Osteoporosis International

, Volume 16, Issue 12, pp 2129–2141

Exercise frequency and calcium intake predict 4-year bone changes in postmenopausal women


  • Ellen C. Cussler
    • Department of PhysiologyUniversity of Arizona
  • Scott B. Going
    • Department of PhysiologyUniversity of Arizona
    • Department of Nutritional SciencesUniversity of Arizona
  • Linda B. Houtkooper
    • Department of Nutritional SciencesUniversity of Arizona
  • Vanessa A. Stanford
    • Department of Nutritional SciencesUniversity of Arizona
  • Robert M. Blew
    • Department of PhysiologyUniversity of Arizona
  • Hilary G. Flint-Wagner
    • Department of Health and Nutritional SciencesKasiska College of Health Professions
  • Lauve L. Metcalfe
    • Department of PhysiologyUniversity of Arizona
  • Ji-Eun Choi
    • Department of PhysiologyUniversity of Arizona
    • Department of PhysiologyUniversity of Arizona
    • Department of PhysiologyIna Gittings Building, University of Arizona
Original Article

DOI: 10.1007/s00198-005-2014-1

Cite this article as:
Cussler, E.C., Going, S.B., Houtkooper, L.B. et al. Osteoporos Int (2005) 16: 2129. doi:10.1007/s00198-005-2014-1


The aim of this study was to examine the association of exercise frequency and calcium intake (CI) with change in regional and total bone mineral density (BMD) in a group of postmenopausal women completing 4 years of progressive strength training. One hundred sixty-seven calcium-supplemented (800 mg/day) sedentary women (56.1±4.5 years) randomized to a progressive strength training exercise program or to control were followed for 4 years. Fifty-four percent of the women were using hormone therapy (HT) at baseline. At 1 year, controls were permitted to begin the exercise program (crossovers). The final sample included 23 controls, 55 crossovers, and 89 randomized exercisers. Exercisers were instructed to complete two sets of six to eight repetitions of exercises at 70–80% of one repetition maximum, three times weekly. BMD was measured at baseline and thereafter annually using dual-energy X-ray absorptiometry. Four-year percentage exercise frequency (ExFreq) averaged 26.8%±20.1% for crossovers (including the first year at 0%), and 50.4%±26.7% for exercisers. Four-year total CI averaged 1,635±367 mg/day and supplemental calcium intake, 711±174 mg/day. In adjusted multiple linear regression models, ExFreq was positively and significantly related to changes in femur trochanter (FT) and neck (FN), lumbar spine (LS), and total body (TB) BMD. Among HT users, FT BMD increased 1.5%, and FN and LS BMD, 1.2% ( p <0.01) for each standard deviation (SD) of percentage ExFreq (29.5% or 0.9 days/week). HT non-users gained 1.9% and 2.3% BMD at FT and FN, respectively, ( p <0.05) for every SD of CI. The significant, positive, association between BMD change and ExFreq supports the long-term usefulness of strength training exercise for the prevention of osteoporosis in postmenopausal women, especially HT users. The positive relationship of CI to change in BMD among postmenopausal women not using HT has clinical implications in light of recent evidence of an increased health risk associated with HT.


Dietary calciumExerciseHormone therapyMenopauseOsteoporosisStrength training BMD


In October 2004, the Surgeon General of the United States issued a report on the state of the nation’s bone health [1]. The report warned that, by 2020, 50% of the population older than 50 would be at risk for bone fracture. As the number of older Americans increases, the issue of age- and menopause-related fractures takes on greater significance, both in terms of quality-of-life and economic impact. The relationship between bone mineral density (BMD) and hip fracture in postmenopausal women is well established [2], as are the associated costs [3], and these factors have led investigators to seek preventive measures to reduce BMD loss. Randomized controlled trials in humans suggest that strength training exercise may improve BMD at critical fracture sites [4]. Progressive strength training has been shown to maintain or increase BMD in postmenopausal women in studies of at least 1 year, the minimum period considered necessary to produce significant bone changes [59]. However, very few prospective studies have examined the effects of exercise beyond 1 year in postmenopausal women [1013] and only two of these involve strength training.

Sinaki et al. followed women 8 years after the completion of a 2-year back extension program and found less loss in lumbar spine (LS) BMD among exercisers, despite cessation of the exercise intervention [10]. However, after the 2 years of intervention, Sinaki et al. did not find any difference between training and control groups for BMD. In another long-term study, Heikkinen et al. found that femur neck BMD increased in women who were not taking (hormone therapy) HT who performed 1 h of loading exercises per week for 2 years [11]. The results of these studies are a beginning; however, more studies of long-term exercise are clearly needed.

In cross-sectional studies [1416], prospective studies and randomized, controlled trials [1719], calcium intake (CI) predicted less BMD loss among older women. One study [20] reported sustained BMD maintenance over 4 years in a group of postmenopausal women supplemented with 800 mg/day of calcium. Research examining the combined, long-term effects of supplemental calcium intake on BMD in the context of an exercise trial is uncommon [19]. The purpose of the present study was to assess the effects of 4 years of progressive resistance exercise and calcium intake on BMD in postmenopausal women supplemented with calcium. In previous reports we showed significant increases in BMD at several sites after 1 year of progressive strength training [9] that were associated with total weight lifted [21]. In the present study, the analysis was extended to assess changes in BMD after an additional 3 years and to examine effects of 4-year calcium intake. We hypothesized that frequency of strength training exercise and level of calcium intake during the 4-year intervention would be positively associated with the changes in BMD of the hip, the lumbar spine, and total body.


