Figure 1. Forest plot of primary outcome, unadjusted difference in mean change in body mass index (BMI) (calculated as weight in kilograms divided by height in meters squared). Heterogeneity, τ2 = 0.02; χ25 = 8.05 (P = .15); I2 = 38%. Test for overall effect, Z = 1.0 (P = .32). CI indicates confidence interval.
Figure 2. Forest plot of secondary outcome, unadjusted difference in mean change in screen time measured in hours per week. Heterogeneity, τ2 = 8.45; χ28 = 23.69 (P = .003); I2 = 66%. Test for overall effect, Z = 0.69 (P = .49). CI indicates confidence interval.
Figure 3. Forest plot of subgroup analysis of secondary outcome, screen time, by age, measured in hours per week. CI indicates confidence interval. For age 6 years or younger, heterogeneity, τ2 = 0.00; χ21 = 0.95 (P = .33); I2 = 0%; and test for overall effect, Z = 2.07 (P = .04). For age older than 6 years, heterogeneity, τ2 = 9.61; χ26 = 20.58 (P = .002); I2 = 71%; and test for overall effect, Z = 0.13 (P = .90). For total heterogeneity, τ2 = 8.45; χ28 = 23.69 (P = .003); I2 = 66%; and test for total overall effect, Z = 0.69 (P = .49). Test for subgroup differences, χ21 = 2.16 (P = .14); I2 = 54%.
Wahi G, Parkin PC, Beyene J, Uleryk EM, Birken CS. Effectiveness of Interventions Aimed at Reducing Screen Time in ChildrenA Systematic Review and Meta-analysis of Randomized Controlled Trials. Arch Pediatr Adolesc Med. 2011;165(11):979-986. doi:10.1001/archpediatrics.2011.122
Author Affiliations: Division of General Pediatrics, Department of Pediatrics (Dr Wahi), and Program in Population Genomics, Department of Clinical Epidemiology & Biostatistics, Faculty of Health Sciences (Dr Beyene), McMaster University, Hamilton, Ontario, Canada; Division of Pediatric Medicine and the Pediatric Outcomes Research Team (Drs Parkin and Birken), and Hospital Library (Ms Uleryk), Hospital for Sick Children, Toronto, Ontario, Canada; Child Health Evaluative Sciences (Drs Parkin, Beyene, and Birken), Hospital for Sick Children Research Institute, Toronto; and Departments of Pediatrics and Health Policy, Management and Evaluation, University of Toronto Faculty of Medicine, Toronto (Drs Parkin and Birken).
Objective To evaluate the impact of interventions focused on reducing screen time.
Data Sources Medline, Embase, Cochrane Central Register of Controlled Trials, PsycINFO, ERIC, and CINAHL through April 21, 2011.
Study Selection Included studies were randomized controlled trials of children aged 18 years or younger with interventions that focused on reducing screen time.
Intervention Efforts to reduce screen time.
Main Outcome Measures The primary outcome was body mass index (BMI); the secondary outcome was screen time (hours per week).
Results A total of 1120 citations were screened, and 13 studies were included in the systematic review. Study samples ranged in age (3.9-11.7 years) and size (21-1295 participants). Interventions ranged in length (1-24 months) and recruitment location (5 in schools, 2 in medical clinics, 1 in a community center, and 5 from the community). For the primary outcome, the meta-analysis included 6 studies, and the difference in mean change in BMI in the intervention group compared with the control group was −0.10 (95% confidence interval [CI], −0.28 to 0.09) (P = .32). The secondary outcome included 9 studies, and the difference in mean change from baseline in the intervention group compared with the control group was −0.90 h/wk (95% CI, −3.47 to 1.66 h/wk) (P = .49). A subgroup analysis of preschool children showed a difference in mean change in screen time of −3.72 h/wk (95% CI, −7.23 to −0.20 h/wk) (P = .04).
Conclusions Our systematic review and meta-analysis did not demonstrate evidence of effectiveness of interventions aimed at reducing screen time in children for reducing BMI and screen time. However, interventions in the preschool age group hold promise.
