How Many Women Lose Bone Mineral Density While Taking Hormone Replacement Therapy?Results From the Postmenopausal Estrogen/Progestin Interventions Trial

Background  The frequency of bone loss among women using postmenopausal hormone therapy is unknown.

Methods  We used data from the Postmenopausal Estrogen/Progestin Interventions Trial to address the frequency of bone loss among women using postmenopausal hormone replacement therapy. Of 701 women randomized to active treatment (conjugated equine estrogens alone or in combination with 1 of 3 progestins), 538 (76.7%) were adherent and had replicate bone mineral density (BMD) measures at baseline, 12 months, and 36 months. Of 174 placebo-assigned women, 132 (75.9%) were similarly eligible. Replicate BMD measures were used to calculate within-person measurement errors, which were then used to delineate cut points that defined bone losers with 97.5%, 95.0%, 90.0%, or 75.0% confidence.

Results  At the lumbar spine, during the first 12 months, 1.5% of hormone users lost BMD with 97.5% confidence, corresponding to a decline of −3% per year; during months 12 to 36, only 0.6% of treated women lost spinal BMD to this degree. An annual loss of −1% or more was the criterion for spinal bone loss at the 75.0% confidence level; 5.1% and 8.0% of hormone users met this criterion in the first year and in months 12 to 36, respectively. For the total hip, during the first 12 months, 2.3% of hormone-adherent women lost −3.0% per year or more, the 97.5% confidence definition of loss; 0.4% were so classified during months 12 to 36. To be 75.0% confident of hip BMD loss, a −1.0% per year decline in BMD was required; using this criterion, 14.5% and 11.8% of hormone users lost total hip BMD between 0 to 12 and 12 to 36 months, respectively. Among hormone-adherent women, at the spine and hip, there was virtually no overlap between women classified as bone losers in the first 12 months and those classified as such in the last 24 months. With 95.0% certainty, corresponding to an approximate loss of −2.5% at the spine and hip, 31.3% and 11.7% of placebo-adherent women lost spinal BMD in the first 12 and last 24 months, respectively. Parallel figures for the hip were 32.3% and 7.9%, respectively.

Conclusion  Bone loss while taking postmenopausal hormones is rare, and bone loss among untreated women is far from universal.

INFORMAL ESTIMATES of the percentage of women who lose bone while taking postmenopausal hormone therapy range from 10% to 30%, but formal evidence for these appraisals is lacking. If true, a high frequency of bone loss despite treatment would have important implications for clinical practice.

It is difficult to be certain of whether an individual's bone mineral density (BMD) has truly increased or decreased over time. When longitudinal measurements are made to assess possible change in BMD, the within-person measurement error must be considered.¹ Measurement error can be estimated by calculating the variation of 2 replicate BMD readings after repositioning taken on the same person by the same technologist on the same day. Using the observed within-person variability calculated from the replicate measures, and specifying the degree of "certainty" required (eg, 90% sure that the observed change is beyond the range of measurement variation), it is possible to estimate how much BMD loss over time would be required to be confident that the observed loss is not attributable to measurement error.

We used this theoretical framework to quantify a definition of "true" BMD loss and analyzed data from the Postmenopausal Estrogen/Progestin Interventions (PEPI) Trial to address 2 main questions: (1) How many postmenopausal women taking estrogen or estrogen/progestin therapy (hormone replacement therapy [HRT]) lose BMD? and (2) If bone loss while taking HRT is observed, are there factors that predict it? We also estimated the frequency of bone loss among placebo-adherent women.

Between December 1989 and February 1991, the PEPI Trial enrolled 875 postmenopausal women, aged 45 to 64 years, at 7 US clinical centers. The eligibility criteria and study design have been reported in detail.² In brief, women were between 1 and 10 years postmenopause; not taking estrogens or progestins for at least 2 months before screening; if treated for hypothyroidism, taking a stable dose of thyroid hormone for at least 2 months; without major contraindication to estrogen or progestin treatment; free of cancer within the prior 5 years; and without previous spine or hip fracture. Protocols were approved by each center's institutional review board, and each participant signed written, informed consent.

