Myocardial Infarction and Use of Low-Dose Oral Contraceptives
A Pooled Analysis of 2 US Studies
Background—Population-based case-control studies to assess the relationship of low-dose oral contraceptive (OC) use with myocardial infarction (MI) were performed at 2 sites in the United States (California and Washington state). The purpose of the present study was to estimate risk of MI in relation to use of low-dose OCs in a pooled analysis combining results from the 2 sites.
Methods and Results—The study included as cases women aged 18 to 44 years with incident MI who had no prior history of ischemic heart disease or cerebrovascular disease. Women in the case and control groups were interviewed in person regarding OC use and cardiovascular risk factors. The analysis included 271 MI cases and 993 controls. Compared with noncurrent users, the adjusted pooled odds ratio for MI in current OC users was 0.94 (95% CI, 0.44, 2.20) after adjustment for major risk factors and sociodemographic factors. Compared with never users, the adjusted pooled odds ratio for MI was 0.56 (0.21, 1.49) in current OC users and 0.54 (0.31, 0.95) in past OC users. Among past OC users, duration and recency of use were unrelated to MI risk as was current hormone replacement therapy. There was no evidence of interaction between OC use and age, presence of cardiovascular risk factors (hypercholesterolemia, hypertension, diabetes), obesity, or smoking.
Conclusions—We conclude that low-dose OCs as used in these populations are safe with respect to risk of MI in women.
Concern about the potential increased risk of cardiovascular disease (CVD) associated with oral contraceptive (OC) use has existed since soon after the introduction of OCs in 1960. Although several1 2 3 4 5 6 but not all7 8 9 10 of the early studies showed evidence of increased risk of myocardial infarction (MI) with current OC use, these findings were based on the use of OCs containing higher doses of estrogen than those now commonly prescribed. More recent reports have generally shown modest (≤2-fold) increases in risk that were not statistically significant in current OC users,11 12 13 14 15 although an ≈5-fold increase in risk of MI in current OC users was shown in the study conducted by the World Health Organization (WHO).16 Most studies have reported no increased risk of MI associated with past OC use,1 2 4 16 17 although complete agreement on this issue is lacking.5
In 1991, the National Institute of Child Health and Human Development funded population-based case-control studies at 2 sites (California and Washington state) to assess the relationship of low-dose OC use with MI. The study designs were developed independently by investigators at both sites with the expectation that key variables would be collected in a manner that would allow pooling of data. In a previous report15 of the results of the Kaiser Permanente Medical Care Program (KPMCP) study in California, the odds ratio (OR) for MI was 1.67 (95% CI, 0.48, 5.85) in current OC users compared with noncurrent users; when compared with never users, OR was 1.14 (0.48, 4.72) for current OC users and 0.60 (0.25, 1.44) for past users. The results of the University of Washington (UW) study have not previously been published. The primary purposes for a pooled analysis of the data of the 2 studies were to determine whether the OR for MI associated with current OC use was consistent between the 2 study sites; to obtain a higher degree of precision in estimating the OR for MI associated with OC use than could be achieved at either study site alone; and to determine whether the apparent protective effect of past OC use found in the Kaiser Permanente (KP) study was also present in the UW study and, if so, to perform analyses to attempt to better understand this association.
All MIs occurring among 18- to 44-year-old female members of the KPMCP were identified from May 1991 to August 1994 (northern California) and from July 1991 to August 1994 (southern California). For the UW study, incident MIs were identified from July 1991 through February 1995 among women aged 18 to 44 years who lived in King, Pierce, and Snohomish counties. In both studies, the medical record for each potential case was reviewed to establish the diagnosis of MI with diagnostic criteria adapted from those of the Cardiovascular Health Study.18 Included as cases were women with events classified as definite or probable MI on the basis of chest pain (presence or absence), cardiac enzymes, and ECG findings. Women with a history of coronary heart disease (CHD) or cerebrovascular disease before the date of identification by the study were excluded.
All women identified as cases in the KP study were required to have been admitted to the hospital or seen in the emergency department, whereas the UW study also included subjects who died of primary cardiac arrest outside the hospital; these subjects (n=16) were excluded from the pooled analysis. Proxy interviews were conducted with a close relative or friend when the women with MI died or were unable to communicate orally. Concerns about the accuracy of data on past OC use led to the exclusion of proxy interview data (n=26) from the pooled analysis.
For each woman with MI in the KP study, 3 controls, matched for year of birth and location of care facility, were randomly selected from female members of the KPMCP. Controls who could not be located, declined to be interviewed, or spoke neither English nor Spanish were replaced with other randomly selected controls until 3 controls had been enrolled for each women with MI or 2 replacement controls had been selected.