Over the past 2 decades, the prevalence of overweight and obesity in children has steadily risen around the world.1- 3 It has been suggested that the epidemic of childhood obesity will lead to a phenomenon seen for the first time in recent history whereby children will have a shorter life expectancy than their parents.4 Unfortunately, successful interventions for the prevention and treatment of childhood obesity have been elusive and complex.5- 7
Principles for preventing and treating childhood obesity include reducing energy intake, increasing physical activity, and reducing sedentary behaviors.1 Sedentary behaviors are activities that do not involve physical exertion, including television and computer use, schoolwork, reading, and playing or listening to music.8 The most prevalent form of sedentary behavior is time spent in front of a screen, screen time, which includes television, videos and/or DVD, computer, and videogames. Screen time has steadily increased among youth, and approximately 1 in 4 children living in the United States watch an average of 4 hours of television per day.1,9 Television viewing has been associated with important health outcomes in children, including delayed language development, aggressive behavior, and cigarette smoking.10- 12 The focus of our study is to review the impact of interventions aimed at reducing screen time on change in body mass index (BMI, calculated as weight in kilograms divided by height in meters squared) in children.13
Interventions aimed at reducing screen time have been a focus of childhood obesity prevention and treatment for the past decade.8 A systematic review of interventions aimed at reducing sedentary behaviors (including screen time) published in 200714 noted great variation in types of interventions and concluded that these interventions were effective in reducing sedentary behaviors and improving indices of body composition. To our knowledge, our systematic review is unique because it focuses on a specific sedentary behavior that has been the focus of many randomized controlled trials (RCTs) in children, namely, screen time. Also, the previously published systematic review did not include a meta-analysis, and several additional RCTs have subsequently been published. A meta-analysis provides a point estimate of a treatment effect and helps to summarize evidence to guide clinicians, researchers, and policy makers.15 The primary objective of our systematic review and meta-analysis was to evaluate the impact on children of interventions aimed at reducing screen time on the outcome of BMI. The secondary objective was to evaluate the impact of interventions aimed at reducing screen time on screen time itself.
We searched 6 databases through April 21, 2011: OVID–Medline (from 1948), EMBASE (from 1980), Cochrane Central Register of Controlled Trials (from first quarter 2011), Psycinfo (from 1967), ERIC (from 1965), and EBSCOHost–CINAHL (from 1982). The strategy used terms customized for the database, including RCTs and derivations of the terms Television* OR Videogame* OR Computer* AND Overweight* OR Obesity* OR Physical activity* (eTables 1-6). To identify unpublished studies, databases of registered clinical trials (clinicaltrials.gov) and conference proceedings (PapersFirst and ProceedingsFirst) were hand searched. Reference lists of all retrieved manuscripts were searched for any relevant articles not previously identified. Two of us (G.W. and C.S.B.) independently reviewed articles for eligibility, and disagreements were resolved through discussion with a third reviewer (P.C.P.). The inclusion criteria were RCT (study design), participants aged 18 years or younger (population), and interventions that included a reduction of screen time (ie, television, videogames, and/or computer use). Two of us (G.W. and C.S.B.) screened all study titles and abstracts, and full-text articles were retrieved for any studies deemed relevant. The same 2 investigators (G.W. and C.S.B.) assessed articles for eligibility, and a κ coefficient was calculated to reflect the measure of agreement between the reviewers.16
Two of us (G.W. and C.S.B.) independently abstracted information, in duplicate, from eligible manuscripts using standardized data collection forms. Information was collected on characteristics of study populations, interventions, and outcomes. The primary outcome of interest was body composition, as measured by BMI, which is considered an appropriate measure of adiposity in children.17,18 The secondary outcome was amount of screen time, as measured by reported hours per week. For the purpose of our meta-analyses, we used the outcomes reported at the end of the study intervention period in the analyses. We contacted the authors of the trials included in the systematic review to obtain further data if the study did not report outcome measures in a manner that could be combined with other trials.
