NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Humphrey LL, Takano LMA, Chan BKS. Postmenopausal Hormone Replacement Therapy and Cardiovascular Disease [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2002 Aug. (Systematic Evidence Reviews, No. 10.)

  • This publication is provided for historical reference only and the information may be out of date.

This publication is provided for historical reference only and the information may be out of date.

Cover of Postmenopausal Hormone Replacement Therapy and Cardiovascular Disease

Postmenopausal Hormone Replacement Therapy and Cardiovascular Disease [Internet].

Show details


Literature Search

We sought to review all published English-language literature with quantitative data evaluating the relationship between HRT use and CVD, CAD, and stroke in postmenopausal women. We conducted 2 MEDLINE searches. First, the topic of HRT and CVD was searched in the MEDLINE database from 1966 to December 2000. Second, the search was narrowed to specifically look for stroke and cerebrovascular disorders. Full search strings are listed in Appendix 1. The Cochrane Library was also reviewed. Two investigators reviewed all abstracts to identify papers for full-text review and also evaluated editorials, letters, and reviews to ensure that no key papers were missed in the MEDLINE and Cochrane searches. We searched the bibliographies of all review papers, meta-analyses, and original research studies to ensure that all papers evaluating the epidemiologic relationship between HRT use and CVD outcomes were retrieved from the medical literature, reviewed, and abstracted, including those predating the search dates.

Criteria for inclusion in the systematic review were that the study included postmenopausal women and that it was in the English language or in a key non-English journal. We included meta-analyses, randomized controlled trials, and observational cohort and case-control studies if they reported CVD (as defined above), stroke, or CAD incidence or mortality. We did not review articles dealing with the use of estrogen in men, or in association with pregnancy or lactation.

Appendix 2 summarizes the results of the literature searches. A total of 1,926 abstracts were identified and reviewed: 1,668 in the CVD search and 258 in the stroke search. Sixty-five studies about HRT and CVD met criteria for full text review. This includes 34 cohort studies, 24 case-control studies, 4 angiography studies of secondary prevention of CAD, 2 randomized controlled trials of secondary prevention of CAD with HRT, and preliminary findings from the Women's Health Initiative. Of these 65 studies, 24 cohort and 8 case-control studies describing stroke and HRT met criteria for full text review. Fourteen of the cohort studies and 3 of the case-control studies overlapped and were used in evidence tables for both the CVD/CAD and stroke sections.

Literature Synthesis and Preparation of Systematic Evidence Review

We abstracted study data onto data-collection forms prepared at the beginning of the review, and then created and organized evidence tables by study type. One difficulty in summarizing and interpreting this literature is that analyses and results are reported differently among studies. The methods used to assess HRT use and to define categories of use in case-control and cohort studies are shown in Tables 1 and 2, respectively. Most studies report point estimates comparing “ever” to “never” or non-use of HRT. Notably, “ever” use can be of any duration, and may reflect as little use as 1 to 2 months or just filling a prescription for HRT, or can be many years of use. However, some studies report their findings only as “current” or “recent” use, or after a specific duration of use. Current or recent use may reflect a broad range of durations of use, which often are not characterized within the studies, and may reflect weeks to years of therapy. These are important differences to note, since some studies have identified reduced risks only in association with current or ever use, and not past use. When possible, we differentiate these findings on the evidence tables and in the text of this report. Since many of the studies only evaluate HRT use on one occasion, there is substantial room for misclassification in many of the studies.

Table 1. Characterization of hormone replacement therapy use in case-control studies: assessment and definitions.


Table 1. Characterization of hormone replacement therapy use in case-control studies: assessment and definitions.

Table 2. Characterization of hormone replacement therapy use in cohort studies: assessment, prevalence and definitions.


Table 2. Characterization of hormone replacement therapy use in cohort studies: assessment, prevalence and definitions.

Hormone use was classified in each study as unopposed estrogen (ERT) or combined estrogen and progesterone (CHRT) when it was specified, which was infrequent. When the type of estrogen or progesterone therapy was not specified, or the data were analyzed and/or reported together, the exposure was categorized as HRT.

In general, well-conducted randomized controlled trials provide more valid results than observational studies evaluating causal relationships. However, with the exception of preliminary data from the Women's Health Initiative, there are no randomized controlled trials of primary prevention of CVD with HRT. Therefore, our review involves predominantly observational studies, which are limited by lack of randomization and the potential for substantial selection biases.

