multicomponent intervention (eg, nutrition education arm versus nutrition education + digital component). Studies were excluded if the primary aim of the intervention(s) was treatment, therapy, and/or management of existing chronic disease (ie, weight loss or treatment of diagnosed overweight and obesity). Only interventions that specifically measured our primary outcomes of interest (screen time and sedentary behavior, as author defined) were included. Although we are aware that physical activity-focused interventions may address sedentary behavior in terms of increases in physical activity or aerobic performance, these metrics were not primary outcomes of interest for this review.

Statistical analysis was conducted using Review Manager 5.4 software. Randomized controlled trials and cluster-randomized controlled trials were analyzed separately from quasi-experimental study designs. Meta-analyses were conducted for each outcome of interest, only when there were data fora minimum of 3 studies. Where multiple measures were reported for an outcome in a single study, we used the most commonly reported measure across all included studies. To mitigate heterogeneity within included studies, a random effects meta-analysis was used for all pooled outcomes. Overall effect estimates were considered statistically significant if the associated P value was <.05.

Because of variation in when studies evaluated outcomes after intervention, when given the choice between after intervention and an alternative, and longer follow-up period, we reported the time point that immediately followed the end of the intervention. This was done where possible across all studies for more consistent and generalizable synthesis. Where possible and appropriate, unit conversions were conducted; this was largely done for screen time and sedentary behavior outcomes where screen time was measured differently (ie, hours/day versus minutes/day). We did not adjust estimates for clustering if cluster-randomized-controlled trials did report adjusted estimates. Sensitivity analyses were not conducted given the lack of studies that could be isolated and provide any meaningful or valuable additional synthesis.

Assessment of risk of bias for included studies were conducted according to criteria and tools outlined in the Cochrane Handbook for Systematic Reviews of Interventions²⁶  and the Cochrane Effective Practice and Organization of Care guidelines²⁷  for randomized trials, nonrandomized trials, controlled before-after and interrupted time series. C.O. and B.C. independently assessed risk of bias for each study. These scores were compared and a final score decision was made.

Specifically, randomized trials were assessed using the Cochrane Risk of Bias tool^26,28  across the following domains: randomization process, deviations from the intended interventions (blinding of personnel, participants, and outcome assessment), missing outcome data, outcome measurement, the selection of the reported result, and disclosure of funding and conflicts of interest. Studies were assigned an overall risk of bias judgement accordingly (low risk, high risk, or some concerns).

Quasi-experimental study designs were assessed using the Risk of Bias tool for Nonrandomized Studies of Interventions (ROBINS-I) tool.^26,29  Studies were assessed according to the following domains: bias because of confounding, bias in selection of study participants, bias in classification of interventions, bias because of deviations from intended interventions, bias because of missing data, bias in measurement of outcomes, and bias in selection of the reported result. Each study was assigned an overall risk of bias judgement (low, moderate, serious, and critical risk).

A summary of the intervention effect and a measure of quality for all outcomes were produced using the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) approach.³⁰  The GRADE approach considers 5 domains (study limitations, consistency of effect, imprecision, indirectness, and publication bias) to assess the quality of the body of evidence for each outcome. The evidence was downgraded from “high quality” by one level for serious (or by two levels for very serious) limitations, depending on assessments for risk of bias, indirectness of evidence, serious inconsistency, imprecision of effect estimates, or potential publication bias.

A database search produced 29 301 records and hand searching revealed another 168 records. After removal of 9132 duplicates, 20 337 records were screened at the title-abstract stage, which identified 680 records for full-text review. Of these, 51 studies (146 articles) met our inclusion criteria for nondigital interventions and 23 were included in the meta-analysis. We excluded 407 records at the full-text review stage for reasons including wrong intervention type, wrong study design, wrong comparator, wrong patient population or wrong outcomes (Supplemental File). Figure 2 shows the study breakdown across exclusion reasons.

Of the 51 included studies, 37 were RCTs,^31–67  4 were nonrandomized controlled trials,^68–71  and 10 were quasi-experimental studies.^72–81  Forty-four studies were conducted in HIC, including two studies that were multicenter (Australia, Belgium, Cyprus, Estonia, France, Germany, Greece, Hungary, Ireland, Italy, Netherlands, New Zealand, Norway, Switzerland, Sweden, Spain, United Kingdom, and United States), and 7 in LMIC (Brazil, Ecuador, Lebanon, China, Mexico, and Iran).

Most studies were conducted in school settings, with the exception of 7 studies that were conducted in the community^{33,38,49,66,69}  or participant’s homes.^37,40  Intervention duration ranged from 8 weeks to 4 years. Sixteen studies conducted interventions that ran for 18 months or longer (up to 7 years), whereas another 17 studies implemented interventions that spanned one school year (typically 8-12 months). The remaining studies (n = 17) implemented interventions for a duration of <6 months, and 1 study was unclear in its duration⁴⁸  (see Table 2 for characteristics of included studies).