The control group for the UW study contained women who served as a common control group for the study of the relationship of OC use to MI and a concomitant study of stroke. A representative sample of female residents of King, Pierce, or Snohomish counties, frequency matched to the combined age distribution of cases for the studies of MI and stroke, was selected during the case diagnosis period by random-digit telephone dialing.
For both studies, eligible cases and controls were interviewed in person by trained interviewers who used a standardized instrument. Informed consent was obtained in person before the interview. All information was collected for the period before a predefined reference date, which was the date of MI for cases. For controls, the reference date was either the same date as the case to which she was matched (KP) or a randomly assigned date selected from among the possible MI diagnosis dates (UW). In the KP study, a woman was considered a current OC user if she reported that she was taking OCs in the month before the reference date. In the UW study, OC use was ascertained according to responses given in calendar months with current OC use defined as use of OCs in either the same calendar month as the reference date or the calendar month before the reference date. In each study, a woman was classified as a past OC user if she had ever used OCs but did not meet the current user definition. All other women were classified as never users of OCs. While the interview instruments were not identical, comparable self-reported data were obtained for key analytic variables, including history of OC use, sociodemographic characteristics, obstetrical and gynecological history, and medical and lifestyle cardiovascular risk factors.
The exposure OR was used to estimate relative risk. The pooled analysis accounted for the difference in design of control selection strategies at the 2 study sites by an adaptation of logistic regression.19 Estimation of pooled ORs was accomplished by the method of maximum likelihood with the contribution to the likelihood function from the matched case-control study (KP) assessed by conditional logistic regression and the contribution from the unmatched study (UW) assessed by unmatched logistic regression. Interval estimation as well as tests of significance were accomplished with standard large-sample likelihood theory. The combined sample size of 271 cases and 993 controls provided approximately 90% power to detect an OR of 2.0 with a 2-sided test and significance level of 0.05. Tests of heterogeneity across study sites in the effects of confounders and exposure were performed with Wald’s χ2 test for equality of the regression coefficients for the 2 sites. All computations were performed with software developed in the GAUSS programming language.20 We adjusted for race/ethnicity and major established risk factors for CHD, including cigarette smoking, hypertension, hypercholesterolemia, and diabetes. In addition, we adjusted for variables that were apparent confounders in these data, ie, if there was an appreciable change (>5%) in the exposure coefficients with the addition of the potential confounder. Appropriate cross-product terms were included in the models to test for the statistical significance of potential interactions and to obtain stratum-specific estimates of association between current OC use and MI.
A total of 187 incident confirmed MIs occurred in women aged 18 to 44 years who were members of the KPMCP during the period of case ascertainment. One was not eligible for interview because she did not speak English or Spanish. Interviews were completed for 178 (95.7%) of the remaining 186 cases or their proxies; reasons for nonparticipation included physician refusal (n=1) or patient refusal (n=7). Data from 11 of the interviews were excluded (Table 1⇓), leaving 167 cases for analysis.
A total of 192 eligible patients were identified in the UW catchment area (excluding the 16 out-of-hospital deaths). Interviews were completed for 126 (65.6%) of the 192 cases; reasons for nonparticipation included physician refusal (n=4), patient or proxy refusal (n=46), inability to locate patient or proxy (n=14), and inability to interview patient before close of fieldwork (n=2). Data from 22 cases were excluded (Table 1⇑), leaving 104 cases for analysis.
Seventy-seven percent of all KP controls were selected from the initially chosen 3 controls, and 95.5% of the interviewed case-control sets had 3 controls. Altogether, of the eligible women identified for the KP study, an attempt was made to enroll 758, of whom 7 were excluded because of prior history of major CVD (n=1), inability to speak English or Spanish (n=3), or database error in age, gender, or address (n=3). Of the remaining 751 women, 545 were interviewed for an overall response rate of 72.6%. Of the 492 controls in the matched sets for the 167 cases included in this analysis, 11 were excluded because of pregnancy, leaving 481 controls in the analysis set.
In the UW study, a household census to ascertain women meeting eligibility criteria was completed for 94.9% of the residences contacted during the random-digit dialing for control identification. Among the eligible women identified, an attempt was made to enroll 691, frequency matched to the combined age distribution of all MI and stroke patients recruited for the study. Seven of the 691 women were excluded because of prior history of major CVD (n=6) or inability to communicate in English (n=1). Of the remaining 684 women, 526 were interviewed for an estimated overall response rate of 73.0% (526/684×94.9%). Of the 526 controls interviewed, 14 were excluded because of pregnancy as of their reference date, leaving 512 controls in the analysis set.
Table 2⇓ shows the characteristics of the MI cases and controls overall and stratified by OC use status. Controls were unlikely to be treated for hypertension or diabetes regardless of OC use status, and treatment for high cholesterol among controls was very rare. Current and former OC users were more likely to be smokers than never users. Younger age and white race/ethnicity were associated with current use. Former use was associated with frequent heavy exercise, education less than college graduation, and being married or living as married.
