Lean Mass and Fat Mass as Mediators of the Relationship Between Physical Activity and Bone Mineral Density in Postmenopausal Women

Abstract

Background:

The relationship between physical activity (PA) and bone health is well known, but the role of lean mass (LM) and fat mass (FM) in this relationship remains uncertain. Therefore, the aim of this study was to examine the mediating effect of LM and FM on the relationship between PA and bone mineral density (BMD) in postmenopausal women.

Materials and Methods:

This cross-sectional study involved 282 postmenopausal women aged between 50 and 65 year, who were randomly selected from Hongqi community of Harbin City in China. PA was measured using an International PA Questionnaire. Body composition, BMD of the lumbar spine, hip, and total body were measured using dual-energy X-ray absorptiometry. Mediation analysis was performed to investigate the mediating effect of LM and FM on the relationship between PA and BMD.

Results:

In partial correlation analysis, PA, LM, and FM were positively related to BMD. Positive correlation was found between PA and LM. There were significant differences in BMD between different categories of PA, but the differences disappeared after adjusting for LM. Mediation analysis showed that LM and FM played a mediating role in the relationship between PA and BMD. LM appeared to mediate the effect of BMD in the spine, hip, and total body by 26.91%, 19.55% and 47.98%, respectively; and FM was 22.23%, 27.97%, and 33.02%, respectively.

Conclusion:

LM and FM affected the relationship between PA and BMD as mediator. Postmenopausal women with high LM and FM had more BMD.

Introduction

Osteoporosis in postmenopausal women was a major public health issue due to its high risk for fragility fractures. Protecting bone mass during adulthood may be a key determinant of bone health and decreasing fracture risk.¹ Bone mineral density (BMD) declines with aging and the rate of decline is more pronounced after menopause.^1,2 Meanwhile, falling level of 17-β-estradiol may accelerate bone loss, which was considered as the predictor of primary osteoporotic fracture.³ Many studies reported that the bone loss was affected by a number of factors: body composition, years since menopause, age of menophania, dietary patterns, genetics, nutrition, lifestyle factors, or the prolonged use of certain medication.^3,4

A large number of cross-sectional studies have found a significant relationship between bone mass and body composition in postmenopausal women. Some studies suggested that lean mass (LM) or fat mass (FM) was associated with BMD or bone mineral content (BMC).^5
–7 Still other studies have found that both LM and FM were significant predictors of BMD, with FM being more important than LM in premenopausal women, and LM being more important than FM in postmenopausal women.^4,8,9 The inconsistency of findings may relate to the expression of bone mass as an aerial BMD or apparent BMD, and the colinearity between FM and LM.^4,10,11 However, as the better casual effect approach, some prospective studies have confirmed that LM, not FM, was the significant predictor of BMD.^{4,12

–17}

Lifestyle factors, such as physical activity (PA) may exert influence on bone mass in both pre- and postmenopausal women. Physical activity has been considered as a major potential modifiable factor linked to bone mass or osteoporosis. There was evidence that regular PA minimized bone loss in postmenopausal women.¹⁸

Although the association of PA, LM, and FM with BMD in postmenopausal women has been extensively described, there were few studies to investigate the mediating effect of LM and FM on the relationship between PA and BMD, especially in postmenopausal women in China. Furthermore, to adjust for potential confounders, most studies have been conducted using statistical methods such as analysis of covariance (ANCOVA), multiple linear regression, or logistic regression. But these statistical procedures did not report the partial and semi-partial correlations and failed to distinguish between confounding and intermediate variables. Therefore, the aim of this study was to examine whether LM, FM, and habitual PA were related with BMD, and to investigate whether the relationship between PA and BMD was mediated by LM or FM in postmenopausal women with no clinical evidence of disease in Northeast China.