All interventions employed a behavioral modification component including classroom education (ie, didactic, peer-to-peer, or exercise activities), family and community engagement and counseling (ie, newsletters and other media) to promote the benefits of physical activity, the risk of sedentary behavior, and excessive screen use. Some interventions also included other components, such as school and home environment modifications (ie, greater access to healthy foods in the cafeteria, improved physical activity spaces and equipment at school, and implementation of school wellness policies).^{32,43,49,51,55,57–60,80}  None of the studies disaggregated outcome data based on discrete behavioral and environmental components, providing a limited ability to analyze and understand the specific component effects on outcomes in this age group. The mean age of participants ranged from 5.0 to 18 years of age. Approximately one-half of the studies (n = 23) reported a mean age <10 years, 11 studies included both school-aged children and adolescents, and the remaining studies reporting mean ages between 11 and 18 years.

The majority of nondigital based RCTs (28 of 37) were high risk of bias, six had some concerns,^{31,32,43,52,66,67}  and three were of low risk.^48,51,53  Randomization was considered adequate in 24 trials. A common reason for downgrading study quality was concerns with risk of bias due to deviations from the intended interventions, involving allocation concealment blinding processes, and outcome assessment. Allocation concealment was unclear in most studies (26 of 37). Blinding of participants and personnel was considered poor or unclear, with only four trials blinding participants,^48,51–53  9 trials blinding personnel,^{31–33,46,48,51–54}  and 7 trials blinding outcome assessment.^{32,48,51–53,59,67}  Other reasons for downgrading study quality included attrition bias, disclosed funding and conflicts of interest. Attrition bias was considered high risk in 6 trials, with loss to follow-up ranging from 22%^44,56  to 32%.⁶⁰  The majority of studies disclosed funding, except for three,^35,38,47  whereas 9 studies did not declare their conflicts of interest.^{34,35,53,55,58,60,62,63,78} 

Of the nonrandomized controlled trials and QES, the majority of studies (9 of 14) were judged as having a moderate risk of bias because of poor adjustment of confounding variables, missing outcome data, subjective outcome assessments, and selected reported results. Three studies had an overall low risk of bias,^69,70,76  whereas two studies had serious risk.^74,78 

Eighteen studies were included in the RCT meta-analyses for nondigital based interventions,^{31,32,36–38,40, 43,47–53,56,57,60,62,68}  whereas 5 studies were included in QES meta-analyses.^{68,70,71,73,81} 

When compared with control groups, nondigital interventions probably results in a slight reduction of TV-specific screen time (minutes per day) (mean difference [MD], −12.46; 95% confidence interval [CI], −20.82 to −4.10; participants = 6097; studies = 6; I² = 59%; P = .004; moderate quality of evidence). Additionally, nondigital interventions may result in a reduction in screen time (all media types) (minutes per day) (MD −11.45; 95% CI, −19.18 to −3.73; participants = 7070; studies = 9; I²= 38%; P = .004; low quality of evidence) (Figs 3 and 4).

However, these interventions may make little to no difference on reducing screen time specific to computer gaming or video gaming (minutes per day) given the high uncertainty of the evidence (MD, −3.51; 95% CI, −9.02 to 2.01; participants = 5365; studies = 5; I² = 56%; P = .21; very low quality of evidence). With regards to sedentary behavior, nondigital interventions probably result in a slight reduction of sedentary time (minutes per day) as compared with controls (MD, −3.86; 95% CI, −6.30 to −1.41; participants = 8920; studies = 8; I² = 0%; P = .002; moderate quality of evidence) (Figs 5 and 6).

The effects of nondigital interventions on MVPA (minutes per day) as compared with control groups may make little to no difference on increasing MVPA (MD, −0.07; 95% CI, −1.83 to 1.69; participants = 5540; studies = 6; I² = 31%; P = .94; low quality of evidence). Two RCTs reported accelerometer data for weekdays and weekends (counts per minute or steps per day).^52,69  However, both studies found nonsignificant differences between intervention and control groups at follow-up, after adjustment (Fig 7).

With regards to QES, nondigital interventions may make little to no difference on reducing screen time of all media types (minutes per day) (MD, −26.76; 95% CI, −67.31 to 13.79; participants = 1984; studies = 3; I² = 97%; P = .20; very low quality of evidence) or sedentary behavior (minutes per day) (MD, −9.65; 95% CI, −41.05 to 21.75; participants = 1010; studies = 3; I² = 90%; P = .60; very low quality of evidence), given the high uncertainty of the evidence (Supplemental Information).

