Current Use of Unopposed Estrogen and Estrogen Plus Progestin and the Risk of Acute Myocardial Infarction Among Women With Diabetes
The Northern California Kaiser Permanente Diabetes Registry, 1995–1998
Background— Little is known about hormone replacement therapy (HRT) and risk for myocardial infarction (MI) in diabetic women. We examined associations of current HRT, estrogen dosage, and time since HRT initiation with risk of acute MI in diabetic women.
Methods and Results— Cox proportional hazards models, with current HRT modeled as a time-dependent covariate, were used to assess the 3-year risk of MI. Among 24 420 women without a recent MI (mean age 64.9 years), 1110 incident MIs were identified. After adjustment for cardiovascular risk factors, current HRT was associated with reduced MI risk (relative hazard [RH] 0.84, 95% CI 0.72 to 0.98). The RH for MI associated with current estrogen plus progestin use was 0.77 (95% CI 0.61 to 0.97), and the RH for MI associated with current unopposed estrogen use was 0.88 (95% CI 0.73 to 1.05). Women were at reduced MI risk if they were taking a low or medium dose of estrogen (equivalent to <0.625 or 0.625 mg of conjugated estrogen, respectively) but not a high dose (>0.625 mg of conjugated estrogen or its equivalent). Among those whose current use of HRT was <1 year, the RH for MI was 1.03 (95% CI 0.74 to 1.44), whereas among users for ≥1 year, the RH was 0.81 (95% CI 0.66 to 1.00). Among 580 women with a recent MI (mean age 69.2 years), 89 recurrent MIs were identified. An increased risk of recurrent MI was observed among current HRT users (RH 1.78, 95% CI 1.06 to 2.98), which was higher among those with current use for <1 year (RH 3.84, 95% CI 1.60 to 9.20).
Conclusions— In women without a recent MI, use of estrogen plus progestin was associated with decreased risk of MI. However, HRT was associated with increased risk of MI in women with history of a recent MI. Data from clinical trials in diabetic women are needed.
Received June 28, 2002; revision received September 30, 2002; accepted October 1, 2002.
Diabetes confers approximately a 2- to 4-fold increased risk of coronary heart disease (CHD).1 In the United States, the prevalence of diabetes among women aged 50 years and older is at least 15%.2 Little is known about the possible associations of hormone replacement therapy (HRT) and the risk of CHD in diabetic women.
See p 2
For women without diabetes, a recent meta-analysis3 reported that risk of CHD was decreased by ≈20% in women who were currently using HRT but not in women who were past HRT users. The Heart and Estrogen/Progestin Replacement Study (HERS),4 a randomized trial of HRT and CHD in women with coronary disease, did not show a protective effect of HRT on the risk of CHD. The Women’s Health Initiative (WHI),5 a large-scale randomized trial of HRT and primary prevention, has reported that among women assigned to estrogen plus progestin, there was an increased risk of CHD. In addition, estrogen plus progestin increased the risk of several adverse outcomes, including stroke, pulmonary embolism, and breast cancer.5 Secondary analyses among small subgroups of women with diabetes from case-control studies have been equivocal, some reporting a nonsignificant reduced risk6,7⇓ and others a nonsignificant increased risk8,9⇓ of CHD with exposure to HRT.
We examined the risk of acute myocardial infarction (MI) associated with current use of different hormone regimens, estrogen doses, and time since HRT initiation in a 3-year follow-up observational study of a cohort of 25 000 diabetic women aged 50 years and older identified by the Northern California Kaiser Permanente Diabetes Registry.
The setting for this study was the Kaiser Permanente Medical Care Program (KPMCP) of Northern California, a group practice, prepaid health plan that provides comprehensive medical services to more than 2.7 million members. The Kaiser Permanente membership closely approximates the population living in the same geographic area except with regard to income, where KPMCP members underrepresent the very poor and the very wealthy.10 A total of 30 315 diabetic women aged 50 years and over who were members of the Northern California Kaiser Permanente Diabetes Registry11 were invited to complete a written or telephone survey between January 1995 and March 1997. Of these, 26 037 (86%) responded to the survey. After exclusion of 1037 responders who stated they did not have diabetes, 25 000 women formed the study population.
The survey included questions on ethnicity, education, weight, height, age at diabetes diagnosis, hypoglycemic therapy, diet, exercising as part of treatment for diabetes, history of hypertension, smoking, and alcohol consumption. Body mass index was calculated as weight (in kilograms)/height (in meters)2. Baseline glycosylated hemoglobin A1c (HbA1c) levels were obtained from computerized laboratory records.
