Androgenicity of Progestins in Hormonal Contraceptives and the Risk of Gestational Diabetes Mellitus

The setting of the study was the KPMCP program, which provides comprehensive medical services to >3 million members located in a 14-county region in Northern California. The KPMCP program membership represents ∼30% of the population and is demographically, ethnically, and socioeconomically representative of the area, except that membership underrepresents the very poor and the very wealthy (14). From 1 January 1996 to 30 June 1998, ∼94% of all pregnancies that reached the third trimester were screened for GDM by a 50-g, 1-h oral glucose challenge test (GCT). If abnormal (1-h plasma glucose levels ≥140 mg/dl), this test was followed by a diagnostic 100-g, 3-h oral glucose tolerance test performed in the morning after a 12-h fast (15).

We searched the KPMCP program hospital discharge and billing claims databases to identify all singleton live births and searched the KPMCP program laboratory database to obtain all plasma glucose values measured during screening GCTs and diagnostic 3-h oral glucose tolerance tests (16). We restricted our cohort to women who delivered between January 1996 and June 1998 without previously recognized diabetes (n = 72,946). We excluded pregnancies not screened for GDM by a 50-g, 1-h GCT in the laboratory database (n = 4,560), pregnancies screened for GDM outside of the recommended 24–28 weeks’ gestation (n = 21,297), and women with a diagnosis of GDM in a prior pregnancy (n = 362). Finally, we excluded women who were not continuous health plan members for 5 years before the index pregnancy (n = 32,492), leaving 14,235 eligible pregnancies from which to select case and control subjects. Compared with the entire cohort of 72,946 women who delivered during the study period, 14,235 women who were eligible to be selected as case and control subjects for the current study were more likely to be aged >35 years at pregnancy (23.6 vs. 15.7%) and to be non-Hispanic white (57.1 vs. 47.0%), but there were no differences in parity and education. Trained medical records abstractors completed chart reviews and confirmed that the inclusion criteria were met and that none of the exclusion criteria were present.

Women were classified as having GDM if two or more of the four plasma glucose values obtained during the 100-g, 3-h oral glucose tolerance test were abnormal according to the National Diabetes Data Group criteria (2) (fasting ≥105 mg/dl; 1 h ≥190, 2 h ≥165, and 3 h ≥145 mg/dl). If a woman had more than one diagnostic test performed, we used the latest test performed in pregnancy.

We performed a medical chart review of all of the 437 potential GDM case subjects we identified electronically. Medical chart review revealed that 26 (5.9%) were ineligible because their screening test was performed outside of the 24–28 weeks’ gestation window, 12 (2.7%) were ineligible because they had GDM in a prior pregnancy as noted in their chart, and 10 (2.3%) had insufficient information in their medical charts to determine eligibility, leaving 391 eligible case subjects.

Eligible control subjects were women without GDM, by the National Diabetes Data Group criteria, whose records confirmed the absence of exclusion criteria. A total of 310 control subjects were previously randomly selected and had their medical records abstracted for a case-control study of maternal hyperglycemia and infant complications (hypoglycemia, hyperbilirubinemia, and macrosomia) among women without GDM (17). Because of the selection criteria of that study, only 1.8% had at least one of the infant complications of interest, in contrast to 8.3% of potential control subjects. Therefore, to ensure that the control subjects for this study were representative of the entire cohort, we randomly selected an additional 28 control subjects from among pregnant women whose infants had at least one of four complications and an additional 30 women whose infants did not have any of the complications. Thus, there were a total of 368 control subejects, and 8.9% of them had an infant with one or more of the complications listed above. To ensure that the control group we selected was representative of the 13,798 potential control subjects, we compared our control subjects with the entire sample of potential control subjects and found no significant differences in age, race, and infant complications.

Medical chart abstracters recorded all information on hormonal contraceptive use found in medical charts or electronic databases during the 5 years before the women's index pregnancies. First, they recorded the date, type, and duration of each prescription found in the chart. Second, they searched the KPMCP program's computerized pharmacy, laboratory, and outpatient database, which records all drugs prescribed by KPMCP program physicians and dispensed in KPMCP program outpatient pharmacies. For each prescription, the database includes the date it was filled, the dosage, and the formulation. Abstractors recorded the date and type of any prescriptions. Finally, any additional information found in the medical charts pertaining to hormonal contraceptive use was recorded, such as the type of hormonal contraceptive, months of use, discontinuation, and the last menstrual period dates of any pregnancies occurring in the 5-year window. They also abstracted information on other medical conditions that are indications or contraindications for various hormonal contraceptive regimens and associated with glucose tolerance, including amenorrhea, polycystic ovary syndrome (PCOS), infertility, fibroids, depression, hypothyroid, smoking status, hypertension, and elevated cholesterol (≥200 mg/dl). Information on date of last menstrual period for the index pregnancy, marital status, parity, and height also were abstracted from the form completed at the first prenatal visit. Prepregnancy weight was defined as the last recorded weight found in the chart before the woman's last menstrual period for the index pregnancy. For 14.4% of women for whom these data were not available, the self-reported weight on the prenatal form was used. Prepregnancy BMI was calculated as prepregnancy weight (in kilograms) divided by height (in meters squared). Women's self-reported race/ethnicity and education were obtained by linkage to the electronic birth certificate database.