This review provides a comprehensive appraisal of 51 studies conducted in 24 countries, evaluating nondigital interventions aimed at minimizing screen time and sedentary behavior in healthy children and adolescents of normal BMI. This review analyzed over 16 000 children and adolescents and included 19 new trials, conducted in the last 5 years.^{17,31,36,39,40,44–46, 48–50,56,65–67,69–71,81}  Our review suggests that nondigital interventions indeed resulted in a small, but significant reduction in sedentary behavior. This review also indicates that nondigital interventions were most successful at reducing TV screen time compared with other forms of screen time, such as computer and gaming. Although there are many previous reviews that evaluated both single and multicomponent interventions in a variety of populations, including overweight and obese participants, it was important that this review focus solely on healthy children and adolescents, to provide greater insight into the possible effectiveness and potential of these interventions in public health prevention initiatives.

Previous systematic reviews^16–23  found similar results, whereby screen time from all media types was reduced by 10 to 20 minutes per day in intervention groups when compared with control groups. However, these meta-analyses varied in their inclusion criteria of participants (ie, exclusively overweight and obese populations) and in some cases, both preschool and adult populations. For example, van Grieken et al¹⁹  reported adolescent screen time use was reduced by a mean of −17.95 minutes per day (95% CI, −26.61 to −9.28) in a pooled analysis of 13 studies including overweight and obese adolescents. Likewise, Wahi et al²⁰  found in a pooled analysis of 9 studies in children (aged 3.9 to 11.7 years), intervention groups reduced screen time by a mean of −0.90 hours per week (95% CI, −3.47 to 1.66), however these results were not significant (P = .49). Albeit in the long-term, this small reduction does equate to some improvement in public health.

Interestingly, most interventions recruited young children, under the age of 13 years; perhaps as an effort to prevent excessive screen time and social media use in their later years, and to instill positive habits and long-term behavior change. Furthermore, a common observation of this review and previous systematic reviews is that a majority of the nondigital interventions targeting sedentary behavior and screen time are multicomponent and are often delivered through schools. Although this makes it difficult to evaluate the true effect of the screen time or sedentary behavior components, this observation suggests that addressing behavioral change in school-aged children and adolescents are most effective when used as a comprehensive and multifaceted strategy rather than a singular-component intervention. Similarly, school-based interventions alone may not be enough to counteract the trend of increasing screen time and sedentary behavior.

Although we included the largest number of studies to-date in a systematic review on screen time and sedentary behavior in SACA, most studies were conducted in HIC. It is possible that this finding is attributed to the stark inequality in digital connectivity in SACA living in LMIC. In the recent COVID-19 report, The International Telecommunication Union and United Nations Children’s Fund (UNICEF) highlight that 1.2 billion children and adolescents (aged 3-17 years) do not have internet access at home, and primarily reside in South Asia, West, East, Central, or Southern Africa.⁸³  Likewise, disparities exist between HIC and LMIC in mobile phone ownership, although this gap is closing among youth. Physical inactivity, however, remains consistent across world regions.¹¹  Thus, as the world becomes more connected, we expect preventive interventions, policies, and programs to become more prevalent.

Unfortunately, this review and meta-analysis present similar gaps in the evidence and methodology as a previous scoping exercise of existing systematic reviews conducted by the authors. An overwhelming majority of interventions were implemented in high-income settings and the heterogeneity of available data because of diverse interventions, a lack of standardization of screen time metrics, vague and diverse methodologies, and use of subjective tools such as self-reported screen use limit the findings of this review. Thus, generalizability of these findings proves difficult. Furthermore, some of the findings of this review should be interpreted with caution, considering the quality of the evidence. Despite a robust number of studies included, very few were rated as high-quality. Moreover, although many studies reported a randomized-controlled design, the majority of included RCTs lacked description and/or implementation of more robust methods. Consistent with existing literature, the risk of bias in some areas was notable across the majority of studies; the most common risks of bias among included studies were failure to blind participants and personnel, attrition bias, and selective reporting. This limits and introduces a level of uncertainty regarding the efficacy of these types of interventions.

With the rise of digital technologies, the proliferation of technology and connectivity have led to increased sedentary behaviors and poorer lifestyle behaviors in this age group. We know that increasingly poor lifestyle behaviors among youth and adolescents are no longer population health issues relegated to high-income settings. Thus, utilizing nondigital interventions to promote universal health, including physical activity and minimizing screen time are critical for long-term gains in human health and development. Future research should examine screen time as a proportion of sedentary time, as well as use standardized and objective measures of screen use and sedentary time. Policies and programs which reduce sedentary time and excessive screen use will be critical, especially in the post-COVID 19 reality.

BMI

body mass index

CI

confidence interval

COVID-19

coronavirus disease 2019

HIC

high-income countries

LMIC

low- and middle-income countries

MD

mean difference

MVPA

moderate-to-vigorous physical activity

SACA

school-aged children and adolescents