Materials and Methods

Subjects

In Harbin City, China, Hongqi Community Health Center was randomly selected as the target community from all teaching community hospitals of Harbin Medical University. Three hundred seventy-three volunteers residing in this community were randomly recruited and those who met the inclusion criteria would enter into the study. Subjects with BMD-related discords that might affect calcium metabolism and calcium absorption, and the behaviors resulting in the noncompliance were excluded. The details of inclusion and exclusion criteria were described in the previous study.⁴ Finally, 282 women were included in the study. The study protocol was approved by the Research Ethics Committee of Harbin Medical University. All subjects provided written informed consent.

All subjects were interviewed to collect information including demographic and socioeconomic background, PA, body composition, BMD, and BMC. The data of PA was collected using the International PA Questionnaire (IPAQ), which was designed in case report form. The anthropometrics measurements included weight, height, BMI, and WHR, which were detailed in elsewhere.⁴

Body composition and BMD measurements

For all subjects, body composition and BMD were measured using dual-energy X-ray absorptiometry (DXA, Norland Corp.) in Harbin Orthopedics Hospital. DXA equipment accuracy was checked on a daily basis before each scanning session by the manufacturer, and all measurements were performed at high resolution by the same person. BMD (g/cm²) were measured including the left hip (femoral neck, trochanter, and Ward's triangle), the lumbar spine (L2–L4), and the total body. LM (kg) [body mass− (FM + bone mass)] and FM (kg) were obtained from the total body measurement. LM and FM were categorized as follows: low (first quartile), medium (second and third quartiles), and high (fourth quartile). And, the cutoff points were 33.87, 36.48, 40.07 kg and 18.89, 21.48, 24.81 kg, respectively.

PA assessment

The PA levels of the postmenopausal women were assessed at the time of recruitment using the “long last 7 days self-administered format” of the IPAQ.¹⁹ This questionnaire includes about three specific types of activity: walking, moderate-intensity activities, and vigorous-intensity activities.¹⁹ The minutes spent every week on each type of activity were computed separately by multiplying the duration and frequency of activity. A continuous activity score was calculated by multiplying the selected metabolic equivalent (MET) value and weekly minutes of activity. Thus, PA was expressed as MET-min per week. The subjects were divided into low, moderate, and high levels of PA on the basis of their total PA (MET min/week) and the frequency of the activity. The MET values and the levels of PA were calculated according to the guidelines for data processing and analysis of the IPAQ.^19,20 PA was categorized as low (first quartile), medium (second and third quartiles), and high (fourth quartile). And the cutoff points were 2553.25 MET-min/week, 2887.45 MET-min/week, and 3024 MET-min/week, respectively.

Statistical analysis

Both statistical (Kolmogorov–Smirnov test) and graphical methods (normal probability plots) were used to examine normality for each continuous variable. Since the distributions of these variables were normal, participants' characteristics were described using mean and standard deviation.

Partial correlation coefficients controlling for confounders were calculated to assess the relationships among variables. ANCOVA was used to test the mean differences of bone-related variables among different LM, FM, and PA categories. For LM and FM, age, height, and PA were adjusted for; for PA, age, height, and LM were adjusted for. Bonferroni was used for multiple comparisons.

Mediation analysis was conducted to examine whether the relationship between PA and BMD was mediated by LM or FM. Linear regression analysis was performed to test whether the potential mediating effect of LM and FM on the relationship between PA and BMD followed the criteria outlined by Baron and Kenny²¹: (i) the independent variable must be significantly related to the mediator; (ii) the independent variable must be significantly related to the dependent variable; (iii) the mediator must be significantly related to the dependent variable; and (iv) the relationship between the independent and dependent variables must be attenuated when the mediator is included in the regression model.

In addition, mediation analysis was tested using the steps outlined by Sobel²²: First, the attenuation or indirect effect was estimated (i.e., the effect of the independent variable on the mediator from the first regression model, multiplied by the effect of the mediator on the dependent variable obtained from the third regression model). Second, the indirect effect was divided by its standard error and Z test was performed under the null hypothesis that the indirect effect equaled to zero. Thus, the last regression model examined whether the relationships between PA and BMD of the spine, hip, and total body were mediated by LM or FM. All regression models adjusted for age.

