Introduction

Low-trauma fractures related to osteoporosis are a major problem in our aging society. Considering the demographic change in Europe, the number of osteoporotic fractures will quite likely increase by about 25% during the next 10–15 years [1]. A large variety of pharmaceutic agents target osteoporosis, most of which are very cost intensive, have potential negative adverse effects, and focus predominately on the bone. In contrast, physical exercise is a low-cost approach providing positive effects on fall risk [2] and bone strength [3, 4] without causing relevant adverse effects [5]. Thus, exercise might be an excellent strategy for combatting fractures in older adults. Reviewing the literature shows that there is indeed some evidence for a fracture-preventing effect of exercise in older adults [6,7,8,9]. However, with the exception of an older systematic review and meta-analysis that focused on low-trauma overall fractures [7], all the other studies [6, 8, 9] focused on data regarding fall-related fractures. In a recent systematic review and meta-analysis, we determined significant positive effects of exercise on overall and major osteoporotic fracture incidence [10]. Nevertheless, due to the considerable heterogeneity between the trial results, it is important to identify key components of promising exercise protocols. While their close interaction might prevent a meaningful sub-analysis of many exercise parameters (e.g., exercise intensity), we focus on intensity progression during the trial and study duration, as these may well be more independent training parameters. Thus, besides providing evidence for a (osteoporotic) fracture-preventing effect of exercise, we concentrated on the corresponding effect of (1) the progression of intensity during exercise intervention and (2) the duration of the study intervention,Footnote 1 in order to derive reliable exercise recommendations.

Methods

This systematic review and meta-analysis is part of the Austria/German/Swiss (DACH) S3 Guideline “körperliches Training zur Frakturprophylaxe” (AWMF: 183—002).

Literature search

We adopted the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [11]. Briefly, we checked six electronic databases (PubMed, Scopus, Web of Science, Cochrane, Science Direct, and ERIC) without language restrictions for articles published from January 1, 2013 (last search [7]) to May 2021. We applied keywords and their synonyms around the queries “Bone mass” or “Osteopenia” or “Bone turnover” or “Bone metabolism” or “Bone mineral content” or “Skeleton” or “Bone Mineral Density” or “BMD” or “Bone Density” or “Osteoporoses” or “Osteoporosis” or “Bone structure” or “Bone status” or “Bone Tissue” or “bone”) AND (“Bone fracture” or “Fracture” or “fragility fracture” or “Broken Bone”) AND (“Exercise” or “physical activity” or “Physical training” or “Exercise training”) AND (“clinical trial”) AND (“45 years and older”). We also checked reference lists of eligible studies and systematic reviews/meta-analysis that focused on fracture and fall reduction and bone-related outcomes (e.g., BMD). Studies without full texts were not considered.

Eligibility criteria

Briefly, randomized and non-randomized clinical studies were included that fulfilled the following eligibility criteria: (a) exercise studies on fracture prevention, fall reduction, and bone strength with (b) at least one exercise (EG) versus one control group (CG) that (c) reported the number of hip, lumbar spine, forearm, and/or humerus fractures (d) separately for EG and CG, independently of (e) whether fractures were defined as primary or secondary outcome, observation, or adverse event, and (e) female and male cohorts older than 50 years on average that were observed (f) for at least 3 months (i.e., study length ≥ 3 months).

Studies that supplied (a) pharmaceutic agents (e.g., glucocorticoids, bisphosphonates) or treatments (e.g., chemo- and/or radiotherapy) with relevant impact on bone metabolism, (b) trials/study groups with mixed interventions other than exercise and low-dosed calcium/cholecalciferol were excluded. We also excluded review articles, case reports, editorials, conference abstracts, letters, preliminary data, or duplicate studies. For the present subgroup analyses, studies (i.e., [12,13,14]) that terminated their intervention 6 months ago and longer were not considered.

Data extraction

During the first step, two reviewers (IH, MS) independently reviewed the titles and abstracts for eligible articles. Subsequently, full-text articles were reviewed by IH and WK. Eligible articles were extracted by IH and WK using a detailed extraction form that asked for study characteristics, study protocol, participant and exercise characteristics, supplementation, and fractures in the EG and CG. In the case of missing information, the authors in question were contacted (n = 6).

