Osteoporosis and cardiovascular disease are leading causes of morbidity and mortality in postmenopausal women.1, 2 Novel lifestyle interventions are needed to address these health problems. Exercise training has a beneficial effect on bone in postmenopausal women, but the effects are small, with bone mineral density (BMD) of the proximal femur or lumbar spine increasing 1% to 2% per year.3, 4 Likewise, exercise has a small beneficial effect of improving lipid profile in women, reducing low density lipoprotein (LDL) and total cholesterol by 2% to 3%, and increasing high density lipoprotein (HDL) by about 3%.5 Estrogen deficiency after menopause may increase the set point for bone to detect loads, causing bone to be less sensitive to mechanical force and making loading less effective for increasing bone mass.6 The optimal intervention for maintaining bone mass in postmenopausal women may therefore be a combination of increased loading and estrogen replacement. Several studies of postmenopausal women indicate hormone replacement therapy (ie, a combination of estrogen and progestin, or unopposed estrogen) and exercise training are additive or synergistic for increasing bone mineral density,7–9 supporting the theory that estrogen enhances bone responsiveness to loading.
Hormone replacement therapy, at the time of initiating this study, was reported to result in greater risks (ie, breast and endometrial cancers) than benefits (ie, reduced hip fractures).10 For that reason, many women have been hesitant to start hormone replacement therapy and have turned to alternative therapies, including phytoestrogens. Phytoestrogens are found in plant products including soybeans (as isoflavones).11 Isoflavones are structurally similar to estradiol and compete with estradiol for estrogen receptor sites, acting as estrogen antagonists in some cases and estrogen agonists in others.11 Isoflavones have not been shown to pose increased risk for cancer12 and may also reduce total cholesterol and LDL levels.13
Isoflavone supplementation and exercise have a cooperative effect on bone mineral and cholesterol in postmenopausal animal models (ie, ovariectomized mice and rats). The combination of exercise training and isoflavones was greater than the individual interventions for enhancing femoral, lumbar spine, or whole-body BMD, and architectural properties of bone (ie, increased trabecular bone thickness, and decreased trabecular separation), and decreasing total cholesterol levels.14–16 The effects of exercise (walking three times per week) combined with isoflavone supplementation (75 mg/d) in postmenopausal women were studied over 1 year.17 The combined interventions were more beneficial than the single interventions for maintaining total hip BMD. HDL level was increased with the exercise training, but there were no other changes in blood lipids. This study did not use an intent-to-treat analysis; bone measurements were restricted to BMD at the hip, the exercise prescription was relatively light, and assessment of breast or endometrial tissue was not done to determine whether isoflavones have the same deleterious effect as estrogen.
The purpose of our study was to investigate the combined intervention of isoflavones and higher-volume exercise training (resistance training and walking program) in postmenopausal women, using an intent-to-treat analysis, with BMD at the lumbar spine and proximal femur as the primary outcome measures. Geometry around the proximal femur, bone speed of sound (as a predictor of bone architecture18), and blood lipids were secondary outcome measures. Hip geometric assessments were included because along with hip BMD these might be related to physical activity patterns.19 We also used a long duration of intervention (2 years) and a high dose of isoflavones (ie, 165 mg total isoflavone/d), and included assessments of breast tissue (by mammography) and endometrial tissue (by ultrasound). We hypothesized that the combined treatment of exercise training and isoflavone supplementation would be additive (ie, more effective than either treatment alone) for improving bone outcomes, and blood lipids. Based on epidemiological studies,12 we hypothesized that isoflavones would have minimal adverse effects on breast and endometrial tissue.
Subjects and Methods
The study involved a double blind, parallel-group randomized controlled trial design to compare the independent and combined effects of isoflavones and exercise training. The setting was the community of Saskatoon, SK, Canada, and involved one center. Participants were randomized on a 1:1:1:1 basis to four unique groups after exclusion criteria were applied: Group 1 was exercise training (combined weight training and walking) plus isoflavone placebo (Ex); Group 2 was exercise training placebo (flexibility program) plus isoflavone therapy (Iso; 165 mg total isoflavone/d or 105 mg aglycone equivalent/d); Group 3 was exercise training plus isoflavone therapy (ExIso); and Group 4 was exercise training placebo plus isoflavone placebo (control). A program combining resistance training and walking was chosen because mixed exercise programs seem to have the most benefit on BMD.4 A higher dose of isoflavone than most other studies was chosen as isoflavones are most effective at higher doses for increasing BMD.20 Randomization was done with a computer-generated allocation schedule through our research pharmacy, which was independent from the rest of the study staff. Randomization was performed using a fixed block size of eight (using a permuted block design with a computer random number generator). Isoflavone and placebo were administered in a double-blind fashion in the form of tablets that were identical in taste, color, and appearance. They were prepacked in bottles and placed into opaque study kits that were sequentially numbered for each woman according to the randomization schedule. Instructions for either the actual exercise program or the placebo exercise program were also placed into each kit. The allocation sequence was concealed from the research assistant enrolling and assessing participants. Corresponding kits were opened only after the enrolled participants completed all baseline assessments and it was time to allocate the intervention. Although the participants could not be blinded to the exercise assignment, they were blinded to the hypothesis that the resistance exercise combined with walking would be superior to the placebo-flexibility exercises. All researchers and those involved in outcome assessment (ie, the individuals performing exercise tests, bone densitometry/sonography, medical examinations, and data analysis) were blinded to group assignment. Exercise tests were run by a research assistant blinded to group assignments. Training was supervised by two additional research assistants who were not involved in any other aspect of the study and who were blinded to whether subjects were on isoflavone or placebo. Another research assistant was in charge of all data entry, during which she remained blinded to group allocations (groups were coded). Our study statistician performed a blinded statistical analysis. The study was approved by the Biomedical Research Ethics Board at the University of Saskatchewan and all participants signed an informed consent form before participating. The study complied with the World Medical Association Declaration of Helskinki—Ethical Principles for Medical Research Involving Human Subjects. We adhered to the Consolidated Standards of Reporting Trials (CONSORT) guidelines for reporting on randomized clinical trials21, 22 including trials with factorial designs.23 This trial was registered with clinicaltrials.gov (http://clinicaltrials.gov/show/NCT00204425. Effect of Exercise Training and Soy-Based Nutritional Supplementation on Prevention of Osteoporosis).
