Osteoporosis and associated fractures are a major health concern in Western industrialized nations. Exercise during growth is suggested to oppose the involutional bone loss later in life by increasing peak bone mass. The primary aim of the present meta-analysis was to provide a robust estimate of the effect of weight-bearing activities (WBAs) on bone mineral content (BMC) and areal bone mineral density (aBMD), during childhood and adolescence. To locate relevant studies up to June 2012, computerized searches of multiple bibliographic databases and hand searches of key journals and reference lists were performed. Results were extracted by two independent reviewers. The quality of the included trials was assessed via the Physiotherapy Evidence Database (PEDro) score. The study group effect was defined as the difference between the standardized mean change for the treatment and control groups divided by the pooled pretest SD. From 109 potentially relevant studies, only 27 met the inclusion criteria. The analyzed training programs were capable of significantly increasing BMC and aBMD during growth. However, the weighted overall effect sizes (ESs) for changes in BMC (ES 0.17; 95% confidence interval [CI], 0.05–0.29; p < 0.05) and aBMD (ES 0.26; 95% CI, 0.02–0.49) were small. Stepwise backward regression revealed that more than one-third of the observed variance (r2 = 0.35) between subgroups of the BMC dataset could be explained by differences in the amount of habitual calcium intake per day (beta 0.54, p < 0.01) and the maturational stage (beta −0.28, p < 0.01) at baseline. No significant moderators were identified for aBMD, possibly due to the small number of trials investigating WBAs on aBMD. The results of this meta-analysis conclude that WBAs alongside high calcium intake provide a practical, relevant method to significantly improve BMC in prepubertal children, justifying the application of this exercise form as an osteoporosis prophylaxis in this stage of maturity. © 2014 American Society for Bone and Mineral Research.
Despite the lack of direct evidence, it is assumed that optimizing peak bone mass in childhood and adolescence is beneficial in the prevention of osteoporosis later in life. That is, the maximal bone mineral content (BMC) attained in these years may compensate for age-associated bone loss and therefore may reduce the risk of fractures and bone fragility associated with osteoporosis. Furthermore, maximizing age-specific bone mass may also prevent fractures during growth, especially those of the upper limb.[1, 2] In order to minimize the risk of fractures, strategies to improve bone health should be adopted very early in life.
In addition to calcium intake and hormonal factors, mechanical loading of the human skeleton seems to be a key factor to increase bone mineral accrual in children and adolescents. In this context, it seems reasonable that weight-bearing activities (WBAs), defined as force-generating exercises placing higher mechanical stress on the human skeleton than daily living, such as jump-training interventions or resistance-training programs, are beneficial in the improvement of bone health. This assumption is supported by the conclusions of numerous reviews and position statements.[5-7]
As claimed by Heaney, BMC, not areal bone mineral density (aBMD) should be the parameter of choice when assessing bone mineral accrual of children and adolescents in growth studies. This claim is based on the fact that aBMD factors out most of the bone accumulation associated with growth-dependent changes in bone size. However, not all pediatric bone researchers agree with Heaney and consider aBMD to be a more consistent predictor of fractures in children than BMC. To our knowledge there has been no previous meta-analysis providing a robust estimation of the effects of WBAs on BMC and aBMD in children and adolescents. Narrative reviews, by their very nature, lack objectivity. Therefore, the primary aim of the present work is to apply objective statistical procedures to calculate standard mean effect sizes (ESs) by combining data from multiple studies. Furthermore, training program variables and subject characteristics were analyzed in order to reveal a closer insight into the relationship of exercise stimulus, the individual response matrix, and the observable adaptations of bone tissue.
Materials and Methods
The study methods of the present analysis and the inclusion criteria were defined in advance and documented in the study protocol (Supporting Information Document 1). The guidelines of the Preferred Reporting for Systematic Reviews and Meta-Analyses (PRISMA) statement[9, 10] were followed and a PRISMA checklist is provided (Supporting Information Document 2).
An in-depth systematic computerized search of several databases led to the accumulation of relevant data for this meta-analysis. Dating back to their respective inceptions, PubMed (1966), Sport Discus (1975), ERIC (1966), Web of science (1945), and EBMZ (1917) databases were searched until the end of our data collection phase in June 2012. Further, hand searches of key journals and reference lists from key studies were also performed. German studies were included in the literature search to reduce the risk of oversampling of significant studies that are often published in the English language. The following subject headings, key words and text words, in English and German, respectively, were included: children, adolescents, youth, athletes; resistance, strength, weight, power, jump, plyometric, weight-bearing, impact, load; training, exercise, program, activity, physical; bone, mineral, content, mass, density. Direct contact to authors was initiated for instances in which incomplete information was provided or clarification was needed.