Outcome measures

As per FRAX [15], low-trauma fractures of the arm, forearm, or wrist and hip and vertebral fractures were summarized into major osteoporotic fracture as the primary study outcome of the present study. Fractures induced by falls from levels higher than standing and car or bicycle accidents were not included. However, in a minor variation from FRAX, all types of humerus and vertebral fractures were included.

Quality assessment

The Physiotherapy Evidence Database (PEDro) scale risk of bias tool [16] and the TESTEX (Tool for the assEssment of Study qualiTy and reporting in Exercise) score [17] specifically dedicated to evaluate the methodologic quality of physiotherapy and exercise trials was used to rate methodologic quality of the exercise trials.

Data synthesis

Of importance, we pooled the three different exercise groups of Karinkanta et al. [18] into one exercise group. With respect to the study of Bischoff-Ferrari et al. [19], we included the isolated exercise group (without vitamin D) with data provided by the authors. As stated, we focused on two research issues, intensity progression and duration of the exercise study in the sub-analyses. Two reviewers (IH, WK) independently categorized the trials into the subgroups, with full consensus for classification.

Statistical analysis

We used the mixed-effects conditional Poisson regression model suggested by Stijnen et al. [20] for our analysis. We applied R packages metafor [21] included in the statistical software R [22]. The incidences were transformed into incidence rate ratios (IR) along with 95% confidence intervals (95% CI). Heterogeneity between studies was checked using I2 statisticsFootnote 2 [23] in combination with a Wald and likelihood ratio test, respectively. Funnel plots with Kendall’s τ statistic were applied to explore potential small study/publication bias. Subgroup analyses were applied for subgroups as described in data synthesis above. All tests were 2-tailed, and significance was accepted at p < 0.05.

Results

Our search identified 11 eligible studies [18, 19, 24,25,26,27,28,29,30,31,32] (Fig. 1) with a pooled number of participant years of n = 9715 in the EG and n = 9592 in CG. All studies included community dwelling middle-aged to older cohorts.

Fig. 1
figure 1

Flow-chart of the present systematic review according to PRISMA [11]

Full size image

Table 1 gives a summary of the study and participant characteristics. In summary, no relevant between group differences (EG vs. CG) were observed for baseline participant characteristics of the individual studies. Initial sample sizes varied from 27 to 3279 participants/group. All but two studies [24, 32] included Caucasian cohorts on average between 54 ± 3 {Chan, 2004 #8453} and 80 ± 4 {Sakamoto, 2013 #15970} years of age. Seven studies focused exclusively on women. Six studies defined fracture risk as the primary outcome (Table 1).

Table 1 Study and participant characteristics of the included studies
Full size table

Exercise characteristics

Table 2 displays exercise characteristics of the included studies. The exercise program of three studies focused predominately on combined fall prevention/bone strengthening [25, 27, 28] or fall prevention protocols [26, 30, 32], while four studies [18, 24, 29, 31] concentrated on bone strengthening only. Length of the exercise intervention ranged from 6 months [32] to 16 years [28]. Unfortunately, not all of the studies reported the exercise intensity applied for the respective training component adequately and comprehensively (Table 2). With respect to physical interventions in the control group, at least three studies [19, 25, 27] implemented an “active control group.”

Table 2 Exercise characteristics of the included studies
Full size table

Supplementation with vitamin D and/or calcium

Two studies [27, 28] provided calcium (up to 1000 mg/d) supplements for the EG and CG.

Methodological quality

The methodological quality is listed in Table 3. Score points applying PEDro vary between 3 and 9 from a maximum of 10 (9) points and 6–14 from a maximum of 15 when applying the TESTEX score. Of importance, blinding of instructors (i.e., treatment providers) is not applicable in exercise studies; consequently, the maximum score for PEDro should be considered 9 points. In contrast, TESTEX did not score blinding of treatment providers and participants.

Table 3 Methodologic quality of the exercise studies
Full size table

Altogether we observed 151 major osteoporotic fractures (MOF) in the exercise and 196 fractures in the control group. Excluding the follow-up studies, 126 MOF were observed in the EG versus 162 MOF in the CG. In detail, 44 versus 58 hip fractures were recorded in the EG vs. CG; in parallel 62 (EG) vs. 52 (CG) forearm and wrist fractures were reported. Unfortunately, some studies did not report vertebral fractures; thus, the number of 25 fractures in the EG vs. 49 in the CG might be considerably underreported.