Of 1175 participants who were recruited, 351 eligible postmenopausal women were stratified as either 1 to 9 years postmenopause or over 9 years postmenopause then randomized into four groups, as described above in Study Design (see Fig. 1 for flowchart of participants). Participants were stratified this way because loss of BMD is most rapid in the early years postmenopause and then the rate of loss levels off.24 Participants were recruited from November 2004 to January 2006 and all had completed the intervention by June 2008. Participants in all intervention groups received supplemental calcium and vitamin D (1200 mg and 800 IU, respectively) each day. All participants were postmenopausal, as indicated by questionnaire about their last menstrual period. If women reported that they were less than 2 years postmenopause, menopausal status was verified by determining levels of follicle stimulating hormone and luteinizing hormone. Exclusion criteria were the following: previous fragility fractures (defined as fractures resulting from minimal trauma); having taken bisphosphonates, hormone replacement therapy, selective estrogen receptor modulators (Rolaxofene), parathyroid hormone, or calcitonin within the past 12 months; currently taking corticosteroids or thiazide diuretics; Crohn's disease; Cushing's disease; kidney disease, allergy to soy; severe osteoarthritis; currently involved in vigorous exercise training (defined as jogging or resistance training for more than 20 minutes per session, more than twice per week); planning to travel outside of Saskatoon for an extended period during the study; osteoporosis (lumbar spine or proximal femur BMD 2.5 SD below the young adult mean (ie, T-score of −2.5 or lower); current or previous breast cancer; and current or previous endometrial cancer. All exclusion criteria were determined by questionnaire except osteoporosis, breast cancer, and endometrial cancer, which were evaluated by bone densitometry, mammography, and ultrasound, respectively. Participants who developed osteoporosis or started taking medications for low BMD during the study were removed from the study. This was required for ethical reasons (ie, women who developed osteoporosis were referred to their family physicians for treatment). Participants were recruited via newspaper advertisements and posters from the city of Saskatoon.
Isoflavone or placebo was administered orally in tablets. Isoflavone (Novasoy; Archer Daniels Midland, Inc., Decatur, IL, USA) was administered at a dose of 165 mg per day, containing 105 mg aglycone equivalents per day (35 mg aglycone equivalents taken three times per day). This isoflavone supplement contained genistin, daidzin, and glycitin in a ratio of 1:1:0.2. The placebo tablets contained dicalcium phosphate, magnesium stearate, and sorbitol. Analysis of placebo tablets indicated a minimal amount of isoflavone (0.001 mg per tablet). Contents of the isoflavone and placebo tablets were verified by testing in an independent laboratory (Douglas Laboratories, Pittsburgh, PA, USA). The exercise training intervention consisted of weight training combined with walking. The exercise training placebo consisted of a home-based flexibility program (a type of training that has no effect on bone mineral). Exercise training involved strength (resistance, weight) training twice a week and brisk walking four times per week. The strength training and walking were combined two days per week, at which time exercise sessions were fully supervised. On 2 other days of the week, participants were required to perform walking exercise on their own. Exercises during strength training included: hack squat, hip abduction, adduction, flexion, and extension (on a multihip machine), hamstrings curl, quadriceps (knee) extension, back extension, abdomen flexion, bench press, lat-pull down, shoulder press, biceps curl, and triceps extension (presses). Two sets of eight repetitions for each exercise were done at intensities corresponding to approximately 80% of 1-RM (1 repetition maximum; ie, the maximal amount of weight a participant could lift one time) for hack squat and bench press and at about 8-RM (ie, the maximal amount of weight that could be lifted eight times) for the other exercises. The walking program involved 20 to 30 minutes of brisk walking per session at an intensity corresponding to 70% of age-predicted maximal heart rate (220–age). Women were instructed on how to take their radial pulse for a 15-second count to ensure they were exercising at the proper intensity. Intensity and duration of exercise were increased progressively on an individual basis. Pedometers were given to the women to track walking distance and to encourage compliance. The exercise placebo groups were given a home-based flexibility program that involved stretching exercises for all major muscle groups, requiring 20 to 30 minutes of stretching 4 days per week. The exercise placebo groups were telephoned periodically to assess compliance to the flexibility program (and to decrease the difference in amount of attention given to the exercise and exercise placebo groups). Duration of the interventions was 2 years.
