br Controls were identified through random digit dialing and
Controls were identified through random-digit dialing and were frequency-matched to cases by 5-year age groups and 3-digit telephone
Institutional Review Board approval for the study was obtained from hospitals in which cases were identified and from the University of Pittsburgh. All participants provided written, informed consent.
2.2. Data collection and exposure assessment
Trained interviewers conducted a standardized 2-hour in-person in-terview to obtain detailed information on reproductive, medical, and demographic data from birth until a reference date. To aid recall, a life events calendar with milestones, such as marriages, births, and deaths, was used . Each pregnancy was denoted on the calendar by coloring the month of pregnancy MDMB-CHMCZCA until pregnancy end. Breastfeeding episodes were similarly noted on the calendar. For each pregnancy, a woman was asked the outcome (live birth, still birth, miscarriage, abor-tion). For each live birth, she was asked detailed information, including “Did you ever breastfeed this baby?” If she responded “yes”, the total number of months breastfed was elicited by asking “For how many months did you nurse?”
The reference date was calculated as 9 months prior to diagnosis (cases) or interview (controls) to ensure that exposures occurred before ovarian cancer diagnosis in cases and within a similar time frame for controls.
Pathology data were extracted from pathology reports by two inde-pendent readers. Differences were reviewed by study staff to assign final pathology data.
2.3. Statistical analyses
Analyses were limited to the 1572 controls and 689 cases who re-ported at least one live birth. A woman was classified as having ever breastfed if she responded positively to the question “Did you breastfeed this baby?” for any live birth she reported. Total duration of breastfeeding was calculated by summing the number of months breastfed across all breastfeeding episodes. Age at first breastfeeding episode was calculated using the month and year of the pregnancy end corresponding to the first breastfeeding episode and the month and year of a subject's birth. Age at last breastfeeding episode was sim-ilarly calculated using the month and year of the pregnancy end corre-sponding to the last breastfeeding episode. Time since the first breastfeeding episode was obtained by calculating the time elapsed since the end of the first breastfeeding episode until the reference date. Time since the last breastfeeding episode was similarly obtained using the time elapsed since the end of the last breastfeeding episode.
We assessed total number of breastfeeding episodes, breastfeeding some or all offspring, average duration of breastfeeding per breastfeeding episode (defined as total breastfeeding duration divided by total number of offspring breastfed), and the duration of breastfeeding for the first and last breastfeeding episodes.
Case-control differences in demographic and other factors were ini-tially assessed using X2 tests. Multivariable unconditional logistic re-gression was used to calculate odds ratios (ORs) and 95% confidence intervals (CIs) for the association between each breastfeeding factor and EOC (separate models for each factor). Age at reference date, total OC duration, parity, race, education, tubal ligation status, hysterectomy status, and family history of breast or ovarian cancer were selected a priori as potential confounders. A variable was retained if its removal re-sulted in at least a 10% change in the effect estimate. Family history of breast and ovarian cancer, tubal ligation, and hysterectomy did not af-fect the relationship between breast feeding and ovarian cancer risk and were not included in the final models. Sensitivity analyses showed estimates were not changed when including factors that may influence a woman's decision to breastfeed in our models (body mass index (BMI), smoking history, ever use of alcohol, and ever use of aspirin or
NSAIDs); therefore, they were not included in the final models. Sensitiv-ity analyses showed estimates were also unchanged by adjusting for age at first live birth, age at last live birth, decade of subject's birth, and year of birth as being pre vs post 1950 and therefore we not included in the final models. Thus, the final models included age at reference date, total OC duration, and parity as continuous variables, and race and education as categorical variables. Tests for trend were performed by coding the ex-posure of interest as a grouped linear variable. Analyses were repeated examining only women age 50 or over and restricting cases to women with invasive disease only. Further analyses examined the associations limiting to high grade serous EOC (the most common histotype); limited cases of other histotypes precluded meaningful analyses. All p-values were two-sided and considered statistically significant at p b 0.05. Analy-ses were conducted using Stata version 9.1 (StataCorp).