Study design

The Bone Estrogen Strength Training (BEST) Study was a block-randomized clinical trial designed to examine the relationship of strength training exercise to BMD in early postmenopausal women. Women were asked to take calcium supplements, primarily calcium citrate (800 mg/day), and were randomized to either exercise or control conditions within groups stratified by HT use. The BEST study was reviewed and approved by the University of Arizona Human Subjects Review Committee, and written informed consent was obtained from all subjects prior to study entry.

For the present study, women who completed baseline and at least the fourth year of annual testing were included ( n =167). At the end of the first year of intervention, controls ( n =78) were allowed to begin or “cross over to” the prescribed exercise regimen. Fifty-five women began exercising in the second year (crossovers); 23 women remained controls, and 89 comprised the original randomized exercise group.

Of the 177 women who returned for BMD testing in their fourth year (55.3% retention from baseline ( n =320)), ten women were excluded: six women had started taking Fosomax (alendronate); one woman used steroids; one woman was hospitalized with a broken leg (unrelated to exercise) for 6 months; and two women developed cancer. The final sample was 167 women recruited in six cohorts at approximately 6-month intervals over 3 years from 1995–1998. Four years of data were accumulated from 1995 to 2002.

Recruitment and entry criteria

Subjects were recruited using selected zip codes for direct mailing, medical clinics, community organizations, and media advertisement. Initial telephone screening was followed by small group meetings, during which study requirements were explained, informed consent procured, and initial demographic data collected. Inclusion criteria for the first year were as follows: 40–65 years of age; surgical or natural menopause (3.0–10.9 years); body mass index (BMI) greater than 19.0 kg/m2 and less than 33.0 kg/m2; nonsmoking; no history of osteoporotic fracture and an initial BMD greater than Z -score of −3.0; undergoing HT (1.0–5.9 years) or not undergoing HT (>1 year); no weight gain or loss greater than 13.6 kg (30 lbs) in the previous year; free of cancer and cancer treatment for last 5 years (excluding skin cancer); not using BMD-altering medications, beta-blockers, or steroids; dietary calcium intake >300 mg/day; performing less than 120 min of low-intensity, low-impact exercise per week and no weightlifting or similar physical activity. Participants agreed to be randomized to exercise or control and continue their usual dietary practices, maintain their HT status, and take daily calcium citrate supplements for the duration of the intervention period.

Hormone therapy

Women using HT were asked to continue to follow the regimen prescribed by their physicians and to report any changes every year. A variety of hormone combinations was used; at baseline, most women took estrogen plus progesterone (80%), oral estrogen (10%), or estrogen and/or progesterone by patch (10%).

Of the 167 women, 90 used HT at baseline. One woman discontinued use in the middle of the second year. Of the 77 non-users, 25 started using HT during the study. Twelve began in the first 2 years and 13 began taking HT in the third or fourth year.

Dietary calcium intake

Dietary calcium intake was assessed using 8 days of diet records (DR) in the first year and, for each of the 4 years, optically scanned Arizona Food Frequency Questionnaire (AFFQ) based on the Block Model [22]. Dietary intake was measured from 8 randomly assigned days of DRs collected at baseline (3 days), 6 months (2 days) and 12 months (3 days). The DRs were analyzed by trained technicians using the Minnesota Nutrient Data System, versions 2.8–2.92. For a complete description of the DR methods see Maurer et al. [23]. AFFQs were distributed with verbal and written instructions at the end of the first year. During years 2 through 4, the participants were mailed the AFFQ annually with written instructions. AFFQs were reviewed with the participants by trained staff for completeness and accuracy prior to analysis.

Calcium supplements

In their first and second years, all participants received 800 mg/day of elemental calcium (calcium citrate) (Citracal, Mission Pharmacal, San Antonio, TX, USA). Starting in the third year, women were asked to continue calcium supplementation on their own, purchasing either Citracal or a comparable type of calcium supplement. Calcium-supplement compliance and type were monitored by tablet counts and written quarterly self-reports of calcium supplement intakes.


At baseline and annually thereafter, standing height and weight were measured with participants wearing lightweight clothing and no shoes. Height (cm) was measured to the nearest 0.1 cm with maximal inhalation with a Schorr measuring board. Weight (kg) was measured using a calibrated scale (SECA, model 770, Hamburg, Germany) accurate to 0.1 kg. The average height and weight from three trials was used to calculate body mass index (BMI) in kilograms: Weight (kg)/ Height (m2).

Blood sample collection

Blood specimens were collected at baseline after an overnight (12 h) fast. Each subject was asked to refrain from medications including HT, calcium supplements, or multivitamins the morning of their blood draw. Blood specimens were collected at the laboratory between 0600 hours and 0900 hours and were aliquotted into cryovials and transferred into −80°C for storage until assayed. Serum steroids were quantified by radio immunoassay (RIA) methods. Estrone (E1) and estradiol (E2) were measured with RIA kits from Diagnostic Systems Laboratories (Webster, TX, USA).