This cohort was followed up from January 1995 until an acute MI (n=1199), death not related to acute MI (n=2010), departure from the health plan (n=1975), or December 1998 (n=19 916), whichever came first, by use of the health plan’s computerized hospitalization and membership files and State of California death certificate files. Incident acute MI was defined by the International Classification of Diseases–9th Revision (ICD-9) as code 410.x in the primary hospital discharge diagnosis or as underlying cause of death. This approach has been shown to be highly specific (98%) for clinical diagnosis of acute MI.11
The health plan’s computerized pharmacy system, which was fully implemented in August 1994, was used to assess baseline use of lipid-lowering medications and to identify all prescriptions of HRT, along with the number of pills or patches and dose for each prescription. When a new prescription was filled before the supply from the previous one had elapsed, we assumed that remaining pills were carried over. The sensitivity and specificity of the computerized pharmacy system database for assessing HRT versus self-reported HRT were 86% and 93%, respectively.12
Age-adjusted comparisons of baseline characteristics by HRT status were performed by the Mantel-Haenszel method.13 The Cox proportional hazards model14 was used to provide point and interval estimates of the relative hazard (RH) of fatal and nonfatal acute MI associated with current HRT, current HRT regimens, or current estrogen dose, all entered as time-dependent covariates that allowed for changes in current status over time. For the purpose of defining HRT exposure in time-dependent models, women were defined as current HRT users at a given date if their cumulative medication supply lasted through that date or had ended less than 3 months earlier; otherwise, they were defined as not being current HRT users at that date. Current estrogen dose at a given point of time was classified as follows: (1) low dose: <0.625 mg of oral estrogens or <0.02 mg of estradiol; (2) medium dose: 0.625 mg of oral estrogens or 0.05 mg of estradiol; or (3) high dose: >0.625 mg of oral estrogens or 0.1 mg of estradiol.8 Two models were considered; the first model adjusted only for age, and the second adjusted for age, self-reported ethnicity, education, body mass index, hypoglycemic therapy, diabetes duration, hypertension, smoking, alcohol consumption, exercise, baseline use of lipid-lowering medications, and HbA1c. For each covariate, the category “unknown” was entered for missing data in the Cox models (percentages of missing data are shown in Table 1).
Cox proportional hazards models were also used to examine whether the risk of acute MI varied by time since HRT initiation. In time-dependent models, women who were not current HRT users at a given point in time were classified according to HRT exposure before that point as previously exposed if they filled at least 1 HRT prescription before that time or not previously exposed (referent group) if they did not fill any HRT prescription before that time. Women who were current HRT users were further classified by time since HRT initiation (<1 year, ≥1 year, or unknown). HRT initiation was defined as follows: (1) for women with 1 or more gaps in HRT use (a gap was defined as a period of >6 months between the date the cumulative supply ended and the date of the next filled prescription), initiation was the date of the first HRT prescription filled after the last gap; (2) for women with no gaps in HRT use, HRT initiation was the date of the first filled HRT prescription found in our database if it occurred after the baseline; and (3) if there were no gaps in HRT use and the first HRT prescription was filled before the baseline date, time since initiation was classified as unknown if duration was <1 year; otherwise, it was classified as 1 year or longer. Because the pharmacy system was not in operation before August 1994, these analyses were restricted to women whose baseline dates were on or after March 1, 1995. This allowed a minimum of 6 months of observation in all women to distinguish HRT initiation from continuing use.
Tests for interaction in the Cox models were performed to assess whether the association between HRT and acute MI risk differed by the occurrence of a hospitalization for MI during the year before baseline. Because the interaction term was statistically significant (P=0.02), results are presented separately for the 24 420 women who did not have such a history (women without a recent MI) and the 580 women who did (women with a recent MI).
All statistical analyses were conducted with SAS version 6.11.15 All probability values are for 2-tailed tests. The Human Subjects Committee of the Kaiser Foundation Research Institute approved this study.
Table 1 shows baseline characteristics of the 25 000 diabetic women by baseline HRT use. Women using unopposed estrogen and those using estrogen plus progestin had lower HbA1c levels and were younger, better educated, and more likely to be non-Hispanic white and to report exercising, alcohol consumption, and cigarette smoking than women not using HRT. Although statistically significant, no meaningful differences in hypoglycemic therapy, diabetes duration, hypertension, and use of lipid-lowering medications by baseline HRT were observed. Adjustment for age did not alter any of the associations shown in Table 1, except that women using estrogen plus progestin were leaner than women not using HRT.