Compliance with the isoflavone and placebo, and the calcium and vitamin D supplement were assessed with logs. Compliance with the exercise program was assessed by attendance at the supervised exercise sessions and logs for the exercises done at home. Participants were surveyed after the study to assess the effectiveness of our blinding by asking if they thought they were on the isoflavone supplement, the placebo, or were not sure.
All outcome measurements were made at baseline, 1 year, and 2 years. Primary outcomes were lumbar spine and proximal femur BMD. Height and mass were measured by a standard stadiometer and a calibrated scale, and recorded to the nearest 0.1 cm and 0.1 kg, respectively, and used to determine body mass index (BMI, kg/m2). Waist circumference was measured at the superior border of the iliac crest (National Institutes of Health protocol).25 Blood pressure was measured using a mercury sphygmomanometer, with the cuff placed over the bare arm about 2 cm above the elbow after the participant was in a comfortable seated position for 5 minutes. Blood pressure was recorded to the nearest 2 mmHg. If systolic blood pressure was greater than 144 or diastolic blood pressure greater than 94, then blood pressure was measured again after the participant remained seated for 5 minutes. In these cases, the second blood pressure measurement was recorded.
Body composition was assessed with dual-energy X-ray absorptiometry (DXA). BMD and bone mineral content of the whole body, lumbar spine (L1–L4 vertebrae), and proximal femur (including the femoral neck, trochanter, Ward's, and total hip) were measured by DXA in array mode (QDR Discovery Wi; Hologic, Inc., Bedford, MD, USA) using QDR software for Windows XP (QDR Discovery, Hologic, Inc.). The coefficients of variation for these measures from our laboratory were 0.5% for the whole body, 0.7% for the lumbar spine, and 1.0% for the proximal femur. Lean tissue and fat mass were assessed from the whole-body scans. In our laboratory, the coefficients of variation for these measurements are 0.5%, and 3%, respectively. Hip structural analysis as described by Beck and colleagues19 was used to assess structural characteristics of three regions of the proximal femur from the DXA scans: the narrow neck region, which is located across the narrowest segment of the femoral neck; the intertrochanteric region along the bisector of the neck-shaft angle; and the femoral shaft, which is located 2 cm distal to the midpoint of the lesser trochanter. For each region the distribution of the bone mass across the bone is extracted, then subperiosteal width (SPW), bone cross-sectional area (CSA), which is equivalent to cortical area, cross-sectional moment of inertia (CSMI), and BMD (grams of bone divided by bone area) were measured. Section modulus (Z), a measure of bone bending strength, is determined by dividing the CSMI by one-half of the SPW. Like bone mineral content, CSA measures the amount of bone within the cross-section but expresses the quantity in terms of cortical equivalent surface area (important for axial bone strength) rather than mineral mass. The coefficients of variation for narrow neck, intertrochanteric, and femoral shaft regions, respectively, were as follows: BMD (1.7%, 1.3%, and 1.3%); SPW (5.3%, 1.8%, and 1.2%); CSA (2.6%, 2.2%, and 1.8%); CSMI (7.2%, 4.3%, and 3.7%); and Z (3.5%, 3.4%, and 2.1%). BMD and geometric measurements of multiple areas of the proximal femur were included as outcomes because each has been associated with fracture risk.26, 27
Ultrasound measurements were made over the distal radius and tibial shaft using a multisite bone sonometer (Sunlight, Omnisense, 7000S; BeamMed Ltd., Petah Tikva, Israel). This gives a measurement of bone speed of sound, which reflects the architecture and density of the bone.18 The coefficient of variation for this measurement in our laboratory is 1.5%.
Overnight fasting blood samples were drawn into Vacutainer serum separator tubes for the analysis of glucose, total cholesterol, HDL cholesterol, LDL cholesterol, and triacylglycerides. Blood was centrifuged at 2160 × g for 10 minutes at 20°C. An LX20 Beckman Coulter analyzer (Beckman Coulter Canada, Inc., Mississauga, ON, Canada) was used to analyze glucose, total cholesterol, HDL cholesterol, and triacylglycerides by enzymatic kits. LDL cholesterol concentrations were calculated using the Friedwald formula.
Analog film mammograms were collected at a breast cancer screening clinic. Quality control with a phantom was performed weekly. A technician, certified through the Canadian Association of Medical Radiation Technologists, reviewed the films. Mammograms were assessed for cancers, cysts, and any other abnormalities.
Transvaginal ultrasonography was used to assess endometrial thickness.28 High-resolution Ultramark 9 and ATL HDI 5000 ultrasound machines (Advanced Technologies Laboratories, Bothell, WA, USA) with 5-MHz to 90-MHz multifrequency convex array transducers were used to acquire imaging data. Endometrial thickness was measured as the distance from the anterior stratum basalis–myometrial junction to the posterior stratum basalis–to the myometrial junction, in the mid-sagittal plane. The transverse and sagittal planes of section that represented the largest dimensions of the fundal aspect of the endometrium were used for all measurements.
Measurements to determine effectiveness of intervention implementation
Strength in the lower body was assessed by determining the 1-RM during the hack squat; whereas upper body strength was determined by the 1-RM during the bench press, as described.29 Dynamic balance was measured as time taken to perform backward tandem walking (ie, toe to heel) over a distance of 6 m on a board that was 10 cm in width and raised about 4 cm off the ground. Number of errors (ie, number of times the participant stepped off the walking board) during the test was also recorded as a measure of dynamic balance. This test is sensitive to the effects of exercise training.30 Walking speed was assessed by timing walking over an 80-m course at a fast pace.31 The coefficients of variation for squat strength, bench press strength, dynamic balance (backward tandem walking time), and walking time over 80 m, were 31.3%, 8.2%, 19.3%, and 4.5%, respectively.
