Serum Sex Steroid Levels and Longitudinal Changes in Bone Density in Relation to the Final Menstrual Period

Abstract

Context:

The associations of serum sex steroid and FSH levels with change of bone mineral density (BMD) across the complete menopausal transition are incompletely understood.

Objective:

The objective of the study was to examine the associations of annual serum levels of FSH, estradiol (E₂), T, and SHBG with the rates of bone loss in 3 phases: pretransmenopausal [baseline to 1 year before the final menstrual period (FMP)], transmenopausal (1 year before to 2 years after the FMP), later postmenopausal (≥ 2 years after the FMP).

Design:

The design of the study was a repeated-measures, mixed-effects regression.

Setting:

This was a community-based observational study, with a 10-year follow-up.

Participants:

A total of 720 participants of the Study of Women's Health Across the Nation Bone Study participated in the study.

Outcome Measures:

Annualized lumbar spine (LS) and femoral neck (FN) BMD decline was measured.

Results:

The mean annual change in BMD was slowest in pretransmenopause (0.27%/year in FN) and fastest in transmenopause (2.16%/year in LS). In the pretransmenopausal phase, for every doubling of FSH level, LS BMD change was faster by −0.32%/year (P < .0001). In the transmenopausal phase, for every doubling of FSH level, LS BMD change was −0.35%/year faster (P < .0001); for every doubling of SHBG level, LS BMD change was −0.36%/year faster (P < .0001). In the later postmenopausal phase, for each doubling of the E₂ level, the LS BMD change was slower by +0.26%/year (P = .049); for each SHBG doubling, the LS BMD change was 0.21%/year slower (P = .048). The FN associations were weaker and inconsistent.

Conclusions:

Higher E₂ levels and lower FSH levels were associated with lower rates of LS bone loss in some but not all menopausal transition phases.

Because natural menopause is defined as 12 consecutive months of amenorrhea without other obvious pathological or physiological cause, it can be diagnosed only retrospectively (1). Bone mineral density (BMD) declines over the menopausal transition (MT) (2–12), with an accelerated rate of decline in the late perimenopausal period (13–16). Because the MT is characterized by changes in serum sex steroid and FSH levels (17), sex steroid and/or FSH-related changes likely contribute substantially to bone loss (9, 11, 12). The most rapid increases in FSH are within 2 years prior to the final menstrual period (FMP), whereas the most rapid decrease in estradiol (E₂) occurs at the FMP (17). This pattern of hormonal changes over the MT bears a striking resemblance to the pattern of declines in BMD over the MT. Lumbar spine and femoral neck BMD losses are greatest beginning 1 year before the FMP until 2 years after the FMP, a phase termed the transmenopause (18). In contrast, rates of bone loss between 2 and 5 years after the FMP, a phase termed the later postmenopausal phase, are lower than those during the transmenopause.

In prospective studies of the MT, lower BMD and/or greater BMD loss are associated with higher serum FSH levels and lower serum E₂ levels (4, 6, 9, 13, 14, 19–24). No longitudinal studies have examined associations of annual values of serially measured serum sex steroid and FSH levels with changes in BMD across the complete MT using the date of the FMP as the reference point. Cross-sectional and case-control studies of pre-, peri-, and postmenopausal women (25–33) and longitudinal studies of postmenopausal women (34) have found that high concentrations of SHBG, which binds to T, and low levels of bioavailable and/or total T are associated with low hip and spine BMD as well as increased risk of hip fracture (20, 35, 36). Among postmenopausal and elderly women, low E₂ concentrations, low T concentrations, and high SHBG concentrations are associated with a higher risk of hip, vertebral, and nonvertebral fractures, supporting the existence of significant sex steroid-related endocrinological influences on bone loss (36–39).

We used data from the Study of Women's Health Across the Nation Bone Health Study to describe the associations of annually-assessed sex steroid and FSH levels with rates of bone loss in 3 (mutually exclusive) FMP-anchored phases: pretransmenopause, transmenopause, and later postmenopause.