Participants were randomly assigned to one of the following: (1) placebo; (2) conjugated equine estrogens, 0.625 mg/d; (3) a combination of conjugated equine estrogens and medroxyprogesterone acetate, 10 mg, days 1 to 12; (4) a combination of conjugated equine estrogens and medroxyprogesterone acetate, 2.5 mg/d; or (5) a combination of conjugated equine estrogens and micronized progesterone, 200 mg, days 1 to 12.

Because the present analysis is primarily concerned with the possibility of BMD loss while taking active treatments, only treatment-adherent women were included. The PEPI Trial defined adherence as having taken at least 80% of pills (assessed by pill count) at each of the 6-month clinic visits for 3 years.³ The final PEPI Trial visit was attended by 847 women (97% of the participants). Of these, 613 (72%) were adherent to their assigned treatment. Compared with other treatments, adherence was relatively lower (63%) in women with a uterus assigned to treatment with conjugated equine estrogens only; this was due to protocol-mandated cessation of the study drug when women developed endometrial hyperplasia.^3,4 For the other active and placebo regimens, adherence ranged between 79% and 84% and did not vary significantly by treatment.³ In addition, participants must have had replicate BMD measures at baseline and at the 12- and 36-month visits. Because no differences in BMD outcomes were found previously among active treatments,⁵ we combined all active treatments into a single group for this study analysis. There were no exclusions based on baseline BMD.

Demographics, medical history, current medications, calcium supplement use, cigarette smoking, and use of noncontraceptive estrogens were assessed using standardized questionnaires. Self-reported physical activity was ascertained using an intensity-based questionnaire.⁶ Usual daily intakes of dietary calcium, energy from alcohol, and calcium supplement use were assessed using the Block Food Frequency Questionnaire.⁷ Height and weight were measured with participants wearing light clothing and no shoes. Baseline and 12-month estrone concentrations were measured with radioimmunoassay⁸; the sensitivity of this assay is 25.9 pmol/L, and interassay and intra-assay coefficients of variation are 15% and 16%, respectively. The amount of non–sex hormone–binding globulin–bound estradiol was measured using the method of Tremblay and Dube.⁹

The PEPI Trial BMD protocol has been described in detail.¹⁰ Briefly, dual-energy x-ray absorptiometry scans of the lumbar spine (L2-4), total hip, and hip subregions were done at baseline, 12 months, and 36 months with standard instruments (model 1000 QDR; Hologic Inc, Waltham, Mass). The BMD spine and hip quality control program included the following: scanning a spine phantom daily, reviewing daily phantom scans by the quality control center to ensure that all were within 1% of the standard BMD value, reviewing all spine and hip scans by the quality control center to identify morphological abnormalities, and reanalyzing unacceptable scans. Replicate BMD measures, with repositioning, were performed on each participant at each visit. For the present analysis, only lumbar spine and total hip BMD values are considered. The in vivo replicate measurement errors for percentage change in BMD are presented in the "Results" section.

The average of 2 measures, with repositioning, at the spine and hip was used as the BMD value for each woman at baseline, 12 months, and 36 months. Annualized percentage change values (ie, actual percentage change from the baseline to the 12-month visit [period 1] and the 12-month to the 36-month visit [period 2] divided by the time in years between visits) are the response variables for this analysis.

To create categorical definitions of "bone losers," we calculated compound errors of measurement for each period and applied a set of decreasingly stringent 1-sided confidence levels (97.5%, 95.0%, 90.0%, and 75.0%). True bone loss for each of these levels of certainty occurs when the observed BMD value falls below the lower bound (2.5%, 5.0%, 10.0%, and 25.0%, respectively) of the compound error range of confidence placed around the individual mean percentage change for each period in each participant. A complete description of compound measurement error calculations and confidence interval estimation is available from the authors.

