Eligibility Criteria

For this meta-analysis, we accepted RCTs with an exercise duration of at least 6 months that used (p)QCT, MRI or DXA HSA to assess whole bone strength. The 6-month follow-up limit was selected because the entire bone remodelling cycle typically takes around 6 months, and thus it is unlikely that any true physiological exercise-induced skeletal changes would occur prior to this period. Examiner blinding or participants' ethnicity were not eligibility requirements. Throughout this paper, we will use a common term 'bone strength' to encompass all specific biomechanical terms of bone strength, such as bone strength index (BSI), stress-strain index (SSI), maximal moment of inertia (Imax), cross-sectional moment of inertia (CSMI) and section moduli (Z). We focused on papers published on the topic in English until October 2009. We included studies from the prepubertal period (≥6 to 10 years of age) to older age (≥50) and analysed the results by age groups (childhood, adolescence, adulthood and older adulthood) to reduce the clinical heterogeneity since the effects of exercise on bone strength is an age- and maturity- dependent phenomenon [12]. For instance, it is hypothesized that growth is a time when exercise may be associated with a marked increase in bone strength. In contrast, adulthood is a period when exercise may help to maintain bone strength, while exercise in older adulthood is most likely to reduce (or attenuate) the natural decline in bone strength.

We included healthy female and male children, and adult women and men not using anti-osteoporotic medication, unless equally distributed between study arms in a given trial. We excluded quasi RCTs, trials for people with disabilities and all animal experimental studies. Intervention exercise or physical activity trials were defined as weight-bearing impact, resistance, endurance training or a combination of these types of training. Habitual recreational activity without any specific intervention or supervised activity known not to affect bone (sham exercise) was accepted as activity for the control participants.

Information Sources and Search

For this systematic review and meta-analysis, an experienced medical librarian conducted a search of four databases (PubMed, Sport Discus, Physical Education Index, and Embase) using a comprehensive combination of keywords describing exercise: exercise movement techniques, exercise, exercise therapy, physical education and training, sports, physical fitness, physical activity, physical training, exercise training, physical exercise, physical education and motor activity. The keywords describing bone strength included bone structure, bone strength, bone rigidity, bone geometry, bone and bones. The authors RD, AH and RN also used cross-referencing from retrieved articles and conducted a hand search to identify possible missed RCTs in database searches.

Data Collection Process and Quality Analysis

Two authors (AH and RN) independently reviewed all screened abstracts. Then AH and RN, together with RD, independently evaluated all potential full-text articles. If there was disagreement whether to include a full-text article, all authors evaluated the article to come to a joint consensus. Similarly, the methodological quality was individually assessed by AH and RN together with RD. AH then extracted the data from the original full-text articles, and RN and RD rechecked all relevant data, including all outcome values needed for the analysis. Two RCTs where participants were taking either anti-osteoporotic medication or hormone replacement therapy were included because the study arms were equally distributed. Only exercise arms and non-exercise arms without treatment were analysed in these trials because interaction effects were not reported [13, 14]. However, two other exercise trials including multiple but heterogeneous exercise groups in their exercise regimen were treated as independent in our analysis [15, 16].

Data Synthesis

In the data synthesis of the meta-analysis, Review Manager software (5.0.16; Thomson ResearchSoft, Carlsbad, CA, USA) was used to calculate the pooled effect estimates for the combinations of single effects of RCTs using an ITT approach if possible. ITT data were replaced with per-protocol data in cases where ITT data were not available [13]. To calculate standardized mean differences (effect size), we used the follow-up values of bone strength measures adjusted for baseline bone values. An effect size around 0.2 was considered a small effect, around 0.5 a medium effect and ≥0.8 a large effect. For all analyses, we used an inverse variance weighted random effects model that incorporates heterogeneity into the model. In multiple comparisons with two or more exercise groups, the number of controls was divided among comparisons to ensure that we counted control participants only once in the meta-analysis. Results were considered to be statistically significant at an alpha level of <0.05. Heterogeneity was calculated using the Cochran's Q-test and an alpha level of <0.10 for statistical significance. In addition, I ² was used to examine inconsistency in the study findings. In general, values of <25%, 25%-50%, 50%-75% and >75% were suggestive of low, moderate, high and very high inconsistency.