Uncontrolled intervention factors
A food frequency questionnaire, modified to include only Canadian fortification levels, was used to assess individual calcium, vitamin D, and total calories, and isoflavone intake (Block 98.2 FFQ; Block Dietary Data Systems, Berkeley, CA, USA). This food frequency questionnaire has been validated for assessment of isoflavones.32 The food frequency questionnaire was filled out on computer sheets by filling in bubbles beside appropriate food items, with pencil. This computer sheet was then sent for computer scanning to determine nutrient values (Nutrition Quest, Berkeley, CA, USA; www.nutritionquest.com). Physical activity levels outside of the training program were assessed by questionnaire.33 An arbitrary physical activity score is derived from this questionnaire based on frequency of physical activities performed in a typical week and classified as mild, moderate, and strenuous in intensity. This questionnaire has good reliability (test-retest correlation of 0.62–0.74) in middle-age to older adults.33, 34 Validity of this questionnaire has been confirmed in middle-age to older adults by correlating the physical activity score with treadmill time (r = 0.57) and maximal aerobic capacity (r = 0.56).34 We have previously shown that older men (59–76 years of age) who perform poorly on tests of muscular strength and power also have lower amounts of activities classified as “strenuous” from this questionnaire.35
Adverse events were collected on adverse event forms throughout the trial. Adverse events were probed for each time research assistants had contact with participants. This included a description of the adverse event, its relationship to the intervention (not related, unlikely, possibly, probably, definite), whether it was serious (ie, resulted in death, life-threatening, required hospitalization, or resulted in persistent disability) or nonserious, and its intensity (mild, moderate, severe, life-threatening). Endometrial and breast cancers, coronary heart disease, stroke, and pulmonary embolism were assessed for 2 years after the intervention, by questionnaire.
Sample size calculation
Sample size was determined based on the expected change in lumbar spine BMD. The baseline lumbar spine BMD for postmenopausal women is approximately 0.92 g/cm2 with an SD of 0.13 g/cm2.36 The expected annual increase in lumbar spine BMD for postmenopausal women taking isoflavones versus those not taking isoflavones is approximately 5% (based on the published literature at the time the study was started).37, 38 The expected annual increase in lumbar spine BMD for postmenopausal women on aerobic or strength exercise programs versus those not on exercise programs is approximately 1.8%.3 Assuming that the therapies are additive for increasing BMD, as shown in animal studies,14, 15 the predicted increases in the combined therapy group are clinically significant as an increase in lumbar spine BMD of 5% is associated with a relative risk reduction for fracture of 25%.39 With an alpha level of 0.05 and a power of 80% the required sample size was 78 per group (ie, total of 312). We expected a loss to follow-up of about 10%.36 Our recruited sample size of 351 was therefore deemed sufficient.
BMD variables from the standard DXA analyses (ie, total hip, trochanter, femoral neck, Wards, lumbar spine, and whole-body BMD) were analyzed by a three-factor multivariate analysis of variance (MANOVA), with between-group factors for exercise (exercise groups versus non-exercise groups) and isoflavone (isoflavone groups versus placebo groups) and a within-subjects factor of time (baseline versus 1 year versus 2 years). When a significant interaction was detected from this MANOVA, univariate tests were performed. When a significant interaction was detected from univariate tests, a Bonferroni post hoc test was used to determine differences between means. There were unequal subject numbers assessed for the remaining dependent variables; therefore, each of these were analyzed by separate three-factor analyses of variance, rather than MANOVA, with between-group factors for exercise and isoflavone, and a within-subjects factor of time. A Bonferroni post hoc test was used to assess differences between means when main effects or interactions were found. All analyses were done using Statistica version 7 (Statsoft, Chicago, IL, USA). Missing observations were assumed to be missing completely at random. Data were collected at baseline, 1 year, and 2 years. The 1-year data were used to account for any missing data at the 2-year mark (ie, data were carried forward). To verify that year 1 data were an appropriate substitute for year 2 data, all analyses were also run using only participants who completed all testing time points (ie, those that dropped out between years 1 and 2 were excluded). Data were analyzed on an intent-to-treat basis; ie, an attempt was made to follow up participants that did not adhere to the exercise or supplementation. Our intent-to-treat analysis included women who were removed during the study when they developed low BMD or were treated for low BMD. To evaluate this effect on our results, we also ran all analyses excluding these women. Adverse events across groups were compared by chi-square analysis. Baseline data are presented as means (SD). All other data are presented as absolute change scores and their 95% confidence intervals. Significance was set at alpha = 0.05.
