Secondary Logo

Journal Logo

Research Series Article: Meta-Analysis

Resistance Training and Bone Mineral Density in Women

A Meta-Analysis of Controlled Trials

Kelley, George A. DA; Kelley, Kristi S. MEd; Tran, Zung Vu PhD

Author Information
American Journal of Physical Medicine & Rehabilitation: January 2001 - Volume 80 - Issue 1 - p 65-77
  • Free

Abstract

Osteopenia and osteoporosis are major public health problems in the United States, affecting primarily lean, white, postmenopausal women. 1 Currently approximately 26.2 million white, postmenopausal women in the United States have either osteopenia or osteoporosis. 1 More specifically, osteopenia, defined as bone density that is 1 to 2.5 SD below the young adult reference range, affects an estimated 16.8 million (54%) of postmenopausal white wo-men in the United States, whereas osteoporosis, defined as bone density >2.5 SD below the young adult reference range, affects another 9.4 million (30%) women. 1, 2 Low-bone density increases the risk for fractures, particularly at the hip, spine, and distal forearm. Currently, the estimated lifetime risk for fracture in 50-yr-old white women in the United States is 17.5% at the hip, 15.6% at the vertebrae, and 16.0% at the distal forearm. 1 In terms of the mortality rate, the survival rate at 5-yr follow-up relative to those of like age and gender is 0.83 for those who have experienced a hip fracture, 0.82 for vertebral fractures, and 1.00 for fractures of the forearm. 3 In the United States, the estimated cost of fractures can be as high as $20 billion per year, with hip fractures accounting for more than a third of the total cost. 4 Furthermore, because of increased life expectancy, the number of women with low-bone density and subsequent fractures is expected to increase substantially in future years.

Physical activity has been suggested as a nonpharmacologic intervention for maximizing bone density during the younger years and preventing the bone loss during the later years. 5 Recent meta-analyses 6, 7 demonstrated the positive effects of aerobic exercise on both lumbar spine and hip bone mineral density (BMD) in postmenopausal women. Another potentially valuable type of physical activity is resistance training. Resistance training is a low-cost, nonpharmacologic intervention that is available to most of the public. Besides the positive effects on the bone, resistance exercise increases lean-body mass, decreases body fat, and increases muscular strength in both adult men and women.

Unfortunately, traditional narrative reviews on the effects of progressive resistance exercise on BMD have led to conflicting results. For example, seven reviews 8–14 have suggested that progressive resistance exercise may have a positive effect on BMD, although nine reviews 15–23 have suggested that progressive resistance exercise does have a positive effect. These discrepancies are not surprising given the fact that intervention studies 24–52 examining the effects of resistance exercise on BMD in adults have led to less than overwhelmingly positive results. For example, for the BMD sites assessed in previously mentioned studies, only 20% were reported as statistically significant. One of the possible reasons for the lack of statistically significant findings may be the result of the low statistical power leading to an increased risk of type 2 errors in some studies. Meta-analysis is a quantitative approach in which individual studies addressing a common problem are statistically aggregated. 53, 54 It is especially useful with a small number of subjects in the studies. 54

As part of a larger study, we 55 previously showed a weight-training-induced improvement of approximately 1% in BMD at all sites combined in postmenopausal women. However, a detailed examination of the effects of progressive resistance exercise on BMD was not conducted. This is also the case with another meta-analysis 56 that combined BMD results from both progressive resistance and aerobic exercise studies. To date, we are unaware of any meta-analysis that has provided a detailed examination of the effects of resistance training on BMD in women. Given the healthcare consequences of low BMD, especially among women, it is important to gain a better understanding of the effects of resistance training on BMD. Thus, the purpose of this study was to use the meta-analytic approach to examine the effects of resistance training on BMD in women.

METHODS

Data Sources

We performed computerized literature searches of articles indexed between January 1966 and December 1998 using MEDLINE, Current Contents, Sport Discus, and Dissertation Abstracts International databases. The following keywords were used either alone or in combinations for computer searches: bone, bone density, bone mineral density, exercise, physical activity, women, females, physical fitness, fitness, weight training, resistance exercise, resistance training, osteoporosis, and osteopenia. The titles and abstracts of studies identified in the computerized searches were examined to exclude irrelevant studies. We retrieved the full text of the remaining articles and we read each paper to determine whether it contained information on the topic of interest. Because computer searches have been shown to yield less than two-thirds of relevant articles, 57 the reference lists from both original and review articles were also reviewed to locate studies that had not been previously identified and which seemed to contain information on the topic of interest. In addition, we also hand searched selected journals. Furthermore, three experts on exercise and BMD (Drs. Charlotte Sanborn, David Nichols, and Christine Snow) reviewed our reference list and coding sheet for thoroughness and completeness.

