The study population and data collection methods for this study have been described elsewhere (Epplein et al., submitted for publication, 2007); a brief description follows.
The setting was Group Health (GH), a Washington state integrated health plan with over 500,000 enrollees. Cases included all female members of GH, over the age of 18 years, who were diagnosed with possible complex EH or possible EH with atypia between January 1, 1985 and December 31, 2003. Potential cases were excluded if they had a previous diagnosis of endometrial cancer. Because of the known ambiguities in the diagnosis of EH, study cases were also required to have a diagnosis of complex hyperplasia or hyperplasia with atypia assigned in a separate review. Specifically, all pathology slides from the potential cases identified through GH pathology text searches were independently reviewed by two outside pathologists, utilizing standardized diagnostic criteria. When the two pathologists disagreed (22.2 percent of the time), the slides were read by a third pathologist. If the third pathologist disagreed with both of the initial diagnoses, the diagnosis was assigned by the senior pathologist.
Controls were selected randomly from the membership files of GH from 1985–2003, excluding women who had a prior hysterectomy or who had a previous diagnosis of EH or endometrial cancer. One control was selected for each case. Controls were matched to cases by year of birth (+/- one year) and were required to have been enrolled at GH at the time of the case’s diagnosis of EH. To ensure that the controls actually used GH as their primary care institution, we sought to choose as controls women who had received GH services (i.e., seen a medical professional there, whether for a well-adult visit or any other reason) at least once in the three years prior to the time their matched case was diagnosed.
The Group Health automated pharmacy database provided information on oral contraceptives and post-menopausal hormones dispensed to study subjects. It contains data on all prescriptions for medications dispensed to enrollees through GH pharmacies since 1977. The data included the specific drug and drug class, date and amount dispensed, and dosing instructions, where available. For each case and her matched control, hormonal medications dispensed were examined prior to the one month before the case’s onset of abnormal vaginal bleeding (“reference date”). For those missing information on abnormal bleeding (n=72), the reference date was assigned as three months prior to the biopsy date (the median duration of abnormal bleeding among the cases).
Trained medical abstractors obtained the following variables of interest from the paper and electronic medical records of cases and controls: parity (defined as the number of pregnancies of at least 20 weeks gestation); race; smoking status at reference date; medical history, including a diagnosis of diabetes or hypertension (presence of elevated or high blood pressure that was treated with medications or lifestyle changes); height and weight (kg/m2) as of the reference date; and a previous weight, at least one year prior to that time.
The Group Health Enrollment database was queried to determine dates of enrollment for each study subject, and residential zip code at reference date (or the closest time thereof). Study subjects’ zip codes were then used to find the median income in their geographical region, as determined by the 1990 US Census (Summary Tape File 3). Median income was not available for zip code areas where the population was deemed too small by the US Census Bureau, so as to abide by confidentiality laws, resulting in missing information on residential median income for 19 (4.1%) cases and 23 (5.0%) controls.
Of the 548 women with diagnoses of complex EH (n=332) or EH with atypia (n=216), 40 were excluded because of missing medical charts. Also excluded were 19 controls and 30 cases, and their matched counterparts, who were not enrolled at GH for at least 6 months prior to the reference date (and thus lacked sufficient pharmacy data for analysis), and one case and her matched control whose onset of symptoms was 8 years prior to diagnosis, resulting in a total of 462 cases (289 complex and 173 atypia) and 462 controls. While we would have preferred to investigate hormone use in the two (or even three) years prior to the reference date, rather than limiting our analyses to only the prior 6 months, we found that 26% of our subjects (or their matched counterparts) were not enrolled for at least two years prior to the reference date, and thus would have had to been dropped from our analyses.
The data analysis was performed with the goal of determining the association between complex EH and EH with atypia and the following exposures: combined estrogen and progestin oral contraceptive (OC), estrogen-only post-menopausal hormone therapy (ET), and combined estrogen plus progestin post-menopausal hormone therapy (EPT). OC users were defined as women who were dispensed one of these preparations for more than 1 month. ET users were defined as women who were dispensed ET for 60 days or more. For EPT, users were defined as women who were dispensed estrogen for 60 days or more and progestin for 20 days or more during the same time period. Women who had been dispensed both estrogen and progestin within the 6 months prior to the reference date but whose total number of days of intended progestin use was less than one-third of their total number of days of intended estrogen use (16 cases of complex hyperplasia, 11 cases of hyperplasia with atypia, and 6 controls) were excluded from the analyses. The rationale for this exclusion was that we were not able to determine whether these women were ET users who then took progestin, or EPT users who took a smaller amount of progestin than the 10 days per month generally recommended. The comparison group for each of these analyses consisted of women who were prescribed no exogenous hormones.
The variables age, race, zip code (as a proxy for socioeconomic factors), years enrolled in GH, body mass index, parity, diabetes, hypertension, and smoking were evaluated as potential confounders. The decision was made a priori to include in the models only those variables that were shown in the data to be related both to the exposure of interest and the outcome of EH.
We used conditional logistic regression (23
) to test the association of exogenous hormones with: 1) complex hyperplasia, 2) hyperplasia with atypia, and 3) either complex or atypical hyperplasia. Results are presented as odds ratios with 95% confidence intervals.
All analyses were performed with the STATA version 8.0 software package (College Station, Texas). To determine an upper confidence interval in the case of an odds ratio of 0.0, we used the PAIRS software within PEPI, version 4.0 (Salt Lake City, Utah).