Baseline data are presented in Table 1. Compliance with the exercise program was 77% of sessions completed for both the Ex and ExIso groups. Compliance with the isoflavone or placebo supplement was 70, 65, 74, and 64% for Ex, Iso, ExIso, and control groups, respectively (p > 0.05). Compliance with the calcium and vitamin D supplement was 90, 87, 91, and 89% for Ex, Iso, ExIso, and control groups, respectively (p > 0.05). The participant flow, along with losses and exclusions, is displayed in Figure 1. The final analysis included 298 women with the remainder lost to follow-up. Loss to follow-up was similar between intervention groups (p = 0.60). Final numbers analyzed for each outcome variable are presented in the outcome tables. There was some variation in the number of women analyzed across outcome variables. For bone ultrasound, some participants had too much subcutaneous fat to derive a valid measurement (n = 3). For hip geometric properties, some scans could not be analyzed due to poor patient positioning during the scan process (n = 4). For exercise tests, some participants could not complete the tests due to injury (n = 3 for bench press, n = 6 for hack squat). For blood analyses, incomplete tests were due to errors during the blood collection process or analyses (n = 3). Women with hysterectomies were not required to undergo testing for endometrial thickness (n = 15) and some women refused follow-up measurement due to the invasiveness of the procedure (n = 15). Incomplete questionnaire data (ie, dietary records, physical activity records) were due to lack of time or refusal by some participants (n = 10). At the end of the study, 17%, 13%, 14%, and 11% of women in the Ex, Iso, ExIso, and control groups, respectively, were able to correctly identify whether they were taking isoflavone or placebo (p = 0.72), with the remaining participants stating they either did not know which group they were in or guessing the incorrect group.
Table 1. Baseline Data by Intervention Group
Exercise (n = 86)
Isoflavone (n = 90)
Exercise and isoflavone (n = 87)
Control (n = 88)
All values are means (SD).
BMD = bone mineral density; CSA = cross-sectional area; CSMI = cross-sectional moment of inertia; Z = section modulus; SPW = subperiosteal width; SOS = speed of sound; HDL = high density lipoprotein; LDL = low density lipoprotein; TG = triglycerides; DBP = diastolic blood pressure; SBP = systolic blood pressure.
Lumbar spine BMD (g/cm2)
Total hip BMD (g/cm2)
Femoral neck BMD (g/cm2)
Trochanter BMD (g/cm2)
Wards BMD (g/cm2)
Whole body BMD (g/cm2)
Narrow neck BMD (g/cm2)
Narrow neck CSA (cm2)
Narrow neck CSMI (cm4)
Narrow neck Z (cm3)
Narrow neck SPW (cm)
Shaft BMD (g/cm2)
Shaft CSA (cm2)
Shaft CSMI (cm4)
Shaft Z (cm3)
Shaft SPW (cm)
Intertrochanteric BMD (g/cm2)
Intertrochanteric CSA (cm2)
Intertrochanteric CSMI (cm4)
Intertrochanteric Z (cm3)
Distal radius SOS (m/s)
Tibia SOS (m/s)
Total cholesterol (mmol/L)
Leisure physical activity score (arbitrary units)
Total energy intake (kcal/d)
Calcium intake (mg/d)
Vitamin D intake (µg/d)
BMD outcomes from the standard DXA analyses
Changes in BMD from the standard DXA analyses are presented in Table 2. There was an exercise × isoflavone × time interaction from the MANOVA (p = 0.043). Univariate tests indicated that there were significant exercise × isoflavone × time interactions for total hip (p = 0.0006) and trochanter (p = 0.004) BMD. ExIso experienced a greater decline in total hip and trochanter BMD than either Ex or Iso alone (Table 2). There were no main effects for exercise or isoflavone on any of the variables.
Table 2. Mean Absolute Changes (95% CI) From Baseline to 12 Months and From Baseline to 24 Months for Bone Mineral Density Measures Within Groups
Exercise (n = 77)
Isoflavone (n = 76)
Exercise and isoflavone (n = 72)
Control (n = 73)
Exercise main effect
Isoflavone main effect
All values are g/cm2. Main effects (95% CI) for exercise and isoflavone are presented in the last four columns.
CI = confidence interval.
Within-group change is significantly different from baseline (Bonferroni post hoc on the univariate interaction; p < 0.001).
Within-group change is significantly different from baseline (Bonferroni post hoc on the univariate interaction; p = 0.016).
Hip structural analyses from DXA scans and bone ultrasound
There were no exercise × isoflavone × time interactions for any variables from the hip structural analyses or ultrasound measures, although interactions approached statistical significance for several variables (ie, narrow neck and shaft BMD, shaft CSA, shaft Z, and distal radius speed of sound (SOS); p = 0.051 to 0.07; Table 3). There were no exercise or isoflavone main effects, except for an isoflavone main effect for shaft Z (p = 0.02). Non-isoflavone groups increased shaft Z from baseline to year 1 (p = 0.017) and from baseline to year 2 (p < 0.001), with no change in isoflavone groups (Table 3).
Table 3. Mean Absolute Changes (95% CI) From Baseline to 12 Months and From Baseline to 24 Months for Hip Structural Analysis and Bone Ultrasound Measures Within Groups
Exercise and isoflavone
Exercise main effect
Isoflavone main effect
Main effects (95% CI) for exercise and isoflavone are presented in the last four columns. Number of subjects per group for narrow neck characteristics: Exercise = 76; Isoflavone = 76; Exercise and isoflavone = 71; Control = 73. Number of subjects per group for shaft characteristics: Exercise = 77; Isoflavone = 74; Exercise and isoflavone = 71; Control = 72. Number of subjects per group for intertrochanteric characteristics: Exercise = 76; Isoflavone = 76; Exercise and isoflavone = 71; Control = 73. Number of subjects per group for distal radius characteristics: Exercise = 76; Isoflavone = 75; Exercise and isoflavone = 71; Control = 73. Number of subjects per group for tibia characteristics: Exercise = 76; Isoflavone = 75; Exercise and isoflavone = 71; Control = 73.
