The impact of long-term school-based physical activity interventions on body mass index of primary school children – a meta-analysis of randomized controlled trials

Keywords in English, including physical activity, physical education, exercise or active break, body mass index or BMI or obesity, and school children were used individually or in combination to retrieve related articles published between January 1990 and March 2015 from major databases (PubMed, Web of Science, the Cochrane Library and Google Scholar). The retrieve protocol was like ((((physical activity [MeSH Major Topic]) OR physical education [MeSH Major Topic]) OR exercise [MeSH Major Topic]) AND body mass index [MeSH Major Topic]). The bibliographies of relevant meta-analysis and systematic reviews were also manually investigated to retrieve additional relevant original articles that met the inclusion criteria indicated as follow. The preliminary retrieved references were carefully examined to avoid duplications and omissions.

The analysis was limited to the studies published in English and with human subjects. The following criteria were used for paper selection: 1) primary school-aged children (6 to 12 year-old) regardless of their weight status, 2) RCT designed, 3) intervention conducted in the school setting, 4) PA intervention duration ≥12 months, 5) available data on the mean and standard deviation (SD) of BMI or the BMI changes from baseline to the final follow-up in both intervention and control groups.

The initial screening of titles and abstracts was performed by one investigator with a randomly selected 10 % of the sample checked by a second investigator. Data were independently extracted by one investigator using a self-made data-collection form developed based on the inclusion criteria and was independently checked by a second investigator. The following information of studies including basic information (authors, publishing year, and study area), study design and sample size, grade and gender of students, intervention type and duration, and PA measurement were extracted from the eligible studies. Another two investigators conducted internal and overall validity assessments independently using the Jadad Scale scoring algorithm [35]. The sum of the scores of all five items forms the final Jadad Scale score (JSS) for each study, ranged from 0 to 5. Studies with JSS equal to or greater than 3 were considered as high quality studies. If the score for the same item differed between the two investigators, a third investigator had to reassess the study and at last come to an agreement on all the items of the included studies. The JSS has been used in previous RCT studies [36].

The eligible studies were originated in European, Asian, and African countries and in the United States. The study’s design was RCT or cluster RCT with randomization at the school level. The intervention type was stratified as physical activity only (PA) and physical activity plus nutrition (PA&N). Intervention duration was presented in months as a dichotomous variable (12 ~ 24 months and >24 months). PA level was electronically monitored and/or collected by using questionnaires, therefore the measurement was classified as electronic instrument only (I), questionnaire only (Q) or both (I & Q). PA contents detailed the frequency and duration of weekly PA at school. Weekly PA intervention time was calculated as the original physical education time plus the additional PA intervention time, presented as minutes per week (min/week) and was classified as ≤100 min/week and >100 min/week. The final BMI was measured immediately after the completion of the intervention. The Chi-square test was used to compare the difference of JSS between RCT studies and cluster RCT studies.

The primary analysis was focused on the long-term effect of PA interventions on the BMI of primary school children. The mean change in BMI (^△BMI) for both intervention groups (IG) and control groups (CG) was calculated as BMI at final follow up minus BMI at baseline. For each study, the effect size was the difference in ^△BMI between control and intervention groups (^△BMI_IG-^△BMI_CG) and then was expressed as the standardized mean difference (SMD) with a 90 % confidence interval (CI). A crude forest plot was conducted to graph the effect size of studies on △BMI using random effect model. In the forest plot, the SMD of each study was displayed, along with a combined estimated variance of the overall effect. I ² was used to assess the heterogeneity of SMD across the included studies, while Begg’s funnel plotting was adopted to visualize the publication bias. Thereafter, one-way sensitivity analysis was performed to assess the robustness of the results by removing each study individually and assessing the I ² impact on the summary estimate. After removing the studies that highly impact the summary estimate, stratified analyses were used to detect the contribution of study design, origination area, duration, intervention type, PA measurement method, and weekly PA intervention time on the origin of heterogeneity.

All statistical analyses were performed using STATA version 12.0.

There were 11,175 potentially relevant articles identified according to the search protocol, and another 156 articles were manually retrieved from the reference lists of relevant meta-analysis and systematic reviews. With duplicates of records searched in both search protocol and manually retrieved, there was 11,158 left. Of those, 10,751 studies were excluded for being non-obesity prevention studies (10,203 records), non-English (225 records), and without available outcome measures (323 records). An additional 389 studies were excluded due to the following reasons: non-school-based studies (76 records), not from primary school (95 records), non-RCT (87 records), no BMI outcome available (46 records), and intervention duration <12 months (85 records). As a result, 18 eligible studies were retained for the meta-analysis. The flow chart is shown in Fig. 1.