Study selection, data extraction, and quality assessment
To examine the effects of weight-bearing activities on BMC and aBMD in children and adolescents the following inclusion criteria were used: (1) the study design must have included a training intervention with weight-bearing activities; (2) the effects of the training intervention on BMC and/or aBMD must have been examined and reported in means and SDs for the treatment (T) and control group (C) for pretests and posttests; (3) the participants must have been age 17 years or younger; (4) subjects must not have received calcium supplementation during the study period; and (5) research must have been conducted on healthy male and/or female subjects. Overweight individuals were considered healthy unless reported otherwise. In addition to BMC and aBMD values, information from included studies on baseline subject characteristics (age, maturity, weight, height, and habitual calcium intake per day) and program variables (frequency, duration, intensity, and resistance type) were extracted by two independent reviewers. Inclusion criteria, methods of data extraction, and analyses were specified in advance.
Using the Physiotherapy Evidence Database (PEDro) score, the risk of bias of the selected studies were assessed by two independent investigators. Any discrepancies were resolved by a third investigator. Points were only awarded to each of the 11 methodological components if a criterion was clearly fulfilled. The PEDro scale has been shown to be a valid and reliable measure of methodological quality of clinical trials, and even summarizing scores into a single number has been stated to be valid. However, according to the PRISMA statement for Reporting Systematic Reviews and Meta-Analyses of Studies, scales that numerically summarize different items into a single number are misleading. Therefore, both the summarized PEDro score (Table 1) and the individual methodological components assessed (Supporting Information Document 3) are reported in the present meta-analysis. The risk of bias of analyzed studies was not an inclusion criterion. The quality of evidence was assessed using the “Strength of Recommendation Taxonomy.”
|Study||Subgroup||Age (years)||Weight (kg)||Height (cm)||StD (weeks)||SeD (min)||Intervention||Freq (n/week)||Gen||Ml||PS|
|Bianchini and colleagues (2012)||12.0a||82.4||160||16||60||Sit-ups, squats, medicine ball-exercises, walking/jogging and basketball||3||M+F||Nda||4|
|Blimkie and colleagues (1996)||16.3||57.7||164.3||26||–||Whole body workout in a circuit system: 13 exercises, 7 resistance training machines, 2–4 sets of 10–12 repetitions and 10–12 1RM, enhancement every 6 weeks||3||F||Intra/Post||5|
|Bradney and colleagues (1998)||10.4||36.9||143.9||32||30||Different weight-bearing activities: weight training, aerobics and gymnastics||3||M||Pre||6|
|Fuchs and colleagues (2001)||7.5||27.1||125.1||28||20||5 min warm-up, 10 min jumping (50–100 jumps in a unilateral direction off of 61 cm high boxes) and 5 min cool down||3||M+F||Pre|
|Heinonen and colleagues (2000)||Premenarcheal||11.7||39.3||147.9||39||50||10 min warm-up, 15 min aerobic exercises (non-impact exercises), 20 min high-impact training (100–150 both legs and 25–50 one-leg jumps on and off 30-cm-high boxes) and 5 min cool down||2||F||Pre||5|
|Iuliano-Burns and colleagues (2003)||9||30.5||132.2||34||20||Moderate-impact sessions: hopping-based, jumping-based, and skipping-based activities: ground reaction forces between 2 and 4 times body weight||3||F||Intra/Post||6|
|Johannsen and colleagues (2003)||10.3||39||138||12||-||25 both leg jumps from a 45-cm-high box||5||M+F||Nda||5|
|Kontulainen and colleagues (2002)||12.8||46.4||154.6||39||50||10 min warm-up, 15 min aerobic exercises (non-impact exercises), 20 min high-impact training (100–150 both leg and 25–50 one-leg jumps on and off 30-cm-high boxes) and 5 min cool down||2||F||Intra/Post||3|
|Laing and colleagues (2005)||6||21.5||115||104||60||15 min warm-up (stretching and light activities) and gymnastic training: rotations of approximately equal time on uneven bars, vault, balance beam, and floor exercises||1||F||Pre||3|
|Linden and colleagues (2006)||7.6||27.8||128.1||104||40||A variety of ball games, running, jumping, and climbing||5||F||Pre||5|
|Löfgren and colleagues (2012)||Boys||7.8||28.2||129.3||208||28||A variety of ball games, running, jumping, and climbing||5||M||>Intra/Post year||5|