The mean age of the 24 420 diabetic women without a recent MI was 64.9 (SD 9.1) years. At baseline, there were 2526 women (10%) using unopposed estrogen and 2088 (9%) using estrogen plus progestin. After a mean of 3.04 years of follow-up, there were 256 fatal acute MIs and 854 nonfatal acute MIs, for a total of 1110 events. In an age-adjusted Cox proportional hazards model, current use of estrogen plus progestin was associated with a statistically lower (22% lower) rate of fatal and nonfatal acute MI, whereas in women currently using unopposed estrogen, the rate of acute MI was 11% lower, which was not statistically significant (Table 2). Similar RH estimates for current use of estrogen plus progestin and current use of unopposed estrogen were obtained in Cox models that predicted fatal acute MI and nonfatal acute MI separately (Table 2). Estimates of RHs changed little after additional adjustment for CHD risk factors, including ethnicity, education, body mass index, hypoglycemic therapy, diabetes duration, HbA1c, hypertension, smoking, alcohol, exercise, and baseline use of lipid-lowering medications (Table 2). The difference between the RHs for unopposed estrogen and estrogen plus progestin use were not statistically significant (P=0.35), and we therefore combined the 2 HRT regimens and calculated the multiple-adjusted RH for current HRT predicting fatal and nonfatal acute MI (RH 0.84; 95% CI 0.72 to 0.98).
Among women taking estrogens (with or without progestins) at baseline, 96% were taking oral estrogens (90% of which were conjugated equine estrogens) and 4% were using transdermal estradiol. Overall, 8% were taking low-dose estrogen, 78% were taking medium-dose estrogen, and 14% were taking high-dose estrogen. In a Cox model with multiple adjustments, with current estrogen dose modeled as a time-dependent covariate, women currently using low and medium doses of estrogen had a statistically significant lower risk of fatal and nonfatal acute MI than women who were not currently using estrogen (Table 3). Current use of high-dose estrogen was not associated with lower risk of acute MI (Table 3).
We explored the possibility that risk of acute MI among current HRT users might vary by time since HRT initiation by examining current HRT as function of time since HRT initiation among a subgroup of 17 947 women whose baseline dates were on or after March 1, 1995 (Table 4). When women who were not currently using HRT and who had not been previously exposed to HRT were used as the referent group, the multiple-adjusted RH for fatal and nonfatal acute MI among women currently using HRT for <1 year was ≈1.00, whereas the risk of acute MI among women using HRT for ≥1 year was 19% lower (P=0.052) than among the referent group (Table 4).
Of the 580 diabetic women with a recent MI, the mean age was 69.2 (SD 9.1) years, 15% were using HRT at baseline, and mean follow-up was 2.53 years. In a Cox model with multiple adjustments, women currently using HRT had a significantly higher risk of recurrent acute MI (RH 1.78, 95% CI 1.06 to 2.98; 23 events) than women not currently using HRT (66 events). We then examined the association between current HRT and the risk of recurrent acute MI by time since HRT initiation among a subset of 439 women whose baseline dates were on or after March 1, 1995. Compared with women not currently using HRT and with no prior exposure to HRT (43 events), women currently using HRT for <1 year had a statistically significant increased risk of recurrent acute MI (multiple-adjusted RH 3.84; 95% CI 1.60 to 9.20; 7 events), whereas in women currently using HRT for ≥1 year, there was no association between current HRT and risk of recurrent acute MI (multiple-adjusted RH 0.87; 95% CI=0.36 to 2.15; 6 events).
Among diabetic women who did not have a recent MI, current HRT use was associated with a significant 16% lower risk of acute MI. The incidence of acute MI was 22% lower for current users of estrogen plus progestin, whereas current use of unopposed estrogen showed a nonsignificant lower risk of acute MI of 11%. Lower risk of acute MI was observed among women using low or medium doses of estrogen but not among those using a high dose. The risk of acute MI appeared to be lower after 1 year of HRT use but was not lower during the first year of use. Among women with a recent MI, current HRT use was associated with an 80% higher rate of recurrent acute MI, and the rate of recurrent events was ≈4-fold higher during the first year of HRT.
The observed decreased risk of acute MI associated with current HRT in diabetic women without a recent MI is consistent with results from observational studies in nondiabetic women without CHD.3 However, the present results are inconsistent with recent results from the WHI,5 a randomized trial in women without CHD, which reported a 30% increased risk of CHD in women assigned to estrogen plus progestin compared with women on placebo. The increased CHD risk in the treatment group appeared soon after randomization and persisted through 6 years of follow-up. In observational studies, such as the present study, the association between MI risk and HRT might be due to uncontrolled confounding or bias. It has been suggested that a possible bias in the association between HRT and reduced risk of CHD found in observational studies is that HRT may be a marker for better CHD risk profile.16,17⇓ Although we carefully adjusted for the recognized CHD risk factors measured in the present study, it is possible that unmeasured potential confounders for the observed association between estrogen plus progestin and reduced MI risk might have been missed.