The 18 studies were published from 1999 to 2014, with a total of 22,381 children included. Two of the included articles were from the same research project (one presented the medium term effect and one showed the final term effect) [37, 38]. Nine studies were conducted in Europe [33, 39‐46], 5 studies were from the United States [34, 47‐50], 4 studies were conducted in Africa [37, 38] and Asia [51, 52] (2 each). Among the 18 studies, 7 studies (38.9 %) included more than 1000 children; 3 (16.7 %) studies have only recruited roughly 100 children, the rest ranged from 346 to 646 children. Most of the included studies targeted both genders with a gender ratio (boy to girl) that ranged from 0.79 to 1.18; while only one targeted boys. Five interventions (27.8 %) lasted less than two years, and 8 of them (44.2 %) continued for at least three years. The majority of the studies (14, 77.8 %) were cluster RCT, and the reminding 4 (22.2 %) studies were RCTs. Five studies (27.8 %) only included PA intervention and the rest 13 (72.2 %) involved both PA and nutrition components. For the measurement method of PA, 11 studies (61.1 %) only used questionnaires, 5 (27.8 %) adopted electronic instruments, and 2 study (11.1 %) used both questionnaires and electronic instruments. There were 8 studies that indicated PA intensity indicators. Both Mackelvie et al. [39] and Ahamed et al. [43] used the scores from Physical Activity Questionnaires (PAQ-C) for Children to assess PA intensity, while only Ahamed et al. [43] stressed the PAQ-C scores for both baseline and terminal line. The other two studies from Manios et al. [33] and Angelopoulos et al. [41] used the same PA intervention and same questionnaire to assess the moderate-to-vigorous PA (MVPA). While the former one shown significant effect of MVPA intervention on reducing BMI increment, the latter one shown nonsignificant results. There were 4 studies that used electronic instruments to assess PA intensity. The instruments differed between these studies (Caballero et al. [47] used the Tritrac R3D, Hemokinetics, Iowa City; Donnelly et al. [48] used Actigraph, 7163, Pensacola, FL; Dzewaltowski et al. [50] used Actigrap GT1M accelerometers, Shalimar, FL; Kriemler et al. [40] used MTI/CSA 7164, Actigraph, Shalimar, FL). There were 12 studies conducted with the PA intervention time ≤100 min/week and 6 studies with PA intervention time ranged from 120–450 min/week. Details can be seen in Table 1.

Two investigators individually assessed the quality of 18 studies using the Jadad Scale and yielded almost identical JSSs, i.e., only 6 items were scoring differently out of 126 items. A third investigator reexamined the eligible studies on each item especially the 6 inconsistent items, and then the final JSS were given to each study. Five studies received scores of 3, 12 studies attained scores of 2, and the remaining one received 1 score. No significant difference in quality was found between RCTs and cluster RCTs by the Chi-square test. Results were presented in Table 2.

Children’s ^△BMI was significantly different (p < 0.05) between the PA intervention group and the control group (SMD: -2.23 kg/m^2, 90 % CI: -2.92, -1.56) (Fig. 2). High heterogeneity (I ²  = 99.8 %) was identified cross the studies. The Begg’s Funnel plot was asymmetric with some outliers (p < 0.05) (Fig. 3).

The meta-influence plot showed 7 studies (from Jiang et al. [52], Tarro et al. [46], Angelopoulos et al. [41], Manios et al. [33], Llargues et al. [38], Lohman et al. [49], and Magnusson et al. [42]) have significant impacts on the overall effect. When these studies were removed from the analysis, the difference of ^△BMI between the intervention group and the control group increased from -2.23 kg/m² to -2.00 kg/m² (90 % CI: -0.30, -0.09). Simultaneously, the heterogeneity among the remaining studies was reduced (I ²  = 90.5 %).

With the seven studies excluded, stratified analysis showed that intervention type (PA and PA&N) did affect the summary estimate of ^△BMI. The difference of △BMI for PA group and control group was -0.13 (N = 5, 90 % CI: -0.29, 0.04; I ²  = 89.8 %); while ^△BMI between PA&N group and control group was -0.26 kg/m² (N = 6, 90 % CI: -0.30, -0.09; I ²  = 92.1 %). The SMD for studies with the weekly PA time ranged from 36–100 min was -0.22 (90 % CI: -0.38, -0.05, I ² = 94.0 %), and for weekly PA interventions >100 min, the SMD was -0.18 (90 % CI: -0.35, 0.00; I ² = 87.1 %). Overall intervention duration of studies also impacted the BMI change. The difference of ^△BMI was significantly different between the interventions that lasted for 12–24 months groups and control groups (SMD: -0.20 kg/m², 90 % CI: -0.39, -0.02); and the difference of ^△BMI in intervention lasted for over two years and the control group was -0.19 kg/m² (90 % CI: -0.35, -0.03).