1.1. METHODS

We used the 27‐item Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) statement as a basis to report our systematic review. The systematic review protocol was registered at PROSPERO: CRD42015025036 (available from http://www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42015025036).

1.2. Study eligibility

A study was eligible for inclusion if it met the following criteria: (a) randomized controlled trial (RCT) or longitudinal cohort studies of healthy children and adolescents 2 to 18 years of age and exposures reported separately for boys and girls at each age. For RCTs, only data from control subjects were considered if they did not receive any physical activity interventions; (b) MVPA was measured in minutes per day using accelerometers at least twice with at follow‐up of 1 year between measures; (c) the outcome of interest was annual percentage change in MVPA or absolute annual change in MVPA; and (d) published in English, due to resource constraints. These criteria were selected to minimize any seasonal influences on MVPA and because previous longitudinal studies have shown that the observed MVPA change from childhood to adolescence is generally not linear. Data collected at more than two time points were included in the systematic review and synthesis if the gap between each time point was 1 year. Studies were excluded if the participants had a disease or disability expected to impact MVPA. Studies where participants were part of physical activity interventions were excluded. Studies that used pedometers or mobile phone devices instead of accelerometers were not considered, since MVPA cannot be estimated accurately. For the meta‐analysis, in addition to the above criteria, studies with exactly 1 year of follow‐up were eligible to be able to quantify the annual MVPA change at each age.

1.3. Study selection and search strategy

Up to 31 December 2018, eligible studies were identified in the following databases: MEDLINE, Embase, PsycINFO, Scopus, PsycLit, Science Citation Index, SciTech Collection, Biological Sciences, Physical Education Index, and SPORTDiscus. The search strategy included medical subject headings and free‐text terms for the domains: study design, measurement tool, population and exposure. No publication date filters were applied. An example search strategy for the MEDLINE (OVID) platform is provided (Table S1).

Titles and abstracts were initially screened for eligibility by A. F.. The full‐text versions of potentially eligible studies were obtained and screened independently by two reviewers (A. F. and J. J. R.). Disagreements regarding study eligibility were resolved through discussion and/or additional information obtained from the study authors. For studies that did not report MVPA by age or for both sexes separately, corresponding authors were contacted by email to obtain this information. If no response was received after at least 3 reminders and 2 months or more, the study was deemed ineligible. Of the authors contacted, 32 of 42 eligible studies replied and shared the requested information. The response rate from corresponding authors was 76.2%. Of those authors who did not respond, 3 had changed institutions so no longer had access to the data and 7 were unable to respond due to their busy schedule.

1.4. Data extraction

An Excel spreadsheet was prepared to facilitate data extraction based on data items identified (Tables S2A and S2B). To reduce errors, duplicate data extraction was performed, in which AF extracted the data from all included studies on two occasions separated by ~3 months. All data extracted were checked and reviewed by two independent reviewers (A. M. G. and J. J. R.) for consistency and accuracy prior to data analysis.

From each paper, the following information was extracted and coded: (1) author and year of publication, (2) site/city/country where the study was performed, (3) year of baseline data collection, (4) length of follow‐up, (5) year of data collection at each follow‐up time point, (6) number of participants at baseline, (7) mean age of the participants at baseline, (8) number of participants at follow‐up, and (9) MVPA expressed as mean or median at baseline and at follow‐up time points.

In addition, the methods used for physical activity data collection using accelerometers were collected, based on Tanaka et al21: (a) brand of accelerometer used, (b) number of days accelerometer was worn, (c) minimum duration of wear time per day to be included in study, (d) minimum number of valid days of accelerometer data to be included in study, (e) whether accelerometer was worn on weekends, (f) the length of the recording interval duration used by the accelerometer to integrate data, and (g) cutoff points used to define MVPA.