CI = confidence interval; BMD = bone mineral density; CSA = cross-sectional area; CSMI = cross-sectional moment of inertia; Z = section modulus; SPW = subperiosteal width; Intertroch = intertrochanteric; SOS = speed of sound.
Significant isoflavone main effect (p = 0.02). Bonferroni post hoc testing indicated that non-isoflavone groups increased shaft Z from baseline to 12 months (p = 0.017) and from baseline to 24 months (p < 0.001), whereas there were no changes in isoflavone groups.
There was an isoflavone × time interaction for LDLs (p = 0.03; shown as an isoflavone main effect on the change score in Table 4). LDLs decreased from baseline to 2 years in groups receiving isoflavone (p = 0.003). There were no significant differences between groups for changes in other lipid measurements, glucose, or blood pressure (Table 4).
Table 4. Mean Absolute Changes (95% CI) From Baseline to 12 Months and From Baseline to 24 Months for Lipids, Glucose, and Blood Pressure Within Groups
Exercise and isoflavone
Exercise main effect
Isoflavone main effect
Main effects (95% CI) for exercise and isoflavone are presented in the last four columns. Numbers in each group for total cholesterol were as follows: Exercise = 77; Isoflavone = 74; Exercise and isoflavone = 72; Control = 73. Numbers in each group for cholesterol/HDL ratio, HDL, and glucose were as follows: Exercise = 77; Isoflavone = 74; Exercise and isoflavone = 72; Control = 73.
Numbers in each group for LDL were as follows: Exercise = 77; Isoflavone = 73; Exercise and isoflavone = 72; Control = 73. Numbers in each group for diastolic and systolic blood pressure were as follows: Exercise = 77; Isoflavone = 76; Exercise and isoflavone = 72; Control = 73.
CI = confidence interval; HDL = high density lipoprotein; LDL = low density lipoprotein; TG = triglycerides; DBP = diastolic blood pressure; SBP = systolic blood pressure.
Isoflavone main effect (p = 0.03); Bonferroni post hoc test indicated that LDL decreased from baseline to 24 months in groups receiving isoflavone (p = 0.003).
Balance, exercise tests, anthropometrics, and body composition
There were exercise × time interactions for time to complete the backward tandem walking (dynamic balance) test (p = 0.032), hack squat strength (p < 0.001), bench press strength (p < 0.001), self-paced walking time (p = 0.049), and lean tissue mass (p = 0.013). There was also an isoflavone × time interaction for hack squat strength (p = 0.024). Post hoc analyses on change scores indicated that compared to the non-exercise groups, the exercise groups had a greater decrease in backward tandem walking time at year 2 (p = 0.017), a greater increase in hack squat and bench press strength at years 1 and 2 (p < 0.001), and a decrease in self-paced walking time at year 1 (p = 0.019) (data not shown). Exercise groups maintained lean tissue mass compared to a decrease in the non-exercise groups from baseline to 2 years (p = 0.007) (data not shown). The isoflavone groups increased hack squat strength to a greater extent than the non-isoflavone groups between years 1 and 2 (p = 0.021) (data not shown).
Uncontrolled intervention factors: physical activity level and dietary intake
There were no differences between groups during the intervention for leisure time physical activity (ie, physical activity outside of the intervention), total energy, calcium, vitamin D, and isoflavone intake (ie, excluding supplements given during the intervention) (Table 5).
Table 5. Mean Absolute Changes (95% CI) From Baseline to 12 Months and From Baseline to 24 Months for Uncontrolled Intervention Factors (Physical Activity Score and Dietary Intake) Within Groups
Exercise (n = 75)
Isoflavone (n = 73)
Exercise and isoflavone (n = 70)
Control (n = 70)
There were no differences between or within groups for changes from baseline.
CI = confidence interval.
Values only include nutrients from dietary intake and do not include the supplements given during the study.
Out of 351 women randomized there were 34 serious adverse events (SAEs) reported before, during, and 2 years after the intervention. Two SAEs occurred before the intervention (ie, between the time the participants were randomized and started the intervention), including hemorrhoid surgery in 1 Ex participant and a broken wrist in 1 Iso participant (both unrelated to our testing). The number of serious adverse events in each group, relative to original number randomized was as follows: Ex = 7 of 86 (8%); Iso = 7 of 90 (8%); ExIso = 13 of 87 (15%); Control = 8 of 88 (9%). There were no differences in rates of SAEs between individual groups, between exercise and non-exercise groups, or between groups receiving isoflavone and those not receiving isoflavone. Most SAEs were classified as “not related” or “unlikely” to be related to the interventions. Five SAEs were classified as “possibly” related to the intervention. These included a hysterectomy in the Ex group, a hysterectomy in the Iso group, a malignant spot that was removed from the face of a participant in the ExIso group, a case of ductal carcinoma in the ExIso group, and a stroke in the ExIso group.