Materials and Methods

Study sample

The Study of Women's Health Across the Nation (SWAN) is a community-based longitudinal cohort study of 3302 women. At baseline, participants were aged 42–52 years, were premenopausal (menstruated in the past 3 months with no change in menstrual regularity in the past year) or early perimenopausal (menstruated in the past 3 months with decreased regularity in the past year), had an intact uterus with 1 or 2 ovaries, were not pregnant or lactating, and were not using exogenous reproductive hormones (40). Each site enrolled Caucasian women and women of 1 other self-identified racial/ethnic group: African-American women (Boston, Massachusetts; Detroit, Michigan, area; Chicago, Illinois; and Pittsburgh, Pennsylvania); Japanese women (Los Angeles, California); Hispanic women (Newark, New Jersey); and Chinese women (Oakland, California). At baseline and annually, participants were asked to complete questionnaires and provide fasting blood samples. Participants gave written informed consent. Sites obtained institutional review board approval.

Five of the SWAN sites participated in the SWAN bone substudy: Boston, Pittsburgh, Detroit, Oakland, and Los Angeles. Of 2413 potential participants, 2335 participants were enrolled in the bone cohort at baseline, 2181 of whom underwent at least 1 pair of consecutive BMD measurements. We excluded data from participants who were surgically menopausal (n = 183 participants), for whom the FMP date could not be definitively ascertained (n = 1622 participants) and who were pregnant/lactating (n = 1 participant). We also excluded observations from participants reporting the use of osteoporosis medications (risedronate, alendronate, calcitonin, raloxifene, teriparatide), exogenous sex steroids, aromatase inhibitors, or GnRH agonists by censoring their data beginning at the first reported use of the medication (n = 877 participants). For pre- and early perimenopausal women, we excluded data from observations collected outside the early follicular phase (87 of 2123 premenopausal and 1135 of 1397 early perimenopausal observations). Our final analytic sample consisted of 720 participants (4588 observations).

BMD measurements

At baseline and through annual follow-up visit 10, lumbar spine and femoral neck BMD were measured using Hologic 2000 (Pittsburgh and Oakland sites) or 4500A (Boston, Detroit, and Los Angeles sites) densitometers (Hologic, Inc, Bedford, Massachusetts). Positioning devices (Osteodyne, Research Triangle Park, North Carolina) were used to optimize reproducibility of measurements (41). Each study site used daily anthropomorphic spine phantoms, cross-site and cross-time calibration with a Hologic spine phantom, and on-site review of all scans for predesignated criteria. Five percent of scans, and all scans with potential problems, were reviewed by Synarc, Inc (Waltham, Massachusetts) (23). Two sites upgraded from 2000 to 4500A models at follow-up visit 8. These sites scanned 40 volunteers on both their old and new machines to develop cross-calibration regression equations.

Sex steroid assay methodology

Venipuncture was scheduled to occur before 10:00 am in the follicular phase (between days 2 and 5) of a spontaneous menstrual cycle occurring within 60 days of recruitment at the baseline visit and annually thereafter (17). If a follicular phase sample could not be obtained after 2 scheduling attempts, a random fasting sample was taken within a 90-day window of the anniversary of the baseline visit. Hormone assays were conducted as they were received at the SWAN Central Ligand Assay Satellite Services Laboratory, University of Michigan (Ann Arbor, Michigan). Annual quality assurance/quality control, conducted using the SWAN Central Ligand Assay Satellite Services Laboratory manual of operations, controlled for assay drift.

E₂ concentrations were measured in duplicate with a modified, off-line ACS-180 (E2–6) immunoassay (Bayer Diagnostics Corp, Norwood, Massachusetts). Inter- and intraassay coefficients of variation averaged 10.6% and 6.4%, respectively, over the assay range (17, 42). Serum FSH concentrations were measured in singlicate with a 2-site chemiluminometric immunoassay ACS-180 automated analyzer (Bayer Diagnostics). Inter- and intraassay coefficients of variation were 12.0% and 6.0%, respectively (17). The lower limits of detection ranged from less than 1.0 to less than 7.0 pg/mL for E₂ and less than 0.4 to less than 1.05 mIU/mL for FSH.