1.5. Extraction of MVPA data

Often MVPA data for boys and girls were expressed as minute day⁻¹ and was directly available from the studies at each age. In other studies, MVPA data were provided for population subgroups (e.g., by body mass index categories) or by weekend and weekday. In such cases, corresponding authors were contacted to provide MVPA data for each age and by sex. When data were not directly available, it was pooled using formulae and methods recommended in the Cochrane handbook for extracting data for continuous outcomes, as long as the sample size in each group and means were available.22 When MVPA data were subdivided into weekday and weekends, a 5:2 ratio weight was used to compute aggregate MVPA minute day⁻¹ for the group. Two of the 52 included studies presented findings as figures with MVPA on the y‐axis. Here data were extracted using Plot digitizer software, which has a reliability of up to 93% for a precision of 0.1 unit.23

1.6. Risk of bias assessment

The Cochrane Collaboration tool for assessing risk of bias in longitudinal observational studies or RCTs was used that includes five domains of bias (selection, performance, attrition, reporting, and other)21, 24 (Table S3). Author A. F. paired with A. M., A. H., or X. J. to appraise the included studies for the presence of risk of bias and judged the level of bias for each of the five domains (high bias, low bias or unclear bias). Any disagreement in a domain was resolved by discussion to reach a consensus rating.

1.7. Summary measures

The primary summary measure for this systematic review was the annual relative change in participant's accelerometer‐assessed MVPA in minute day⁻¹. Participants' MVPA at baseline and at subsequent follow‐up of one or more years was recorded. A variety of accelerometry cut‐off points have been used to define MVPA25; we did not exclude studies based on the particular cut‐off point used to define MVPA. To synthesize data, we calculated relative change in MVPA as the percentage change from the baseline value (calculated as [follow‐up MVPA minus baseline MVPA]/baseline MVPA × 100). When there was a follow‐up duration of more than 1 year, relative change was divided by the number of follow‐up years, as described previously.18

2.2. Study characteristics

Table 1 shows the characteristics of all studies included in the systematic review. Studies with girls were overrepresented and a total of 13,234 girls and 8,857 boys provided data for the subsequent analysis. The data were mostly from children living in high‐income regions (99.1% of the study participants).

In at least half of the eligible studies, participating children and adolescents wore the accelerometer for at least 7 days. Almost all studies included weekend days. The definition of MVPA thresholds varied across studies, but the majority of studies used the Actigraph and Evenson's cut‐off of 2,296 counts per minute to define MVPA. In 29.4% of included studies, the Epoch length was 60s. For the accelerometer data to be considered valid, almost all cohorts required 8 hours or more of data each day and at least 3 days of valid accelerometer data (Table 2).

2.3. Risk of bias within studies

In the risk of bias assessment, around one fifth of the included studies were judged as high risk of bias, because of lack of random selection and convenient sampling in the population recruited (Table S3). Performance bias was rated as unclear (24%) or high (20%) when studies used 2 days or 1 day of valid wear time (>8 h), respectively, to estimate MVPA. The highest bias was reported in the attrition domain (55%). Around 33% confirmed that characteristics of the participants with missing and without missing MVPA data were similar at baseline. There were no studies with selective reporting bias or other sources of bias due to the strict inclusion criteria of our study.

2.4. Mean MVPA change per year

The data extracted on average time spent in MVPA at baseline and at each consecutive follow‐up measurement is presented in Figure 2. Mean MVPA is often below the 60 minute day⁻¹ recommendation, and it declines across most of the paediatric age range. There was considerable variation between studies in the magnitude of the decline, in part related to methodological differences and in part due to real differences between samples and settings: mean duration of annual change in daily MVPA ranged from −63.0 to 20.7 minutes with an average of −3.4 minutes, 95% CI (−4.6 to −2.2) and a SD of ±8.8 minutes.