In reviewing the quality of the evidence evaluating the relationship between HRT and CVD, cohort studies are methodologically stronger since they assess exposure prior to the onset of disease and are not as dependent on the recall of HRT use as case-control studies. In addition, case-control studies are limited by the refusals of some patients with disease to participate, as well as by a frequent inability to evaluate patients with severe disease or those who have died. This can be an especially important problem in CVD epidemiology because a significant number of incident CVD events result in death. Women who have died of CVD may have had different HRT exposures from those of the women who did not, which might bias the results of these types of studies. In addition, case-control studies are prone toward recall bias, where cases remember or report exposures differently than controls. Finally, HRT use may be different among those who choose to participate and those who do not, which could bias the findings of the study. These issues will be discussed below. In this review, therefore, we assign more importance to the results from cohort studies. We did not review cross-sectional studies, since they are limited by prevalence/incidence/survivor bias and are often not helpful in evaluating potentially etiologic relationships. In ranking the quality of both cohort and case-control studies, we gave significant weight to adequate control of potential CVD risk factors because of known differences in risk profiles among women who use HRT and those who do not.11, 46 Criteria for evaluating study quality were created by the third US Preventive Services Task Force and are described in Appendix 3. Table 3 shows each study's rating by quality; reasons for study quality ranking are described in the evidence tables.

Table 3. Studies contributing to CVD review by quality ranking.


Table 3. Studies contributing to CVD review by quality ranking.

After reviewing and rating all of the epidemiological studies displayed in the evidence tables, we limited our formal review and meta-analyses to studies meeting two criteria: 1) the study is a population-based, case-control study evaluating the risk of incident CVD, CAD, or stroke, or death from CVD, CAD, or stroke, with adequate control of major CVD risk factors, or a cohort study with internal controls evaluating CVD, CAD, and/or stroke incidence or mortality with adequate control for major CVD risk factors and at least 3 years of followup, and 2) the study was rated as fair or good quality based on the criteria described in Appendix 3.47

Only studies of fair or good quality are described in detail in our report and included in the meta-analyses, although all are summarized in the evidence tables. Some studies might be of fair or good quality in one type of analysis and of poor quality in another analysis. For example, some studies age-adjusted certain findings and used multivariate analyses for others. When this occurred, only the information of fair or good quality from the study was included in the results, discussion, and meta-analyses sections of our report. Unless stated, all relative risk estimates discussed are adjusted for some important covariables. In studies with multiple publications from the same cohort or population, only data from the most recent publication were included in the discussion and meta-analyses, with reference in the text to older publications if they presented unique findings.

Three other studies discussed include the HERS trial, the Estrogen Replacement and Atherosclerosis (ERA) study, and preliminary information from the Women's Health Initiative (WHI) study. In the last 2 years, data have been published from the HERS and ERA studies, both randomized controlled trials of HRT in women with known CAD. Although these data may not be generalizable to women without CAD, it is important to include these data in this review, since the findings are concerning and provide the only data on secondary prevention among women randomized to HRT or placebo. In addition, preliminary data from the Women's Health Initiative were released in the spring of 2000 in the lay press, and although these data have not yet been published, the preliminary findings are included in this systematic review.48


We performed meta-analyses using a random-effects model to determine whether HRT has any impact on the risk of total CVD, CAD, and stroke mortality and incidence (see Appendix 4). All studies reported relative risk estimates. The logarithm of the relative risk (logRR) was assumed to have a normal distribution. If confidence intervals or p-values were reported, then standard errors for the logRR were calculated. The adjusted logRR and the corresponding standard errors were the data points for the meta-analyses.

The model used allows for stratification by three categories of HRT therapy (HRT unspecified, ERT alone, and combined ERT+HRT). Mean relative risks and confidence intervals are estimated for each HRT type. Separate models were fit on each outcome. When the data were sufficient, summary risks were determined by exposure type (past, current, ever) if the summary relative risks differed by exposure type. Finally, a global measure of use-- “any use”-- was determined based on mutually exclusive data points in the above 3 categories. In the stroke meta-analysis, the estimated relative risks of stroke associated with current, ever, past, or any use of HRT were similar, and therefore only one summary estimate was determined for each stroke outcome. A model was only fit if there were 3 or more data points available.

The Bayesian data analytic framework was used to fit the model. Inference on the parameters was done via posterior probability distributions. WinBUGS was used to analyze the data; this software uses a method of Markov Chain Monte Carlo called Gibbs Sampling to simulate posterior probability distributions. Noninformative prior probability distributions were used. Inference was made on 10,000 simulated draws (2,000 draws from 5 chains) from the posterior distribution after adequate convergence.

PubReader format: click here to try


  • PubReader
  • Print View
  • Cite this Page

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...