The Northern California Kaiser Permanente Diabetes Registry
OBJECTIVE—In women with diabetes, the changes that accompany menopause may further diminish glycemic control. Little is known about how hormone replacement therapy (HRT) affects glucose metabolism in diabetes. The aim of this study was to examine whether HbA1c levels varied by current HRT among women with type 2 diabetes.
RESEARCH DESIGN AND METHODS—In a cohort of 15,435 women with type 2 diabetes who were members of a health maintenance organization, HbA1c and HRT were assessed by reviewing records in the health plan’s computerized laboratory and pharmacy systems. Sociodemographic and clinical information were collected by survey.
RESULTS—The mean age was 64.7 years (SD ± 8.7). The study cohort comprised 55% non-Hispanic whites, 14% non-Hispanic blacks, 12% Hispanics, 11% Asians, 4% “other” ethnic groups, and 4% with missing ethnicity data. Current HRT was observed in 25% of women. HbA1c levels were significantly lower in women currently using HRT than in women not using HRT (age-adjusted mean ± SE: 7.9 ± 0.03 vs. 8.5 ± 0.02, respectively, P = 0.0001). No differences in HbA1c level were observed between women using unopposed estrogens and women using opposed estrogens. In a Generalized Estimating Equation model, which took into account patient clustering within physician and adjusted for age, ethnicity, education, obesity, hypoglycemic therapy, diabetes duration, self-monitoring of blood glucose, and exercise, HRT remained significantly and independently associated with decreased HbA1c levels (P = 0.0001).
CONCLUSIONS—HRT was independently associated with decreased HbA1c level. Clinical trials will be necessary to understand whether HRT may improve glycemic control in women with diabetes.
- GEE, Generalized Estimating Equation
- HRT, hormone replacement therapy
- KPMCP, Kaiser Permanente Medical Care Program
- OHA, oral hypoglycemic agent
- SMBG, self-monitoring of blood glucose
In women with type 2 diabetes, the changes in sex-hormone levels, abdominal fat, and insulin metabolism that accompany menopause (1,2) may represent additional impediments in achieving good glycemic control. There is evidence that exogenous estrogens might reduce some of these adverse changes. Recent small randomized placebo-controlled trials (3,4) conducted among 40 diabetic women (or less) have shown that short-term hormone replacement therapy (HRT) with estrogen alone improves HbA1c level and reduces hyperandrogenicity in postmenopausal women with type 2 diabetes. In women without diabetes, the Postmenopausal Estrogen/Progestin Intervention trial (5) has shown that either estrogen alone or in combination with progestins slightly decreases fasting glucose levels compared with placebo, whereas estrogen in combination with medroxyprogesterone acetate increases 2-h glucose levels. As a result, there is an increasing demand by women with diabetes for information on the usefulness of HRT (6).
An association among HRT, lower fasting glucose, and insulin level has also been found in several large observational studies (78910). None of these studies included women with clinically diagnosed diabetes, and only one (7) included U.S. race-ethnicity minorities (African-Americans).
We report the HbA1c levels by current HRT use among a multiethnic population of 15,435 women with type 2 diabetes aged ≥50 years, who were identified by the Northern California Kaiser Permanente Diabetes Registry, and consider the possible effects of age, hypoglycemic therapy, diabetes duration, self-monitoring of blood glucose (SMBG), BMI, and education on the observed differences between women using and not using HRT.
RESEARCH DESIGN AND METHODS
The setting for this study was the Kaiser Permanente Medical Care Program (KPMCP) of Northern California, which is a group practice, prepaid health plan. The sociodemographic characteristics of KPMCP members are generally representative of the population living in the same area (11).
The population for this study was drawn from the Northern California Kaiser Permanente Diabetes Registry. The diabetes registry was started in 1993 by identifying probable diabetic individuals from automated health plan sources (including pharmacy prescriptions for diabetic medications), abnormal HbA1c values (≥6.7%), and any inpatient or outpatient diagnosis of diabetes (12). As of January 1996, the diabetes registry included 91,018 people aged ≥19 years and current KPMCP members. At that time, the registry was estimated to have 96% sensitivity for diagnosed diabetes when matched against two large mailed surveys of random membership samples conducted in 1990 and 1993. The registry has also been found to contain ∼3% false positives.