Dual energy X-ray absorptiometry

Regional and total body (TB) BMD (g/cm2) and percentage body fat (%Fat) were measured at baseline and annually thereafter by dual energy X-ray absorptiometry, using a total-body densitometer (model DPX-L; Lunar Radiation, Madison, WI, USA). Subject positions for total body, anteroposterior lumbar spine (LS), and femur (neck and trochanter) scans were standardized according to manufacturer’s recommendations, as previously described [24]. Each subject was scanned twice at baseline and at follow-up, and the mean of the two measurements was used in analyses. Initial scan analysis, including the placement of baselines distinguishing bone and soft tissue, edge detection, and regional demarcations, was done by computer algorithms (Version 1.3y, Lunar). One certified technician inspected all scans at all intervals and adjustments were made when necessary. Calibration of the densitometer was checked daily against a standard calibration block supplied by the manufacturer. A spine phantom was scanned daily to account for potential BMD variations due to machine error. The coefficient of variation (CV) of the phantom (BMD, LS) was 0.6%. Technical errors, expressed as a percentage of mean BMD, were ≤ ±1.8%, ≤ ±2.4%, ≤ ±2.4% and ≤ ±0.8% for LS, FN, FT, and TB BMD, respectively, estimated from study subjects with repeat scans over 4 years. Precision error is comparable to errors found in other research [25].

Exercise program

Participants randomized to the exercise intervention were asked to attend training sessions 3 days per week, on non-consecutive days, in one of four community facilities under the supervision of study on-site trainers. Sessions lasted 60–75 min and included stretching, balance, and weight-bearing activities for warm-up, weightlifting, an additional weight-bearing circuit of moderate-impact activities (e.g., walk/jog, skipping, hopping), and stair-climbing/step boxes with weighted vests. Exercise frequency (ExFreq), weightlifting loads, sets and repetitions, steps with weighted vests, and minutes of aerobic activity were recorded in exercise logs that were monitored regularly by on-site trainers.

The participant-to-trainer ratio was five-to-one in the first year. Supervision was reduced during the second year; and in the third and fourth years, trainers were available at each facility one morning or afternoon per week. Crossover exercisers received supervision comparable to randomized exercisers because new cohorts with trainers were present in all facilities during the entire study.

Weightlifting was done using free weights and machines. Eight core exercises focused on major muscle groups with attachments on or near BMD measurement sites. These exercises included the seated leg press, lat (latissimus dorsi) pull down, weighted march, seated row, back extension, one-arm military press (right and left), squats (wall squats initially, progressing to Smith or hack squats), and the rotary torso machine.

Women completed two sets of six to eight repetitions (four to six repetitions for the military press to decrease injury to the shoulder) at 70% (2 days per week) or 80% (1 day per week) of the one-repetition maximum (1-RM), determined by monthly testing. A detailed description of the exercise program can be found elsewhere [26].

Statistical analysis

The relationship between exercise participation and BMD was examined for 167 participants who completed baseline and at least the fourth year of annual DXA measurements. The sample included 23 controls, 55 crossovers and 89 randomized to exercise. Baseline and 4-year data were stratified and reported by HT status because of the well-documented association between current hormone use and BMD [27].

Although the women were asked to maintain their HT status throughout the study, 26 women changed their HT use: 25 women started taking HT and one woman stopped. Twenty-four of the women who started HT during the study used HT for at least 1 year, a time period considered adequate to impact BMD [28] and were combined into one group with the women who used HT throughout the study ( n =115). Those women who did not use any HT throughout the 4 years (plus one woman with HT use of 0.3 years) ( n =52) formed the second group in HT-stratified analyses.

In order to achieve the most accurate measurement of dietary calcium and energy intakes, 4-year mean daily intakes were derived from the average of 1 year of DR assessments and 4 years of estimates from the AFFQ. For each of these two variables, the values from the first year of DRs and AFFQs were averaged. This average was then averaged with the values from years 2, 3, and 4 of the AFFQ to produce 4-year means. Supplemental calcium intake was assessed over 4 years from tablet counts or quarterly self-reports. Mean daily tablet intake (mg) was estimated by multiplying tablet counts by mg per tablet and dividing by the number of days in the tablet count period, giving mg/day for each period. Self-reports of calcium supplement intake were calculated as the number of tablets reported on logs multiplied by the mg per tablet divided by the number of days in the reporting period. The mg/dayfrom tablet counts and self-reports for all periods were averaged over 4 years. Total calcium intake was calculated as the sum of dietary and supplemental intakes for each year and averaged for 4 years. Calcium intake variables (mg) were considered as continuous variables in descriptive statistics and regression analyses. For analyses of the association of supplemental calcium intake and BMD, a binary variable of mean 4-year intake was created, grouping subjects into those who took less than 800 mg/day ( n =125) and those who took at least 800 mg/day ( n =42). In multiple linear regression looking at the relationship between change in BMD and ExFreq, unstandardized residuals from regressing 4-year mean dietary calcium intake on 4-year mean energy intake were used. Although residuals and separate variables for mean dietary calcium and energy intake produced similar results, these two variables were correlated (r=0.74), and thus we used a single variable in analyses to avoid collinearity.