|Macdonald and colleagues (2008)||Boys||10.2||40.7||141.7||47||15||Short bouts of high-impact jumping: 5 two-foot landing jumps (or 10 one-foot landing jumps), 36 two-foot landing jumps||5||M||Pre||5|
|MacKelvie and colleagues (2001)||Prepubertal||10||31.2||138.6||30||10–12||5 jumping stations in a circuit system: jumping jacks, lunge jumps, hopping, jumping over various obstacles and drop jumps from a platform: progressive impact-loading during the school year||3||F||Pre||6|
|MacKelvie and colleagues (2002)||Asian||10.2||31.9||138.9||28||10||5 jumping stations in a circuit system: jumping jacks, lunge jumps, hopping, jumping over various obstacles and drop-jumps from a platform: progressive impact-loading during the school year||3||M||Pre||6|
|MacKelvie and colleagues (2003)||9.9||34.6||139.6||86||1.5–2||5 jumping stations in a circuit system: plyometric jumps, alternating-foot jumps and 2-foot obstacle jumps: ground reaction forces between 3.5 and 5 times body weight||3||F||Pre||4|
|MacKelvie and colleagues (2004)||10.2||36.9||140.8||20||10–12||5 jumping stations in a circuit system: plyometric jumps, alternating-foot jumps and 2-foot obstacle jumps: ground reaction forces between 3.5 and 5 times body weight||3||M||Pre||5|
|McKay and colleagues (2000)b||8.6||30.8||135.1||32||10 tuck jumps 3 times weekly and jumping, hopping, and skipping twice weekly during physical education classes||2–3||M+F||Intra/Post||5|
|McKay and colleagues (2005)||10.1||35.7||139.8||56||3||10 counter movement jumps (2-foot takeoff, clutch knees, 2-foot landing), 3 times each school day: ground reaction forces of 5 times body weight||5||M+F||Intra/Post||5|
|Meyer and colleagues (2011)||Prepubertal||8.3||27.8||129||43||45||At least 10 min of jumping activities: hopping, jumping up and down stairs, rope skipping etc., and motor skill tasks: jumping around on one leg, balancing on one leg, power games or coordinative tasks||2||M+F||Pre||4|
|Morris and colleagues (1997)||9.5||33.4||139.7||10||30||A variety of vigorous, high-impact aerobic workouts: aerobics, Australian rules football, skipping, ball games and weight training||3||F||Pre||6|
|Nichols and colleagues (2001)||16.1||57||157.6||16||30–45||Whole-body workout with 15 different exercises: combination of free weights and machines: sets/reps-relation progressed from 1/14 to 3/10||3||F||Intra/Post||5|
|Petit and colleagues (2002)b||Prepubertal||10.1||31.2||138.6||28||10–12||Five stations of diverse jumping exercises. Number of jumps was increased from 10 to 20 and height from 10 to 50 cm.||3||F||Pre||7|
|Specker and Binkley (2003)||3.8||16.3||100.6||50||30||Warm-up (5 min), jumping, hopping, and skipping activities (20 min) and cool-down (5 min)||5||M+F||Pre||6|
|Van Langendonck and colleagues (2003)||8.74||28.6||132.23||35||10||5 high-impact exercises: ground reaction forces between 3.5 and 5 times body weight with increasing number of jumps (from 10 to 20) for each exercise||3||F||Pre||6|
|Weeks and colleagues (2008)||Boys||13.8||55||163.7||35||10||Jumps, hops, tuck-jumps, jump-squats, stride jumps, star jumps, lunges, side lunges, and skipping (∼300 jumps per session) and upper-body strengthening activities: push-ups and exercises with resistive latex bands||2||M||Intra/Post||5|
|Witzke and Snow (2000)||14.6||61||164.3||36||30–45||Two-footed landings from a 12-inch box: the training progressed from 2 sets/10 repetitions of 5–7 exercises to 2–3 sets/2–20 repetitions of 12–20 exercises||3||F||Intra/Post||5|
|Yu and colleagues (2005)||10.3||54.6||146||6||75||Warm-up (10 min), strength training: whole-body workout: 8 exercises in a circuit system with a combination of free weights and machines (30 min), aerobic exercise (10 min), agility training (10 min) and cool-down (5 min)||3||M+F||Nda||7|
ESs were calculated as the difference between the standardized mean change for the treatment and control groups divided by the pooled pretest SD. As ESs of small samples tend to be overestimated, the correction factor suggested by Hedges was used to calculate an effectively unbiased estimate (Hedges' g). The random effects model was used for the meta-analysis procedure. This model assumes that there is not one true ES, rather a distribution of true effects, and that the combined effect represents the mean of the population of true effects. This model incorporates statistical variability caused by sampling error and substantive variability.