T assays were performed in singlicate using a modified rabbit polyclonal anti-T ACS-180 immunoassay (43). Serum SHBG concentrations were assayed in singlicate using de novo 2-site chemiluminescent assays based on competitive binding of dimethylacridinium ester-labeled SHBG to a commercially available rabbit anti-SHBG antibody. Intra- and interrassay coefficients of variation were 9.7% and 11.3% for T and 9.9% and 6.1% for SHBG (43). Lower limits of detection ranged from less than 1.9 to less than 3.2 nM for SHBG and less than 2.0 to less than 2.2 ng/dL for T.

If an FSH or sex steroid assay result was below the lower limit of detection for a given hormone, it was set equal to a unique random number between 0 and the lower limit of detection (24 observations for E₂, 0 observations for FSH, 9 observations for T, and 34 observations for SHBG).

Questionnaire-based and anthropometric measures

Baseline and annual follow-up questionnaires requested information regarding age, race, reproductive and menstrual history, medication use, and lifestyle factors. Menstrual cycle characteristics (frequency, regularity, duration, date of most recent menstrual period, date of FMP) were assessed on annual questionnaires.

At baseline and at annual follow-up visits 5 and 9, using a modified 1995 Block Food Frequency Questionnaire, we assessed typical weekly alcohol consumption (drinks per week) (44, 45). At visits in which the Food Frequency Questionnaire was not administered, information regarding usual alcohol consumption was collected using self-assessment questionnaires. Calcium and vitamin D supplement use were assessed annually and recorded as none, 1–3 days per week, 4–6 days per week, or daily.

Participants underwent body weight and height measurements at baseline and annually; body mass index (BMI) was calculated as body weight in kilograms divided by the square of height in meters.

Statistical analysis

We used a repeated-measures, mixed-effects regression to examine the associations of each hormonal predictor with annualized percentage change in BMD. Specifically, for each pair of consecutive annual BMD measures year_k and year_k+1, we examined the association of the hormone value at the first visit (year_k) with the percentage change in BMD over the ensuing year (year_k to year_k+1). Because consecutive annual follow-up visits were not always 365 days apart, the change in BMD between the visits was converted to an annualized percentage change as 100 ([BMD_{year k+1} minus BMD_{year k}]/BMD_{year k}) (365/number of days between the visits).

Prior work in SWAN identified 3 distinct phases of bone loss relative to FMP date (18). These phases of bone loss were termed pretransmenopause (baseline to 1 year before FMP), transmenopause (1 year before to 2 years after FMP), and postmenopause (2 years after FMP and thereafter) (18). Therefore, we examined associations of hormone predictors with annualized rates of bone loss in each of these 3 phases by including cross-product terms.

Serum levels (at the first of each pair of consecutive annual BMD assessments) of FSH, E₂, T, and SHBG were examined as primary predictors of annualized change in BMD, first individually in separate regression models and then mutually adjusted for each other in a single model. Due to skewed distributions, all hormonal predictors were log transformed to log base 2. This allows a more intuitive interpretation than does using natural log; each 1-U increment in log base 2 of the hormone value translates to a doubling of the untransformed level. To account for correlation among repeated measurements of the outcome within the same participant, we included random intercepts in each of the 3 phases.

Models were adjusted for clinical site, BMI, BMI², race, weekly alcohol intake, calcium supplement use, vitamin D supplement use, smoking, age, initial BMD, chemotherapy use, antiepileptic use, fertility medication use, number of visits using corticosteroid (1, 2, or ≥3 visits), and BMI race interaction. For pre- and early perimenopausal women, we excluded observations that were collected outside the early follicular phase; for late perimenopausal women, we included a covariate indicating whether serum sampling occurred during the follicular phase (yes/no). Time-varying covariates were measured at the same visit at the hormonal predictors. For the pretransmenopausal and transmenopausal phases, we included binary out-of-window indicator variables that flagged participants who were not sampled within days 2–5 of the menstrual cycle as well as cross-product terms to allow the effect of out-of-window sampling to be different in the 2 phases.