Pooled analyses were conducted with the measure of effect size as the unadjusted difference in mean (SEM) change in BMI and screen time between the intervention and control groups. For studies in which results were not reported as the difference in mean change, we used the published literature to identify appropriate correlation coefficients to calculate the variance needed for this calculation. For example, we assumed a correlation coefficient of 0.9 for BMI as used by previous authors.5,19 For screen time, we calculated the variance assuming a correlation coefficient and performed a sensitivity analysis using values of 0.4 and 0.7.20 A generic inverse method was used to account for the pooling of appropriately analyzed cluster randomized trials with trials that randomized at the level of the individual.20,21 The summary estimate for difference in mean change of BMI and screen time was calculated using a generic inverse method, random-effects meta-analyses. A random-effects model is conservative because it calculates the error term by taking into account between-study and within-study differences.20 Weights were assigned to the studies included in the meta-analyses based on the inverse variances of their effect estimates, as described by Higgins et al.20 For example, studies with less precise results, and therefore wider confidence intervals, are given less weight.20
Heterogeneity was defined by statistical measures of χ2 and I2. Heterogeneity is the variability among studies that is due to true differences between studies rather than sampling error.22 A priori, we defined P < .10 as significant heterogeneity; for I2, we defined 25% as low heterogeneity, 50% as moderate heterogeneity, and 75% as high heterogeneity.20,22 We established a priori hypotheses for heterogeneity among studies, including both population (age, baseline BMI) and intervention (setting, presence of cointerventions). If significant heterogeneity was observed, subgroup analyses and tests of interactions were planned. RevMan software, version 5.0.18 (The Cochrane Collaboration, Oxford, England) was used for all pooled analyses.
The Cochrane Collaboration checklist was used for assessing risk of bias.20 Components of the assessment included sequence generation, concealment of allocation, blinding of participants and outcome assessors, and percentage of participants lost to follow-up. Recruitment bias is a form of bias unique to cluster-randomized trials and therefore was considered in those trials.20 A summary score has been shown to be inaccurate when used among qualitative scales; therefore, it was not included in our risk of bias assessment.23,24
The Grading of Recommendations Assessments, Developments and Evaluation (GRADE)25 was used to provide a systematic approach to the assessment of the quality of evidence and strength of recommendations. It is a widely accepted tool endorsed by many organizations, including the World Health Organization and the Cochrane Collaboration.26 Criteria for consideration included assessment of methodology, precision and consistency of results, directness, and risk of publication bias.25
We identified 1120 potentially eligible articles from the 6 databases (PsycINFO, 125; Embase, 300; ERIC, 128; Medline, 184; Cochrane CENTRAL, 239; and CINAHL, 144). There were 226 duplicate articles, and 804 were excluded as not relevant after title and abstract review. Ninety full-text articles were retrieved and independently assessed for inclusion criteria. After full-text articles were assessed, 73 additional articles were excluded. Seventeen articles met eligibility criteria, but only 13 were unique trials, and 4 articles27- 30 were secondary publications reporting outcomes not previously addressed in the primary publications. None of these 4 articles identified outcomes of interest for the purposes of this review. The chance-adjusted, between-reviewers agreement (κ) on the application of study inclusion criteria to full-text articles was 0.92; almost perfect agreement.16
Descriptions of the 13 included trials8,12,21,31- 40 are listed in Table 1. The average age of participants in trials ranged from 3.9 to 11.7 years. The sample sizes of the included studies ranged from 21 to 1295 participants (median sample size, 90 participants), for a total of 3133 participants in the 13 trials. The duration of the interventions ranged from 1 to 24 months (median duration, 6 months). The settings varied: 5 trials recruited from schools8,12,21,31,32; 2 recruited from medical clinics33,34; 1 from a community center35; and 5 recruited from community settings.36- 40 There were 5 cluster-randomized trials and 8 parallel-group RCTs. All of the cluster-randomized trials accounted for this study design in the analysis.
We were unable to include several trials in the meta-analysis for the following reasons: 4 trials did not measure the outcome of interest (BMI or screen time)33,34,37,38; 3 reported results as the difference in the proportion of patients at particular BMI percentiles31,37,38 and hours of screen time31,38; 1 reported change in age- and sex-standardized BMI39; and 1 reported only adjusted analyses.21
Table 2 summarizes the assessment of methodologic quality (Cochrane Collaboration checklist for assessing risk of bias). Among 9 of the 13 trials, sequence generation was adequate. Allocation concealment was adequate in 6 trials. Blinding of participants or outcome assessors was not reported in 11 trials. Two trials lost greater than 20% of participants to follow-up. Two of the 5 cluster-randomized trials reported adequate recruitment methods.