The I2-index was examined in order to assess the proportion of the observed variance that reflected real differences in ES due to heterogeneity rather than sampling error. I2-values <25% were suggestive of low inconsistency, 25% to 50% moderate, 50% to 75% high, and >75% very high inconsistency.
Subgroup meta-analyses, z-tests, and meta-regressions (bivariate and multiple regression with backward variable elimination) were used to examine the relationships between ESs and moderator variables. The hypothesized categorical moderator variables were: maturity (prepubertal versus intrapubertal/postpubertal), sex (male versus female), and resistance type (machine versus body weight versus mixed). The duration of intervention, training frequency per week, age, height, habitual calcium intake, and weight of participants were the quantitative independent variables tested. Data from continuous moderator variables was excluded from meta-analysis if no mean values were available or computable from raw data. Statistical significance was set to p < 0.05 for all analyses. Funnel plot analyses were used as a visual aid to test for the likely presence of publication bias. Data were analyzed in Statistica version 7.1 (StatSoft, Inc., Tulsa, OK, USA) and IBM SPSS version 188.8.131.52 (SPSS Inc., Chicago, IL, USA).
From 109 potentially relevant studies, only 27 met the inclusion criteria for the present meta-analysis (Fig. 1). Two of the included studies presented aBMD data only, whereas the remaining studies presented both BMC and aBMD data. Eight out of the 27 investigations provided multiple subgroups, giving a total of 35 combined ESs. Assessing methodological quality of the included investigations by using the PEDro scale revealed a mean score of 5.22 ± 1.01 out of 10.
Altogether, 2985 subjects (T = 1640; C = 1345) were included in the meta-analysis and the average sample size per group was 60.7 subjects for T and 49.8 subjects for C. The mean duration of interventions was 47.1 ± 46.4 weeks. There was a distinct sex imbalance, with a total of 1757 female (T = 868; C = 815) and 847 male subjects (T = 440; C = 337). The gender of the remaining 381 subjects was not reported. Sex-specific bone data were unavailable for three studies, in which only the combined results of male and female participants were presented. The mean age of the included subjects was 10.3 ± 2.7 years. A total of 1591 subjects were classified as prepubertal (Tanner stage 1), and 899 as intrapubertal and postpubertal (Tanner stage 2–5). The mean height and weight of all subjects were 141.4 ± 14.1 cm and 38.8 ± 13.2 kg, respectively. Characteristics of the training interventions applied are summarized in Table 1.[18-44]
The weighted overall ES for changes in BMC, based on 32 ESs, was 0.17 (95% CI, 0.05–0.29). By contrast, the weighted overall ES for changes in aBMD, based on only 17 ESs, reached an estimated value of 0.26 (95% CI, 0.02–0.49). Forest plots of the respective meta-analyses can be found in Figures 2 and 3. The performed Z-test revealed that both ESs were significantly greater than zero at p < 0.05. Because the heterogeneity index I2 was small for both the BMC (I2 = 16.4%) and the aBMD values (I2 = 21.22), the pooled estimate based on fixed-effect models revealed similar results (ES for BMC, 0.15; 95% CI, 0.07–0.23; p < 0.05; ES for aBMD, 0.23; 95% CI, 0.11–0.35). No significant differences in ESs (p > 0.05) could be found between randomized controlled trials (RCTs) and nonrandomized controlled trials (nRCT) for both the BMC dataset and aBMD dataset. Further, resistance types (ie, body weight, resistance training machines, or a mix of both) applied within the analyzed studies did not differ significantly in regard to the calculated effect (Table 2).
|Z||p||ES (SE)||n||Z||p||ES (SE)||n|
|Tanner 1||0.28 (0.01)||15||0.33 (0.19)||10|
|Tanner 2–5||0.02 (0.00)||12||0.16 (0.10)||5|
|Male||0.32 (0.04)||7||0.00 (0.31)||3|
|Female||0.11 (0.00)||17||0.24 (0.09)||8|
|Randomized||0.18 (0.01)||21||0.17 (0.09)||10|
|Nonrandomized||0.15 (0.01)||9||0.55 (0.35)||4|
|MIX versus BW||1.22||0.22||1.27||0.20|
|MIX versus MA||1.39||0.16||nc||nc|
|BW versus MA||0.32||0.75||nc||nc|
|MIX||0.90 (0.38)||3||0.90 (0.35)||3|
|BW||0.14 (0.00)||26||0.14 (0.01)||12|
The observed differences between ESs of male (ES BMC, 0.32; 95% CI, −0.07 to 0.71; ES BMD, 0.0008; 95% CI, −0.61 to 0.61) and female (ES BMC, 0.11; 95% CI, −0.01 to 0.22; ES BMD, 0.33; 95% CI, −0.07 to 0.72) subjects were insignificant. By contrast, statistically significant differences (p < 0.05) of weighted mean ESs could be found between BMC gains in prepubertal (ES, 0.28; 95% CI, 0.04–0.51) and intrapubertal and postpubertal subjects. Changes in aBMD values pointed toward the same direction, but failed to reach statistical significance (prepubertal: ES, 0.33; 95% CI, −0.04 to 0.71; intrapubertal and postpubertal: ES, 0.16; 95% CI, −0.04 to 0.36) (Table 2). Bivariate meta-regressions for continuous moderators of BMC ESs revealed significant coefficients of determination for age (r = 0.02, p < 0.05), body weight (r = 0.02, p < 0.05), PEDro scale (r = 0.06, p < 0.05), and calcium intake (r = 0.27, p < 0.05). By contrast, training frequency (r = 0.06, p = 0.05) and body height (r = 0.10, p = 0.07) only approached statistical significance. No relation could be found between study duration and BMC gains (r = 0.005, p = 0.28). None of the investigated continuous moderator variables reached statistical significance for the aBMD data (Table 3).