Open in a separate window

Figure 2

Trajectories of moderate‐to‐vigorous‐intensity physical activity among boys and girls from age 3 to age 18 years

2.5. Mean percentage MVPA change per year

Based on MVPA data from all 52 included studies (Figure 2), the percentage MVPA change per year was determined and is presented in Table 3. Girls showed greater % decline per year in MVPA (−5.3%; 95% CI, −7.6 to −3.0) than boys (−3.5%, 95% CI, −6.0 to −0.9). Although the percentage MVPA change per year was always negative from age 6 among girls and from age 8 among boys, significant declines were observed in girls at ages 6 years (−6.4%, 95% CI, −10.0 to −2.7), 9 years (−10.2%, 95% CI, −14.2 to −6.3), 10 years (−7.9%, 95% CI, −15.0 to −0.7), and 13 years (−8.5%, 95% CI, −12.8 to −4.3). As for the boys, percentage MVPA change per year was statistically significant at age 9 years (−7.8%, 95% CI, −11.2 to −4.4). Combining data of boys and girls, the annual declines in MVPA peaked twice across the entire age range; once at age 9 years (−9.0%, 95% CI, −11.5 to −6.6) and then at 13 years (−8.4%, 95% CI, −13.2 to −3.7). The percentage MVPA change per year was higher and statistically significant on weekends (−5.3%, 95% CI, −9.5 to −1.0) compared with weekdays (−3.1%, 95% CI, −7.5 to 1.3) based on available weekday‐weekend data from 13 of 52 studies.

2.6. Meta‐analysis of annual change in MVPA minutes (paired mean difference)

A subset of 31 studies with exactly 1 year of follow‐up data were included in the REs meta‐analysis to quantify the annual change in MVPA at each age (Figure 3). The bubbles represent SMD of annual MVPA change. Larger bubbles show that studies with more power and bubbles below zero indicate declines. While there was clear evidence of declining MVPA across childhood and adolescence from an early age, the decline was generally nonlinear in both boys and girls (Figure 3).

Open in a separate window

Figure 3

Scatter plot of standardized mean annual difference in moderate‐to‐vigorous‐intensity physical activity against age at baseline in boys and girls

Legend: The bubbles represent the estimates of effect sizes using random effects derived from meta‐analysis with covariates, age at baseline, and gender. Size of bubbles are proportionate to the level of precision of each effect in the study; the projected 95% CI shaded region is a loess function curves for boys and girls respectively

The forest plots of the RE model for boys (Figure 4A) and girls (Figure 4B) provides both actual changes in minutes of MVPA and SMD separately (n=31 studies). Combined annual change in MVPA across all ages, expressed as the paired SMD, was −0.03 (95% CI, −0.09 to 0.04) among boys and −0.10 (95% CI, −0.15 to −0.04) among girls. Age‐specific subgroup analysis indicated a statistically significant decline in MVPA among boys age 9 (SMD: −0.20, 95% CI, −0.31 to −0.09). Among girls, the meta‐analysis (Figure 4B) revealed multiple ages at which MVPA declined significantly starting as early as ages 6, 9, 10 and again at age 13 corresponding to reduction of MVPA SMDs: −0.21, 95% CI (−0.39 to −0.02); −0.29, 95% CI (−0.41 to −0.18); −0.22, 95% CI (−0.36 to −0.08), and −0.16, 95% CI (−0.26 to −0.06) respectively.

Open in a separate window

Figure 4

A, Random effects meta‐analysis of the annual mean difference in moderate‐to‐vigorous physical activity (min day⁻¹) based on age level among boys. B, Random effects meta‐analysis of the annual mean difference in moderate‐to‐vigorous physical activity (min day⁻¹) based on age level among girls

In terms of heterogeneity, meta‐analysis provided a heterogeneity statistics I ² of 86.0% for boys and I ² of 84.2% for girls. This indicated substantial inconsistency across studies and reflective of different differences found at each age. Funnel plots in Figure S1 show acceptable levels of symmetry suggesting the absence of publication bias. This was confirmed by Egger's tests that provided nonsignificant values for effect sizes representing studies from boys (P=.921) and girls (P=.455).