Study Selection

Inclusion criteria for this study were as follows: (1) randomized or nonrandomized trials that included a comparative nonexercise control group or control period; (2) resistance training, defined as any external resistance added while performing exercises, as the only intervention; (3) adult female humans (mean study age, ≥18 yr) as subjects; (4) journal articles, dissertations, and master’s theses published in the English-language literature; (5) studies published and indexed between January 1966 and December 1998; (6) BMD (relative value of bone mineral per measured bone area) assessed at the femur, lumbar spine, or radius; (7) training studies lasting a minimum of 16 wk. Only information that met the above criteria was included in our analysis. Thus, for example, if BMD was also assessed in women performing aerobic exercise, we did not include this information because it did not meet our inclusion criteria. We limited our analysis to the femur, lumbar spine, and radius because they are the most often studied and these areas are the most vulnerable to fracture. Because dissertations may eventually become full-length journal articles, we cross-referenced between the two to avoid duplication. We did not include abstracts and conference papers from national meetings because of the paucity of data provided as well as the inability to obtain complete data from the authors. Studies published in foreign language journals were also not included because of the potential error in the translation and interpretation of findings. Studies that met our inclusion criteria were also examined to ensure that the same subjects were not included in more than one study. 58 For studies that met our inclu-sion criteria but did not provide appropriate information on changes in BMD, 59, 60 we personally tried to contact the authors to retrieve such information.

Data Extraction

Coding sheets that could hold 242 items were developed and used in this investigation. In addition, coding instructions that described how to code each item on the coding sheet were developed and used. To avoid coding bias, all data were extracted independently by two authors. The authors then met and reviewed every item for accuracy and consistency. Disagreements were resolved by consensus. Blinding of coders to study information in relation to the identity and institutional affiliation of the study authors, as well as study results, were not performed because, according to a recent work, 61 these procedures have neither a clinically nor statistically significant effect on the results. The major categories of variables coded included study characteristics, physical characteristics of subjects, and primary and secondary outcomes.

Statistical Analysis

Primary Outcomes.

The primary outcomes in this study were changes in BMD at the femur, lumbar spine, and radius. Because of the various ways in which the authors reported data on changes in BMD and because we also wanted to maximize the number of studies and outcomes that could be included in our analysis, we used the standardized difference approach as our effect size (ES) measure. 62 This was calculated by subtracting the change outcome in the exercise group from the change outcome in the control group, and then dividing this difference by the pooled standard deviation of the exercise and control groups. 62 The ES was then corrected for small-sample bias. 62 For studies that included multiple outcomes because of more than one group (for example, an exercise group that trained at a higher intensity vs. one that trained at a lower intensity), net changes in bone mineral density were treated as independent data points. 63 In general, an ES of 0.20 was considered a small effect, 0.50 a moderate effect, and 0.80 a large effect. 64 An ES of 0.20, for example, means that the exercise group differed from the control group by two-tenths of a standard deviation in favor of the exercise group. Because of the small-sample size in this study, especially for subgroup analyses, bootstrap resampling (5,000 iterations) was used to generate 95% BCIs around mean ES changes for BMD. 65 The bootstrap technique is a computer-intensive, nonparametric method of estimating the reliability of the original sample estimate, in this case, ES changes in BMD. By randomly drawing from the available sample, with replacement, samples the same size as the original are generated. Each time an observation is selected for a new sample, each of the elements of the original sample has an equal chance of being selected. This is similar to replicating each member of a sample 5,000 times (iterations). The main advantage of this approach is that the estimate desired is not based on some theoretical distribution, but rather, on the sample itself. This approach frees one from the constraints of the central limit theorem. The number of iterations chosen was based on previous research demonstrating that improvement of estimation accuracy was limited beyond 5,000 iterations. 66 If the 95% confidence interval included zero (0.00), it was concluded that there was no statistically significant effect of exercise on BMD.

Heterogeneity of ES changes in BMD was examined using the Q statistic. 62 A random-effects model was used when changes were significantly heterogeneous (P < 0.05), whereas a fixed-effects model was used in the absence of significant heterogeneity. 53 For studies that included multiple outcomes because of more than one group, net changes were treated initially as independent data points. However, to examine the influence (sensitivity) of each study on the overall results, analyses were performed with each study deleted from the model.

Publication bias (the tendency for journals to publish studies that yield statistically significant results and/or authors to only submit studies that yield statistically significant results) was examined using Kendall’s tau statistic (τ). 67 A statistically significant result (P < 0.05) was considered to be suggestive of publication bias.

Study quality was assessed using a three-item questionnaire designed to assess bias, specifically, randomization, blinding, and withdrawals/dropouts. 68 The number of points possible ranged from a low of 0 to a high of 5. All questions were designed to elicit yes (1 point) or no (0 point) responses. The questionnaire took less than 10 min per study. The questionnaire has been shown to be both valid (face validity) and reliable (researcher interrater agreement, r = 0.77, 95% confidence interval = 0.60–0.86). 68

Subgroup Analyses.

For categorical variables, subgroup analyses for primary outcomes were performed using analysis of variance-like procedures for meta-analysis. 62 These procedures provide statistics for both within (Qw) and between (Qb) group differences. If statistically significant within-group (Qw) heterogeneity existed (P < 0.05), a random-effects model was used. If no statistically significant within-group (Qw) heterogeneity existed, a fixed-effects model was used. ES changes in BMD were examined initially when the data were partitioned according to study design (randomized vs. nonrandomized), country in which the study was conducted (United States vs. other), study quality (0–2 vs. 3–5), menopausal status (pre vs. post), calcium supplementation, changes in dietary intake during the study, drugs that could affect BMD, and physical activity habits of subjects. For the femur site, we also examined changes in BMD with data partitioned according to the femoral neck, trochanter, intertrochanter, and Ward’s triangle. We were unable to examine specific sites at the lumbar spine and radius because of insufficient data. Bootstrap resampling (5,000 iterations) was used to generate 95% confidence intervals around ES changes for all subgroups. Randomization tests (5,000 iterations) were used to generate probability values for between-group differences. 69 Randomization tests using 5,000 iterations can detect a probability as low as 0.002. 69

Regression Analysis.