Results

Participant characteristics (Table 1)

Table 1.

Selected Characteristics of the Analytic Sample at Baseline and Rates of Change in BMD Over the Study Period^a

	Mean	SD	n	Frequency
Age, y	46.21	2.53
BMI, kg/m2	27.14	6.83
Alcohol, drinks per week	1.61	3.76
Race
African American			183	26%
Caucasian			330	46%
Chinese			106	15%
Japanese			97	13%
Menopausal stage
Premenopausal			422	59%
Early perimenopausal			290	41%
Current smoking			116	16%
Routine use of vitamin D supplement
None			645	96%
1–3 d/wk			9	1%
4–6 d/wk			3	1%
Daily			12	2%
Routine use of calcium supplement
None			489	73%
1–3 d/wk			52	8%
4–6 d/wk			34	5%
Daily			94	14%
BMD lumbar spine, g/cm2	1.06	0.14
BMD femoral neck, g/cm2	0.83	0.13
Annualized change in BMD
Annualized change, lumbar spine BMD
Pretransmenopause, %/yb	−0.28	2.45
Transmenopause, %/yc	−2.16	3.14
Later postmenopause, %/yd	−1.00	2.82
Annualized change, femoral neck BMD
Pre-transmenopause, %/year	−0.27	3.28
Transmenopause, %/year	−1.80	3.66
Later postmenopause, %/year	−0.99	3.47

^aSample sizes for each baseline characteristic do not total 720 due to missing data, and because some women entered the analytic sample after the baseline visit.

^bPretransmenopause refers to the period from participants' baseline measures until 1 year before the final menstrual period.

^cTransmenopause refers to the period from 1 year before the final menstrual period to 2 years after the FMP.

^dLater postmenopause refers to the period beginning 2 years after the FMP.

At baseline, the mean age of the participants was 46.2 years, 59% of participants were premenopausal, 41% of participants were early perimenopausal, and 54% of participants were non-Caucasian. The analytic sample participants were similar to the complete SWAN Bone Study cohort with respect to age, BMI, alcohol intake, race, menopausal stage, smoking, and calcium and vitamin D supplement use (data not shown). The median numbers of visit pairs per participant were 3 for the pretransmenopausal phase, 2 for the transmenopausal phase, and 1 for the later postmenopausal phase (data not shown).

Median and interquartile ranges of initial hormone levels and selected time-varying covariates for each of the 3 FMP-anchored phases are shown in Table 2. Compared with the beginning of the pretransmenopausal phase, FSH values were markedly higher and E₂ values were markedly lower in the beginning of the transmenopausal phase, but T and SHBG values were similar in the beginning of the 2 phases. Compared with the beginning of the transmenopausal phase, FSH values were higher still and E₂ values even lower in the beginning of the later postmenopausal phase, but SHBG and T values were similar.

Table 2.

Median and Interquartile Range of Initial Value of Predictors and Time-Varying Covariates at the Beginning of Each of the 3 FMP-Defined Phases^a