At both baseline and annually thereafter, two measurements of BMD were taken approximately 1 week apart. If there was a difference of 5% or more between the two values, scans were re-inspected and reanalyzed if necessary. The mean of the two values was then used in analyses. Change in BMD at 4 years was calculated as the difference between the average BMD (g/cm2) at the end of 4 years minus baseline.

Exercise frequency (ExFreq) was computed from monthly exercise cards filled out by the women at each session. Annual frequency was calculated as a percentage, by dividing the number of sessions performed by the number prescribed for that year and multiplying by 100. ExFreq for controls was set at 0%. Crossovers were assigned 0% frequency in the first year and then their actual percentage frequency for the 3 subsequent years. BMD change over 4 years was assessed in terms of frequency averaged over the entire 4 years regardless of intervention-group assignment.

Baseline characteristics and 4-year changes for women using HT and not using HT were compared using Student’s t -tests. The significance of 4-year changes in BMD, body weight, lean soft tissue (LST), and %Fat was tested using paired t -tests. Pearson correlations were conducted to assess the associations among exposure and outcome variables and covariates. Multiple linear regression models were constructed for each BMD site, with 4-year change in BMD as the dependent variable and mean percentage ExFreq or mean daily total calcium intake (mg) as the independent variables of interest. Models were stratified by HT status and, depending on the independent variable of interest, adjusted for age, baseline BMD, baseline body weight, change in body weight, percentage ExFreq, 4-year mean daily dietary calcium intake, 4-year mean daily supplemental calcium intake, and 4-year mean daily total energy intake. Because of the large number of independent variables pertaining to body composition, the most parsimonious set of variables was chosen to predict change in BMD.

Tertiles were produced for ExFreq, and BMD changes across these tertiles were compared using analysis of variance (ANOVA) statistics with Bonferroni post hoc tests. Adjusted mean BMD changes for each tertile of ExFreq were computed using the general linear model (GLM-random effects) procedure analysis of covariance (ANCOVA). Type 1 error was set at α =0.05 (two-sided) for all analyses, which were conducted using SPSS 12.0 [29].


Characteristics at baseline

Baseline characteristics by HT status are given in Table 1. As expected, the women showed significant differences by HT use in blood estrogen levels ( p <0.001). Women using HT were younger ( p <0.01), had fewer years past menopause ( p <0.001) and had higher BMD at baseline (FT, p <0.05; LS, p <0.01; TB, p <0.05).
Table 1

Baseline characteristics of 167 women completing 4 years of study (BMD bone mineral density, BMI body mass index, HT hormone therapy)

No HT (n =77)

HT (n =90)





Age (years)**





Years postmenopausal***





Estrone (pg/ml)***





Estradiol (pg/ml)***





BMI (kg/m2)





Body fat





Lean soft tissue (kg)





BMD (g/cm2)

Femur neck





Femur trochanter*





Lumbar spine**





Total body*





*p <0.05

**p <0.1

***p <0.001

Characteristics after 4 years

Tables 2 and 3 and Fig. 1 show 4-year characteristics for the 167 women. Table 2 presents findings not stratified by HT use for all women and by tertiles of ExFreq. Although asked to maintain their weight, women gained an average of 0.9±4.7 kg (Table 2). The range was −17.0 kg to +14.0 kg. On average, the women gained a small but significant amount of LST (0.5±1.5 kg; p <0.001). The average change in %Fat (0.1±3.7%) was not significant; change ranged from −13.4% to 11.3%. Four-year mean daily dietary calcium intake averaged 924±347 mg/day and ranged considerably, from 112–1,990 mg/day; 4-year mean daily calcium from supplements averaged 711±174 mg/day and ranged from 82–1,426 mg/day. Total calcium intake averaged 1,635±367 mg/day (range=581–2,772 mg/day). There were significant differences in 4-year characteristics of the 167 women when analyzed by tertiles of ExFreq (Table 2). On average, women in the two lower tertiles of frequency gained less LST and more weight, increasing their %Fat, versus women who attended the most exercise sessions ( p <0.01). Women in the highest tertile also gained more BMD at the FT, LS, and TB than women in the lower tertiles ( p <0.01). The FN BMD showed a similar trend, but the differences were not significant. The percentage of HT use (76.8%) was somewhat higher among women in the highest tertile of ExFreq but not significantly (chi-square p =0.3). Mean supplemental calcium intake was somewhat greater in the highest ExFreq tertile compared with the lowest tertile ( p <0.01), while mean dietary calcium intake was not associated with ExFreq.
Table 2

Four-year characteristics of all subjects and by tertiles of exercise frequency (n=167) (BMD bone mineral density, HT hormone therapy, LST lean soft tissue)


Tertiles of exercise frequency













% Exercise frequency



0.0 to 93.6







Change in body weight (kg)†



−17.0 to 14.0







Change in LST (kg)†



−7.6 to 4.3







Change in % body fat



−13.4 to 11.3







Mean daily dietary Ca (mg/day)



112 to 1,990







Mean daily supplement Ca (mg/day)



82 to 1,426







Mean daily total Ca (mg/day



581 to 2,772







% Using HT










Change in BMD (g/cm2)

Femur neck (FN)



−0.120 to 0.102







Femur trochanter (FT)



−0.100 to 0.081







Lumbar spine (LS)