|n||B (SE)||R2||p||n||B (SE)||R2||p|
|Age, years||31||−0.022 (0.009)||0.02||0.02a||16||−0.025 (0.026)||0.01||0.35|
|Height, cm||32||−0.003 (0.002)||0.01||0.07||17||−0.003 (0.005)||0.01||0.5|
|Weight, kg||32||−0.004 (0.002)||0.02||0.02a||17||−0.003 (0.004)||0.01||0.48|
|Duration, weeks||32||0.000 (0.000)||0.01||0.28||17||−0.003 (0.002)||0.04||0.08|
|Frequency, days/week||32||0.020 (0.019)||0.00||0.28||17||0.029 (0.064)||0.00||0.65|
|Calcium intake, mg/day||25||0.001 (0.000)||0.27||<0.01a||11||0.000 (0.000)||0.02||0.39|
|PEDro score||32||0.094 (0.024)||0.06||<0.01a||17||0.203 (0.051)||0.18||<0.01a|
A stepwise regression with backward variable elimination was performed for the BMC data using significant continuous (ie, age, body weight, and calcium intake) and categorical moderator variables (ie, “dummy coded” maturity). Because none of the subject or program characteristics reached significance for aBMD data, no stepwise regression was performed for this dataset. The final model of the regression analysis revealed that the daily calcium intake and the maturity level of subjects (prepubertal versus intrapubertal and postpubertal) at baseline were significant predictors of the BMC gains during growth (Table 4).
|Variables in the model||B (SE)||Standardized beta||p||R2 of the final model|
|Ca-intake, mg/day||0.001 (0.000)||0.541||<0.001a||0.35|
Funnel-plot analysis of the BMC dataset revealed an asymmetrical appearance. Two of the less precise studies (with comparatively high standard errors) presented the highest effects on BMC. By contrast, less precise studies with concomitant low effects were not found by the literature search performed. The funnel plot created from the aBMD dataset presented a more symmetrical appearance. However, the highest ES in this dataset also comes from a less precise study with no comparable counterpart at the opposite side of the plot (Fig. 4).
The purpose of the present meta-analysis was to objectively quantify the effects of weight-bearing activities on BMC and aBMD in children and adolescents. We included both RCTs and nRCTs. The observed weighted overall ESs for changes in BMC (based on 32 ESs and 2664 subjects) and aBMD (based on 16 ESs and 1376 subjects) were significantly greater than zero, indicating that structured exercise programs of different types of loading are capable of improving BMC and aBMD during growth. These results are consistent with previously published narrative reviews and position statements which have concluded that weight-bearing activities are effective in enhancing bone health parameters during growth.[7, 45-47] Nevertheless, it should be noted that the observed overall weighted mean ES for WBA on BMC (ES, 0.17; 95% CI, 0.02–0.32) and aBMD (ES, 0.26; 95% CI, 0.02–0.49) were small.
Because strains, resulting from mechanical forces placed on the skeleton, are thought to be the essential signal sensed by bone cells as osteogenic, one might speculate that gravitational forces and muscle pulls resulting from the applied interventions were insufficient to induce higher adaptations. This assumption is supported by numerous cross-sectional studies that found junior athletes participating in impact sports to have significantly higher BMD values when compared those participating in non-impact sports.[49-51] These studies clearly point toward a dose-response relationship for intensity or at least a threshold of mechanical loading that is needed to be crossed to induce a biopositive osteogenic adaptation.