The androgenicity of the progestin component in each oral contraceptive was determined by compiling data from previous studies assessing the androgenic potential of progestins, while also factoring in the dose of progestin per oral contraceptive formulation (18). The overall androgenic activity of a progestin also depends on the pharmokinetics of the progestin and the dose. A higher-potency progestin may be used in a much smaller dose and be equivalent to a larger dose of a less-potent progestin in terms of androgenicity.

Table 1 shows the androgenicity of the oral contraceptives most commonly used by study participants. We used Dickey's (19) classification of androgenic activity. The androgenic activity was determined by the rat ventral prostate assay, with methyl testosterone used as the standard (19,20). We classified any oral contraceptive with an androgenic activity of 0.47 mg of methyl testosterone equivalents per 28 days (19) or higher as “high androgen.” There is no clearly defined cutoff for what constitutes a high-androgen hormonal contraceptive because most hormonal contraceptives contain some degree of androgenicity. This cutoff encompased oral contraceptives in the highest quartile of androgenicity among oral contraceptives used by the study population.

For hormonal contraceptives that are not orally administered, we classified Norplant as high androgen because it contains levonorgestrel, a progestin with high androgenic activity. For DPMA (depo-medroxyprogesterone acetate), which contains medroxyprogesterone, a progestin with low androgenic activity, we assigned the classification “low androgen.”

Using the information obtained from medical charts and pharmacy databases, we calculated the duration of use and time since discontinuation of hormonal contraceptives. We categorized women as having taken a high-androgen hormonal contraceptive if they took any high-androgen hormonal contraceptive for at least 6 months (regardless of whether they also took a low-androgen hormonal contraceptive). Low-androgen hormonal contraceptive users were women who took only low-androgen hormonal contraceptives for at least 6 months, and nonusers were women who did not take either type. Because women who discontinue pill use shortly after starting may be different from those who continue use, women who used either type of hormonal contraceptive for <6 months were excluded.

Of the eligible 391 case and 368 control subjects, we excluded women who used a hormonal contraceptive for <6 months (29 case and 25 control subjects). Women who could not be classified into categories of high- or low-androgen hormonal contraceptive use because of missing information on hormonal contraceptive formulations or duration (three case and seven control subjects) also were omitted from our analyses.

Statistical analyses were performed using unconditional logistic regression. Odds ratios were computed as estimates of the relative risk of GDM as a function of each category of hormonal contraceptive use. Potentially confounding factors were entered individually as covariates in the models, and those that changed the odds ratio estimates of the relationship between hormonal contraceptive use and GDM by >10% were included in the covariate-adjusted models. These factors included age (continuous), race/ethnicity (non-Hispanic white, Asian, Hispanic, African American, other, or unknown), and a history of infertility (yes/no). We ran a third model further adjusting for prepregnancy BMI. Trends across levels of exposure to hormonal contraceptives (e.g., duration of use) were assessed examining P values for a single-trend variable coded as the category of exposure (1, 2, 3, etc.). This study was approved by the human subjects committee of the Kaiser Foundation Research Institute.

Table 2 compares case and control subjects with regard to demographic factors, reproductive factors, and selected medical conditions. Relative to control subjects, case subjects were more likely to be aged >30 years, to be from a race/ethnic minority group, and to be married, but they were similar with respect to education. Case subjects also were more likely to have two or more prior live births, to have a high prepregnancy BMI, to have a first-degree family history of diabetes, and to have certain other medical conditions, such as infertility, amenorrhea, fibroids, and elevated serum cholesterol.

Table 3 presents odds ratios and 95% CIs for GDM by type of hormonal contraceptive as well as duration and timing of hormonal contraceptive use. Compared with women who had not used hormonal contraceptives, women who used low-androgen hormonal contraceptives for at least 6 months during the 5 years before the index pregnancy had a 16% reduced risk of GDM (covariate-adjusted odds ratio 0.84 [95% CI 0.58–1.22]). In contrast, there was a suggestion that women who used a high-androgen hormonal contraceptive for at least 6 months during the 5-year period before pregnancy had an increased risk of GDM (1.43 [0.92–2.22]). Among women who used Loestrin, the highest androgen oral contraceptive, the risk was greater (1.99 [0.91–4.32]) than for women who used other high-androgen hormonal contraceptives (1.21 [0.75–1.97]). These results did not change appreciably when prepregnancy BMI was added to the multivariate models (Table 3). Results were similar when women who breastfed an infant (n = 209) during the 5-year period were excluded (data not shown). A subanalysis examined the risk of GDM among women who had taken only low-androgen hormonal contraceptives, only high-androgen hormonal contraceptives, or both types of hormonal contraceptives. Odds ratios were 0.84 (95% CI 0.58–1.22), 1.29 (0.77–2.15), and 1.70 (0.90–3.19), respectively.