	Pretransmenopause (Baseline to 1 Year Before FMP)b Median (Interquartile Range)	Transmenopause (1 Year Before to 2 Years After FMP) Median (Interquartile Range)	Later Postmenopause (2 Years After FMP and Thereafter) Median (Interquartile Range)
FSH, mIU/mL: initial value	23.53 (17.24, 33.93)	83.38 (59.49, 114.65)	103.32 (75.60, 126.53)
E2, pg/mL: initial value	55.94 (36.48, 91.49)	24.75 (15.85, 48.95)	15.91 (12.40, 21.25)
T, ng/dL: initial value	36.59 (27.65, 48.00)	36.86 (27.39, 48.78)	34.75 (25.78, 47.73)
Serum SHBG levels, nM: initial value	43.98 (29.81, 59.53)	42.28 (29.04, 59.31)	39.75 (27.25, 58.29)
BMI, kg/m2: initial value	25.30 (22.16, 30.89)	26.31 (22.90, 31.75)	26.43 (23.00, 31.49)
Current smoker, %	14.69%	13.91%	11.67%
Routine use of vitamin D supplement, %
None	95.18%	92.53%	92.70%
1–3 d/wk	1.32%	1.51%	1.40%
4–6 d/wk	0.85%	1.51%	1.50%
Daily	2.64%	4.45%	4.40%
Routine use calcium supplement, %
None	68.89%	68.03%	68.16%
1–3 d/wk	8.50%	7.18%	8.24%
4–6 d/wk	6.37%	5.96%	4.87%
Daily	16.24%	18.82%	18.73%
Lumbar spine BMD, g/cm2: initial value	1.05 (0.97, 1.15)	1.00 (0.91, 1.11)	0.95 (0.88, 1.05)
Femoral neck BMD, g/cm2: initial value	0.81 (0.73, 0.91)	0.79 (0.70, 0.89)	0.75 (0.68, 0.84)

^aIn the pretransmenopausal phase, 635 observations were excluded due to measurements being made outside the follicular phase. In the transmenopausal phase, 411 observations were excluded due to measurements being made outside the follicular phase.

^bThe baseline measure refers to the first BMD measure obtained in each participant.

Adjusted associations of hormone levels with rates of lumbar spine bone loss by phase of bone loss

The adjusted associations of initial serum hormone levels (the value at the start of each pair of annual BMD measures) with annualized (absolute) percent changes in lumbar spine BMD within each of the 3 bone loss phases are displayed in Table 3.

Table 3.

Associations of Serum FSH and Sex Steroid Levels (Singly and in Combination) With Annualized Percentage Rates of Change in Lumbar Spine BMD Before, Around, and After the Final Menstrual Period (FMP)^a

	Annualized Rate of BMD Change (%/Y)b
	Pretransmenopause (Baseline to 1 Year Before FMP)c			Transmenopause (1 Year Before to 2 Years After FMP)d			Later Postmenopause (2 Years After FMP and Thereafter)e
	BMD Change (%/Y)	95% CI	P	BMD Change (%/Y)	95% CI	P	BMD Change (%/Y)	95% CI	P
Models with each hormonal predictor entered individually
FSH (per doubling)	−0.28	−0.41, −0.16	<.0001	−0.33	−0.50, −0.16	<.0001	−0.08	−0.37, 0.21	.59
E2 (per doubling)	0.10	0.00, 0.20	.04	0.05	−0.07, 0.18	.38	0.24	0.00, 0.48	.05
T (per doubling)	0.03	−0.14, 0.21	.70	0.13	−0.07, 0.32	.21	0.07	−0.16, 0.29	.56
SHBG (per doubling)	0.01	−0.12, 0.15	.85	−0.38	−0.55, −0.20	<.0001	0.21	0.00, 0.41	.048
Models with FSH, E2, T, and SHBG together
FSH (per doubling)	−0.32	−0.48, −0.16	<.0001	−0.35	−0.55, −0.15	<.001	−0.10	−0.40, 0.20	.52
E2 (per doubling)	−0.05	−0.17, 0.08	.46	−0.05	−0.20, 0.10	.52	0.26	0.00, 0.51	.049
T (per doubling)	0.02	−0.15, 0.19	.83	0.09	−0.11, 0.29	.37	0.01	−0.22, 0.24	.94
SHBG (per doubling)	0.02	−0.12, 0.16	.79	−0.36	−0.54, −0.18	<.0001	0.21	0.00, 0.42	.048

Abbreviation: CI, confidence interval.

^aModels estimate the association of the initial hormonal measurement for each year on the subsequent change in BMD in that year, in which change in BMD is modeled as 100 ([BMD_{year k+1} − BMD_{year k}]/BMD_{year k}) [365/number of days between the visits]. Regression models are adjusted for clinical research center site, BMI, BMI², whether serum sampling occurred during follicular phase, race, alcohol intake, frequency of calcium intake, frequency of vitamin D intake, smoking, age, baseline BMD, use of chemotherapy, epilepsy medication, fertility medication, or corticosteroids, and BMI race interaction.