Although it is currently impossible to easily quantify exercise intensity with respect to bone-loading forces, it is assumed that these forces generally increase in proportion to exercise intensity when quantified by conventional methods (eg, percent of the one-repetition maximum). However, due to insufficient descriptions of training intensity within the selected studies, it cannot be confirmed that any intensity-related dose-response relationship is responsible for the observed low ES.
To our knowledge, the only other meta-analysis concerning exercise induced bone mineral accrual during childhood and adolescents was published in 2010 by Nikander and colleagues. Rather than focus on BMC and aBMD as in the present meta-analysis, Nikander and colleagues focused on bone strength (defined as maximal moment of inertia, bone strength index, stress-strain index, and cross-sectional moment of inertia or section moduli). They reported significant but small exercise effects for children and adolescents. The estimated effect in prepubertal and early pubertal boys was 0.17 (95% CI, 0.02–0.32), in prepubertal and early pubertal girls –0.01 (95% CI, −0.18 to 0.17), in adolescent boys 0.10 (95% CI, −0.75 to 0.95), and in adolescent girls 0.21 (95% CI, −0.53 to 0.97). The results for male subjects are in agreement with the maturity effect found by the present meta-analysis (Table 4), whereas data presented on female subjects seem to paint a different picture (ie, no effects within early phases of maturity). However, it should be noted that the data presented by Nikander and colleagues are based on a small sample size with the data for prepubertal and early pubertal boys and girls, which were underlaid by three studies and two studies, respectively, and for adolescent boys and girls, only one study. Therefore, the outcome variables (bone strength) chosen by Nikander and colleagues that differ from the present meta-analysis and the mentioned small sample sizes make it difficult to compare the results with those of the present meta-analysis.
The weighted overall ESs for changes in BMC and aBMD could also be affected by the amount of calcium and/or vitamin D intake. According to a double-blind randomized placebo-controlled trial conducted by Courteix and colleagues, exercise programs are only completely effective if high calcium intake is provided. Similar conclusions were drawn by Bass and colleagues, who found 2% to 3% greater exercise-induced BMC gains in children supplemented with calcium (392 ± 29 mg/d) than in controls. Vitamin D plays a critical role in the gut absorption of calcium. It is assumed that inadequate vitamin D levels decrease gut absorption of calcium by about 15% to 30%, leading to the suggestion that vitamin D with or without calcium, but not calcium alone improves skeletal mineralization in children and adolescents.
Interestingly, according to bivariate meta-regressions, the amount of habitual calcium intake per day at baseline best predicted training-induced BMC (but not aBMD) gains in children and adolescents (Fig. 5) when compared to the remaining continuous moderator variables analyzed in the present meta-analysis (Table 3). Effects of the training programs applied were higher in subgroups that featured greater calcium intake per day. Almost one-third of the observed variance between subgroups could be explained solely by this factor (r2 = 0.29, p < 0.05). Based on the present data, daily calcium intake should exceed 850 mg/d if aiming for exercise-induced BMC gains. That is, calcium intake according to values of the Recommended Dietary Allowance (RDA), which are 1000 and 1300 mg/d for 4–8-year-old and 9–18-year-old subjects, respectively, would be sufficient to facilitate exercise-induced bone mineral accrual. The highest effects in the present meta-analysis were found for trials in which habitual calcium intake approached 1300 mg/d.
It should be taken into account that the observed correlation between calcium intake and BMC gains was largely affected by one positive outlier; excluding this study from the analysis resulted in a lower correlation coefficient of r = 0.34 (r2 = 0.11, p < 0.05). However, under a random effects model, we assume that there is a distribution of true effects, with each trial estimating a different true ES—based on real differences between studies. Therefore, the observed magnitude of effect is not necessarily indicative of a measurement error but might reflect an advantageous covariate combination in this specific study. In this case, the combination of covariates consisted of prepubertal subjects (Tanner stage 1), a high habitual calcium intake at baseline (1286.9 mg/d), and a high-impact training program (50–100 jumps per session; three sessions per week). Because we are estimating the mean of a distribution of true effects, the aim is to include all of the enrolled populations in the weighted mean ES. Nevertheless, given that ESs >1 are rarely seen in human biology and that our meta-analysis found no effects of calcium intake on BMD values, the results should be interpreted with caution.
Unfortunately, corresponding vitamin D intake was only recorded by five trials, so no robust estimation concerning the effect of Vitamin D intake on exercise-induced BMC gains can be drawn from the present data. Nevertheless, in all probability, data on vitamin D intake in combination with the amount of calcium intake per day would explain even higher proportions of the observed variance.