For continuous variables, potential associations with ES changes in BMD were conducted using meta-regression procedures, calculated with each ES weighted by the reciprocal of its variance, according to procedures described by Hedges and Olkin. 62 This model yields a test of the significance of each predictor (QR) as well as a test of model specification (QE) which assesses whether systematic variation remains unexplained in the regression model. Thus, a statistically significant QR value means that the variables included in the regression are significantly related to the variable of interest, whereas a nonsignificant QE value means that the model is well specified. Continuous variables that were examined included percentage of dropout (number of subjects who did not complete the study), age, height, initial as well as changes in body weight, body mass index, percentage of body fat and lean-body mass, changes in muscular strength, initial BMD, calcium intake, years postmenopausal, length and intensity of training, number of exercises performed, and compliance, defined as the percentage of exercise sessions attended by the subjects.

Secondary Outcomes.

Secondary outcomes (changes in body weight, body mass index, percentage of body fat, and lean-body mass) were calculated as the difference (exercise minus control) of the changes (initial minus final) in these mean values. The original metric was used for all secondary outcomes. For those studies in which variance estimation was necessary, these were accomplished using the same procedures as those for estimating variances for BMD. 63 Fixed and random effects models were used following the same procedures as those previously described for BMD. Percentage of changes in muscular strength (one repetition maximum) were reported separately for exercise and control groups.

Unless otherwise noted, all results are reported as mean ± SD. The α level for statistical significance was set at P < 0.05. Values between 0.05 and 0.10 were considered as a trend toward statistical significance. Bonferroni adjustments were not made because of the increased risk of a type 2 error.

RESULTS

Study Characteristics

Thirty-one studies met the criteria for inclusion. 24–52, 59, 60 However, we were unable to include two studies 59, 60 because of the inability to obtain data necessary for the calculation of an ES. Thus, we had a 6% loss that met our inclusion criteria. One study 70 was excluded because it included some of the same subjects from another study that we included. 40 A general description of the 29 included studies is shown in Table 1 and the physical characteristics of the exercise and control group subjects are described in Table 2. The per person time to code each study once ranged from 0.58 to 4.67 hr (1.26 ± 0.79 hr). Study quality ranged from 1 to 4 (2 ± 1). The 29 included studies represented 94 ES (femur = 53, lumbar spine = 24, radius = 17) from 61 groups (32 exercise, 29 control). Twenty-three studies were published in journals, 25, 27–32, 34–38, 40, 42–50, 52 five were dissertations, 24, 26, 33, 39, 41 and one was a master’s thesis. 51 Twenty studies were conducted in the United States, 24, 26–28, 33, 34, 36–39, 41–50 three in Finland, 30, 31, 52 two each in Austria 29, 40 and Canada, 25, 51 and one each in Australia 32 and France. 35 Percentage of dropout, defined as the number of subjects who did not complete the study, ranged from 0 to 63% in the exercise groups (28 ± 17%) and 0 to 69% in the control groups (17 ± 18%). Thus, pre and post measures of BMD were available for 572 subjects who served as exercisers and 551 subjects who served as controls. The minimum and maximum number of subjects in the exercise groups was 6 and 46 (18 ± 10), respectively, whereas the minimum and maximum number of subjects in the control groups was 7 and 42 (19 ± 10), respectively. For the 14 studies that reported information on race, 12 reported that all of the subjects were white, 26, 28, 33–36, 40, 43, 45–48 one study reported that the subjects were white and black, 42 and another reported that the subjects were white and Asian. 51 For the 18 studies that reported information on calcium supplementation during the study, eight studies reported that some subjects were taking supplements, 33, 36–38, 43, 46, 48, 51 seven reported that all of the subjects were taking supple ments, 24, 26–28, 34, 44, 49 and three reported that none of the subjects were taking supplements. 35, 39, 47 For the 23 studies that reported on whether subjects were taking any type of pharmacologic interventions that could affect BMD, 14 reported that none were taking any pharmacologic interventions, 26, 27, 32–37, 40, 42, 44, 46, 47, 51 eight reported that some were, 25, 28, 30, 31, 39, 43, 45, 48 and one study reported that all were. 38 Ten studies reported that none of the subjects smoked cigarettes, 25, 31, 33, 36, 39, 40, 44–47 whereas four reported that some subjects smoked.28, 35, 48, 51 Two studies reported that some of the subjects consumed alcohol.25, 48 Ten studies reported that none of the subjects had been previously active, 25, 26, 29, 31, 33, 34,36, 39, 40, 43 eight reported that some were, 24, 28, 32, 35, 44, 48–50 and five reported that all were. 30, 37, 46, 51, 52 Five studies reported that none of the subjects had suffered previous fractures, 29, 39, 43, 46, 47 whereas three reported that some had. 33, 36, 40 Compliance, defined as the percentage of resistance training sessions that the exercise groups attended, ranged from 44% to 96% (79 ± 13%). Reliability for BMD assessment (coefficient of variation) ranged from approximately 0.6% to 4% at the femur, 0.6% to 5.0% at the lumbar spine, and 0.5% to 5% at the radius.