Mediation analyses were conducted using SPSS.^23,24 To estimate the serial mediation models, the order of independent, dependent, and mediating variables must have been previously predetermined. All analyses were conducted using BMD of the spine, hip, and total body as dependent variables. Statistical analyses were performed using IBM-SPSS (Software, v.19.0 SPSS, Inc.), and the level of significance was set at α = 0.05.

Results

Table 1 shows the descriptive characteristics of the study participants. The subjects were, on average, 56.11 years old. Weight and height were 61.84 kg and 158.06 cm, respectively. The mean BMI was 24.72 kg/m². BMD of the total body, the lumbar spine, and the left hip were 0.91 ± 0.09, 0.90 ± 0.16, and 0.76 ± 0.11. LM and FM were 37.13 and 21.90 kg. Average PA level was 2887.45 MET-min/week.

Table 1.

Descriptive Characteristics of All Subjects

Characteristics	Mean	Standard deviation
Age (years)	56.11	3.83
Height (cm)	158.06	5.30
Weight (kg)	61.84	8.40
BMI (kg/m²)	24.72	2.91
Body BMD (g/cm²)	0.91	0.09
Spine BMD (g/cm²)	0.90	0.16
Hip BMD (g/cm²)	0.76	0.11
LM (kg)	37.13	4.90
FM (kg)	21.90	4.71
PA (MET-min/week)	2887.45	520.92

BMI, body mass index; BMD, bone mineral density; FM, fat mass; LM, lean mass; MET, metabolic equivalent; PA, physical activity.

Table 2 shows partial correlations among LM, FM, PA, and BMD of all regions after adjusting for age and height. Positive correlation was found between LM and FM and BMD of the spine, hip, and total body (r = 0.260, 0.347, 0.336 and 0.205, 0.210, 0.328, respectively; and all p < 0.001). There were significantly positive correlations between PA and BMD of the spine, hip, and total body (r = 0.242, 0.270, 0.296, respectively; and p = 0.025, <0.001, <0.001, respectively), and between LM and PA (r = 0.196, p = 0.033) but negative correlation between FM and PA (r = −0.176, p = 0.041).

Table 2.

Partial Correlation Among LM, FM, PA, and BMD of the Spine, Hip, and the Total Body (Adjusting for Age and Height)

	PA		LM		FM
Variables	r	p	r	p	r	p
Spine BMD	0.242	0.025	0.260	<0.001	0.205	<0.001
Hip BMD	0.270	<0.001	0.347	<0.001	0.210	<0.001
Total body BMD	0.296	<0.001	0.336	<0.001	0.328	<0.001
PA	1.000	—	0.196	0.033	−0.176	0.041

Table 3 shows the mean-adjusted differences in BMD of the spine, hip, and total body among different LM, FM, and PA categories. Postmenopausal women with high LM and FM had higher BMD of the spine, hip, and total body after controlling for confounders in Model 1 and Model 2. Moreover, postmenopausal women with high PA had significantly higher BMD of the spine, hip, and total body (p < 0.001, p = 0.010, and p = 0.029, respectively) after controlling for age and height in Model 1. However, when adjusting for age, height, and LM in Model 2, these significant differences disappeared (p = 0.133, p = 0.214, and p = 0.088, respectively).

Table 3.

Analysis of Covariance Models Comparing Means of BMD Among LM, FM, and PA Categories