Menopausal symptoms that were classified as adverse events included breast tenderness, decreased concentration, depression, fatigue, hot flashes, insomnia, numbness, vaginal dryness, vaginal itching, night sweats, subjective weight gain, migraine headaches, anxiety, irritability, and changes in bleeding patterns. These classifications are based on the index developed by St. Germain and colleagues.40 The rates in each group were as follows: Ex = 35 of 86 (41%); Iso = 15 of 90 (17%); ExIso = 9 of 87 (10%); and Control = 23 of 88 (26%). The groups taking isoflavone had significantly lower (24/177; 14%) adverse events related to menopausal symptoms compared to the groups not taking isoflavone (58/174; 33%) (p < 0.01). All menopausal symptom adverse events were rated as mild-moderate, except for one severe migraine headache in the ExIso group. None of these adverse events were considered serious. Fifteen participants in the Ex group, 6 in the Iso group, 2 in the ExIso group, and 6 in the control group quit taking isoflavone/placebo supplements because of adverse events. One participant in the Iso group, 2 in the ExIso group, and 5 in the control group dropped out of the study because of adverse events.
Six adverse events were detected with mammography. In the exercise group this included one mastectomy of both breasts after a lump was found after the intervention (serious, severe, not related), and one abnormal mammogram requiring biopsy that resulted in benign results (mild, possibly related). In the Iso group 1 participant had a cyst on her breast (mild, unrelated). In the ExIso group, 1 participant had calcification of breast tissue (mild, unrelated), and in the control group 1 subject had breast calcification and1 had a ruptured breast implant (both mild, unlikely).
There were no differences between groups with respect to changes in endometrial thickness throughout the study. The change (SD) in endometrial thickness in the groups who received isoflavone was from 3.0 (2.5) to 2.3 (1.4) cm and in the groups who did not receive isoflavone from 2.5 (1.6) to 2.2 (1.3) cm. Endometrial adverse events were reported by 4 participants: One in the Ex group (hysterectomy, SAE, moderate, possibly related), 1 in the Iso group (abnormal pap test, mild, possibly related), and 2 in the control group (abnormal polyp, mild, possibly related, and abnormal pap test, mild, unlikely related).
Our original hypothesis was that isoflavone supplementation and exercise training would be additive for increasing lumbar spine and proximal femur BMD. Each therapy preserved total hip BMD; however, the combination of therapies resulted in decreased total hip BMD along with the control group (Table 2). Thus, contrary to our hypothesis, there was a negative interaction between the two therapies for total hip BMD. This trend was also apparent for some of the BMD and geometric measurements at specific sites on the proximal femur (Table 3), with negative interactions approaching statistical significance for narrow neck BMD (p = 0.058), and femoral shaft BMD (p = 0.060), CSA (p = 0.051), and Z (p = 0.061). A negative interaction was also close to significance (p = 0.070) for the SOS measurement at the distal radius (Table 3), a predictor of bone architecture.18 Our results indicated a beneficial effect of exercise training or isoflavone by itself for preserving BMD at the total hip. Some of the exercise group changes at the specific femur sites, although not statistically significant, might be considered clinically significant (ie, the approximate 4.8% increase in femoral shaft BMD). A 5% increase in BMD is associated with about a 25% reduction in fracture risk.39 This beneficial effect was lost when exercise training and isoflavone supplementation were combined (ie, the group combining the two interventions did not differ over time compared to the control group). Neither exercise, isoflavone supplementation, nor their combination affected lumbar spine BMD (Table 2).
Our hypothesis that exercise and isoflavone supplementation would be additive for increasing bone mineral was based on a systematic review of exercise training that indicated a positive benefit of exercise for increasing lumbar spine BMD3 and a number of small studies indicating a beneficial effect of isoflavone supplementation.37, 38 Two recent meta-analyses also indicate that isoflavone supplementation is beneficial for increasing lumbar spine BMD.20, 41 Our results do not agree with these meta-analyses. Our study is one of the longest and largest studies in postmenopausal women and we were careful to follow all aspects of the randomized controlled design to avoid bias. This may account for the difference between our study and other smaller, less well-controlled studies. Recently published randomized controlled trials with relatively large sample sizes (n > 200) and long duration (2–3 years) found no effect of soy isoflavones on lumbar spine or hip BMD.42, 43
One explanation for the apparent interference of isoflavone supplementation with exercise involves the types of estrogen receptors on bone for the detection and transduction of mechanical strains to increase biochemical signals for bone formation (ie, activation of osteoblasts). Estrogen receptor-α, when activated by exercise, will increase osteoblast proliferation,44, 45 whereas estrogen receptor-β seems to block the beneficial effects of exercise on bone and has been termed the “anti-mechanostat” (because it downregulates the mechanical strain on bone induced by exercise).46, 47 Phytoestrogens, such as isoflavones, have a stronger affinity for estrogen receptor-β than for estrogen receptor-α.48, 49 The activation of estrogen receptor-β by isoflavones may cause a downregulation in the detection of strains from exercise; thereby reducing the effectiveness of exercise loading on bone. Our study is in contrast to a number of animal studies that demonstrated a beneficial effect of consuming isoflavones during exercise training.14–16 It may be easier to elevate blood levels of phytoestrogens in animal studies because of higher doses relative to body weight. At these higher blood levels, estrogen receptor-α may be activated along with estrogen receptor-β.49 On the other hand, one previous study using one-half the dose of isoflavones compared to our study reported a beneficial effect of combining exercise training and isoflavone supplementation on hip BMD.17 This raises the possibility that perhaps lower doses of isoflavone are more beneficial.