−0.131 to 0.153







Total body



−0.169 to 0.039







*Includes 23 controls

†Mean positive change from baseline for body weight (p<0.05) and LST (p<0.001) for all women; for FT (p<0.01) and LS (p<0.001) BMD for highest tertile

a/b combinations are significantly different, 0.001<p<0.05

Table 3

Four-year characteristics of all subjects by HT status ( n =167) (HT hormone therapy, LS lumbar spine, LST lean soft tissue)

No HT (n =51)

HT (n =116)





Exercise frequency





Change in body weight (kg)





Change in LST (kg)





Change in percentage body fat





Mean daily dietary calcium (mg/day)





Mean daily suppl. calcium (mg/day)





Mean total calcium (mg/day)





Change in BMD (g/cm2)

Femur neck





Femur trochanter*





Lumbar spine**





Total body***







***p<0.001 between HT and no-HT groups; significant gain from baseline for LS BMD among HT users (p <0.05)

Fig. 1

Four-year percentage change in BMD by tertiles of frequency of exercise ( n =167) ( BMD bone mineral density, FN femur neck, FT femur trochanter, LS lumbar spine, TB total body)

In Fig. 1, average adjusted percentage changes in BMD for tertiles of ExFreq are illustrated. The lowest tertile experienced BMD loss of at least 1.0% across bone sites (significant decrease from baseline for TB BMD in both the lowest and middle tertiles; p <0.01; data not shown). The highest tertile showed significant increases from baseline in FT and LS BMD ( p <0.01 and p <0.001, respectively) and differences from the lowest tertile of 3.3%, 1.9%, and 2.8% at FN, FT, and LS ( p <0.001, p <0.05, p <0.001, respectively), after adjusting for covariates. The highest ExFreq was associated with less BMD loss for the TB compared with the lowest tertile ( p <0.1.)

Table 3 stratifies 4-year results by HT status. No significant differences were found for women using HT in ExFreq, body weight, body fat and LST, as compared with no HT. Mean daily calcium dietary, supplement, and total intake were similar between HT groups. For HT users, there was a change in LS BMD from baseline ( p <0.05) and changes in FT ( p <0.05), LS ( p <0.01), and TB ( p <0.001) BMD were significantly greater than those of non-users.

Table 4 shows changes in 4-year BMD regressed on ExFreq (model 1) and mean total daily calcium intake (model 2), stratified by HT status. ExFreq was positively, independently, and significantly associated with change in FT, FN, LS, and TB BMD among HT users ( p <0.01) and associated with change in LS BMD among non-users ( p <0.05), after controlling for age, baseline BMD and body weight, change in weight, mean daily dietary calcium intake, and mean daily supplemental calcium intake. Among HT users, FT BMD increased 1.5%, LS and FN BMD, 1.2%, and TB BMD, 0.4% ( p <0.01) for every standard deviation (SD) of ExFreq (29.5%). Among HT non-users, ExFreq was significantly associated with change in LS BMD only (1.2% increase per SD (26.7%); p <0.05).
Table 4

Results from linear models of change in BMD, regressed on exercise frequency and total calcium intake stratified by HT use (n =167) (BMD bone mineral density, HT hormone treatment)

Model 1: mean percentage exercise frequency†

Model 2: mean total calcium (supplemental calcium)

BMD site

Percentage change in BMD per SD of percentage exercise frequency

Adj R2

Standardized beta-coefficients

Percentage change in BMD per SD of calcium intake

Adj R2

Standardized beta coefficients


Femur trochanter




1.9 (1.5)

10.3 (19.3)%

0.29* (0.33*)

Femur neck




2.3 (1.8)

20.7 (22.0)%

0.33* (0.33*)

Lumbar spine




0.4 (1.1)

33.0 (37.2) %

0.07 (0.25*)

Total body




0.7 (0.8)

0.9 (4.4)%

0.19 (0.32*)

HT (some or 4 years)

Femur trochanter




0.5 (0.4)

13.0 (12.1)%

0.07 (0.07)

Femur neck




−0.4 (−0.5)

8.6 (7.9)%

−0.06 (−0.11)

Lumbar spine




0.1 (0.3)

8.2 (6.5)%

0.02 (0.07)

Total body




0.0 (0.3)

7.7 (9.6)%

−0.01 (0.15)

*p <0.05

**p <0.01

***p <0.001

†Adjusted additionally for age, baseline BMD and body weight, weight change, dietary calcium intake regressed on energy intake and supplemental calcium intake

††Adjusted additionally for age, baseline BMD and body weight, exercise frequency, and energy intake

Mean total daily calcium intake was positively and significantly associated with change in BMD at the FT and FN ( p <0.05) for women not using HT (model 2), after adjusting for age, baseline BMD and body weight, ExFreq, and mean daily energy intake. Percentage changes in BMD per SD of total calcium intake (399 mg/day) were 1.9%, 2.3%, for the FT and FN, respectively. Calcium intake (CI) was not significantly associated with BMD changes among HT users (Table 4).

Mean daily supplemental calcium intake was positively and significantly associated with change in BMD at all sites and for the TB ( p <0.05) for women not using HT (model 2—values in parentheses). Percentage changes in BMD per SD of calcium supplemental intake (218 mg/day) were 1.5%, 1.8%, and 1.1% for FT, FN, and LS BMD, respectively.