	LM				FM				PA
Region	Low (n = 70)	Medium (n = 142)	High (n = 70)	p	Low (n = 62)	Medium (n = 125)	High (n = 95)	p	Low (n = 93)	Medium (n = 122)	High (n = 67)	p
Spine
Model1	0.838 ± 0.018	0.904 ± 0.011	0.961 ± 0.018^a	<0.001	0.880 ± 0.018	0.902 ± 0.013	0.924 ± 0.018^a	<0.001	0.892 ± 0.013	0.896 ± 0.018	0.923 ± 0.018^a	<0.001
Model2	0.839 ± 0.018	0.905 ± 0.012	0.959 ± 0.018^b	0.012	0.884 ± 0.018	0.894 ± 0.012	0.910 ± 0.018^b	0.023	0.895 ± 0.016	0.898 ± 0.012	0.920 ± 0.017	0.133
Hip
Model1	0.699 ± 0.011	0.757 ± 0.009	0.816 ± 0.011^a	<0.001	0.732 ± 0.012	0.759 ± 0.009	0.781 ± 0.012^a	<0.001	0.749 ± 0.013	0.750 ± 0.009	0.781 ± 0.013^c	0.010
Model2	0.700 ± 0.012	0.758 ± 0.008	0.815 ± 0.012^c	0.003	0.734 ± 0.011	0.757 ± 0.009	0.783 ± 0.011^b	0.031	0.750 ± 0.009	0.755 ± 0.013	0.776 ± 0.012	0.214
Total body
Model1	0.876 ± 0.010	0.903 ± 0.006	0.952 ± 0.011^a	<0.001	0.881 ± 0.010	0.910 ± 0.007	0.936 ± 0.010^a	<0.001	0.889 ± 0.010	0.906 ± 0.007	0.924 ± 0.010^a	0.029
Model2	0.877 ± 0.010	0.904 ± 0.006	0.950 ± 0.010^b	<0.001	0.884 ± 0.010	0.910 ± 0.007	0.932 ± 0.010^{a b}	<0.001	0.905 ± 0.010	0.905 ± 0.007	0.919 ± 0.010	0.088

Crude means in Model 1 and estimated marginal means in Model 2.

Covariates for LM and FM: Model 1: age, height; Model 2: age, height and PA. Superscript letter indicates statistical significance (≤0.05) for post hoc hypothesis test using the Bonferroni correction for multiple comparisons.

Covariates for PA: Model 1: age, height; Model 2: age, height, and LM. Superscript letter indicates statistical significance (≤0.05) for post hoc hypothesis test using the Bonferroni correction for multiple comparisons.

High> Medium >Low.

High> Low.

High> Medium.

Mediation analysis

Table 4 shows that, PA was positively associated with LM (p = 0.028, p = 0.020, and p = 0.011 for spine, hip, and total body, respectively) and BMD of three sites (p = 0.035, p = 0.015, and p = 0.023 for spine, hip, and total body, respectively). When LM as the mediating variable, LM was positively associated with BMD (p = 0.002, p = 0.001, and p = 0.007 for spine, hip, and total body, respectively). However, PA was not significantly associated with BMD of the spine, hip, and total body (p = 0.182, p = 0.341, and p = 0.441). The percentage of total effect mediated by LM were 26.91%, 19.55%, and 47.98% for the spine, hip, and total body, respectively (Z = −2.216, p = 0.027; Z = 2.312, p = 0.021; Z = −2.801, p = 0.003).

Table 4.

The Mediating Effect of LM on the Relationship Between PA and BMD of the Spine, Hip, and Total Body

	Coefficient	s.e.	p	Sobel test: Z (p)	Variation due to LM
PA → LM → Spine BMD	—	—	—	−2.216 (0.027)	26.91%
Total effect (c path)	0.012	0.005	0.035
Direct effect (c’)	0.032	0.007	0.182
PA → LM (a)	0.759	0.019	0.028
LM → Spine BMD (b)	0.0003	0.0001	0.002
PA → LM → Hip BMD	—	—	—	2.312 (0.021)	19.55%
Total effect (c path)	0.022	0.007	0.015
Direct effect (c’)	0.020	0.001	0.341
PA → LM (a)	0.746	0.133	0.020
LM → Hip BMD (b)	0.033	0.004	0.001
PA → LM → Body BMD	—	—	—	−2.801 (0.003)	47.98%
Total effect (c path)	0.040	0.001	0.023
Direct effect (c’)	0.047	0.001	0.441
PA → LM (a)	0.112	0.023	0.011
LM → Body BMD (b)	0.018	0.002	0.007

Spine: RMSEA: 0.067, GFI: 0.942, NFI: 0.933, TLI: 0.911.

Hip: RMSEA: 0.060, GFI: 0.937, NFI: 0.926, TLI: 0.914.