Table 3 highlights the reported outcomes of all 13 studies included in the systematic review for both primary and secondary outcome.
Six studies were included in the pooled analysis for the primary outcome.8,12,32,35,36,40 The difference in mean change in BMI in the intervention group compared with the control group was −0.10 (95% confidence interval [CI], −0.28 to 0.09) (P = .32). (Figure 1). The heterogeneity observed in the pooled estimate was demonstrated by an I2 = 38% and P = .20. Since insignificant heterogeneity was observed, the subgroup analyses set a priori to explain heterogeneity were not explored.
Nine studies were included in the pooled analysis for the secondary outcome.8,12,32- 36,39,40 Six studies reported amount of screen time.12,33,35,36,39,40 Five studies reported amount of television viewed.8,32,34- 36 For the purpose of the pooled analysis, the results of the trials were combined, and difference in mean change in screen time is reported. The unit of measure for screen time was converted to hours per week if reported in other units. The results of the pooled difference in mean change from baseline in the intervention compared with the control group was −0.90 h/wk (95% CI, −3.47 to 1.66 h/wk) (P = .49) (Figure 2). The pooled estimate showed a high amount of statistical heterogeneity (I2 = 66%, P = .003). However, when the trials were grouped based on age, there was a statistically significant reduction in screen time among preschool children (younger than 6 years; n = 2 trials) in the intervention group compared with the control group (P = .04) (Figure 3).
Using the GRADE score,25 we found low-quality evidence for both primary and secondary outcomes, BMI and screen time viewing (Table 4). A rating of low-quality evidence implies that future research findings might impact the certainty of the results and change the estimate.25,41 The limitations that impact the quality assessment for this meta-analysis include uncertainty of allocation concealment, uncertainty of blinding of participants and outcome assessors, and number of participants lost to follow-up, as summarized in Table 2. The indirectness of results describes the qualitative differences in the sample populations and interventions across studies. The quality of evidence for the secondary outcome of screen time was further degraded by the inconsistency of results as demonstrated by a significant amount of heterogeneity observed in the pooled analysis.
This systematic review and meta-analysis of RCTs aimed at reducing screen time among children included 13 trials, a substantial number of RCTs in the field of child health research.42 Quantitative pooling of the data for the primary outcome showed no apparent effect of the interventions on reduction of BMI or reduction of screen time overall. However, we did observe a statistically significant reduction in screen time in the subgroup analysis of trials that focused on preschool children.
We applied rigorous approaches to the process of study selection, assessment of methodologic quality, data abstraction, and development of a priori hypotheses for the explanation of heterogeneity. Exploring the methodologic quality and quality of evidence using the GRADE criteria25 revealed deficiencies in reporting of participant and outcome assessor blinding. This is an area that led to the quality of evidence being downgraded to “low.” Although trials of behavioral interventions in the public health domain may have difficulty blinding participants, we believe that the blinding of assessors of outcome should be consistently reported.
Interventions aimed at reducing screen time had no overall effect on the reduction of BMI. Five of the 6 included trials did not have cointerventions, which suggests that interventions that solely deliver a message of reducing screen time may not be effective. We also note that the lack of an observed effect may be due to the short duration of the interventions; the median length of included trials was 7 months. Although BMI is only 1 measure of adiposity, with waist circumference or skinfold thickness as other potential measures, BMI was the most consistently reported measure of adiposity across trials. Higher BMI has also been associated with higher rates of mortality among adults.43 We therefore feel it is the most appropriate outcome to measure as an estimate of change in adiposity. However, there may be other important outcomes on the causal pathway between screen time and obesity that might be affected by these interventions, such as meals in front of the screen and exposure to food advertising. In addition, evaluating other important screen time–related outcomes such as language development may be an important next step in understanding the effects of reducing screen time.
We were unable to include 7 trials in the pooled analysis for BMI, and the estimate of effect might have been further strengthened if data from all available trials were included. Randomized controlled trials were excluded when unadjusted outcomes were unavailable. Also, different scales of measurement between studies, for example continuous (BMI) vs categorical (normal weight and obese) data, prohibited data from being combined across trials. Publication of unadjusted results along with adjusted results in RCTs, as well as consensus on clinically important outcomes for trials with similar foci, would prove useful for providing point estimates of effect in meta-analyses. As an example, the 2 largest studies excluded from the meta-analysis had differing results and, if included, might have altered the result of the meta-analyses. The study by Gortmaker et al31 (n = 1295) showed a significant reduction in television viewing and obesity (measured by a composite measure of BMI and triceps skin folds) in female participants compared with the study by Kipping et al21 (n = 679), in which no differences in adjusted BMI or screen time were found.