In the results section of this systematic review and meta-analysis, we present our findings from meta-analysis using forest plots of the standardized mean differences for children and adolescents, young adults, and older adults. We then discuss and summarize the findings in the discussion section. In the final two sections of this review, we discuss the structural basis underlying exercise-induced improvements in bone strength, as well as the characteristics of exercise loading that have been shown to be best associated with structural improvements in bone. For these two sections, pertinent cross-sectional studies were reviewed.

Exercise Effects on Bone Strength: A Meta-Analysis of Randomised Controlled Trials

We identified and screened 1252 potential abstracts, of which 1224 were excluded either because they were unrelated to the specific topic, nonrandomised studies or duplicate studies from different databases. We then reviewed the remaining 28 full-text articles of RCTs in which exercise effects on bone were investigated. Seventeen of these 28 trials were excluded because of evaluation of BMD rather than estimates of bone strength. Finally, 11 RCTs investigating exercise effects on bone strength were thoroughly reviewed, one of which was excluded as a duplicate because it used the same sample and a similar outcome as in the investigators' previous trial [18]. Thus, 10 remaining trials were analysed (Figure 1).

From the 10 analysed exercise RCTs that evaluated the effects of different modes of exercise training on estimates of bone strength at different skeletal sites, five were conducted in pre-/early pubertal children and/or adolescents [19–23], one in premenopausal women [24] and four in postmenopausal women [13–16]. To our knowledge, there are no published RCTs examining the effects of exercise on bone strength in adult men. From these 10 trials, the data needed for the effect size estimations were reported in six articles, and the authors provided the data required for three more original articles identified for this review. The findings from the remaining RCT could not be entered into the model because follow-up values were not reported in the original article and could not be obtained from the authors [22]. However, the results were qualitatively analysed, and the findings are discussed in the text. Intervention characteristics of the 10 evaluated studies are summarised in Tables 1 and 2.

Even though we did not exclude articles that may have been biased from our meta-analysis because of the limited number of available RCTs on this topic, the quality evaluation of individual studies revealed several possibilities for systematic bias (Table 2). The major concerns were that 3 of the 10 analysed RCTs were <12 months in duration, and 4 included <100 participants. Moreover, the prescribed type and amount of weekly exercise varied considerably between the different trials. One trial also reported findings on the basis of a per-protocol approach rather than ITT [13]. In addition, some studies included small (albeit not statistically significant) between-group differences in baseline outcome variables despite the randomization. However, we used baseline-adjusted values in our meta-analysis to control for this possible bias. Also, outcome variables describing structural strength varied between the studies, and comparisons were made with controls continuing their habitual activity rather than offering them nonosteogenic (sham) exercise interventions. Other minor concerns were related to the lack of details comparing participants who declined to participate and those who participated in some trials.

Effects of exercise on bone strength in children and adolescents

Bone Strength

Our meta-analysis of published RCTs evaluating exercise effects on bone strength during growth indicate that there was a small but significant effect on the lower extremities in young boys (effect size 0.17; 95% CI, 0.02-0.32), but not in young girls (Figures 2 and 3). Furthermore, when the authors of the original articles conducted per-protocol analyses, which accounted for physiologically meaningful covariates such as weight and height, they reported that weight-bearing exercise can enhance bone strength at the distal tibia in prepubertal but not early pubertal boys [19]. These findings suggest that the response of bone to loading may be maturity-dependent, but further long-term RCTs are still needed to test this hypothesis.

The results of one RCT were excluded from our meta-analysis because the follow-up bone values were not reported in the original article and could not be obtained from the authors [22]. In this RCT that included pre- and early pubertal girls, the authors reported a 4% exercise effect on femoral neck strength in early pubertal girls. However, our meta-analysis including values of the ITT approach did not find statistically significant exercise effects on femoral neck strength in another RCT including postpubertal adolescent boys or girls. In contrast, per-protocol analysis in this study revealed that there was a trend toward a greater improvement among those who were compliant (19% vs. 11%) in both sexes [23] (Figure 4).

Effects of exercise on bone strength in adults

Bone Strength

With regard to the effects of exercise on bone strength during early adulthood, only one RCT was recently conducted in premenopausal women [24]. Our analysis from this 12-month progressive impact exercise trial with additional home training did not indicate any effect on bone strength at either the proximal tibia or femoral shaft (Figure 5). However, the authors' subgroup analysis of the exercise group revealed that those women who were most compliant (the highest quartile, >66 exercise sessions during the 12 months) had a 0.5%-2.5% greater gain in bone size, cortical thickness and bone strength at the proximal tibia than those who were in the least compliant quartile (<19 sessions).