To assess factors that might be related to bone loss while taking HRT, we proposed a set of candidate predictors based on the literature and plausible hypotheses. All variables were those measured at baseline, except change in estrone concentration. We examined the following: age (5-year increments), body mass index (BMI) (calculated as weight in kilograms divided by the square of height in meters, per SD unit), calcium supplement use (yes or no), total calcium intake (diet plus supplement), physical activity (highest compared with lower 2 tertiles), current alcohol consumption of 1 drink or more daily (yes or no), current cigarette smoking (yes or no), ever use HRT (yes or no), discontinued HRT to join the PEPI Trial (yes or no), estrone concentration (measured as picomoles per liter, per SD unit), estradiol concentration (measured as picomoles per liter [picograms per milliliter], per SD unit), and percentage change in estrone from baseline to 12 months (per 10% increase). In 32 women (6.0%), estradiol levels were below 11.0 pmol/L (<3.0 pg/mL), the lower limit of detection of the assay. Therefore, relations between estradiol and bone loss were examined 2 ways: first, with below-detection cases excluded, and then substituting a value of 10.3 pmol/L (2.8 pg/mL), a value two tenths of a unit lower than the detection level for those cases in which estradiol was undetectable. Because conclusions based on these 2 methods did not differ, we present the results of the latter.

To have sufficient bone loss outcomes to model predictors of categorical bone loss, we used the 75.0% confidence interval definition of bone losers. This definition corresponds to lumbar spine BMD loss of −1.0% per year during months 0 to 12 and −1.2% per year during months 12 to 36. During months 0 to 12 and 12 to 36, annual BMD losses of −1.0% and −1.1%, respectively, were the cut points that defined hip BMD loss (Table 1). First, we examined contingency tables of categorical candidate predictors of bone loss; those variables that were associated with at least a 50% increment or decrement in the odds of bone loss were retained for further consideration. We ran simple logistic regressions to assess the crude relations between the outcome of bone loss and continuous variables; those associated with bone loss at P≤.20 were retained for consideration. This screening procedure to narrow the list of candidate variables was performed separately for the spine and hip. We included this narrowed list of candidate predictors in logistic regression models for bone loss at the spine or hip during each period. We used 2 modeling strategies: first, we "forced" all candidate predictors into the multivariable model; second, we used backwards stepwise regression with P = .50 for enrollment and retention set at P = .10. Estimated effects and P values were not substantially different in the stepwise model from those in the forced model; we, therefore, present only the results of the latter method. All analyses were performed using statistical software.¹¹

Of 701 PEPI Trial participants who were assigned to active treatment, 538 (76.7%) were adherent and had replicate BMD measurements at all study visits. Among 174 placebo-assigned women, 132 (75.9%) were similarly eligible for this report. The average age at baseline of participants adherent to active treatment (56.1 years) was similar to that of placebo-adherent women (56.3 years). The BMI (SD) averaged 25.9 (4.4) and 26.5 (4.6) in active treatment– and placebo-adherent women, respectively. Mean (SD) values of lumbar spine BMD at baseline were 0.970 (0.153) g/cm² in the active treatment group and 0.965 (0.156) g/cm² in the placebo group. Average baseline values of total hip BMD were 0.859 (0.122) and 0.858 (0.125) g/cm² in these respective groups.

Table 1 presents the categorical cut point definitions of bone loss by study period and treatment category. The lower the confidence level, the lower the percentage change in BMD required to define true loss. For example, to be 97.5% confident that an individual woman has lost spinal BMD in period 1, her BMD must have declined by at least 3.0%; in contrast, if one requires 75.0% certainty that spinal bone loss has occurred during the first 12 months, then a 1.0% decline will suffice. The measurement errors of replicate hip BMDs are similar to measurement errors for the spine. Thus, the percentage changes that are necessary for the observed loss to exceed the error of measurement, with varying degrees of confidence, are close to those estimates for the spine. As the degree of certainty becomes less stringent, the change in BMD that defines hip bone loss during the first 12 months ranges from −2.9% to −1.0%.

As expected, the number of women classified as having lost spinal BMD increases as the criteria used to define loss are relaxed (Table 2). For example, in period 1, among those taking active treatments, only 8 women lost more than 3% of BMD, which defined spinal BMD loss with 97.5% certainty; that number increased to 27 if the 75.0% confidence (1% loss) cut point was used. Notably, no woman classified as a spinal bone loser by any criterion in the first 12 months remained in this category in months 12 through 36. Conversely, all participants who lost spinal bone in period 2 had not lost BMD in the preceding 12 months. While more women lost bone while taking placebo, about half of women using placebo did not lose bone. Similar to the observations in treated women, the overlap between period 1 and 2 bone loss was small.

