Following institutional review board approval from the Group Health Research Institute, this retrospective cohort study was conducted among female enrollees at Group Health (GH), a mixed-model integrated health plan with over 500,000 enrollees in Washington State. Automated pathology, enrollment, pharmacy, inpatient and outpatient databases were linked for data on all women over age 18, diagnosed with complex and/or atypical endometrial hyperplasia, between January 1, 1985 and April 1, 2005. Women were followed from the time of hyperplasia diagnosis until an outcome occurred (endometrial carcinoma and/or hysterectomy), disenrollment, death, or until September 30, 2005. Eligibility criteria included no prior diagnosis of endometrial carcinoma and an intact uterus. Because the most commonly recommended progestin treatment duration for endometrial hyperplasia is at least 8 weeks,(16
) women who had outcomes prior to 8 weeks were excluded. In addition, women who took primarily unopposed estrogen for greater than 6 months at any time during the study period or who left GH for over 2 months during the study period were excluded.
Automated databases were linked through a unique identifier assigned to each woman when she first joined GH, and reassigned upon each subsequent enrollment. Disenrollment was ascertained by computerized membership files.
The pathology database includes unique pathology accession numbers, specimen collection dates, and test results, entered as text fields. Text searches indicating possible diagnoses of complex or atypical hyperplasia were conducted to identify women with the conditions of interest. Details of this methodology have been previously described. (19
) The primary goal of this study was to answer the question, “in a given population of women with a clinical diagnosis of either complex or atypical endometrial hyperplasia, what is the risk of endometrial carcinoma or hysterectomy occurring at least 8 weeks after the endometrial hyperplasia diagnosis, with and without progestin exposure?”
A diagnosis of endometrial carcinoma was ascertained from linkage with records of the Cancer Surveillance System of Western Washington, a population-based cancer registry that participates in the Surveillance Epidemiology and End Results (SEER) cancer registry. Hysterectomy, including date, was determined during review of the medical record (yes/no) and was confirmed by the presence of a uterine specimen in the pathology record.
The exposure of interest, progestin prescriptions, was ascertained from the GH pharmacy database. GH pharmacy databases capture all medications dispensed to enrollees through GH pharmacies including the specific drug, drug class, date and amount dispensed, and dosing instructions. Surveys among female GH members aged 50-80 years have shown that 97% of HT prescriptions are filled at GH pharmacies. (20
All progestin dispensings from one week before the index biopsy up until the outcome or censoring date were identified. Women were classified as progestin “ever users” if they were dispensed at least 15 days of progestin and as “never users” if they had ≤ 14 days of progestin dispensed. Duration of exposure (<56 days and ≥ 56 days of progestin dispensed) was evaluated in sub-analyses.
For time-dependent analyses, days of use was calculated from the time of the index biopsy to the outcome or censoring, within 6-month blocks. Women were classified as progestin “ever users” or as “never users” as described above, within each 6-month block. We created progestin and estrogen exposure variables for each time period that reflected exposure in the previous 6 month period.
Progestins were categorized by type - megesterol acetate, medroxyprogesterone acetate, and norethindrone acetate. Women who were given oral contraceptives were included in the progestin user group as all of these formulations are progestin dominant. We classified women as unopposed progestin users (PT) if they were dispensed progestin without an estrogen or if the number of estrogen pills dispensed was less than 1/3 the number of progestin pills dispensed. Women were classified as estrogen plus progestin users (EPT) if the number of estrogen pills dispensed was at least 1/3 of the number of progestin pills dispensed. We classified women as unopposed estrogen users (ET) if they were dispensed estrogen alone or if the number of progestin pills dispensed was less than 1/3 of the estrogen pills dispensed.
Three trained abstractors reviewed archived paper charts and the electronic medical records. Variables ascertained included: medical and family history; demographic, reproductive, and physical characteristics, including height and weight at the time of the index biopsy; bleeding patterns preceding the biopsy; ultrasound findings; age at menopause; race; parity; history of breast, colon, or ovarian cancer; diabetes; hypertension; and smoking status. Indications for hysterectomy and endometrial biopsies were recorded. Last clinical contact date (including date of death) and if deceased, whether death was related to endometrial carcinoma, were assessed.
Analyses were performed separately for complex and atypical hyperplasia. We computed the proportion of women with each type of hyperplasia (complex; atypical) who subsequently had a diagnosis of endometrial carcinoma and/or a hysterectomy. We calculated the adjusted rates of endometrial carcinoma and hysterectomy by computing the number of events by person years at risk for “ever users versus “never users” and by duration of progestin use (< 56 days, ≥ 56 days). Absolute risk differences were calculated. Time to event was estimated using Kaplan Meier survival functions. We computed adjusted Cox proportional hazard ratios with progestin exposure as a time-dependent variable. The time between index and censoring was divided into 6 month periods. Only cohort members who had been followed for one year or more were included in the analyses using time-dependent variables.
We considered and evaluated confounding factors and adjusted for variables that influenced the risk estimates associated with progestin dispensing by more than 10%, specifically, age (<50, ≥50 years) and body mass index (BMI) (<30, ≥ 30 kg/m2). Analyses were performed using STATA 9.2 (STATA Corporation, College Station, Texas). All reported p-values are 2-sided.