^bTo convert the results per doubling into annualized percent bone loss per 50% increase in hormone level, multiple the results per doubling by 0.585.

^cMean percentage change per year for pretransmenopausal phase was −0.27%/y.

^dMean percentage change per year for transmenopausal phase was −1.80%/y.

^eMean percentage change per year for later postmenopausal phase was −0.99%/y.

Pretransmenopausal phase

The mean annual change in lumbar spine BMD in the pretransmenopausal phase was −0.28%/y. In models with each hormone predictor entered individually, only FSH and E₂ were statistically significantly associated with the rate of decline of lumbar spine BMD in the pretransmenopausal phase. Higher levels of FSH were associated with faster decline and higher E₂ levels with slower decline. Every doubling of FSH was associated with −0.28%/y faster change in lumbar spine BMD (P < .0001), and for every doubling of E₂ level, lumbar spine BMD change was +0.10%/y slower (P = .04). Alternatively expressed, every 50% increase in FSH was associated with −0.16%/y faster change, and every 50% increase in E₂ was associated with +0.06%/y slower change, in lumbar spine BMD. (To convert the results per doubling into annualized percent bone loss per 50% increase in hormone level, multiply the results per doubling by 0.585.) The 10th to 90th percentile range of FSH in the pretransmenopausal phase encompassed 2.5 doublings, and the corresponding range of E₂ included 3.8 doublings. Neither serum T nor SHBG levels were significantly associated with lumbar spine bone loss rates during the pretransmenopausal phase. When serum FSH, E₂, T, and SHBG levels were included in the same model, only FSH was independently associated with the rate of lumbar spine BMD decline (−0.32%/y faster change per doubling of FSH; P < .0001); neither, E₂, T, nor SHBG levels were significantly associated with rates of lumbar spine bone loss.

Transmenopausal phase

The mean annual change in lumbar spine BMD in the transmenopausal phase was −2.16%/y. In single-hormone models, only FSH and SHBG were significantly associated with BMD decline. For every doubling of FSH level, the lumbar spine BMD change was −0.33%/y faster (P < .0001); for each doubling of serum SHBG level, it was −0.38%/y faster (P < .0001). The 10th to 90th percentile range of FSH in the transmenopausal phase included 2.5 doublings, and the corresponding range of SHBG included 2.2 doublings. Neither serum E₂ nor T levels were significantly associated with the rate of lumbar spine BMD loss during the transmenopausal phase. In the model that included FSH, E₂, T, and SHBG levels all together, both FSH and SHBG remained independently associated with faster lumbar spine bone loss in the transmenopausal phase.

Later postmenopausal phase

The mean annual change in lumbar spine BMD was −1%/y in the later postmenopausal phase. Single-hormone models showed that greater E₂ levels and greater SHBG levels in the later postmenopausal phase were associated with slower decline in lumbar spine BMD. For each doubling of serum E₂ level, lumbar spine BMD change was +0.24%/y slower (P = .05), and for each SHBG doubling, it was +0.21%/y slower (P = .048). The 10th to 90th percentile range of E₂ in this phase included 1.7 doublings, and the corresponding range of SHBG included 2.2 doublings. Serum FSH and T levels were not statistically significantly associated with rates of lumbar spine bone loss during the later postmenopausal phase. Both E₂ and SHBG remained independently associated with BMD decline in the model that included FSH, E₂, T, and SHBG levels together.

Adjusted associations of hormone levels with rates of femoral neck bone loss by phase of bone loss

The adjusted associations of initial serum hormone levels (the value at the start of each pair of annual BMD measures) with absolute annualized absolute percent changes in femoral neck BMD within each of the 3 phases are displayed in Table 4.

Table 4.