Our results among diabetic women who have had a recent MI, risk of acute MI associated with current HRT and time since HRT initiation are consistent with those of a previous randomized trial4 and an observational study18 among women with CHD. HERS showed a 50% increased risk in CHD among women assigned to estrogen plus progestin during the first year after randomization and no difference between active treatment and placebo in CHD risk at an average of 4.1 or 6.8 years of follow-up.4,19⇓ The Nurses’ Health Study,18 in an observational analysis of 2489 women with preexisting CHD, also found that HRT was associated with a statistically nonsignificant increased risk of recurrent CHD of 25% during the first year of HRT use.
Only 2 studies8,20⇓ have previously investigated the association of various estrogen doses with the risk of CHD, and neither of them was able to address this question among women with diabetes. The present results suggest an association between estrogen dose and risk of acute MI, which was 51% and 19% lower among women currently using low and medium doses of estrogen, respectively. The lack of association between high-dose estrogen and acute MI risk that we observed might be related to the small number of women using high-dose estrogen.
Strengths of this study were the prospective cohort design, the large sample size, and the ability to assess HRT use over time with computerized pharmacy records. This allowed us to more precisely define current exposure and to treat it as a time-dependent covariate, thereby accounting for changes in HRT over time. The pharmacy database also allowed prospective assessment of HRT, rather than reliance on the person’s recall, which has been found to be inaccurate with regard to type and dose of hormones used.21
A few study limitations are worth noting. Given that we were unable to assess lifetime HRT exposure before baseline, ascertainment of HRT duration was subject to some misclassification. However, the association that we reported between time since HRT initiation and acute MI risk has the same direction and magnitude as the associations reported previously.4,18⇓ An additional limitation is that we did not assess the risk of other adverse outcomes that are more common in women with diabetes and that have been shown to be increased by HRT, such as stroke, pulmonary embolism, and gallbladder disease.5,22⇓
Results from the present observational study do not prove causality. Data from clinical trials in diabetic women are needed to understand the possible risks and benefits of HRT. The Raloxifene Use for The Heart study,23 an ongoing randomized controlled trial that includes a large sample of women with diabetes, will add to our knowledge of the magnitude of the effect of the risks and benefits of HRT in women with diabetes.
This study was supported by a grant (115-9630) from the American Heart Association Western States Affiliate and a grant (115-9675) from Kaiser Foundation Research Institute to Dr Ferrara.
- ↵Wingard DL, Barrett-Connor E. Heart disease and diabetes. In: Harris MI, Cowie CC, Stern MP, et al. Diabetes in America. Bethesda, Md: National Institutes of Health; 1995: 429–448.
- ↵Harris MI, Flegal KM, Cowie CC, et al. Prevalence of diabetes, impaired fasting glucose, and impaired glucose tolerance in U.S. adults: the Third National Health and Nutrition Examination Survey, 1988–1994. Diabetes Care. 1998: 21: 518–524.
- ↵Kaplan RC, Heckbert SR, Weiss NS, et al. Postmenopausal estrogens and risk of myocardial infarction in diabetic women. Diabetes Care. 1998; 21: 1117–1121.
- ↵Lawrenson RA, Leydon GM, Newson RB, et al. Coronary heart disease in women with diabetes: positive association with past hysterectomy and possible benefits of hormone replacement therapy. Diabetes Care. 1999; 22: 856–857.
- ↵Varas-Lorenzo C, Garcia-Rodriguez LA, Perez-Gutthann S, et al. Hormone replacement therapy and incidence of acute myocardial infarction: a population-based nested case-control study. Circulation. 2000; 101: 2572–2578.
- ↵Ferrara A, Karter AJ, Ackerson LM, et al. Hormone replacement therapy is associated with better glycemic control in women with type 2 diabetes: the Northern California Kaiser Permanente Diabetes Registry. Diabetes Care. 2001; 24: 1144–1150.
- ↵Breslow NE, Day NE. Statistical Methods in Cancer Research. Volume 2: The Design and Analysis of Cohort Studies. Lyon, France: International Agency for Research on Cancer; 1987.
- ↵Kalbfleisch JD, Prentice RL. The Statistical Analysis of Failure Time Data. New York, NY: John Wiley & Sons; 1980.
- ↵SAS Institute Inc. SAS/STAT Software. Changes and Enhancements Through Release 6.11. Cary, NC: SAS Institute Inc; 1995.
- ↵Grodstein F, Manson JE, Colditz GA, et al. A prospective, observational study of postmenopausal hormone therapy and primary prevention of cardiovascular disease. Ann Intern Med. 2001; 133: 933–941.