Total body: RMSEA: 0.056, GFI: 0.944, NFI: 0.926, TLI: 0.920.

When FM was taken as the mediating variable, the results were similar to those of LM as the mediating variable. The PA was positively associated with BMD but negatively with FM. FM, as the mediating variable, was positively associated with BMD of all sites (all p < 0.001). However, PA was not significantly associated with BMD (p = 0.114, p = 0.277, and p = 0.202, respectively). The variation of BMD due to FM was 22.23%, 27.97%, and 33.02% for the spine, hip, and total body, respectively (Z = −2.330, p = 0.010; Z = −3.222, p = 0.001; Z = −2.990, p = 0.001) (Table 5). The results suggested that the effect of PA on BMD was fully mediated by LM and FM. Besides, the results of testing the mediating effect of LM and FM on the relationship between BMD and PA were similar to those described above.

Table 5.

The Mediating Effect of FM on the Relationship Between PA and BMD of the Spine, Hip, and Total Body

	Coefficient	s.e.	p	Sobel test: Z (p)	Variation due to FM
PA → FM → Spine BMD	—	—	—	−2.330 (0.010)	22.23%
Total effect (c path)	0.035	0.001	0.022
Direct effect (c’)	0.026	0.001	0.114
PA → FM (a)	−0.940	0.088	0.008
FM → Spine BMD (b)	0.012	0.002	<0.001
PA → FM → Hip BMD	—	—	—	−3.222 (0.001)	27.97%
Total effect (c path)	0.015	0.002	0.045
Direct effect (c’)	0.009	0.001	0.277
PA → FM (a)	−0.524	0.016	0.015
FM → Hip BMD (b)	0.008	0.001	<0.001
PA →FM → Body BMD	—	—	—	−2.990 (0.001)	33.02%
Total effect (c path)	0.123	0.026	0.025
Direct effect (c’)	0.006	0.001	0.202
PA → FM (a)	−0.759	0.098	0.007
FM → Body BMD (b)	0.010	0.002	<0.001

Spine: RMSEA: 0.077, GFI: 0.916, NFI: 0.920, TLI: 0.918.

Hip: RMSEA: 0.074, GFI: 0.918, NFI: 0.923, TLI: 0.922.

Total body: RMSEA: 0.070, GFI: 0.921, NFI: 0.925, TLI: 0.927.

Discussion

This is the first mediation analysis to investigate LM and FM as a total mediator of the relationship between PA and BMD in postmenopausal women. The main findings of this study were as follows: First, postmenopausal women with high PA level had higher BMD in all sites than those with low PA; Secondly, postmenopausal women with high LM or FM had higher BMD in all analyzed regions after controlling for all sets of confounders; Finally, LM and FM as full mediator played a role in the relationship between PA and BMD.

Independent association of PA, LM, and FM with BMD

According to some studies, there have long been controversial on whether LM or FM is associated with BMD in postmenopausal women.^5,16,25,26 Some studies suggested that either LM or FM was associated with BMD. Still other studies have found that both LM and FM were significant predictors of BMD.^6,8
–10 It has recently been suggested that individuals with greater LM had greater BMD in body area and spine, suggesting that LM was indeed independently and positively associated with BMD or BMC.^10,11 In the present study, postmenopausal women with high LM, FM, or PA had higher BMD for the spine, hip, and total body after controlling for age and height. Meanwhile, these differences kept steadily statistically significant after controlling age, height, and PA in all regions. However, when adjusting for age, height, and LM, there was no significant association of PA and BMD. Therefore, it suggested that the significant differences observed in Model 1 might be explained by the effect of LM or FM.