All of the diabetes registry members aged ≥19 years identified before 1996 received a survey questionnaire either by mail or computer-assisted telephone interview between January 1995 and March 1997. The response rate among diabetic women aged ≥50 years was 89% using this combined approach. The survey included questions on ethnicity, education, weight, height, age at diabetes diagnosis, current hypoglycemic therapy and date of starting insulin, diet and exercise as part of their treatment for diabetes, smoking history, and practice and frequency of SMBG. BMI was calculated as weight (kilograms) divided by height (meters) squared and used as a measure of overall obesity.
Type 2 diabetes was defined on the basis of reported treatment according to the following criteria: 1) reported diet and/or exercise and no hypoglycemic medication use, or oral hypoglycemic agents used alone or in combination with insulin; 2) insulin use only and diabetes diagnosis at age ≥30 years; and 3) insulin use only and diabetes diagnosis at age 20–29 years and initiation of insulin therapy >2 years after diabetes diagnosis.
HbA1c values measured between 1 January 1995 and 31 December 1996 were obtained from the computerized laboratory database, which captures all laboratory tests and results performed at the KPMCP regional laboratory. All HbA1c measurements were analyzed at the regional laboratory by a high-performance liquid chromatography analyzer from 1 January 1995 to 4 November 1995 and turbidimetric immunoinhibition assay from 5 November 1995 to 31 December 1996. The two methods were highly correlated (r = 0.986) (13). The majority (88%) of HbA1c levels reported in this study were measured by the turbidimetric immunoinhibition assay.
Use of HRT was assessed by reviewing records in the health plan’s computerized pharmacy system. This database is used for all pharmacy operations; the name and type of the pharmaceutical is entered directly by the pharmacy personnel as the prescription is filled. Computerized health plan benefit records were also reviewed to identify women who had or did not have prepaid medication benefits. Women were defined as current HRT users if they filled a prescription for estrogens and/or progestins, which are indicated for menopausal symptoms between 1–6 months before the date of the HbA1c laboratory test. This time interval was chosen because HbA1c reflects the glycemic control during the 2- to 3-month period before the blood sample was obtained, and most HRT prescriptions (96%) contained a 100-pill supply. All remaining women were classified as not current HRT users.
The assessment of HRT using a computerized pharmacy system database was validated among 295 diabetic women, aged ≥50 years, who also responded to a large mailed survey of random KPMCP membership that was sampled in 1996. In this survey, women were asked if during the past 12 months they had regularly used HRT. There was an agreement of 90% between the two sources. The sensitivity and specificity of the computerized pharmacy system database for assessing HRT used during the same 12-month period versus the self-reported HRT were 86 and 93%, respectively.
As of 1 January 1995, there were 28,837 women aged ≥50 years who were members of the diabetes registry. There were 22,105 women who were continuously enrolled in the health plan from 1 January 1995 through 31 December 1996. Women continuously enrolled in the health plan compared with women who were not continuously enrolled in the health plan were more likely to have HbA1c measured (84 and 33%, respectively) and to have filled an HRT prescription (30 and 14%, respectively). To ensure complete ascertainment of HRT use, only women who were continuously enrolled in the health plan (no gap allowed) were considered eligible for this study. Of these, 19,979 responded to the survey, and 579 (2.9%) stated on the survey that they did not have diabetes and were therefore excluded. Among the 19,400 women who confirmed that they had diabetes, 94.5% had type 2 diabetes, 1.9% had type 1 diabetes, and 4.5% could not be classified for type of diabetes because of missing data. Among women with type 2 diabetes, the 15,435 (85%) women who had HbA1c measured at least once became the final cohort for all analyses in this study.