3.1. Study and evidence‐base strengths and limitations

The main strength of this study is that despite the strict inclusion criteria to only include accelerometer‐measured and longitudinal MVPA data, a relatively large number of studies were identified and included. We contacted authors to provide the data in the format that answered our specific research question with a response rate of 76.2%. This adds credibility to the results. The study also represents an advance on the previous systematic review of the topic18 by extending the age range across childhood and adolescence, including objectively measured change data only, and the focus on MVPA, the most health enhancing intensity of physical activity, and the component of physical activity which is of most importance to the development and maintenance of paediatric obesity.

The present study however had some limitations. The literature search was limited to studies written in English and from countries where accelerometers were readily available. Therefore, the review primarily represents evidence largely from high‐income countries. Secondly, while accelerometers provide valid and acceptable measures of MVPA, they can underestimate MVPA in certain situations e.g. cycling or swimming.84 A weakness often not considered was some of the included studies in the present review represented control group data from RCTs, often with different research objectives to those of the present study. A subgroup analysis was performed to compare if the estimates obtained from RCTs were similar to effects obtained from other study designs and whether employing a different epoch length during accelerometer assessment influenced the parameters estimates. The results across study designs were similar, but epoch length categories revealed different estimates of the annual MVPA decline. Shorter epochs (<15s) were used in studies involving younger children, and this could explain the smaller estimates of change in MVPA per year when compared with longer epochs (60s) used in the older age groups. (Table S6).

There was attrition due to invalid accelerometer data at follow‐up in most studies. Among the eligible studies 55% did not report whether characteristics of the participants lost to follow‐up were similar to those retained. This was not ideal but may have had relatively little impact on our results because in most cases, the numbers of individuals lost to follow‐up were minimal and much of the invalid accelerometer data can possibly be assumed as missing at random. We specifically also asked the corresponding authors to provide valid MVPA change data at every wave of data collection, instead of data for the largest sample at each wave. Moreover, by performing the QE model that gives less weight to studies with high risk of bias with respect to selection, performance, detection, and attrition domain (studies with lower overall quality score), observed effect sizes did not change significantly. (Table S5). However, as per QE model, age 10 boys also showed significant annual decline in MVPA.

We also acknowledge the inability to gather data on explanatory or confounding variables, which may influence changes in MVPA across childhood and adolescence such as adiposity, pubertal maturation, socio‐economic status and other potential predictors like season, weekday/weekend. In most cases, the explanatory and confounding factors were not readily available from the published studies, and if we had requested this information, the response rate from the corresponding authors could have been reduced due to the complexity of the request. By focusing on key variables associated with change (age, sex and MVPA), we were able to address their possible influence on age‐related declines in MVPA, but the influence of other variables remains to be investigated in future research.

Vigorous intensity physical activity (VPA), as distinct from MVPA, may be particularly important in the development and maintenance of child and adolescent obesity,^91,96 and VPA also seems to decline with age from early‐mid childhood.85 However, most of the studies included in this review did not provide VPA separately and so age‐related declines in VPA could not be considered. Alternative approaches to understanding age‐related declines in MVPA or VPA after controlling for seasonality, weekday/weekend, or other factors might be possible utilizing the data from the ICAD or Biobanks.

Our review includes studies across a wide age range from 2 to 18 years, but there were fewer studies of changes in MVPA from early childhood than later in childhood, and far fewer studies of changes in MVPA in late adolescence, above 15 years old (Figure 2). This observation highlights the need for more longitudinal studies in these age groups. Relatively low numbers of studies in these age groups could be due to the common belief among researchers that MVPA changes occur only around adolescence and the erroneous belief that MVPA in early childhood and/or late adolescence is unimportant.86 Research experts in the field of physical activity must overcome the common challenges of working with younger children and older adolescents such as recruitment, compliance with accelerometry and participant retention in order to address these research gaps.

The accelerometry data reduction decisions made in individual studies, such as thresholds to define MVPA, varied by study. Our decision to compute the percentage decline in MVPA per year as done in a previous review18 allowed for a comparison of change in time across studies despite these differences in methodology.

We would like to thank all corresponding authors of the included studies who gave their valuable time and effort to provide requested data for systematic review and meta‐analysis.