Because DPMA (depo-medroxyprogesterone acetate) has been shown to increase fasting glucose, insulin, and weight gain (21), and the effects of Norplant on glucose metabolism are unclear (22), we repeated analyses after excluding women who used these two nonorally administered hormonal contraceptives. Results of the covariate-adjusted models for use of low- and high-androgen oral contraceptives essentially were unchanged (odds ratios 0.83 [95% CI 0.56–1.21] and 1.40 [0.88–2.22], respectively).

After adjusting for covariates, there was some suggestion of a greater decreased risk of GDM with longer duration of use of a low-androgen hormonal contraceptive, but no clear trend was evident with duration of use of a high-androgen hormonal contraceptive (Table 3). The reduced risk of GDM associated with use of a low-androgen hormonal contraceptive was greatest when use was discontinued <6 months before pregnancy (Table 4). In contrast, the increased risk of GDM associated with use of a high-androgen hormonal contraceptive was greatest when use was discontinued ≥1 year before pregnancy (Table 4).

In this study, women with GDM were slightly less likely than control subjects to have used a low-androgen hormonal contraceptive in the 5 years before pregnancy and more likely to have used a high-androgen contraceptive. There was a suggestion of a trend of greater decrease in risk with longer duration of use of a low-androgen hormonal contraceptive, but no clear pattern of GDM risk was observed with high-androgen hormonal contraceptive use. However, the statistical precision of our results was not great, and, given no true associations, chance alone plausibly could have been responsible for those we did observe.

A strength of this study is the use of medical records and the computerized pharmacy database to ascertain contraceptive use. Because the data were recorded in medical records, use of this information eliminates the potential for recall bias. Contraceptives are a covered benefit by the health plan; thus, there is incentive for KPMCP program members to have prescriptions made by a KPMCP program physician and to use the KPMCP program pharmacy. Therefore, the possibility of missing hormonal contraceptive prescriptions is minimized. It also is possible that some women who had prescriptions noted in the medical chart or electronic database did not actually take the pills. Misclassification from these sources likely would be similar for case and control subjects, biasing risk estimates toward the null.

A limitation of the study is that we also were limited to data on hormonal contraceptive use during the 5 years before pregnancy. There is some evidence that hyperglycemia associated with hormonal contraceptives abates when exposure ceases (8). However, the “glucose toxicity” theory suggests that any prolonged exposure to hyperglycemia can cause detrimental effects to pancreatic β-cell function, leading to the progressive worsening of hyperglycemia (23). If this theory is correct, we may have misclassified some women who stopped using hormonal contraceptives >5 years ago as nonusers. The misclassification likely would be nondifferential and, therefore, would have biased our results toward the null.

To our knowledge, this is the first study to examine use of hormonal contraceptives and subsequent risk of GDM. Previous information on hormonal contraceptive use and carbohydrate intolerance largely comes from metabolic studies, which examined changes in glucose and insulin levels in normal women after taking oral contraceptives. Clinical studies conducted on the early high-dose oral contraceptive formulations consistently found that use of these formulations increased serum glucose and insulin levels (24–27). Later studies found that the degree of metabolic alteration varied by type and dose of the progestin component (10,28) and that the higher androgenic formulations consistently had the strongest effects (7–9,29).

Other data support an association between hyperandrogenicity and insulin resistance. PCOS is a reproductive condition characterized by increased serum levels of circulating adrenal androgens, anovulation, and insulin resistance (30). The risk of GDM appears to be higher among women with PCOS compared with women without PCOS (30,31). Women with PCOS often are prescribed oral contraceptives, especially those containing low androgenic activity (32), to treat their condition. Unfortunately, during the study period, PCOS was underdiagnosed in this setting, so we were unable to adequately control for it. The KPMCP program does not have official guidelines regarding the prescription of particular types of oral contraceptives, and their selection was not generally based on characteristics relating to glucose metabolism. If anything, patients with androgen excess (who might be more prone to glucose intolerance) might have received an oral contraceptive with lower androgenicity, not higher.

There have been only two large prospective cohort studies in the U.S. on oral contraceptive use and type 2 diabetes risk. The first Nurses’ Health Study (1976–1988) found that women who used oral contraceptives in the past had only a 10% increased risk of type 2 diabetes (relative risk 1.10 [95% CI 1.01–1.21]) and that current use was not associated with type 2 diabetes risk (0.86 [0.46–1.61]) (18). The second Nurses’ Health Study (1989–1993) observed modest positive associations between current (1.6 [0.9–3.1]) or past (1.2 [0.8–1.8]) use of low-dose oral contraceptives (33) and type 2 diabetes risk. However, neither of these studies examined the effects of oral contraceptives by progestin type.

Our results suggest that the effects of hormonal contraceptive use on risk of GDM may vary by the androgenicity of the progestin component. We found evidence suggesting a 16% reduced risk of GDM associated with use of low-androgen hormonal contraceptives and a 43% increased risk associated with use of high-androgen hormonal contraceptives. However, these observations should be interpreted cautiously pending the results of other studies addressing this question.

This research was supported by the National Institute of Diabetes and Digestive and Kidney Diseases (grant R01 DK 54834) and a Research Award from the American Diabetes Association.