The comparison of characteristics between women currently using HRT and women not currently using HRT was performed using two sample t tests for continuous variables and χ2 tests for categorical variables. Age-adjusted mean levels of HbA1c and BMI by HRT were computed and compared using analysis of covariance. A generalized estimating equation (GEE) model was constructed to assess the association between HRT and HbA1c level after taking into account the clustering of patients’ characteristics treated by the same physician and adjusting for age, ethnicity, education, BMI, hypoglycemic therapy, diabetes duration, SMBG, and exercise. Tests for interaction in the GEE models were performed to assess whether the association between HRT and HbA1c differed by ethnicity and age. Only the interaction term between HRT and age was statistically significant (P < 0.0001). However, in analyses stratified by age, the association between HRT and HbA1c had the same direction and was statistically significant in all age strata, although the association became less pronounced with age (Fig. 1). Confounders were included in the GEE models if their inclusion resulted in appreciable changes in the HRT coefficient or if the variable was shown by previous scientific publications to be associated with both the outcome and the exposure.
Statistical Analysis System version 6.11 was used for all analyses (16). All P values are for two-tailed tests with statistical significance defined as P ≤ 0.05.
Among 15,435 women aged 50–98 years (mean 64.7, SD ± 8.7) who had type 2 diabetes and HbA1c measured during the 2-year study period, 3,852 (25%) were currently using HRT before the HbA1c test. Among women currently using HRT, 62% were using unopposed estrogens, 36% were using opposed estrogen, and 2% were using progestins alone. The study cohort comprised 55% non-Hispanic whites, 14% non-Hispanic blacks, 12% Hispanics, 11% Asian/Pacific Islanders, 4% belonging to “other” race-ethnicity groups, and 4% women having missing data on race ethnicity.
Table 1 shows the characteristics of the study population by current HRT. Women currently using HRT were younger, leaner, better educated, and more likely to be non-Hispanic whites than women not using HRT. These differences were statistically and clinically significant. As shown in Table 1, women currently using HRT were also more likely to self-monitor their blood glucose levels, have a shorter diabetes duration, report exercise as part of their diabetes treatment, and be a past or current smoker versus women who were not using HRT. Hypoglycemic treatment varied very little by current HRT. Although all these differences were statistically significant, given the large sample size, they were not clinically meaningful.
Mean HbA1c levels were significantly lower in women currently using HRT than in women not using HRT, and these differences increased after adjusting for age. The difference in age-adjusted HbA1c mean levels between HRT users and nonusers was ∼0.5 points (Table 1). No differences in HbA1c level were observed between women using opposed estrogens and women using unopposed estrogens (P = 0.48). Both women using estrogens alone or estrogens in combination with progestins had significantly lower HbA1c mean levels than women not using HRT (age-adjusted mean [SE] 8.0 [0.03] and 8.0 [0.04], respectively, vs. 8.5 [0.02]; P = 0.0001).
Because women who were currently using HRT were more likely to have medication benefits for the entire 2-year study period than diabetic women not currently using HRT (Table 1), analyses were repeated after excluding the 2,232 women without continuous medication benefits, and similar differences in HbA1c levels were observed between current HRT users and nonusers (P = 0.0001).
To evaluate whether the observed association between HRT and decreased HbA1c levels varied by age, women were stratified in three age groups (50–59, 60–69, and ≥70 years), and age-specific mean HbA1c levels by HRT were calculated (Fig. 1). Although the association between current HRT and decreased HbA1c levels remained statistically significant in all of the age strata, it was most significant in women aged 50–59 years, intermediate in women aged 60–69 years, and smallest in women aged ≥70 years. To evaluate whether the association between HRT and decreased HbA1c levels varied by levels of covariates of HRT, age-adjusted HbA1c levels by HRT were calculated after stratifying women by type of hypoglycemic therapy, ethnicity, level of education, BMI, and diabetes duration (Fig. 1). The association between HRT and decreased HbA1c levels remained statistically significant and approximately the same magnitude in all of the strata. Similar associations between HRT and decreased HbA1c levels were found when women were stratified by exercise and SMBG practice (data not shown).