Meta-analysis results

The meta-analysis demonstrated a significant (p = 0.006) effect of exercise on major osteoporotic fractures (IR: 0.75; 95% CI: 0.59–0.94) (Figs. 2 and 3, lower part). Heterogeneity between the trials (I2 < 1%) was negligible, and funnel plots and tests for funnel plot asymmetry indicate no relevant evidence for publication/small study bias.

Fig. 2
figure 2

Forest plot of data on the effect of “intensity progression during the trial” on exercise effects on major osteoporotic fracture risk

Full size image
Fig. 3
figure 3

Forest plot of data on the effect of “study/intervention duration” on exercise effects on major osteoporotic fracture risk

Full size image

Subgroup analysis on exercise components

Progression of intensity during the exercise trial

Five studies provided intensity progression in their exercise protocols [18, 26,27,28,29], while another six studies did not change exercise intensity during the intervention (Table 2, Fig. 2). While we observed more favorable effects in the subgroups that applied progression of intensity, in summary, we did not determine a significant difference between the two subgroups (p = 0.133) (Fig. 2). Heterogeneity between the trials was negligible for both subgroups (I2 = 0% and 7.1%).

Duration of the intervention protocol of the exercise trial

Only three studies applied study protocols ≤ 12 months [18, 24, 32], while another eight studies exercised > 12 months to 16 years (Table 2, Fig. 3). In contrast to the shorter studies/exercise interventions (IR: 0.70; 95% CI: 0.23 to 2.15), we observed significant effects for the exercise trials of longer duration (IR: 0.77; 95% CI: 0.61 to 0.97); however, in summary, no relevant differences between the two subgroups (p = 0.883) (Fig. 3) were observed. Heterogeneity between the trials was negligible for both subgroups (I2 = 0%).

Discussion

Apart from generating evidence for the fracture-preventing effect of exercise on low-trauma, major osteoporotic fractures [15], the present systematic review and meta-analysis aimed to determine parameters that might explain the effectiveness of exercise in reducing fractures related to osteoporosis in middle-aged to older adults. In summary the study provided evidence for a significant (major osteoporotic) fracture reducing effect of exercise; however, the sub-analysis on the relevance of intensity progression and exercise duration on this positive interaction did not significantly support the relevance of these important exercise parameters/principles.

Due to the close interaction of exercise parameters [33] and the few exercise trials that focus on definite outcomes of fracture reduction, it is a daunting task to identify key (exercise) parameters for generating meaningful recommendations for promising exercise protocols. This refers especially to the area of fracture reduction with its fundamentally different training strategies on bone strengthening and/or fall reduction [34]. As most exercise parameters (e.g., type of exercise, exercise intensity, training frequency) were confounded by the aspects described above, we focused on the principle of (intensity) progression and the duration of the study intervention because these can be considered superordinate variables of exercise training protocols.

Progression, i.e., the frequent adaptation of training load to persistently apply the overload principle [33], can be realized by changing several parameters including exercise frequency and/or intensity, type of exercise, or exercise duration. However, with few exceptions [27, 28], most of the included exercise trials focused (if at all) on the progression of exercise intensity. In summary, we observed more favorable effects of studies that applied intensity progression (vs. non-progression) on major osteoporotic fracture numbers; nevertheless, differences between the subgroups remained non-significant (Fig. 2). One may argue that progression might be negligible in studies of short duration, but the only study to which this could applied is the 6-month study of Sakamoto et al. [32]. Another reason for our finding might be that progression in particular of balance protocols was rarely reported and the corresponding studies were thus not correctly classified by our approach.

Although this aspect is not negligible for fall prevention studies [2] either, it is outweighed by duration of the study/intervention in exercise programs on bone strengthening due to the length of bone adaptation in adults [35, 36]. Furthermore, along with high sample sizes, study duration is important for generating enough statistical power to address fracture number as a clinical outcome [28]. Of importance only three studies applied exercise protocols of 6 [32] to 12 months [18, 24]. Comparing the latter studies with longer studies (Fig. 3), we observed comparable effects sizes for the two categories.