The age-race/ethnicity adjusted OR for MI in current OC users was 0.55 (95% CI, 0.27, 1.12) compared with noncurrent users. The fully adjusted OR for MI was 0.94 (0.40, 2.20) in current compared with noncurrent users and adjusted for age, treated hypertension, treated diabetes, smoking, race/ethnicity, body mass index (BMI), education, and menopause (Table 3⇓). When current users of OCs were compared with never users, the age-race/ethnicity adjusted OR for MI was 0.45 (0.20, 0.99) and the fully adjusted OR was 0.56 (0.21, 1.49) (Table 3⇓). The OR tended to decrease with increasing duration of use in current OC users compared with both noncurrent users and never users (Table 4⇓), although the test for trend was not statistically significant.
The age-race/ethnicity adjusted OR for MI was 0.79 (0.53, 1.17) in past OC users compared with never users, and the fully adjusted OR was 0.54 (0.31, 0.95). Among past OC users, duration and recency of past OC use were essentially unrelated to risk of MI (Table 4⇑). The adjusted OR for MI was 1.02 (0.42, 2.47) in past users of OCs who were currently receiving hormone replacement therapy (HRT) compared with never users.
We also examined ORs for MI in women >40 years old and <40 years old, in women with and without major cardiovascular risk factors (hypertension, high cholesterol, and diabetes), in obese and nonobese women, in women who were current cigarette smokers and those who were not, and according to progestin use (Table 5⇓). Stratum-specific ORs were close to 1, and there was no evidence of interaction between OC use and age (P=0.28), risk factor presence (P=0.75), obesity (P=0.81), or smoking (P=0.93).
The ORs for current OC preparations containing a progestin derived from 19-nortestosterone (ie, norethindrone family) and those containing a progestin derived from gonane (ie, norgestrel family) were not different (Table 5⇑). Most current OC users (9 of 12 cases, 78 of 87 controls) used formulations containing <50 μg of ethinyl estradiol. Of the remaining 3 cases and 9 controls, 7 (2 cases, 5 controls) used 50-μg preparations, and the estrogen dose of the other 5 preparations was unknown (1 case, 4 controls). The adjusted OR for current OC preparations containing <50 μg was 0.69 (0.26, 1.86) for current compared with noncurrent users of OCs and 0.41 (0.14, 1.24) for current users of OCs compared with never users.
The OC findings in relation to risk of MI were reasonably consistent in the KP and UW studies. In the pooled analysis, there was no evidence of an increase in risk of MI associated with use of OCs. As anticipated, combining the data from the 2 studies narrowed the 95% CI for the relative odds of MI associated with current OC use compared with either study alone. The pooled OR for current use relative to noncurrent use had an upper limit of 2.2, virtually ruling out a larger effect of current OC use.
The results are consistent with most other recent studies, which have not shown statistically significant elevations in risk of MI associated with current OC use.11 12 13 14 15 Although the WHO study16 found an ≈5-fold increased risk of MI associated with current OC use, the authors concluded that the increased risk probably reflects more frequent use of OCs by women with other cardiovascular risk factors and less screening than is currently carried out in the United States. In particular, in the WHO study, there was no increased risk of MI associated with OC use among nonsmoking women at low risk who had a blood pressure check before taking OCs.
Past use of OCs was associated with a statistically significant decrease in risk of MI in the pooled analysis of KP and UW data, and the ORs for the 2 sites were nearly identical. Possible explanations for this apparent protective association, if not a chance finding, include selection bias, unmeasured confounding, and a true protective effect. Selection bias could occur if past users of OCs were prescribed OCs only if they were at low risk for heart disease. However, our data (Table 2⇑) did not suggest that controls who were past OC users had lower levels of major cardiovascular risk factors than never users. However, it is possible that past users and never users differed in unmeasured potential confounders such as diet. A true protective effect of past OC use might also result from beneficial effects on lipids. Although earlier generations of OCs had adverse effects on lipids and lipoproteins,21 some newer formulations of progestin may have beneficial effects on LDL and HDL cholesterol.22 In the absence of knowledge of variations of OC formulations used in the past by study cohort members, we cannot make definitive statements about likely effects on lipids and other cardiovascular risk factors. The absence of variation in the effect of OCs in relation to duration or recency of use is evidence against a biologically mediated protective mechanism. The lack of association of current HRT with risk of MI in past OC users compared with never users suggests that HRT use did not explain the decreased risk of MI associated with past OC use.