Effects of exercise on bone strength in older adults

Bone Strength

Our analysis of the published RCTs examining the exercise effects on bone strength in older women did not find a significant overall effect or any site-specific effects [13–16] (Figure 6). In a recent, yet unpublished, 18-month RCT in middle-aged and older men, high-intensity progressive resistance training combined with moderate-impact weight-bearing exercise increased femoral neck strength by 2% relative to no exercise, but there was no effect of exercise on bone strength at the tibial and femoral midshafts [39]. Consistent with the latter finding, a recent systematic review among postmenopausal women, which included all pQCT studies (RCTs and cross-sectional and prospective studies) in their analysis, concluded that exercise has a positive but modest site-specific effect on bone mass and geometry at the loaded skeletal sites, primarily affecting cortical rather than trabecular bone [40]. The authors also concluded that the mass and geometric changes appear to be largely dependent on continued participation and participants' ability to maintain sufficient exercise intensity.

Structural Basis Underlying Exercise Gains in Bone Strength

Bones can adapt their strength to increased loads through surface-specific changes on the periosteal or endosteal surface either independently or in combination. For example, loading can increase cortical thickness through periosteal apposition, resulting in an increase in bone size and/or via the addition or reduced resorption of bone on the endocortical surface [43].

During growth, there is evidence that the surface-specific responses to loading are sex-specific and maturity-dependent. For instance, in a study using MRI to compare bone structural differences between the playing and nonplaying arms of young female tennis players, Bass et al. [44] reported that exercise-induced gains in humeral bone strength in players prior to puberty were due to periosteal apposition, whereas the predominant effect after puberty was endocortical apposition (Figure 7). Using the same methodology, similar results were recently observed in pre-, peri- and postpubertal boys, but the periosteal gains in prepubertal boys were nearly double those observed in girls and their continued training into peripuberty resulted in further gains in bone size (periosteal apposition) [45]. In both studies, there was heterogeneity in the response to loading at the endocortical surface along the length of the bone. For example, in the prepubertal players, loading resulted in increased expansion at the midhumerus but no change or endocortical apposition at the distal humerus [44–47]. This highlights that there are regional differences in bone adaptation within localized areas of the bone.

In support of the notion that there are regional adaptations to loading, the findings from a recent 16-month school-based exercise intervention revealed that anterior-posterior bending strength at the tibial shaft increased in the active pre- and early pubertal boys [18]. In addition, a recent cross-sectional study involving a range of young adult athletes participating in different sports and a nonathletic reference group revealed that there were surface-specific differences in bone adaptations at the distal tibia and tibial shaft amongst the different types of athletes. Adult athletes involved in impact-type sports had thicker cortices at the distal tibia and tibial shaft, but larger cross-sectional area (bone size) only at the tibial shaft. This suggests that the cortex at the distal tibia may thicken from the endosteum, whereas at the tibial shaft the predominant response was periosteal apposition [48, 49].

In older adults, the results from cross-sectional studies indicate that exercise can enhance cortical thickness and bone strength at loaded sites, which appears to be largely due to an increase in the cross-sectional size of the bone (periosteal apposition) [50–55]. However, these findings are not supported by the limited data from exercise intervention trials [56]. In an early non-RCT using pQCT to assess bone geometric changes at the radius in response to a 6-month upper limb loading program in postmenopausal women, Adami et al. [57] reported a significant training effect on cortical bone area (3%) and cortical BMC (3%), but a decrease in trabecular BMC (-3%). The authors speculated that increased loading resulted in reshaping of the bone cross-section (periosteal expansion) and a redistribution of bone minerals from the trabecular to cortical component (e.g., corticalisation of the trabecular tissue). Similarly, the findings from a 12-month multidirectional jumping intervention in postmenopausal women revealed that exercise seemed to improve bone strength at the distal tibia by increasing the ratio of cortical to total area [14]. There was no effect of exercise on total area, which suggests that exercise reduced endocortical bone loss.

While some RCTs in older women conducted over 6 to 12 months have failed to observe any significant effect of resistance or impact training on bone structural properties at the tibia or radius [13, 16], some beneficial effects on cortical volumetric BMD at both midshafts were observed [16]. Moreover, the results from twin pairs aged 50-74 years who had been discordant for physical activity for at least 30 years indicated that bone strength increased by 18% at the distal tibia, which appeared to be due to an increase in trabecular bone density rather than bone size [51]. At the tibial shaft, however, cortical area and bone strength were 8% and 20% greater in the active compared with inactive twin members, respectively. This was predominantly due to reduced endocortical resorption and not periosteal apposition [51].