The literature consistently identified PA as one of the most important determinants of BMD or BMC in the spine, hip, or total body in premenopausal and postmenopausal women.^1,13,27,28 In the postmenopausal group, there was statistically significant relationship between moderate PA and BMD. Those having moderate activity level on the IPAQ scale were found to have a lower likelihood of bone loss in a 2-year longitudinal study. It indicated that postmenopausal women with low PA level may represent an at-risk group for BMD progression and development of osteoporosis. Nonetheless, our findings were congruent with some observational studies, which have consistently identified weight-bearing physical activities being beneficial for bone mass or bone health.¹

Mediation analysis

It remained unclear whether LM or FM acts as a part of the causal chain in the relationship between PA and BMD in postmenopausal women. This study suggested that LM and FM played the mediator role in the relationship between the PA and BMD of the left-hip, the lumbar spine, and total body. Meanwhile, this research was showing that LM and FM should be intermediate variables as mediator rather than confounding variables in the relationship between PA and BMD for the women in the postmenopausal period. Last but not least, LM and FM might not be the only mediators in the relationship between PA and BMD. Perhaps, there were other mediated variables, such as dietary patterns, environment biomechanical factors, nutrients, genetics, and so on. In the future, longitudinal studies and structural equation models should be employed to confirm the conclusion and clarify more the potential mediating variables. Although previous study from this population confirmed that LM, instead of FM, was the best determinant of BMD,⁴ Pluijm et al. suggested that FM might play a mediating role in the association of weight change and walking activity in women.²⁹ Therefore, LM and FM were both considered as the mediator to investigate their mediating effect on the relationship between PA and BMD. What's more, LM was better than FM in the mediating effect in terms of the model of goodness of fit and variation of BMD due to mediation. Note that, the acceptable ranges of goodness of fit for RMSEA, GFI, NFI, and TLI were less than 0.08 and greater than 0.90, 0.90, and 0.90, respectively. Therefore, the fitness of models with LM and FM as the mediators were both good.

Limitations

This study had several limitations that should be presented. First, although this study is longitudinal, only the baseline data were used to make mediation analysis. In the future, the further longitudinal mediation analysis will be considered. Second, although all analyses adjusted for age and height, other potential variables, such as dietary patterns, calcium intake, or genetic variability, and so on, were not controlled for. Third, since this study was from a longitudinal data, the sample size was relatively small. Therefore, results should be interpreted and extrapolated with caution. Fourth, in this study, since PA was measured using IPAQ, there might be recall bias.

Conclusion

In this study, it showed that LM and FM independently influenced the BMD for Chinese postmenopausal women. And, LM and FM played a pivotal intermediary role as total mediator in the relationship between PA and BMD. Moderate PA, LM, and FM can protect bone health and prevent osteoporosis in postmenopausal women. Therefore, the findings of this study were of important clinical value and epidemiological significance.

Footnotes

Acknowledgments

We thank the staff of Hongqi Community Health Center for their help in selecting data of follow-up period.

Author Disclosure Statement

No competing financial interests exist.

References

Cummings

, Nevitt

, Browner

, et al. Risk factors for hip fracture in white women. Study of Osteoporotic Fractures Research Group. N Engl J Med, 1995; 332:767–773.

Riggs

, Khosla

, Melton

3rd.

A unitary model for involutional osteoporosis: Estrogen deficiency causes both type I and type II osteoporosis in postmenopausal women and contributes to bone loss in aging men. J Bone Miner Res, 1998; 13:763–773.

Guthrie

, Ebeling

, Hopper

, et al. A prospective study of bone loss in menopausal Australian-born women. Osteoporo Int, 1998; 8:282–290.

Chen

, Xiang

, Wang

, et al. Associations of bone mineral density with lean mass, fat mass, and dietary patterns in postmenopausal chinese women: A 2-year prospective study. PLoS One, 2015; 10:e0137097.

Reid

, Ames

, Evans

, et al. Determinants of total body and regional bone mineral density in normal postmenopausal women—a key role for fat mass. J Clin Endocrinol Metab, 1992; 75:45–51.

Reid

, Evans

, Ames

. Volumetric bone density of the lumbar spine is related to fat mass but not lean mass in normal postmenopausal women. Osteoporos Int, 1994; 4:362–367.

Reid

. Relationships among body mass, its components, and bone. Bone, 2002; 31:547–555.