Table 1
Table 1:
Study characteristics
Table 1A
Table 1A:
Continued
Table 1B
Table 1B:
Continued
Table 2
Table 2:
Initial physical characteristics of subjects

Primary Outomes

Initial BMD values for exercise and controls are shown in Table 3, whereas ES changes in BMD are shown in Table 4. BMD values were available for a total of 743 subjects at the femur (392 exercise, 351 control), 870 at the lumbar spine (450 exercise, 420 control), and 441 at the radius (219 exercise, 222 control). Because there was no statistically significant heterogeneity at any of the sites observed, a fixed-effects model was used for overall results at all three sites.

Table 3
Table 3:
Initial BMD values
Table 4
Table 4:
BMD results

Proximal Femur.

Small and statistically insignificant changes in BMD were observed at the femur site. These changes were equivalent to a 0.33% increase in the exercise groups and a 0.05% decrease in the control groups. No evidence of publication bias was observed (r = 0.12, P = 0.26). With each study deleted from the model once, ES changes in BMD at the femur ranged from a low of 0.02 ± 0.37 (95% Bootstrap Confidence Interval [BCI], −0.07–0.11) to a high of 0.09 ± 0.36 (95% BCI, 0.03–0.17). Approximately 90% of the 53 ESs were reported by the authors of the original studies as not being statistically significant.

Lumbar Spine.

Small but statistically significant ES changes in BMD were found at the lumbar spine. These changes were equivalent to a 0.19% decrease in the exercise groups and a 1.45% decrease in the control groups. No evidence of publication bias was observed (r = −0.08, P = 0.62). With each study deleted from the model once, ES changes in BMD ranged from a low of 0.19 ± 0.37 (95% BCI, 0.09–0.33) to a high of 0.27 ± 0.36 (95% BCI, 0.14–0.41). Approximately 67% of the 24 ESs were reported by the authors of the original studies as not being statistically significant.

Radius.

Small and statistically significant ES changes in BMD were observed at the radius. ES changes were equivalent to a 1.22% increase in BMD for the exercise groups and a 0.95% decrease in the control groups. No evidence of publication bias was observed (r = 0.17, P = 0.38). With each study deleted from the model once, ES changes in BMD at the radius ranged from a low of 0.19 ± 0.36 (95% BCI, 0.03–0.45) to a high of 0.33 ± 0.34 (95% BCI, 0.16–0.52). Approximately 65% of the 17 ESs were reported by the authors of the original studies as not being statistically significant.

Subgroup Analysis

Subgroup analyses for those variables in which there were statistically significant differences or trends for statistically significant differences between groups are shown in Table 5.

Table 5
Table 5:
Subgroup analyses

Femur.

There was a trend for greater ES changes in BMD at the femur when studies were of higher vs. lower quality. Higher-quality studies yielded ES changes that were equivalent to a 1.03% increase in BMD in the exercise groups and a 0.16% increase in the control groups. Lower-quality studies yielded ES changes that were equivalent to a 0.21% increase in the exercise groups and a 0.09% decrease in the control groups. There was also a trend for greater ES changes in BMD at the femur when subjects were postmenopausal vs. premenopausal. For postmenopausal women, ES changes in BMD were equivalent to a 0.40% increase in the exercise groups and a 0.21% decrease in the controls. For premenopausal women, ES changes were equivalent to a 0.26% increase in the exercise groups and a 0.13% increase in the control groups. No statistically significant between-group differences were found when data were partitioned according to study design, country in which the study was conducted, calcium supplementation, previous physical activity habits, type of BMD assessment, and different sites at which BMD was assessed. Insufficient data were available to examine differences in BMD at the femur when data were partitioned according to diet as well as drugs that could affect BMD.

Lumbar Spine.

No statistically significant between-group differences were observed for ES changes at the lumbar spine when data were partitioned according to source of study, country in which the study was conducted, study design, menopausal status of subjects, calcium supplementation, previous physical activity, and type of BMD assessment. Insufficient data were available to examine between-group differences in BMD when data were partitioned according to study quality, drugs that could affect BMD, diet, and sites at which the lumbar spine BMD was assessed.

Radius.

There was a trend for greater ES changes in BMD at the radius when studies were of higher vs. lower quality. Higher-quality studies yielded ES changes that were equivalent to a 0.82% increase in BMD in the exercise groups and a 1.87% decrease in the control groups. Lower-quality studies yielded ES changes that were equivalent to a 1.75% increase in BMD in the exercise groups and a 0.23% increase in the control groups. ES changes at the radius were also greater in postmenopausal vs. premenopausal women. For postmenopausal women, ES changes were equivalent to a 1.71% increase in BMD in the exercise groups and a 1.39% decrease in the control groups. For premenopausal women, ES changes were equivalent to a 0.17% increase in the exercisers and a 0.01% increase in the controls. No statistically significant differences were observed when data were partitioned according to source of study, country in which the study was conducted, study design, previous physical activity habits, and type of BMD assessment. Insufficient data were available to examine between-group differences in BMD when data were partitioned according to calcium supplementation, drugs that could affect BMD, diet, and different sites at which BMD of the radius was assessed.

Regression Analyses

Femur.

The only significant predictor for ES changes in BMD at the femur was changes in the percentage of fat (QR = 6.67, P = 0.03; QE = 14.32, P = 0.35). Larger ES changes in BMD at the femur were observed among subjects with smaller changes in the percentage of fat. No other statistically significant associations were observed.