Table 2 also shows that, using the 97.5% and 95.0% degrees of confidence, the proportions of women taking HRT who were classified as having lost hip BMD in period 1 or 2 were similar to the percentage classified as bone losers according to the spine measure. However, with the 90.0% or 75.0% certainty cut points, the numbers of HRT-treated women who manifested hip bone loss were roughly double those who were defined as losing spine BMD. Among placebo-adherent women, rates of hip BMD loss during the first 12 and the last 24 months of the PEPI Trial were not substantively different than those observed for placebo users at the spine. The overlap between total hip bone loss in the first 12 and the final 24 months was also strikingly low among placebo users: at the more stringent levels of confidence (95.0%-97.5%), only 1 woman who lost hip BMD in period 1 continued to do so in the next interval. At the lower levels of confidence, being classified as a hip bone loser in periods 1 and 2 remained rare in the placebo-adherent women, ranging between 4.7% and 13.4%.

The relation between selected factors and the likelihood of being classified as having spinal bone loss while taking HRT is summarized in Table 3. Each factor is controlled for all others listed. In both periods, older women were more likely to lose spinal bone, but this association was statistically significant only in period 2. Those with a higher BMI were less likely to be classified as bone losers in both study periods. The remaining factors examined, including baseline estrone and estradiol levels and change in estrone level from baseline to posttreatment at 12 months, were not predictive of spinal BMD loss while taking HRT. The mean (SD) 12-month value of estrone among treatment-adherent women was 388.35 (236.84) pmol/L compared with the pretreatment mean of 66.53 (43.99) pmol/L. Thus, the mean increase in estrone was 660% (range, −82.5% to 5185.7%). The change in estrone level did not vary substantively among active treatment arms (data not shown).

During the first 12 months of study, older age was marginally associated with being classified as having lost bone at the hip, while discontinuing HRT to join the PEPI Trial appeared to protect against bone loss (odds ratio, 0.53; P = .06). A greater percentage increase in estrone from baseline to 12 months was statistically significantly predictive of hip BMD loss during the first year of the study.

Many randomized, controlled trials^5,12 report that estrogen or estrogen/progestin combinations prevent bone loss. On average, compared with placebo, increases in BMD of between 3% and 10% are observed, depending on the bone measured, the age of the participants, and the duration of therapy.^5,12,13 However, these analyses examined the aggregate change in BMD of treated compared with untreated women, rather than the outcome of bone gain or loss for each woman. The principal findings of this analysis are that bone loss in women who take usual doses of HRT is rare and that there appear to be distinct groups of women who lose bone early (during the first 12 months of treatment) and later (during treatment months 12-36).

Because there is no universally agreed on standard that defines bone loss in a single person, we developed several cut points (percentage loss criteria) that were used to classify bone losers. The PEPI Trial data were well suited to this undertaking, as we measured BMD in duplicate at each visit, which allowed rigorous determination of the measurement error. Since PEPI Trial measurement errors for single BMD tests (1.5%-1.8%) are close to those cited by manufacturers¹⁴ and researchers,¹⁵ others can use our bone loss cut points. To calculate compound error confidence limits for the percentage change based on single BMD values at each visit, the compound measurement errors in Table 1 must first be divided by the square root of 2. Each scientist or clinician must decide the degree of certainty required; the percentage decrease that designates individual bone loss follows directly from that specification.

Is "bone loss while taking estrogen treatment" tantamount to "treatment nonresponse"? We do not believe these 2 concepts are identical. An individual woman who is taking HRT may be observed to lose bone; however, it is not possible to know how much bone she would have lost without it. For example, someone who lost 3% of spinal BMD while taking HRT might have lost 6% without HRT. Perhaps, then, a preferred term is suboptimal treatment response—we cannot really know if it is nonresponse.

Is suboptimal treatment response important? The answer to this question depends on whether the degree of BMD response is related to fracture prevention. This is unknown for estrogen; however, alendronate sodium–induced reduction in fracture risk is related to an increase in BMD.¹⁶ Although the results of treatment with alendronate must be extrapolated to HRT use with caution, they do suggest that the amount of BMD increase predicts actual fracture prevention.