Associations of Serum FSH and Sex Steroid Levels (Singly and in Combination) With Annualized Percentage Rates of Change in Femoral Neck BMD Before, Around, and After the Final Menstrual Period (FMP)^a

	Annualized Rate of BMD Change (%/Y)b
	Pretransmenopause (Baseline to 1 Year Before FMP)			Transmenopause (1 Year Before to 2 Years After FMP)			Later Postmenopause (2 Years After FMP and Thereafter)
	BMD Change (%/Y)	95% CI	P	BMD Change (%/Y)	95% CI	P	BMD Change (%/Y)	95% CI	P
Models with each hormonal predictor entered individually
FSH (per doubling)	−0.13	−0.30, 0.03	.11	−0.24	−0.46, −0.03	.02	−0.12	−0.49, 0.24	.52
E2 (per doubling)	0.03	−0.09, 0.16	.60	0.07	−0.09, 0.22	.39	0.04	−0.26, 0.35	.78
T (per doubling)	0.03	−0.18, 0.25	.77	0.20	−0.05, 0.44	.12	0.11	−0.17, 0.40	.45
SHBG (per doubling)	0.00	−0.17, 0.17	.99	−0.16	−0.38,0.06	.15	−0.04	−0.30, 0.22	.78
Models with FSH, E2, T, and SHBG together
FSH (per doubling)	−0.17	−0.37, 0.03	.10	−0.25	−0.50, 0.00	.05	−0.12	−0.50, 0.27	.56
E2 (per doubling)	−0.05	−0.21, 0.11	.55	−0.03	−0.22, 0.16	.73	0.01	−0.32, 0.34	.95
T (per doubling)	0.03	−0.19, 0.25	.79	0.18	−0.08, 0.43	.17	0.10	−0.20, 0.40	.50
SHBG (per doubling)	0.01	−0.17, 0.18	.93	−0.15	−0.37, 0.08	.21	−0.02	−0.29, 0.24	.86

Abbreviation: CI, confidence interval.

^aModels estimate the association of the initial hormonal measurement for each year on the subsequent change in BMD in that year, in which change in BMD is modeled as 100 ([BMD_{year k+1} − BMD_{year k}]/BMD_{year k}) [365/number of days between the visits]. Regression models are adjusted for clinical research center site, BMI, BMI², whether serum sampling occurred during follicular phase, race, alcohol intake, frequency of calcium intake, frequency of vitamin D intake, smoking, age, baseline BMD, use of chemotherapy, epilepsy medication, fertility medication, or corticosteroids, and BMI race interaction.

^bTo convert the results per doubling into annualized percent bone loss per 50% increase in hormone level, multiple the results per doubling by 0.585.

Pretransmenopausal phase

The mean annual change in the femoral neck BMD in the pretransmenopausal phase was −0.27%/y. When we included serum FSH, E₂, T, and SHBG levels either individually or together in the same model, none of the hormonal predictors were independently associated with rates of femoral neck BMD loss during the pretransmenopausal phase.

Transmenopausal phase

The mean annual change in femoral neck BMD in the transmenopausal phase was −1.80%/y. In single-hormone models, only FSH was significantly associated with BMD decline. For every doubling of initial serum FSH level, the femoral neck BMD change was −0.24%/y faster (P = .02). In the model that included FSH, E₂, T, and SHBG levels all together, greater serum FSH levels remained independently associated with faster femoral neck BMD loss in the transmenopausal phase.

Later postmenopausal phase

The mean annual change in femoral neck BMD was −0.99%/y in the later postmenopausal phase. None of the hormonal predictors was significantly associated with rates of femoral neck BMD loss during the later postmenopausal phase.

Discussion

In this longitudinal study, faster rates of lumbar spine bone loss were associated with higher levels of FSH in the pretransmenopausal phase, higher levels of FSH and SHBG in the transmenopausal phase, and lower levels of E₂ and SHBG in the later postmenopausal phase. Associations of higher E₂ levels with slower rates of bone loss were observed only in the later postmenopausal, not in the transmenopausal, phase. The only hormonal predictor that was significantly associated with rates of femoral neck bone loss was FSH level (in the transmenopausal phase). To our knowledge, no prior study has assessed longitudinal associations of sex steroids and FSH with BMD changes according to FMP-defined phases of bone loss.