Adjusted mean 4-year changes in BMD at regional sites and for the TB are plotted by tertiles of ExFreq and HT status in Fig. 2. Fig. 2a illustrates the relationship of ExFreq with change in FT BMD stratified by HT use. Among HT users, the lowest ExFreq is associated with a 1.4% decrease in BMD, whereas, the highest group gained 2.6% over 4 years ( p <0.001). ExFreq was unrelated to FT BMD among HT non-users. In Fig. 2c, there was a difference in mean change in LS BMD between the lowest and highest tertiles of ExFreq for all subjects combined ( p <0.001). An association between HT use and LS BMD was also evident ( p <0.05). FN (Fig. 2b) and TB (Fig. 2d) BMD showed no significant differences with tertiles; however, HT played a role in preserving TB BMD compared with no HT. An interaction by HT status was not found for the relationship between ExFreq and change in BMD.
Fig. 2

Adjusted BMD changes by tertiles of exercise frequency and HT use ( BMD bone mineral density, FN femur neck, FT femur trochanter, HT hormone therapy, LS lumbar spine, TB total body)

Analyses performed to examine the relationship between supplemental calcium intake groups and change in BMD stratified by HT status are illustrated in Fig. 3. These results were adjusted for potential confounders. At the FT and for the TB (Fig. 3a and d), women taking at least 800 mg/day showed positive changes in BMD, significantly different from changes among women taking less than 800 mg/day for both HT groups combined ( p <0.01 and p <0.01, respectively). Marginally significant differences between high and lower calcium-supplemented women were found at the FN and the LS (Fig. 3b and c; p <0.10). An interaction between calcium supplementation and HT status with respect to change in BMD was not significant.
Fig. 3

Adjusted BMD changes by calcium supplementation intake (mg/day, BMD bone mineral density, FN femur neck, FT femur trochanter, LS lumbar spine, TB total body)


By following a large sample of calcium-supplemented postmenopausal women over 4 years of exercise, this study provided a unique opportunity to address the long-term effects of strength training exercise and calcium intake on BMD in women using and not using HT.

Strength training exercise

The current data showed that BMD was maintained or increased after 4 years in women who, on average, exercised about 70% of the time, approximately two times per week (range 50–94%; the highest tertile of frequency), compared with women exercising the least, less than once per month (on average approximately 5%; range 0–15%; lowest tertile). Crucial questions, which were not addressed and are beyond the scope of this paper, relate to the progression of changes in BMD (changes in slope) over this 4-year period.

From Table 2 and Fig. 1, declines in TB BMD demonstrated an overall decrease in bone density that was probably age- and menopause-related and was prevented only at the highest levels of ExFreq. Results from the FT and LS sites suggested that at least in specific regions, the decline in BMD was counteracted by exercise, promoting gains in BMD. In particular, for women in the highest tertile of ExFreq, FT and LS BMD increased approximately 1.5% compared with a 0.5% decline in TB BMD. The specificity implied by this difference may have been the result of the exercise program designed to target regional sites.

Regional bone sites seemed to respond to exercise differently with consistency over time. Prior results from the first year of the BEST study reported that participation in a progressive strength training program [9] and weight lifting in that program [21] were associated with significant gains in FT BMD but not in FN BMD. This finding was similar after 4 years (Table 2). The continued influence of strength training exercise with muscles with direct attachments to the FT and the lack of such attachments at the FN may have accounted for this sustained difference [30]. The type of program involving primarily weight lifting may also have explained these regional differences. In previous research, Kohrt et al. found significant gains in FN BMD in reaction to a 1-year program of ground reaction forces and no gains with a joint reaction program including weight lifting, while FT BMD responded to both programs [8].

After 4 years of intervention, women who attended the most sessions showed the largest gains in LS BMD, compared with the women in the lowest tertile of measured exercise performance. For LS BMD, weight loaded above the back may stress the bone of the lumbar spine to stimulate new bone formation. While FT BMD improved after the first year of exercise [9] and showed a significant relationship to weight lifted [21], LS BMD among exercisers exhibited no significant improvement compared with BMD among the controls in the first year [9]. After 4 years, however, increases in LS BMD that were related to ExFreq became apparent. There was a 2.5% mean difference in LS BMD between women in the tertile with the greatest ExFreq and those with the least (Fig. 1).

The improvement in FT BMD in the first year and the delayed improvement in LS BMD may be related to a difference between, on the one hand, the effects of strain on BMD due to increased muscle mass and contraction at muscle attachments, and, on the other hand, strains on BMD due to ground reaction forces. In the first year of study, significant gains in lean soft tissue (LST) were found among randomized exercisers. Additionally, change in LST was significantly related to the weight lifted in each of the eight core exercises [31]. An initial increase in muscle mass surrounding the skeletal hip sites during the first year might have contributed to the larger improvements in FT BMD in that shorter period of time. Specifically, among the women in the highest tertile of ExFreq, there was a significant gain in LST from baseline in the first year of exercise ( p <0.001; data not shown). This significance was lost for years 2 through 4 ( p >0.7), although the overall 4-year gain was significant for the total study period (Table 2).