The overall pooled analysis did not show a statistically significant reduction in screen time. The lack of effect on screen time overall may be related to the challenges of measurement of screen time. The included trials used self- or parent reporting to measure screen time, and there is no published validated outcome measure for screen time. Although there is evidence that parent reporting of child screen time provides accurate estimates of television viewing times compared with videotaped observation (r = 0.70),44 parent reporting of screen time might have biased the measurement and provide an explanation for lack of observed effect on screen time reduction.
A significant amount of heterogeneity was observed in the quantitative pooling of screen time data. The a priori study characteristics of the populations that were considered to explain heterogeneity included age. Interventions directed to preschool children may be more effective because parents have more control over lifestyle behaviors at this age. The other a priori study characteristics, including setting of intervention, baseline body composition of sample, and presence of cointerventions, were assessed and found not to be significant (data available on request). Heterogeneity in the screen time data, along with differing results for secondary analysis for age subgroups, makes interpretation of this outcome challenging.
Only 3 trials commented on potential adverse effects of the intervention,29- 31 including issues related to extreme dieting and children's dissatisfaction with body shape. The literature suggests that interventions aimed at reducing weight among adolescents have not been associated with an increased prevalence of eating disorders.34,38 However, including a measure of harm is important for future trials.
Existing tools for evaluating the quality of evidence in RCTs, such as the GRADE system,25 were not specifically designed for RCTs that focus on trials of behavioral interventions. We believe, however, that it is important to apply rigorous methods for evaluating RCTs that include behavioral interventions, and that GRADE is a widely adopted existing system for evaluating quality of evidence.45 In trials of behavioral interventions, where blinding of subjects may not be possible, and measurement of outcomes may rely on self-reporting, it is of utmost importance to use and report all available methods of safeguarding against bias, such as blinding of outcome assessors.
In conclusion, there are a substantial number of RCTs investigating interventions that aim to reduce screen time in children. Reducing screen time is touted as one of the effective interventions to reduce BMI and prevent obesity; however, our meta-analysis did not demonstrate evidence of effectiveness of interventions aimed at reducing screen time in children for reducing BMI and screen time. Although assessing the quality of evidence yielded a result of low quality for both outcomes, this does not dismiss the results as unimportant. This highlights the need for further rigorous investigation and confirms the importance of reporting methods. The interventions in these trials for the most part involved multiple sessions over a prolonged time period, integrated into school curriculum, clinic settings, or home. We propose the evaluation of pragmatic interventions that could feasibly be implemented in fewer sessions, over shorter periods of time, with longer follow-up, and focused on key age groups where behavior change may be sustainable, such as the preschool age group. Given the prevalence of obesity in childhood, and the long-term complications associated with obesity over the life course, testing and implementing effective interventions early in life, including those that focus on screen time, should be a priority for researchers, practitioners, and policy makers.
Correspondence: Catherine S. Birken, MD, MSc, FRCPC, Child Health Evaluative Sciences, Hospital for Sick Children Research Institute, 555 University Ave, Toronto, ON M5G 1X8, Canada (firstname.lastname@example.org).
Accepted for Publication: April 28, 2011.
Published Online: July 4, 2011. doi:10.1001/archpediatrics.2011.122
Author Contributions: Drs Birken and Wahi had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Wahi, Parkin, Uleryk, and Birken. Acquisition of data: Wahi and Birken. Analysis and interpretation of data: Wahi, Parkin, Beyene, and Birken. Drafting of the manuscript: Wahi and Birken. Critical revision of the manuscript for important intellectual content: Wahi, Parkin, Beyene, Uleryk, and Birken. Statistical analysis: Wahi and Beyene. Obtained funding: Parkin and Birken. Administrative, technical, and material support: Parkin. Study supervision: Parkin and Birken. Literature searching: Wahi and Uleryk.
Financial Disclosure: None reported.