It is unclear why the hormones that are associated with bone loss differ in various phases of the MT. It is possible that various molecular regulatory pathways are altered as the MT progresses so that FSH is the primary hormonal regulator of bone loss early in the MT, whereas E₂ and SHBG assume greater importance later. Under this paradigm, FSH must be an independent regulator of bone loss. Although initial reports showing that FSH receptor null mice fail to lose bone despite being hypogonadal suggested that high circulating FSH causes hypogonadal bone loss (46), a subsequent report that these mice have high serum T levels and lose bone after T levels are reduced by ovariectomy (47) cast serious doubt on the notion that FSH causes hypogonadal bone loss. Moreover, pharmacological lowering of FSH levels in hypoestrogenic postmenopausal women does not reduce bone resorption (48). Thus, the hypothesis that FSH independently causes bone loss is not widely accepted. We believe that the reason that FSH levels, rather than E₂ levels, are associated with rates of lumbar spine bone loss before the menopause is because serum FSH is a better measure of estrogenicity than is a single annual measure of serum E₂ during this period of a woman's life. During the menstrual cycle, serum E₂ levels vary markedly; thus, a single annual measure (particularly in the early follicular phase of the cycle) may not adequately reflect integrated estrogen levels. Substantive evidence supports a biological role for E₂ in bone metabolism, including effects on number, activity, and life span of osteoclasts and osteoblasts (49–51). Better measures of ovarian aging are needed to assess hormonal relationships with bone loss, particularly during the MT, when hormone levels are extremely labile.

We found associations of higher SHBG levels with faster rates of bone loss in the transmenopausal phase, independent of E₂ and T levels. Because SHBG levels are increased by estrogens (52), SHBG levels may also serve as a proxy for E₂ rather than exerting an independent effect on bone. It is also possible that SHBG has direct effects on bone loss that are independent of E₂ and T, for example by directly activating the androgen receptor (53).

We found no significant associations of T levels with rates of lumbar spine or femoral neck bone loss. Although cross-sectional studies of pre-, peri-, and postmenopausal women have found that circulating T levels are positively associated with higher BMD (26–29), some longitudinal studies of pre- and perimenopausal women have not found associations of T levels with BMD (21, 54). Androgens serve as estrogen precursors (55), and E₂ probably plays a greater role in maintaining BMD than does T (56).

In the current study, associations of sex steroids and FSH levels with bone loss were stronger at the lumbar spine than at the femoral neck. Although this finding may reflect a true biological difference in the hormonal regulation of bone loss, it may also be related to technical and/or analytic issues. For example, because lumbar spine BMD declines more than femoral neck BMD during the MT (18) and because lumbar spine measurements are more reproducible than femoral neck DXA measurements, associations between hormone levels and lumbar spine BMD should be easier to demonstrate than are associations between hormone levels and hip BMD.

Our study has limitations. The exclusion of participants with higher BMD (menopausal hormone therapy users) or lower BMD (surgically menopausal women) may have introduced bias. Generally, compared with mass spectrometry, the immunoassay has lower sensitivity and decreased specificity, the latter partly due to cross-reacting estrogen metabolites resulting in falsely high estimation of E₂ levels (57). Because over- and underestimation is not unique to either low or high E₂ values with immunoassay vs mass spectrometry (58), our results may be conservative, ie, biased toward the null. SWAN is in the process of conducting a cross-calibration study comparing its RIA results with those of mass spectrometry.

Strengths of our study include prospective assessment of FMP and annually ascertained information regarding osteoporosis risk factors. Our statistical models accounted for differing slopes of bone loss each year, optimizing the precision of our results.

In conclusion, higher E₂ levels and lower FSH levels were associated with slower rates of lumbar spine bone loss in some but not all MT phases. Our FMP-anchored approach is a refinement of prior approaches to the study of hormonal influences on bone loss over the MT. Elucidation of the biology underlying MT-related bone loss will require advances in the characterization of ovarian aging so we can identify women at particular risk of bone loss prior to fracture.