On the basis of these findings and other literature, it appears that exercise-induced improvements in bone strength in older adults are likely due to reduced endocortical bone loss and/or increased tissue density rather than an increase in bone size (periosteal apposition) [44]. As noted above, the latter mechanism is typical for growing bones, which have a greater potential for exercise-induced changes in bone size and strength than their mature counterparts [12, 44–47]. However, this phenomenon appears to differ for adults; thus, further long-term RCTs in all age groups are needed to substantiate these observations.

Trabecular bone architecture can also adapt to increased loading, but owing to the limited resolution of most current imaging techniques, the effects of exercise on the thickness, number, separation and orientation of trabecular elements in human bones are not well known. However, the findings from two small studies using high-resolution MRI to assess bone microarchitecture revealed that trabecular bone volume at the proximal tibia, distal femur, or both were greater in high-level athletes (gymnasts and Olympic fencers) relative to controls, which was due to an increase in trabecular number and not trabecular thickness [58, 59].

Loading Characteristics to Best Improve Bone Strength

There is considerable interest in defining the optimal dose(s) and characteristics of loading to best improve bone strength (i.e., optimal loading type and programme) so that precise exercise prescription guidelines can be developed. Extensive research using animal models has shown that the skeleton's response to loading is regulated by a number of different loading characteristics, including the magnitude, rate, distribution (pattern) and number of loading cycles [60–63].

Consistent with these findings, data from the available cross-sectional studies and limited intervention trials in children that have used (p)QCT or MRI to characterise changes in bone structure and strength indicate that the most effective programmes (or loading characteristics) are those that incorporate a combination of moderate- to high-impact weight-bearing activities that are variable in nature (i.e., multidirectional) and applied rapidly [11]. Specifically, the most successful exercise programmes appear to be those that incorporate a diverse range of weight-bearing activities (e.g., skipping, dancing, jumping, and hopping) ranging in magnitude from three to nine times body weight and which are performed three to five times per week, preferably on a daily basis, for 10-45 minutes per session [19, 64, 65].

In middle-aged and older adults, the optimal type and dose of exercise needed to enhance bone geometry and strength is less definitive. The general consensus from recent intervention trials and meta-analyses with BMD as the primary outcome is that low- to moderate-impact weight-bearing exercise in combination with progressive resistance and/or agility training tends to be the most effective for improving hip and spine BMD (or preventing bone loss) and functional ability in both older men and women [15, 36, 38]. However, further work is still needed to define the specific types of exercises and the associated loads related to these activities that can be safely performed by older adults at varying levels of physical function and fracture risk. While certain moderate- to high-impact exercises are likely to be contraindicated for older adults at high risk of fracture, it is reassuring that few if any adverse effects have been reported from exercise intervention trials that have been conducted in children or in middle-aged or older adults.

On the basis of the available evidence, it is clear that we must await the results of further long-term RCTs before specific exercise prescription guidelines can be developed to maximise bone strength, particularly at the clinically relevant hip and spine. However, a recent cross-sectional study [66] in young adult athletes provided a unique insight into the effects of different modes of loading on femoral neck bone geometry and strength. In this study, MRI was used to examine differences in cortical bone structure and strength in female athletes categorized into five distinct loading groups: high-magnitude vertical impact (volleyball, triple jump, hurdling, and high jump), moderate-magnitude impacts from rapidly varying odd directions (soccer and racket sports), high-magnitude muscle forces (powerlifting), repetitive low impact (running), and repetitive nonimpact (swimming). In comparison to the nonathletic reference group, the authors found that cortical area and bone strength at the femoral neck, but not total cross-sectional area or diameter, were ~15% to 30% greater in athletes representing both the high-impact and odd-impact exercise loading. Further regional analysis examining the cortical thickness at the inferior, superior, anterior and posterior regions of the femoral neck revealed that compared with the reference group, the inferior cortex was ~60% greater in the high-impact group, and the anterior and posterior cortices were 20% greater in both the high-impact and odd-impact loading groups (Figure 8). At the superior cortex, a region subjected to substantial compressive loads from a sideways fall onto the hip [67–69], there was also a trend for a thicker cortex (~15%) in the odd-impact group. These findings suggest that exercise regimes comprising moderate-magnitude impacts from varying, odd directions may represent the optimal mode to enhance bone structure and strength at the clinically important femoral neck [66].