Subgroup analyses for categorical moderator variables revealed that prepubertal children display higher training-induced BMC gains when compared to intrapubertal and postpubertal subjects. Although insignificant, a similar trend could be observed for the analyzed aBMD data. In line with the results from subgroup analyses, it turned out from the performed backward stepwise meta-regression that maturity is an important predictor of BMC gains during growth. Within the final multivariate regression model (Table 4), baseline values of maturity and daily calcium intake were able to explain about 35% of the between-subgroup variance. This supports the assumption that the prepubertal years are an opportune stage of maturation during which the skeleton is inimitably responsive to exercise. In this context, it has been shown that differences in BMC between the playing and nonplaying arm of female tennis and squash players were two times greater if their playing careers began before menarche. However, the underlying biological basis for this observation remains totally unclear. It has been suggested that hormonal factors are responsible for maturation-dependent differences in adaptation. Bass and colleagues argued that growth is largely influenced by the secretion of human growth hormone (HGH) and insulin-like growth factor (IGF-1) and that exercise is a potent stimulus for secretion of these hormones. However, it is unclear why prepubertal subjects would benefit from these hormone responses to a higher extent when compared to intrapubertal or postpubertal children. Exercise-induced HGH responses, relative to baseline values, increase with greater stages of maturity and, with respect to absolute values, these changes even take place at a higher level of blood hormone concentration in later stages of puberty.[61, 62]
It is conceivable that sex steroids explain maturational differences in BMC gains. Besides HGH, blood concentrations of sex steroids increase strikingly throughout puberty in boys and girls. Bone health is known to be positively influenced by testosterone and estradiol, whereas deficiencies of these hormones are associated with osteoporosis.[63, 64] Against this background, one would expect the prepubertal years to be a time period of minor osseous adaptability. However, the positive effects of sex steroids on bone health arise mainly from sex steroids' characteristic of protecting against bone loss by unfolding antiapoptotic effects on osteocytes and osteoblasts as well as their apoptotic effects on osteoclasts. However, the rate of bone remodeling is slowed by sex steroids and the number of remodeling cycles is decreased.[63, 65] Therefore, an unattenuated bone remodeling might abet bone mineral accrual during the prepubertal years and explain some of the observed maturational differences in bone adaptation to exercise. In line with the observed maturational influence, bivariate meta-regression analyses for age of participants points toward greater benefits in younger years for BMC (p < 0.05), albeit not for aBMD (p > 0.05). The low coefficient of determination, r2 = −0.02 (p < 0.05), for BMC likely reflects the idea that maturational status rather than the chronological age of subjects is important for the osteogenic response of bone tissue.
Results from bivariate meta-regressions for study duration and training frequency indicate that long-lasting interventions, and those that applied training stimuli more often during the week, did not result in higher increments in BMC and aBMD. In other words, although self-evident, a dose-response relationship regarding the training volume could not be confirmed based on the available data. This is in contrast to previously published cross-sectional studies that found hours of intensive sports training to be positively correlated with bone mineral accrual in children and adolescents.[59, 66-69] Again, insufficient training stimuli of included trials might explain the results. That is, we would not expect any improvements from increasing duration or frequency of intervention if the training stimulus itself was below a certain threshold.
BMC and aBMD values have been reported to be higher in overweight individuals compared to normal weight subjects. This evidence may have negatively affected exercise-induced effects in one (aBMD dataset) or two (BMC dataset) analyzed trials.[18, 44] However, BMC levels in overweight subjects are also known to be low for their respective body weight. Further, a higher body mass might also increase the exercise stimulus in training regimens using body weight as resistance (eg, jumping interventions). That is, the low BMC-to-body weight ratio in combination with the circumstance that body weight is frequently used as exercise resistance, might outweigh the negative effects of increased baseline levels of analyzed bone parameters. Further, it needs to be taken into account that the majority of the studies included applied absorptiometric measurements, which are known to be affected by the amount of soft tissue surrounding the bone examined. That is, exercise-induced BMC and aBMD changes in overweight individuals may be affected by this additional bias. However, according to the data presented by Wosje and colleagues, the estimated measurement error and absolute and relative smallest detectable differences (SDDs) for bone mass were similar for obese and nonobese children. In summary, it is currently unknown if inducible BMC gains are actually blunted in those subjects. Against this background, studies on overweight children and adolescents were not excluded from the outset. The estimated weighted mean ES of both trials on overweight individuals (ES, 0.07; 95% CI, −0.23 to 0.38) did not significantly differ from remaining subgroups (ES, 0.18; 95% CI, 0.05−0.31). Nevertheless, it cannot be ruled out with reasonable certainty that BMC and aBMD gains would have been higher in the mentioned trials[18, 44] if studies were conducted on normal-weight children.