Lumbar Spine

No significant predictors were observed for ES changes in BMD at the lumbar spine.

Radius.

The only significant predictor for ES changes in BMD at the radius was initial lean-body mass (QR = 6.76, P = 0.009; QE = 9.26, P = 0.41). Smaller ES changes in BMD at the radius were observed among subjects with higher initial levels of lean-body mass. Insufficient data were available to examine the relationship between ES changes in BMD and changes in the percentage of body fat and lean-body mass.

Secondary Outcomes

Statistically significant decreases were observed for the percentage of body fat (−2 ± 2%; 95% BCI, −3 to −1%), whereas there was a statistically significant increase in lean-body mass (2 ± 1 kg; 95% BCI, 1–2 kg). No statistically significant changes were observed for body weight or body mass index. There was a 40% increase in muscular strength in the exercise groups and a 6% increase in the control groups.

DISCUSSION

Implications for Practice

The overall results of this study suggest that across all groups of women included in this analysis, resistance training helps to preserve lumbar spine BMD. Resistance training also seems to increase and preserve BMD at the femur and radius sites in postmenopausal women. Furthermore, with the exception of changes in BMD at the proximal femur, these results were consistent after deletion of each study once from our models.

An interesting finding of this study is the fact that the largest effect on BMD occurred at the radius site in postmenopausal women. One possible reason for this may be the fact that most subjects included in these studies were able to ambulate. Consequently, they may have had greater daily loading placed on the lumbar spine and femur vs. the radius before participation in the studies. Therefore, there may have been an opportunity for resistance training to have a greater effect on BMD at the radius vs. the lumbar spine and femur. However, it may also be that the resistance training programs placed greater relative loads on the radius vs. the lumbar spine and femur sites. The larger changes observed in BMD at the femur when changes in the percentage of fat were smaller as well as the smaller changes at the radius when initial lean-body mass was higher are supportive of the fact that in general, women who weigh more place greater stress on their bones. Thus, heavier women may not experience the same improvements in BMD as leaner women.

Although it seems that postmenopausal women may have the most to gain from a program of resistance training, this form of intervention should almost always be encouraged across all age groups, especially because of other benefits that can be derived from participation in such activities. For example, in this investigation, we saw statistically significant improvements in body composition (decreases in the percentage of body fat and increases in lean-body mass). However, we believe that it is unrealistic to think that any optimal training program (resistance, exercises, sets, repetitions, length of rest intervals, total workload) will ever be developed for maximizing BMD. The best that can occur is some minimal levels to achieve the desired changes. However, even these recommendations are imprecise. For example, despite the various training protocols used in the studies included in this meta-analysis, the deletion of each study once from the analysis had little effect on the overall results. Thus, the best recommendation we can make at this time is to adhere to the general principles of specificity and overload when prescribing resistance training programs aimed at maintaining and/or improving BMD. 5

Although it is encouraging that resistance training seems to have positive effects on BMD at the lumbar spine, femur, and radius, the clinical importance of such small effects is not known, especially as it relates to fracture risk. We are not aware of any randomized trial(s) that have proven that resistance training reduces the risk of fracture. However, it may be that other factors contribute to increases in bone strength and subsequent reductions in fracture risk. For example, a recent animal study 71 found that mechanical loading improves bone strength by reshaping the bone structure with no apparent increase in BMD. Thus, resistance training may have a similar effect in humans.

Because most of the studies included in this meta-analysis examined the efficacy (does the treatment work?) of resistance training for enhancing BMD in women, the effectiveness (does the treatment work in the real world?) of such an intervention could be questioned. This may be especially important given the fact that in the United States only 16% of people between the ages of 18 and 64 yr report regular participation in progressive resistance exercise. 72 It may be that other forms of therapy (calcium and/or vitamin D supplementation, hormone replacement therapy, selective estrogen receptor modulators, bisphosphonates) not only have a greater impact on BMD, but they also reduce the risk of fracture. For example, a recent meta-analysis 73 examined over a 3-yr period the effects of 10 mg of alendronate on BMD in osteoporotic women between the ages of 42 and 85 yr. The authors reported increases in BMD of 8.8% at the spine, 7.8% at the trochanter, and 5.9% at the femoral neck. The estimated cumulative incidence of nonvertebral fractures after 3 yr was 12.6% in the placebo group and 9.0% in the alendronate-treated group. It was concluded that administration of alendronate reduces the risk of nonvertebral fractures in osteoporotic postmenopausal women. Given the former, resistance training in conjunction with other types of nonpharmacologic and/or pharmacologic therapy may be most appropriate, especially for those women with osteoporosis.

Implications for Research

One of the surprising findings of this study was the fact that changes in BMD were greater in studies of higher quality. It is generally believed that studies of higher quality yield less positive results than studies of lower quality. For example, a recent study, 74 using the same quality rating scale as ours, examined the impact of study quality on outcomes in placebo-controlled trials of homeopathy. These authors 74 concluded that studies of higher methodologic quality produced less positive results. However, it may be possible that trials with good designs reduce random variability and allow the intervention to produce a larger ES. This may have been the case with our investigation.

The fact that we included both randomized and nonrandomized controlled trials in our study could be questioned. It is generally felt that randomized trials yield results that are more conservative when compared with nonrandomized trials. However, because we did not find a statistically significant difference between any of our outcomes when the data were partitioned by study design, we felt it was appropriate to include both in our analysis.