A second issue that should be addressed when considering the potential implications of suboptimal BMD response to HRT is whether HRT dosage adjustments provide better efficacy or if there are alternative interventions that might effect a better response. A few dose-ranging studies^17,18 of HRT have reported greater mean BMD gains in relation to higher doses of estrogen and estrogen/progestin combinations. There are also limited head-to-head comparisons of the effects of different pharmacological treatments on BMD. In a randomized, controlled, factorial-designed study¹⁹ of alendronate and conjugated equine estrogens, spinal BMD increased by approximately 6% in the estrogen-only (0.625 mg/d) and the alendronate-only (10 mg/d) groups. In contrast, when examined in separate studies, the cumulative 3-year effect of 10 mg of alendronate daily on spinal BMD (which increased about 9% from baseline)²⁰ appeared to be greater than the spinal BMD effect of 0.625 mg of conjugated equine estrogens daily (which increased by approximately 5%).⁵ Thus, even if change in BMD is predictive of antifracture potency, it will be necessary to consider the BMD-gauged effectiveness of different treatments by direct comparisons, not by examining BMD responses across individual treatment trials.

We explored the impact of selected behaviors, diet, prior hormone treatment, endogenous estrone and estradiol levels, and change in estrone concentration after 12 months of HRT on the probability of losing bone while taking postmenopausal hormones. Because few women lost bone when the most stringent (90.0%-97.5% confidence) criteria were used, it was necessary to use the 75.0% confidence cut point to have sufficient bone loss events for modeling; the results must be interpreted in the context of that limitation. Factors that predicted bone loss while taking HRT differed between the 2 bone sites measured. Among treated women, a higher BMI was protective against spinal bone loss during periods 1 and 2. The mechanism by which a higher BMI protects against loss of spinal BMD in HRT users is uncertain. Although we did adjust for change in estrone level, which might be expected to vary inversely with body size and volume of distribution, we did not measure change in estradiol level during therapy. Estrone levels generally increase more than estradiol levels as a result of treatment with conjugated equine estrogens,²¹ but it is possible that greater adiposity is associated with preferential metabolism of conjugated equine estrogens to estradiol. Endogenous levels of estradiol are more strongly associated with BMD than are endogenous estrone levels.²² One might also postulate that direct skeletal loading due to body weight is more apparent at the spine than the hip.

Older age was the most robust predictor of spinal BMD loss during active treatment, manifesting a statistically significant positive association with spine bone loss during both periods; at the hip, this association was marginally significant and found solely in period 1. These findings are not concordant with results from observational studies,^23,24 which report that in older women even low levels of endogenous estrogens are protective against bone loss and fracture. Recker and colleagues²⁵ also report that, on average, in women older than 65 years, low-dose estrogen (0.3 mg of conjugated equine estrogens) with calcium prevents BMD loss. One might, therefore, posit a positive relation between increasing age and BMD response to treatment. However, the age range of our participants (45-64 years at baseline) is younger than that of women in the studies summarized.

Bone loss among untreated women was not universal. Rather, using the cut point of 75.0% confidence (corresponding to an annual decline of approximately 1% in spine and hip BMD), approximately 40% of women did not experience bone loss at the spine or hip during the 36 months of the PEPI Trial. These BMD results parallel findings from a population-based longitudinal study,²⁶ which described a cohort of individuals who did not have loss of renal function with aging, although the population on average did so. Our results in untreated women underscore the distinction between considering the aggregate experience of a group vs that of individual members.

The limitations of this study must be acknowledged. First, there is no criterion standard for true bone loss in a single person. Therefore, definitions of bone losers used in this analysis are arbitrary. However, we did formulate a reasonably wide range of cut points, based on varied degrees of certainty, so readers can apply their own standards for the degree of proof required. Second, although we attempted to grapple with the difficult concept of treatment nonresponse by examining BMD loss in individual postmenopausal hormone-treated women, it is not proved that the degree of increase in BMD while taking HRT is proportional to the amount of fracture prevention. Thus, the meaning of our findings in terms of fracture risk reduction remains uncertain. We explored possible predictors of loss of BMD among HRT-adherent women. Few predictors were found, and in some cases our findings were not in accord with preexisting literature. The rarity of bone loss in treated women required that we use the 75.0% certainty cut point to define bone losers for multivariable modeling. Even with this least stringent definition, spinal BMD losers numbered only 27 in period 1 and 42 in period 2; there were relatively more hip BMD losers (76 in period 1 and 62 in period 2). Thus, there may be misclassification of losers (because the definition of bone loss is lenient). Particularly for the spine, power to detect associations is limited because bone loss outcomes are few. Our findings may not apply to older women, as the maximum age at study enrollment of PEPI Trial participants was 64 years.