Reid

, Ames

, Evans

, et al. Determinants of the rate of bone loss in normal postmenopausal women. J Clin Endocrinol Metab, 1994; 79:950–954.

Gnudi

, Sitta

, Fiumi

. Relationship between body composition and bone mineral density in women with and without osteoporosis: Relative contribution of lean and fat mass. J Bone Mineral Metab, 2007; 25:326–332.

10.

Ijuin

, Douchi

, Matsuo

, et al. Difference in the effects of body composition on bone mineral density between pre- and postmenopausal women. Maturitas, 2002; 43:239–244.

11.

Khosla

, Atkinson

, Riggs

, et al. Relationship between body composition and bone mass in women. J Bone Miner Res, 1996; 11:857–863.

12.

Namwongprom

, Rojanasthien

, Mangklabruks

, et al. Effect of fat mass and lean mass on bone mineral density in postmenopausal and perimenopausal Thai women. Int J Women's Health, 2013; 5:87–92.

13.

Ho-Pham

, Nguyen

, Lai

, et al. Contributions of lean mass and fat mass to bone mineral density: A study in postmenopausal women. BMC Musculoskelet Disord, 2010; 11:59.

14.

Liu-Ambrose

, Kravetsky

, Bailey

, et al. Change in lean body mass is a major determinant of change in areal bone mineral density of the proximal femur: A 12-year observational study. Calcif Tissue Int, 2006; 79:145–151.

15.

Hsu

, Venners

, Terwedow

, et al. Relation of body composition, fat mass, and serum lipids to osteoporotic fractures and bone mineral density in Chinese men and women. Am J Clin Nutr, 2006; 83:146–154.

16.

Douchi

, Matsuo

, Uto

, et al. Lean body mass and bone mineral density in physically exercising postmenopausal women. Maturitas, 2003; 45:185–190.

17.

Taaffe

, Cauley

, Danielson

, et al. Race and sex effects on the association between muscle strength, soft tissue, and bone mineral density in healthy elders: The Health, Aging, and Body Composition Study. J Bone Miner Res, 2001; 16:1343–1352.

18.

Smidt

, Lin

, O'Dwyer

, et al. The effect of high-intensity trunk exercise on bone mineral density of postmenopausal women. Spine, 1992; 17:280–285.

19.

Craig

, Marshall

, Sjostrom

, et al. International physical activity questionnaire: 12-country reliability and validity. Med Sci Sports Exerc, 2003; 35:1381–1395.

20.

Ainsworth

, Haskell

, Herrmann

, et al. 2011 Compendium of physical activities: A second update of codes and MET values. Med Sci Sports Exerc, 2011; 43:1575–1581.

21.

Baron

, Kenny

. The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. J Pers Soc Psychol, 1986; 51:1173–1182.

22.

Sobel

. Asymptotic confidence intervals for indirect effects in structural equation models. In: Leinhardt

, ed. Sociological methodology. Washington, DC: American Sociological Association, 1982:23.

23.

Preacher

, Hayes

. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behav Res Methods, 2008; 40:879–891.

24.

Preacher

, Hayes

. SPSS and SAS procedures for estimating indirect effects in simple mediation models. Behav Res Methods Instrum Comput, 2004; 36:717–731.

25.

Salamone

, Glynn

, Black

, et al. Body composition and bone mineral density in premenopausal and early perimenopausal women. J Bone Miner Res, 1995; 10:1762–1768.

26.

Wang

, Bachrach

, Van Loan

, et al. The relative contributions of lean tissue mass and fat mass to bone density in young women. Bone, 2005; 37:474–481.

27.

Pines

, Berry

. Exercise in the menopause—an update. Climacteric, 2007; 10 Suppl 2:42–46.

28.

Kohrt

, Bloomfield

, Little

, et al. American college of sports medicine position stand: Physical activity and bone health. Med Sci Sports Exerc, 2004; 36:1985–1996.

29.

Pluijm

, Visser

, Smit

, et al. Determinants of bone mineral density in older men and women: Body composition as mediator. J Bone Miner Res, 2001; 16:2142–2151.