A GEE model was constructed to assess the association between HRT and HbA1c level after taking into account the clustering of patients’ characteristics treated by the same physician and adjusting for age, ethnicity, education, BMI, hypoglycemic therapy, diabetes duration, SMBG, and exercise (Table 2). As shown in Table 2, current HRT remained significantly and independently associated with decreased HbA1c levels in these multiple-adjusted models. Current HRT use was associated with 0.47-point lower HbA1c levels. Other significant independent predictors of decreased HbA1c levels were older age and treatment with diet alone. Insulin treatment compared with oral hypoglycemic agent (OHA) treatment was associated with significantly increased HbA1c levels. Other significant predictors of increased HbA1c levels were U.S. minority ethnic groups, obesity, and longer diabetes duration. There was a significant direct association between SMBG practice less than once a week and HbA1c, suggesting that women who were practicing SMBG had somewhat more severe diabetes. Education was not associated with HbA1c level.
Among the study cohort of 15,435 women with type 2 diabetes, 25% were currently using HRT before their HbA1c test, and HRT was significantly associated with decreased HbA1c levels. The association between HRT and decreased HbA1c levels was independent of age, ethnicity, education, BMI, type of hypoglycemic therapy, diabetes duration, SMBG practice, and exercise.
Women with type 2 diabetes from U.S. minorities were less likely to use HRT than diabetic women who identified themselves as non-Hispanic white. This race-ethnicity difference in HRT has been reported previously (17,18). In this setting, differences in age, ethnicity, and education were the most clinically significant differences between HRT users and nonusers. These results are consistent with recent national surveys (18,19) that found demographic and socioeconomic characteristics to be among the major predictors of HRT use.
The difference in HbA1c levels between women currently using HRT and women not using HRT was ∼0.5 points. Based on published findings from the U.K. Prospective Diabetes Study (20), the observed difference in HbA1c of 0.5 points between women currently using HRT and women not currently using HRT would translate roughly into a 10% reduction for any diabetes complications, an 18% reduction for microvascular complications, and a 7% reduction for myocardial infarction. Therefore, the observed difference in HbA1c between HRT users and nonusers is of clinical relevance not only in relation to glycemic control but also in relation to possible reduction in diabetes complications.
Our findings are consistent with the results of two small randomized placebo-controlled trials of estrogens and HbA1c in postmenopausal women with type 2 diabetes (3,4). Brussard et al. (3) found that the mean change in HbA1c in the 20 women who were assigned to estradiol was significantly greater than the mean change observed in the placebo group (−0.66 vs. −0.34%). Anderson et al. (4) studied 24 postmenopausal women with type 2 diabetes in a randomized double-blind crossover design. Mean HbA1c significantly decreased from 8.7% before estrogen therapy to 7.5% after estrogen therapy.
In nondiabetic postmenopausal women, exogenous estrogens have been found to improve whole-body insulin sensitivity (measured by an insulin tolerance test) (212223) and to increase hepatic insulin clearance (24). A study of women with type 2 diabetes (3) found that improved insulin sensitivity (measured by an euglycemic-hyperinsulinemic clamp) after estrogen therapy was mainly because of increased insulin sensitivity of the liver, resulting in increased insulin-mediated suppression of hepatic glucose production. In animal models, estrogens have also been found to suppress hepatic gluconeogenesis (25). Furthermore, it has been suggested that reduced hepatic glucose production observed after estrogen therapy might also be mediated by the reduced abdominal fat mass observed in HRT users (26,27). Finally, Anderson et al. (4) suggested that estrogen may improve glycemic control in postmenopausal women with type 2 diabetes by reducing their hyperandrogenicity. Reduced androgenecity may improve insulin sensitivity on skeletal muscle (28), which is the major tissue responsible for insulin-dependent glucose disposal during the fed state. However, estrogens have also been found to hamper glucose metabolism in the fed state by delaying the insulin secretion in response to a glucose load (29).