Although it is important to conduct many statistical tests when performing a meta-analysis, some of our statistically significant results may have been the result of chance vs. any real effect. However, we believe that a greater risk existed of committing a type 2 error if Bonferroni adjustments were made to our data. Thus, our data were analyzed without any type of Bonferroni adjustments.

Although some may feel that the inclusion of dissertations and master’s theses which have not been published as journal articles is inappropriate because they lack the same “rigor,” we believe that it is critical, given appropriate resources, to include such because of the reported publication bias that has been shown to exist in the literature. 75, 76 For example, Stern and Simes 76 found that approximately 96% of selected psychology journals and 85% of selected medical journals published studies that yielded a statistically significant result. The inclusion of unpublished data represents a feeling that is shared by the majority of meta-analysts and methodologists, as a study by Cook and colleagues 77 has shown that approximately 80% feel that unpublished material such as dissertations and master’s theses should definitely or probably be included in scientific overviews.

Despite the knowledge that studies can be more objectively evaluated using the meta-analytic vs. traditional, narrative approach, potential problems still exist. In general, the very nature of meta-analysis dictates that the meta-analysis itself inherits those limitations that exist in the literature. Therefore, the meta-analyst must point out these limitations and provide directions for future research. One of the common problems in meta-analysis is the issue of missing data for outcomes other than the primary ones of interest. For example, the fact that insufficient data were available to perform subgroup analysis on BMD at different lumbar and radius sites could have impacted our results. Although the inability to compare BMD at different lumbar and radius sites was more a function of a lack of sample size vs. the absence of reporting such information, additional studies directed at these sites would seem appropriate. In addition, we would suggest that future studies dealing with the effects of resistance training on BMD in women do a better job of assessing and reporting on the dietary habits of their subjects as well as the types of pharmacologic interventions that these subjects may be taking. Furthermore, because few studies included an assessment of the alcohol and calcium intake of the subjects, greater attention to these in the future seem warranted. It is also recommended that future studies include an evaluation of their data using both an analysis-by-protocol as well as an intention-to-treat approach. As a result, one may examine both the efficacy and effectiveness of resistance training for enhancing BMD in women. This will help provide clinicians with more meaningful information regarding the use of resistance training for enhancing BMD in women. Additional information regarding appropriate study design when examining the effects of exercise on BMD may be found in the excellent review of Snow et al. 78 Finally, it would seem plausible to suggest that a need exists for a large randomized trial that examines the effect of resistance training on both BMD and fracture risk. However, a trial of this nature may never be successfully conducted.

In conclusion, the results of this meta-analysis suggest that resistance training has a positive effect on the BMD of all women at the lumbar spine, and in postmenopausal women at the femur and radius.