Aside from small benefits of exercise training by itself on hip BMD, the exercise groups also improved dynamic balance, as assessed by backward tandem walking, which may be important for fall and fracture prevention. Leg strength is associated with greater balance50 and reduced risk of falls,51 and this may be why our resistance-training and walking program improved balance. Along with exercise training, isoflavone supplementation was effective for improving leg strength (ie, hack squat strength). Muscle contains estrogen receptors52 and therefore estrogen or phytoestrogens may have an influence on muscle mass.53 Isoflavone supplementation in postmenopausal or perimenopausal women may increase leg muscle mass54, 55 and this supports our findings for leg strength.
LDL cholesterol (LDL-C) levels were reduced by approximately 6% with isoflavone supplementation (Table 4). This level of reduction is estimated to reduce coronary heart disease related mortality and total events by 4% to 8%.2, 56, 57 The reduction in LDL-C is greater than the 3.6% reduction reported in a recent meta-analysis,13 and may be related to the relatively high dosage and long intervention of our study. Isoflavones are thought to have a cholesterol-lowering effect through activation of estrogen receptors, which affect transcription of genes involved in lipolysis or lipogenesis.58, 59 Our exercise intervention had no effect on lipids. This may be related to the weekly volume of aerobic exercise that was prescribed (ie, 20–30 minutes of brisk walking four times per week). It is estimated that an exercise energy expenditure of between 1200 and 2200 kcal per week is required to affect blood lipids.60 We estimate that the participants in our study would have been expending approximately 475 kcal per week through their walking exercise prescription61; therefore, this would be below the threshold required for improving blood lipids.
We found no differences between groups for cancers during the study or 2 years after the intervention was completed. There were no differences between the isoflavone and non-isoflavone groups for abnormal mammograms or endometrial thickness. This is in agreement with meta-analyses that found no increases in breast abnormalities62 or endometrial thickness63 with isoflavone supplementation and a recent randomized controlled trial that assessed the effect of isoflavone doses of 80 to 120 mg aglycone equivalent (similar to the dose of our study) on endometrial thickness over 2 years.64 Isoflavone supplementation reduced the number of adverse events related to menopausal symptoms by about 58% compared to the women who did not receive isoflavone. Results in previous studies are quite mixed regarding effectiveness of isoflavones for reducing menopausal symptoms.63 Our results may be related to the relatively high dose of isoflavone given and the long intervention, and may also be related to the determination of menopausal symptoms through adverse event reporting. Identification of menopausal symptoms through adverse event reporting, rather than questionnaires given to each participant, would limit participants analyzed to those who experienced only menopausal symptoms that they considered severe enough to report as an adverse event.
There were a number of limitations with our study. Although we performed a MANOVA to account for multiple BMD measurements from the primary assessment of the DXA scans, we did not make any adjustments for the multiple statistical tests of secondary variables; therefore increasing the chance of statistical tests falsely showing significance. Our study was most likely underpowered to detect differences between groups for adverse events, as this requires much larger participant numbers, as was done in the Women's Health Initiative study on hormone replacement therapy.10
The participants assigned to the placebo exercise groups (ie, at-home flexibility training) were contacted every couple of weeks to probe for adverse events; whereas the participants assigned to the supervised training (ie, supervised twice per week) had contact with research assistants twice per week. This might have the effect of skewing the results so that more adverse events may have been recorded in the supervised exercise groups. This would not have an effect on women assigned to the different isoflavone groups because there were approximately equal numbers of women assigned to the isoflavone or isoflavone placebo groups who were contacted every couple of weeks versus twice per week.
Our main analyses included carrying over results from the 12 month time point to the end of the study for women who dropped out between 12 and 24 months. Our analyses also included women who were removed from the study because of low BMD or because they started taking medications to treat low BMD. Results did not differ when analyses were run excluding women who had dropped out between years 1 and 2 or excluding women who had been removed because of low BMD or taking medications for treatment of low BMD, with two exceptions. When analyses were run excluding participants who dropped out between year 1 and year 2, the ExIso group in addition to the control group decreased trochanter BMD from baseline to year 2 (p < 0.05) (data not shown). Also, the exercise × time interaction for backward tandem walking time was no longer significant when women with low BMD or taking medication for treatment of low BMD were removed from the analyses (p = 0.12) (data not shown).
In summary, isoflavone supplementation combined with exercise training appeared to interact negatively on BMD at the proximal part of the femur. Isoflavone supplementation was beneficial for reducing LDL-C levels and adverse events related to menopausal symptoms.
All authors state that they have no conflicts of interest
The study was funded by a grant from the Canadian Institutes of Health Research (application 124322). Archer Daniels Midland supplied the soy isoflavone tablets and placebo. Wyeth Consumer Health Care provided the calcium and vitamin D supplement. Saskatoon Health Region provided support through their blood analysis lab. The Saskatchewan Health Research Foundation provided support for a postdoctoral fellowship for Hassanali Vatanparast.
Authors' roles: Study design: PC, RP, AC, OO, SW, PP, and HB. Study conduct: PC, RP, AC, and OO. Data collection: PC, HV, RP, AC, and OO. Data analysis: PC, HV, RP, TB, and PP. Data interpretation: PC and HV. Drafting manuscript: PC. Revising manuscript content: PC, RP, SW, and TB. Approving final version of the manuscript: all authors. PC takes responsibility for the integrity of the data analysis.