Some studies (21,30313233) suggest that the addition of progesterone to estrogens may reduce the beneficial effect of estrogens on glucose metabolism. Clinical trials (5,34) have shown that although fasting glucose and insulin levels decrease after estrogen therapy (with or without progestins), the addition of progestins worsens postchallenge glucose levels. In our study, 62% of women currently using HRT were using unopposed estrogens. The most common estrogen and progestin used was conjugated equine estrogen (90% of women used estrogen alone or in combination with progesterone) and medroxyprogesterone acetate (98% of women used progestins alone or in combination with estrogens), respectively. No differences in HbA1c levels were observed between women using unopposed estrogen and women using opposed estrogen. These findings suggest that in women with type 2 diabetes, the possible adverse impact of estrogens on glucose homeostasis during the fed state (e.g., higher postchallenge glucose levels) observed in women without diabetes (5) might be offset by improved glucose metabolism during the fasting state, possibly by reducing hepatic glucose production (3,25).
Previous studies (35) have shown that women who were currently using HRT were more likely to have health behaviors that are associated with improved glucose metabolism than women who were not currently using HRT. Similarly, among women with type 2 diabetes in this study, those who were currently using HRT reported exercising and self-monitoring their blood glucose more frequently than women who were not currently using HRT. However, the association between HRT and decreased HbA1c persisted after controlling for exercise and SMBG.
Premenopausal plasma insulin levels have been found to be lower among women who subsequently used HRT than women who subsequently did not use HRT (36). It is possible that HRT is more frequently prescribed to diabetic women with better glycemic control or less severe diabetes. However, no clinically relevant differences were found in the use of insulin (a disease severity indicator) between women using HRT and women not using HRT. Moreover, similar associations between HRT and decreased HbA1c levels were found in women who were treated with diet only, OHAs, or insulin, although these three groups of women have very different degrees of hyperglycemia. Therefore, it seems unlikely that selective prescribing patterns explain our findings.
The association between HRT and decreased HbA1c level was similar in all ethnic groups, as well as in all strata (except age) of education, BMI, hypoglycemic therapy, diabetes duration, and SMBG practice. The association was less pronounced in women aged ≥70 years. This was partially a result of a reduced variability in the outcome variable because HbA1c levels decreased with age.
In observational studies in which treatment is not randomly assigned, differences in the covariates between the treated and untreated group might be so large that they lead to biased estimates of the treatment effect, even after conventional adjustment. The propensity score methods have been proposed as an approach that reduces such potential biases (37). We performed the propensity score analyses (results not shown) in order to balance the covariates (e.g., age, education, hypoglycemic therapy, BMI, diabetes duration, exercise, and SMBG) among HRT users and nonusers. We first calculated the propensity score (defined as the conditional probability of being treated given the covariates) using logistic regression and then adjusted for the propensity score in the GEE models. In the propensity score–adjusted models, the direction and the magnitude of the association between HRT and HbA1c did not vary substantially from the standard GEE models and remained statistically significant (P < 0.0001).
Although such a robust finding in an observational study of the association between HRT and decreased HbA1c does not establish causality, it is consistent with the results of small randomized placebo-controlled trials of women with type 2 diabetes. There is a growing body of literature that points in the direction of a beneficial effect of HRT on glycemic control among women with diabetes. Long-term clinical trials, larger than those conducted thus far among women with diabetes, will be necessary to understand whether and to what extent HRT may improve glycemic control.
Supported by a grant (115-9630) from the American Heart Association Western States Affiliate and a grant (115-9675) from SmithKline Beecham Pharmaceuticals.
Address correspondence and reprint requests to Assiamira Ferrara, MD, PHD, Division of Research, Kaiser Permanente, 3505 Broadway, Oakland, CA 94611. E-mail:.
Received for publication 19 October 2000 and accepted in revised form 27 March 2001.
A table elsewhere in this issue shows conventional and Système International (SI) units and conversion factors for many substances.