In summary, we provided further evidence for the (osteoporotic) fracture-reducing effect of exercises; however, we failed to determine key parameters of promising exercise parameters or training principles in this area. We predominately attribute this unfavorable result to the fact that due to participant characteristics,Footnote 3 two fundamentally different exercise strategies, i.e., bone strengthening or fall reduction (or both), can be applied for reducing fracture risk. Considerable differences in addressing the two training aims confounded a joint analysis for meaningful exercise parameters and training principles. In parallel, against our expectation, with its close interaction of exercise parameters, the complexity of exercise might have also confounded our analysis of intensity progression and study duration. Since the methodologically correct approach for addressing this problem, i.e., trials with two exercise arms that differ only in the given component of interest (e.g., exercise frequency; [37]), was not availableFootnote 4 in the domain of fracture reduction, corresponding exercise recommendations have to be derived from more dedicated meta-analyses in the area of osteoporosis [38,39,40] or fall reduction [2, 41] or even better: from randomized controlled trials with similar or comparable training aims and cohorts.

Due to higher evidence standards, more dedicated inclusion criteria, higher fracture risk, and diverging fracture outcomes, it is difficult to set our results into perspective with data on pharmaceutic studies. However, a (very) rough overview on bisphosphonate (risedronate[42],Footnote 5 zoledronate [43], denosumab [44], and teriparatide [45]) effects on fragility fracture incidence indicates results in the area of 20% (risedronate, non-vertebral fractures) to 54% (teriparatide, overall fragility fractures). Our result falls at the lower range; however, it should be borne in mind that all of these pharmaceutic studies focus on secondary preventions, i.e., subjects with a much higher fracture prevention potential. Evidence for a fracture preventing effect in the area of primary prevention is much lower to negligible (e.g., [42]). From a socioeconomic point of view, on the other hand, it would be wrong to conclude that exercise might be a true alternative to pharmaceutical therapy. A large proportion of frail elderly persons, the most vulnerable group for fractures, demonstrate low affinity to exercise [46] and will be hard to persuade to start exercising frequently. Nevertheless, the combination of bone strengthening drugs and fall prevention exercise will definitely be the most promising fracture reduction strategy for this cohort.

Apart from problems described above, other limitations and/or particularities of the present work might have affected our results. (1) We focused on low-trauma fractures and thus excluded fractures caused, for example, by bicycle or car accidents, or falls from levels higher than standing. However, due to unavailable data, we might have not included only “low-trauma fractures.” However, considering that in osteoporosis not only fragility fractures but also all forms of fractures, including high-impact trauma fractures, occur quite frequently, this limitation might be negligible. (2) We further subsume all types of vertebral and humerus fractures under “major osteoporotic fractures,” which is not consistent with FRAX [15]. (3) We included studies with “active control groups” (Table 3) which might have diluted our exercise effects on fracture reduction slightly. (4) We included one non-randomized controlled trial [28]. Although fully aware of potential sources of bias,Footnote 6 we included this study due to its long duration (16 years), the sufficient power to address fracture as an outcome, and “fracture reduction” being stated as the primary outcome. (5) The latter aspect might be highly relevant since studies that focus on BMD effects (“bone-strength,” Table 3) in older people, for example, did not adequately address all the relevant fracture determinants and thus may have generated suboptimum results. (6) Heterogeneity between the trials was consistently negligible (i.e., I2: 0 to 7.1%) among the subgroups (Figs. 2 and 3). Further funnel plot analyses did not indicate evidence for publication/small-study bias. This finding is noteworthy because the studies vary widely with respect to participant (Table 1) and exercise characteristics (Table 2). (7) Finally, the statistical power to address differences between the subgroups can be considered moderate at best. Nevertheless, a recent (meta-)analysis on major osteoporotic fracture reduction that focused on supervision of the exercise program revealed significant differences in favor of supervised exercise protocols {Hoffmann, 2022 #16145}. Thus, the present analysis does not seem to be “hopelessly underpowered.”

Conclusion

Our systematic review and meta-analysis showed a 23% reduction in major osteoporotic fracture incidence, thus providing further evidence for the significant favorable effect of exercise on (low-trauma) fracture reduction. On the other hand, our joint analysis of exercise protocols that focus on bone strengthening, fall reduction, or both did not indicate high relevance of intensity progression or study duration. Along with others [47], we feel that meta-analyses might not be the best choice for deriving promising exercise recommendations, at least for the area of fracture reduction due to the complexity of exercise and the close interaction of exercise parameters. Well-designed and adequately powered randomized controlled trials might be more suitable to address this issue.