REFERENCES

1. Melton LJ: How many women have osteoporosis now? J Bone Miner Res 1995; 10: 175–7
2. Assessment of Fracture Risk and Its Application to Screening for Postmenopausal Women. WHO Technical Series Report. Geneva, World Health Organization, 1994
3. Cooper C, Atkinson EJ, Jacobsen SJ, et al: Population-based study of survival following osteoporotic fractures. Am J Epidemiol 1993; 137: 1001–5
4. Praemer A, Furner S, Rice DP: Musculoskeletal Conditions in the United States. Park Ridge, IL, American Academy of Orthopaedic Surgeons, 1992
5. Drinkwater BL: Does physical activity play a role in preventing osteoporosis? Res Q Exerc Sport 1994; 65: 197–206
6. Kelley GA: Aerobic exercise and lumbar spine bone mineral density in postmenopausal women: a meta-analysis. J Am Geriatr Soc 1998; 46: 143–52
7. Kelley GA: Aerobic exercise and bone density at the hip in postmenopausal women: a meta-analysis. Prev Med 1998; 27: 798–807
8. American College of Sports Medicine: Osteoporosis and exercise. Med Sci Sports Exerc 1995; 27: i–vii
9. Adachi JD: Current treatment options for osteoporosis. J Rheumatol 1996; 23 (suppl 45): 11–4
10. Bouxsein ML, Marcus R: Overview of exercise and bone mass. Rheum Dis Clin North Am 1994; 20: 787–802
11. Gutin B, Kasper MJ: Can vigorous exercise play a role in osteoporosis prevention? A review. Osteoporos Int 1992; 2: 55–69
12. Inoue T, Kushida K, Kobayashi G, et al: Exercise therapy for osteoporosis. Osteoporos Int 1993; Suppl 1: S166–S168
13. Sinaki M: Exercise and osteoporosis. Arch Phys Med Rehabil 1989; 70: 220–9
14. Snow-Harter C, Marcus R: Exercise, bone mineral density, and osteoporosis. Exerc Sport Sci Rev 1991; 10: 351–88
15. Bellantoni MF: Osteoporosis prevention and treatment. Am Fam Physician 1996; 54: 986–92
16. Birge SJ, Dalsky G: The role of exercise in preventing osteoporosis. Public Health Rep 1989; 104 (suppl 1): 54–8
17. Chilibeck PD, Sale DG, Webber CE: Exercise and bone mineral density. Sports Med 1995; 19: 103–22
18. Forwood MR, Burr DB: Physical activity and bone mass: exercises in futility? J Bone Miner Res 1993; 21: 89–112
19. Lewis RD, Modlesky CM: Nutrition, physical activity, and bone health in women. Int J Sports Med 1998; 8: 250–84
20. Prior JC, Barr SI, Chow R, Faulkner RA: Physical activity as therapy for osteoporosis. Can Med Assoc J 1996; 155: 940–4
21. Recker RR, Dowd R, Gale JR, et al: Patient care of osteoporosis. Clin Geriatr Med 1995; 11: 625–40
22. Rutherford OM: Bone density and physical activity. Proc Nutr Soc 1997; 56: 967–75
23. Sheth P: Osteoporosis and exercise: a review. Mt Sinai J Med 1999; 66: 197–200
24. Bouxsein ML: Physical Activity and Bone Density (dissertation). Palo Alto, CA, Stanford University, 1992
25. Chilibeck PD, Calder A, Sale DG, et al: Twenty weeks of weight training increases lean tissue mass but not bone mineral mass or density in healthy, active young women. Can J Physiol Pharmacol 1996; 74: 1180–5
26. Delaney TA: Association of Insulin-Like Growth Factor-I with Body Composition, Diet and Bone Mineral Indices in Weight Resistance-Trained Premenopausal Women (dissertation). Davis, University of California at Davis, 1991
27. Dornemann TM, McMurray RG, Renner JB, et al: Effects of high-intensity resistance exercise on bone mineral density and muscle strength of 40–50-year-old women. J Sports Med Phys Fitness 1997; 37: 246–51
28. Gleeson PB, Protas EJ, LeBlanc AD, et al: Effects of weight lifting on bone mineral density in premenopausal women. J Bone Miner Res 1990; 5: 153–8
29. Hartard M, Haber P, Ilieva D, et al: Systematic strength training as a model of therapeutic intervention. Am J Phys Med Rehabil 1996; 75: 21–8
30. Heinonen A, Sievanen H, Kannus P, et al: Effects of unilateral strength training and detraining on bone mineral mass and estimated mechanical characteristics of the upper limb bones in young women. J Bone Miner Res 1996; 11: 490–501
31. Heinonen A, Oja PSH, Pasanen M, et al: Effect of two training regimens on bone mineral density in healthy perimenopausal women: a randomized controlled trial. J Bone Miner Res 1998; 13: 483–90
32. Kerr D, Morton A, Dick I, et al: Exercise effects on bone mass in postmenopausal women are site-specific and load-dependent. J Bone Miner Res 1996; 11: 218–25
33. Little KD: Effect of Exercise Mode on Bone Mineral Mass in Recently Postmenopausal Women (dissertation). Kent, OH, Kent State University, 1992
34. Lohman T, Going S, Pamenter R, et al: Effects of resistance training on regional and total bone mineral density in premenopausal women: a randomized prospective study. J Bone Miner Res 1995; 10: 1015–24
35. Mayoux-Benhamou MA, Bagheri F, Roux C, et al: Effect of psoas training on postmenopausal lumbar bone loss: a 3-year follow-up study. Calcif Tissue Int 1997; 60: 348–53
36. Nelson ME, Fiatarone MA, Morganti CM, et al: Effects of high-intensity strength training on multiple risk factors for osteoporotic fractures: a randomized controlled trial. JAMA 1994; 272: 1909–14
37. Nichols JF, Nelson KP, Peterson KK, et al: Bone mineral density responses to high-intensity strength training in active older women. J Aging Physical Activity 1995; 3: 26–38
38. Notelovitz M, Martin D, Tesar R, et al: Estrogen therapy and variable resistance weight training increase bone mineral in surgically menopausal women. J Bone Miner Res 1991; 6: 583–90
39. Payne SG: The Effects of Weight Training on Bone Mineral Density of Premenopausal Females (dissertation). Denton, Texas Woman’s University, 1995
40. Preisinger E, Alacamlioglu Y, Pils K, et al: Exercise therapy for osteoporosis: results of a randomised controlled trial. Br J Sports Med 1996; 30: 209–12
41. Protiva KW: Weighted Vest Exercise Improves Functional Ability in Women Over 75 Years of Age (dissertation). Corvallis, OR, Oregon State University, 1997