Since the health benefits of strength training exercise on BMD are well documented [4], investigators in the BEST study thought it important to allow controls to cross over to the exercise arm of the study after the first year of intervention. This presented a challenge in analysis, since another category of intervention had been created that differed not only in the length of exercise exposure but also as to when the intervention was received. The number of 4-year controls (“pure” controls) was reduced to 23 and made analysis stratified by HT untenable, since there were only eight women among the controls who did not use HT. To maintain a large participant sample size, we chose an analysis based on tertiles of ExFreq, rather than intent-to-treat. The lowest tertile of ExFreq was composed of all “pure” controls, 22 crossovers, and 10 randomized exercisers (HT use=67.3%). In this tertile, the 22 crossovers had a mean ExFreq of 7.8±4.7% and the randomized exercisers, 9.4±3.9%. This contrasts with an average frequency of 70.3±12.6% for the highest tertile of ExFreq. The highest tertile contained nine crossovers (59.3±7.0% average frequency) and 47 exercisers (72.4±12.4% average frequency). Since there was a substantial range in ExFreq among randomized exercisers, and since crossovers were present in all tertiles and crossovers had once been controls, we analyzed BMD change based on frequency in all subjects completing 4 years, regardless of beginning randomization status. When the effect of date of initiation of exercise was examined (binary variable created for exercisers versus crossovers), the crossovers, who started exercising in the second year, showed greater gains in BMD across sites for years 2 through 4 than the women randomized to exercise at baseline and exercising 4 years. (This result supports a nonlinear strength training effect, with more change in the initial years than in the later years.) The exercise-initiation factor added to multiple regression analyses did not alter the relationship of ExFreq with BMD (data not shown)—that is, the potential training time discrepancies between exercisers and crossovers were not a factor in any analyses of BMD.

A further potential bias could have resulted from the accuracy of self-reports of exercise. Since women were asked to record all weights and repetitions for each session on a monthly card, their presence at a training facility was not merely recorded by a trainer with a single check mark. Furthermore, trainers supervised the filling out of these cards and exercisers followed their progress carefully in order to maintain 70–80% of one repetition maximum. Therefore, frequency rates were more likely to be accurate than without such exercise records. If the quality of the self-report was related differentially to the level of ExFreq, there would be reason to suspect an inflated effect. There may be some reason to believe that women who exercised more were more likely to fill out their cards more accurately; however, there was no evidence to suggest that this might have happened.

One of the major issues in long-term exercise intervention studies is participant retention. One hundred seventy-seven of the 266 women who finished the first year went on to complete 4 years, including ten women excluded from analysis. The dropouts (33.5%) were significantly younger ( p <0.01), heavier, had a higher average %Fat ( p <0.05), and lower level of education ( p <0.05) (data not shown). Dropouts may have represented women who were still working, had less time for exercise, and tended to be heavier. This dropout rate was less than expected, given reduced trainer contact in the follow-up years. Permitting women to cross over to exercise may have prevented some subject attrition (62% of controls completing the first year of the study started exercising in the second year). Dropout rates and reasons are a crucial concern in long-term studies of exercise intervention, and more research is needed to understand the factors and mechanisms involved in retention and how they impact results [32].

Hormone therapy

Since it has been documented that estrogen use enhances BMD [27], most analyses were stratified by HT status. Results from these separate analyses confirmed prior research showing that HT use by postmenopausal women who exercise helps maintain or increase BMD [33,34]. In the present study, evidence of the preservative effects of HT on TB BMD indicated that overall bone loss might be retarded through hormone use after 4 years, independent of exercise (Fig. 2d).

Significant differences in BMD change between the lowest and highest tertiles of ExFreq were found for HT users only at the FT and FN. Except for LS BMD, there was less effect of ExFreq on BMD among HT non-users. These data suggested additive effects of HT with exercise on BMD (Fig. 2). Kohrt et al. found that HT was additive for increasing BMD at regional bone sites and that HT and exercise were synergistic for TB BMD increases after 1 year of exercise [35]. In that study, women began taking HT at baseline, and initiation of HT may have influenced the BMD response differently. Recent research has proposed that the bone-enhancing effects of HT may subside after the first few years of use [28]. In the present study, women on HT had been using hormones for an average of 2.7±1.1 years prior to baseline. If women in the BEST study had passed the initial phase of HT benefit to BMD, no synergism or no interaction might have been present between HT and exercise in their impact on BMD. In fact, we found no statistically significant interactions by HT status in the relationship between ExFreq and BMD change at any sites (Fig. 2). The independent effect of exercise on BMD after discontinuing HT has been studied [28] but remains to be examined since the publication of the Women’s Health Initiative (WHI) results. In the 2 years following that report, 42.9% of HT users remaining in the BEST study had discontinued hormones.

Calcium intake

The mean dietary calcium intake reported by women in this study was 924 mg/day and higher than the National Health and Nutrition Examination Survey III average of 686 mg/day for women between the ages of 40 and 59 [36]. Despite the higher mean level, there was a substantial range of values in this group of women (112–1,990 mg/day). Over the 4-year study, women were asked to take 800 mg/day in calcium supplements [37]. Although the mean calcium compliance rates decreased very little over the study, the SD increased threefold by the fourth year, indicating greater variation in calcium compliance with time. In the first year, 92.8% of these women were 80% compliant with protocol; however, by the fourth year, that rate had dropped to 65.1%. On the other hand, some women (22.3%) took more than the recommended amount of calcium supplements in the last year, adding considerable variation in dietary intake for total CI (581–2,772 mg/day). This range was similar to that in a female population examined in a 1997 cross-sectional study [38]. Michaelsson et al. found an association between the highest mean 4-year levels of calcium intake (1,417–2,417 mg/day) and FN, LS, and TB BMD, compared with the lowest levels (400–550 mg/day) in a group of Swedish postmenopausal women.