Subgroup analyses for sex within the BMC dataset revealed ESs of 0.32 (95% CI, −0.07 to 0.71) and 0.11 (95% CI, −0.01 to 0.22) for male and female subjects, respectively. However, the observed difference in ESs was not significant (p = 0.31). By contrast, sex-specific ESs for changes in aBMD were (insignificantly) higher in male subgroups (ES, 0.00; 95% CI, −0.61 to 0.61) compared to female subgroups (ES, 0.33; 95% CI, −0.07 to 0.72). The high standard error of ESs for male subgroups likely prevented the Z-tests from disclosing any significant difference between males and females in the BMC dataset—despite the apparently different levels of ESs. The small number of available trials (BMC: n = 7 subgroups; aBMD: n = 3 subgroups) and associated high standard errors might also explain the inconsistent findings between the BMC and aBMD datasets. The predominance of research conducted on female subjects reflects the well-known gender differences in the prevalence of osteoporosis and associated fractures later in life. However, taking into account that approximately 2.8 million men in the United States are affected by osteoporosis and fractures, this disease should not be considered solely a “woman's disease.” Future research on prophylactic interventions to enhance peak bone mass early in life should broaden their focus by including all-male subgroups in interventional studies.
With respect to the resistance types applied (machine versus body weight versus mixed), no differences in efficacy to induce BMC or aBMD changes could be found. When considering the available data in detail, it becomes obvious that the vast majority of interventions used body weight as the resistance type, whereas resistance training on machines or a mix of both (machine-supported training with body weight exercises) was the exceptional case. Therefore, to provide a robust estimate on the difference between the types of resistance, further studies are needed including machines or mixed as resistance.
By its very nature, the present meta-analysis is limited by the quality of the trials included. The mean PEDro score for the analyzed studies was 5.31 ± 1.00 out of 10, commensurate with a fair methodological quality. As a result of the limited quality and some inconsistent findings, the rated level of evidence, according to the Strength of Recommendation Taxonomy, was 2B. It should be noted, however, that some of the PEDro items, such as blinding subjects or trainer, is almost impossible in most settings associated with training studies. Meta-regression for study quality revealed unexpected results. Even though low-quality trials usually tend to overestimate treatment effects,[75-77] the observed regression coefficient for PEDro score indicates that the quality of trials was positively related to ESs in the BMC (B = 0.09; SE = 0.02) and aBMD (B = 0.20; SE = 0.05) dataset at p < 0.05. That is, high-quality studies presented greater effects when compared with studies that suffered from low PEDro scale scores. This paradoxical outcome might in part be explained by the fact that the outcome of individual studies depends on numerous factors, resulting in a complex and unreliable association between quality measures and ESs.[75, 78] This complex association might also explain the insignificant difference observed between RCTs and nRCTs.
As expected, funnel plot assessment for both BMC and aBMD data showed that larger studies with low SE were tightly clustered and less accurate studies were more widely spread around the estimated overall ESs. However, there was an asymmetrical appearance of BMC data, with small studies showing negative effects missing from the bottom left corner of the plot, making publication bias almost certain (Fig. 4A). Even if the two positive outliers were excluded from the plot, it retains its asymmetrical appearance. The funnel plot created for the aBMD dataset also gives a reason to presume publication bias. That is, the overall ES is somewhat overestimated due to publication bias and results from the present meta-analysis should be treated with reasonable caution (Fig. 4B).
In conclusion, contrary to the widely held belief that exercise is a potent stimulus to increase BMC and aBMD during childhood and youth, significant gains could only be found in BMC in prepubertal subjects. That is, efficacy of training in terms of bone mineral accrual is substantially affected by the maturational status of subjects. The physiological mechanism underlying this observation remains unclear and needs to be determined by future research. Further, habitual calcium intake largely affected the amount of exercise-induced bone mineral accrual (BMC). Therefore, physical activity in form of WBA in combination with high calcium intake should be encouraged in prepubertal years in order to oppose osteoporosis later in life by increasing peak bone mass. Due to the comparatively low number of trials investigating the effects of WBA on aBMD during growth, whether the same conclusion holds true for changes in aBMD, cannot be answered from the present meta-analysis.
All authors state that they have no conflicts of interest. The authors disclose all restrictions on full access for all authors to all raw data, statistical analyses, and material used in the present manuscript.
Authors' roles: Study design: MB, SG, MM, and JM. Study conduct: MB. Data collection: SG and MB. Data analysis: MB and SG. Data interpretation: MB, SG, MM, and JM. Drafting manuscript: MB, MM, and SG. Revising manuscript content: MB, SG, MM, and JM. Approving final version of manuscript: MB, SG, MM, and JM. MB had full access to all of the data in the study and takes responsibility for the decision to submit the publication.