42. Pruitt LA, Jackson RD, Bartels RL, et al: Weight-training effects on bone mineral density in early postmenopausal women. J Bone Miner Res 1992; 7: 179–85
43. Pruitt LA, Taafe DR, Marcus R: Effects of a one-year high intensity versus low-intensity resistance training program on bone mineral density in older women. J Bone Miner Res 1995; 10: 1788–95
44. Rockwell JC, Sorenson AM, Baker S, et al: Weight training decreases vertebral bone density in premenopausal women: a prospective study. J Clin Endocrinol Metab 1990; 71: 988–93
45. Shaw JM, Snow CM: Weighted vest exercise improves indices of fall risk in older women. J Gerontol 1998; 53: M53–M58
46. Sinaki M, Wahner H, Bergstrahl E, et al: Three-year randomized trial of the effect of dose-specified loading and strengthening exercises on bone mineral density of spine and femur in nonathletic, physically active women. Bone 1996; 19: 233–44
47. Sinaki M, Wahner HW, Offord KP, et al: Efficacy of nonloading exercises in prevention of vertebral bone loss in postmenopausal women: a controlled trial. Mayo Clin Proc 1989; 64: 762–9
48. Smidt GL, Lin S-Y, O’Dwyer KD, et al: The effect of high-intensity trunk exercise on bone mineral density of postmenopausal women. Spine 1992; 17: 280–5
49. Snow-Harter C, Bouxsein ML, Lewis BT, et al: Effects of resistance and endurance exercise on bone mineral status of young women: a randomized exercise intervention trial. J Bone Miner Res 1992; 7: 761–9
50. Taaffe DR, Robinson TL, Snow CM, et al: High-impact exercise promotes bone gain in well-trained female athletes. J Bone Miner Res 1997; 12: 255–60
51. Thorvaldson CL: The Effects of a Specific Weight-Training Exercise Program and Hormone Replacement Therapy on Bone Mass in Healthy Postmenopausal Women (thesis). Edmonton, Alberta, University of Alberta, 1990
52. Vuori I, Helnonen A, Slevanen H, et al: Effects of unilateral strength training and detraining on bone mineral density and content in young women: a study of mechanical loading and deloading on human bones. Calcif Tissue Int 1994; 55: 59–67
53. Mosteller F, Colditz GA: Understanding research synthesis (meta-analysis). Annu Rev Public Health 1996; 17: 1–23
54. Petitti DB: Meta-Analysis, Decision Analysis, and Cost-Effectiveness Analysis: Methods for Quantitative Synthesis in Medicine. New York, Oxford University Press, 1994
55. Kelley GA: Exercise and regional bone mineral density in postmenopausal women: a meta-analytic review of randomized trials. Am J Phys Med Rehabil 1998; 77: 76–87
56. Berard A, Bravo G, Gauthier P: Meta-analysis of the effectiveness of physical activity for the prevention of bone loss in postmenopausal women. Osteoporos Int 1997; 7: 331–7
57. Dickersin K, Hewett P, Mutch L, et al: Perusing the literature: comparison of Medline searching with a perinatal trials database. Controlled Clin Trials 1985; 6: 306–17
58. Egger M, Davey Smith G: Meta-analysis: bias in location and selection of studies. BMJ 1998; 316: 61–6
59. McCartney N, Hicks AL, Martin J, et al: Long-term resistance training in the elderly: effects on dynamic strength, exercise capacity, muscle, and bone. J Gerontol 1995; 50A: B97-B104
60. Kohrt WM, Ehsani AA, Birge SJ: Effects of exercise involving predominantly either joint-reaction or ground-reaction forces on bone mineral density in older women. J Bone Miner Res 1997; 12: 1253–61
61. Berlin JA: Does blinding of readers affect the results of meta-analyses? Lancet 1997; 350: 185–6
62. Hedges LV, Olkin I: Statistical Methods for Meta-Analysis. San Diego, CA, Academic Press, 1985
63. Follmann D, Elliot P, Suh I, et al: Variance imputation for overviews of clinical trials with continuous response. J Clin Epidemiol 1992; 45: 769–73
64. Cohen J: A power primer. Psychol Bull 1992; 112: 155–9
65. Efron B, Tibshirani R: An Introduction to the Bootstrap. London, Chapman and Hall, 1993
66. Zhu W: Making bootstrap statistical inferences: a tutorial. Res Q Exerc Sport 1997; 68: 44–55
67. Begg CB: Publication bias, in Cooper H, Hedges LV (eds): The Handbook of Research Synthesis. New York, Russell Sage, 1994, pp 399–409
68. Jadad AR, Moore RA, Carroll D, et al: Assessing the quality of reports of randomized clinical trials: is blinding necessary? Controlled Clin Trials 1996; 17: 1–12
69. Potvin C, Roff D: Distribution-free and robust statistical methods: viable alternatives to parametric statistics? Ecology 1993; 74: 1617–28
70. Preisinger E, Alacamlioglu Y, Pils K, et al: Therapeutic exercise in the prevention of bone loss: a controlled trial with women after menopause. Am J Phys Med Rehabil 1995; 74: 120–3
71. Jarvinen TLN, Kannus P, Sievanen H, et al: Randomized controlled study of effects of sudden impact loading on rat femur. J Bone Miner Res 1998; 13: 1475–82
72. National Center for Health Statistics: Healthy People 2000 Review, 1997. Hyattsville, MD, Public Health Service, 1997
73. Karpf DB, Shapiro DR, Seeman E, et al: Prevention of nonvertebral fractures by alendronate: a meta-analysis. JAMA 1997; 277: 1159–64
74. Linde K, Scholz M, Ramirez G, et al: Impact of study quality on outcome in placebo-controlled trials of homeopathy. J Clin Epidemiol 1999; 52: 631–6
75. Sterling T D, Rosenbaum WL, Weinkam JJ: Publication decisions revisited: the effect of the outcome of statistical tests on the decision to publish and vice versa. Am Statistician 1995; 49: 108–12
76. Stern JM, Simes RJ: Publication bias: Evidence of delayed publication in a cohort study of clinical research projects. BMJ 1997; 315: 630–45
77. Cook DJ, Guyatt GH, Ryan G, et al: Should unpublished data be included in meta-analyses? Current convictions and controversies. JAMA 1993; 269: 2749–53
78. Snow CM, Matkin CC, Shaw JM: Physical Activity and risk for osteoporosis, in Marcus R, Feldman D, Kelsey J (eds): Osteoporosis. San Diego, CA, Academic Press, 1996, pp 511–528
Keywords:

Exercise; Bone; Women; Meta-Analysis

Copyright © 2001 Wolters Kluwer Health, Inc. All rights reserved.