Prospective research has also found relationships between calcium intake and change in BMD over the long term. Riggs et al. followed elderly women not taking HT in a 4-year double-blind controlled intervention study and found 4-year BMD changes at the FN and for TB in women taking 1,600 mg/day supplementary calcium (plus dietary intake of approximately 700 mg/day) that were significantly different from changes in the control group [39]. Similar to the latter study and others [40], the present findings suggested a sustained benefit of calcium supplementation after a continuous longer period of supplemented calcium intake in postmenopausal women. Michaelsson et al. proposed that high calcium supplement compliance over a 4-year period implied the long-term utility of calcium with respect to BMD maintenance in postmenopausal women, even after intestinal absorption and bone turnover had adjusted to higher calcium levels [38]. Devine et al. found that women discontinuing supplements after 2 years of a 4-year study lost more BMD at the ankle than women continuing supplements for the full 4 years [20]. Our results corroborated these findings.

In this research, the associations of total and supplemental calcium intake with change in BMD were examined in women using HT as well as not using HT. For women not using HT, both calcium intake and supplementation were more strongly associated with change in BMD than for women taking the hormones (Table 4), confirming previous research examining the effect of calcium intake on BMD in a 1-year sample from the BEST study [23]. These findings suggested an interaction between HT and calcium with respect to BMD, although no significant interaction was found in regression and ANCOVA models. The interaction of HT and calcium intake has been suggested in other prospective research. One study examining the difference in the calcium/BMD relationship between postmenopausal women on HT and not on HT found that the positive association between change in TB calcium and CI was enhanced by HT use [41]. In one meta-analysis, a synergism between these two factors was discussed [42] and in other work, possible explanations for the relationship between HT and calcium were developed [43,44]. Despite the lack of definitive evidence of an interaction in our data, the finding of the greater impact of total CI and supplementation among HT non-users over 4 years would support a recommendation for continued use of calcium supplementation in conjunction with exercise over longer periods of time, especially as women choose to stop using HT or reduce hormone doses.

A negative correlation between dietary and supplemental calcium intake (r=−0.14, p <0.07), may have accounted for the weaker relationship of total CI than supplementation to change in BMD. This inverse correlation suggested that women taking more supplements were slightly less likely to obtain the mineral through high calcium food choices. Another possible explanation to account for the weaker relationship between total CI and change in BMD might have been the less exact measurement of dietary intake introduced by the AFFQ compared with the more exact supplement assessment through pill counts and tally sheets. BMD changes did not correlate with dietary CI among either HT users or non-users.

Since mean ExFreq was correlated with mean supplemental calcium intake (r=0.3; p <0.001) and a significant difference in mean ExFreq was found between women taking high levels of calcium supplements and those taking low levels ( p <0.05; data not shown; see also Table 2), the relationship between supplementation with calcium and change in BMD may be influenced by ExFreq. Our models showed a positive relationship between CI and change in BMD, after adjusting for ExFreq, and, thus, we found an independent effect of CI.

Summary and conclusion

This study confirms the usefulness of strength training exercise in preserving and increasing BMD in postmenopausal women. The benefits extended to 4 years among women who attended at least two exercise sessions per week and to women using hormone replacement therapy. It appears that FT BMD responded earlier to strength training (first year), whereas, a longer period of exercise was needed to produce increases in the LS. Sustained weight-bearing effects through year 4 may account for longer-term improvement in LS BMD. Furthermore, we found an additive effect of HT use on the relationship between ExFreq and change in BMD.

Among women not using HT, those who took at least 800 mg/day of calcium supplementation had greater improvement in BMD than those taking less calcium supplements. These women received, on average, approximately 900 mg/day calcium from dietary sources. It appears, therefore, that a total CI of at least 1,700 mg/day (greater than the level of 1,500 mg/day recommended by the National Institutes of Health in 1994 [45]) may be more likely to ensure calcium levels high enough to preserve BMD among HT non-users. In light of the fewer number of women using HT since 2002, clinicians should continue to recommend supplementation to postmenopausal women discontinuing HT and/or at risk for osteoporosis. Levels of calcium supplementation need to be sufficient to augment calcium obtained through the diet

Our findings suggested a possible interaction in the relationship between calcium and BMD by HT use. In a regression model predicting BMD change from CI, a significant interaction was not found. The mechanisms that link estrogen and calcium in the metabolism of bone are still unclear.

In conclusion, this study supports the long-term benefits of strength training exercise and calcium intake for the prevention of osteoporosis in postmenopausal women. Combined with exercise, women may choose to continue HT or increase total calcium intake to around 1,700 mg/day to help prevent osteoporosis.


This study was funded by NIH AR39939 and Mission Pharmacal. The authors wish to express their gratitude to Mission Pharmacal (San Antonio, TX, USA) for the generous donation of calcium supplements (Citracal(r)) for the study. The authors wish to express the sincere appreciation to the BEST study participants for their many contributions and dedication to improving women’s health

Copyright information

© International Osteoporosis Foundation and National Osteoporosis Foundation 2005