There was no evidence of interaction between current OC use and cigarette smoking in contrast with earlier studies performed when higher-dose OCs were commonly used. These studies showed higher risks of MI associated with OC use in smokers than in nonsmokers.12 23 The previously reported KP analysis showed no evidence of an interaction between OC use and smoking on risk of MI,15 and most other studies performed as low-dose OC use became common have been unable to address this issue.11 13 14 However, the WHO study showed marked elevations in risk of MI among smokers.16 The discrepancy between the results of the WHO study and these pooled US studies may reflect unmeasured hypertension among smokers in the WHO study or other unmeasured differences in the health of smokers in these studies.
Limitations of this study include possible response, recall, and diagnostic bias. Because recall bias is more likely to result in a greater propensity for reporting exposure by cases compared with controls, its presence would result in an RR that is higher than the true value. Diagnostic bias would occur if diagnosis of MI is more likely to be considered by clinicians in a woman using OCs than in nonusers and also would result in an RR higher than the true value. The use of standardized diagnostic criteria make it unlikely that there was bias in the evaluation of cases. The KP study did not perform surveillance for fatal MIs, whereas in the UW study, 42 cases had died by initiation of recruitment. There is a potential bias in this study if OC use was related to more serious MIs resulting in early mortality. In particular, this might explain the lower risk associated with past OC use in this study, although results were similar in the only prospective study analyzing risk of CVD associated with past OC use.17 Unmeasured potential confounders such as diet that might differ between cases and controls also create potential bias.
Other limitations of this study include the small number of OC users who smoked, had other risk factors for CVD, and were >40 years old. Therefore, the study did not have adequate numbers to offer definitive statements regarding these and other issues (eg, race/ethnicity) relating to potential interactions.
Although caution must always be exercised in generalizing the results of epidemiological studies, the results of this pooled analysis are reassuring in that they reinforce findings from other published studies of little or no increased risk of MI in users of low-dose OCs as taken in these settings. The possible protective association found for past OC users was not the result of HRT and remains unexplained. We conclude that low-dose OC use is safe in relation to risk of MI for healthy women without major CVD risk factors.
This work was supported by the National Institute of Child Health and Human Development, contract numbers N01-HD-1-3107 and N01-HD-1-3108.
- Received January 14, 1998.
- Revision received May 5, 1998.
- Accepted May 10, 1998.
- Copyright © 1998 by American Heart Association
Mann JI, Vessey MP, Thorogood M, Doll R. Myocardial infarction in young women with special reference to oral contraceptive practice. BMJ. 1975;2:241–245.
Mann JI, Inman WHW. Oral contraceptives and death from myocardial infarction. BMJ. 1975;2:245–248.
Rosenberg L, Hennekens CH, Rosner B, Belanger C, Rothman KJ, Speizer FE. Oral contraceptive use in relation to nonfatal myocardial infarction. Am J Epidemiol. 1980;111:59–66.
Inman WHW, Vessey MP. Investigation of deaths from pulmonary, coronary, and cerebral thrombosis and embolism in women of child-bearing age. BMJ. 1968;2:193–199.
Vessey MP, Doll R. Investigation of relation between use of oral contraceptives and thromboembolic disease: a further report. BMJ. 1969;2:651–657.
Maguire MG, Tonascia J, Sartwell PE, Stolley PD, Tockman MS. Increased risk of thrombosis due to oral contraceptives: a further report. Am J Epidemiol. 1979;110:188–195.
Krueger DE, Ellenberg SS, Bloom S, Calkins BM, Maliza C, Nolan DC, Phillips R, Rios JC, Rosin I, Shekelle RB, Spector KM, Stadel BV, Stolley PD, Terris M. Fatal myocardial infarction and the role of oral contraceptives. Am J Epidemiol. 1980;111:655–674.
Rosenberg L, Palmer JR, Lesko SM, Shapiro S. Oral contraceptive use and the risk of myocardial infarction. Am J Epidemiol. 1990;131:1009–1016.
Croft P, Hannaford PC. Risk factors for acute myocardial infarction in women: evidence from the Royal College of General Practitioners’ oral contraception study. BMJ. 1989;298:165–168.
D’Avanzo B, La Vecchia C, Negri E, Parazzini F, Franceschi S. Oral contraceptive use and risk of myocardial infarction: an Italian case-control study. J Epidemiol Community Health. 1994;48:324–325.
Moreno V, Martin ML, Bosch FX, Sanjose S, Torres F, Munoz N. Combined analysis of matched and unmatched case-control studies: comparison of risk estimates from different studies. Am J Epidemiol. 1996;143:293–299.
Abtech Systems. The GAUSS System Version 3.1. Maple Valley, Wash: Abtech Systems, 1992.
Godsland I, Crook D, Wynn V. Coronary heart disease risk markers in users of low-dose oral contraceptives. J Reprod Med. 1991;36(suppl):226–237.