This study has several clinical implications. First, relatively young postmenopausal women treated with estrogen or estrogen/progestin rarely lose BMD. More important, in treated women, bone loss in the first 12 months appears to be independent of bone loss in the latter 24 months, suggesting that it is inappropriate to conclude that an individual is not responding to postmenopausal hormone use based on a 12-month follow-up BMD test. Conversely, one cannot assume that the absence of loss in the first 12 months assures no future loss. These findings should inform the interpretation of BMD measurements. The inverse relation between percentage increase in estrone level and BMD loss is unexplained, but argues against monitoring change in estrone levels to gauge treatment benefit. Finally, using the 90.0% confidence cut point, about half of placebo-treated women did not appear to lose BMD during 3 years of follow-up. They may represent a group of women at lower risk for osteoporosis.

Accepted for publication April 28, 2000.

The Postmenopausal Estrogen/Progestin Interventions Trial is supported by cooperative agreement research grants U01-HL40154, U01-HL40185, UL-HL40195, U01-HL40205, U01-HL40207, U01-HL40231, U01-HL40232, and U01-HL40273 from the National Heart, Lung, and Blood Institute; the National Institute of Child Health and Human Development; the National Institute of Arthritis and Musculoskeletal and Skin Diseases; the National Institute of Diabetes and Digestive and Kidney Diseases; and the National Institute on Aging; and by grant PHS 282-97-0025 from the Iris Cantor–UCLA Women's Center and the UCLA Center of Excellence in Women's Health, Los Angeles, Calif (Dr Greendale). Packaged medication and placebos for the Postmenopausal Estrogen/Progestin Interventions Trial were provided by Wyeth-Ayerst Laboratories, Philadelphia, Pa; the Schering-Plough Research Institute, Madison, NJ; and The Upjohn Company, Kalamazoo, Mich.

We thank Valeri Braun for her help in manuscript preparation.

George Washington University, Washington, DC: Vanessa Barnabei, MD, PhD (principal investigator) (formerly Valery T. Miller, MD, and John LaRosa, MD); Craig Kessler, MD (coinvestigator). The Johns Hopkins University, Baltimore, Md: Trudy Bush, PhD (principal investigator); Howard Zacur, MD, PhD; David Foster, MD; Roger Sherwin, MD (coinvestigators). Stanford University, Stanford, Calif: Marcia L. Stefanick, PhD (principal investigator) (formerly Peter D. Wood, DSc); Robert Marcus, MD; Katherine O'Hanlan, MD; Melissa Ruyle; Mary Sheehan, MS (coinvestigators). University of California, Los Angeles: Howard L. Judd, MD (principal investigator); Gail A. Greendale, MD (coinvestigator). University of California, San Diego: Elizabeth Barrett-Connor, MD (principal investigator); Robert Langer, MD (coinvestigator). University of Iowa, Ames: Susan R. Johnson, MD (principal investigator) (formerly Helmut G. Schrott, MD). University of Texas Health Science Center, San Antonio: Carl Pauerstein, MD (principal investigator); José Trabal, MD (coinvestigator). Wake Forest University School of Medicine, Winston-Salem, NC (Coordinating Center): Mark Espeland, PhD (principal investigator) (formerly H. Bradley Wells, PhD); George Howard, DrPH; Robert Byington, PhD; Claudine Legault, PhD; Beth A. Reboussin, PhD; Sally Shumaker, PhD (coinvestigators).

Reprints: Gail A. Greendale, MD, Division of Geriatrics, University of California, Los Angeles, UCLA School of Medicine, 10945 Le Conte Ave, Suite 2339, Los Angeles, CA 90095-1687.