Methods
The Cohort
The study population was a historic cohort of women pregnant between 1954 and 1963 in Helsinki, Finland (n = 4,090). The cohort was a sample gathered for a study on hormone exposure, including 2,022 exposed, 2,062 controls and 6 women with unknown hormone exposure status. Information on the cohort was collected from the maternity cards of municipal maternity centers, which are used by most pregnant Finnish women. The hormone-exposed women had been prescribed estrogen or progestin drugs during pregnancy to prevent early abortion or preterm delivery. For each exposed woman, a woman next in the maternity center file who gave birth during the same year and had not been prescribed hormones during pregnancy, was chosen as a control. The cohort has been previously prescribed in detail.[22,23] There were no differences in breast or other estrogen-dependent cancers between hormone-exposed and control mothers.[22] Visits to a private doctor were used as an indicator of socio-economic status, since no information on education or occupation at the time of the index pregnancy was available.
Cancer cases were identified through a record linkage to the national cancer registry until June 2001. Mortality and emigration data were obtained from the population registry until August 2001. The linkage between the cohort and the registries was based on a unique personal identification number.
Inclusions and Exclusions
Inclusion criteria were the following: first and last visit at the maternity center between 4-45th gestation weeks, the time between the body weight measurements 3-300 days, and delivery between 22-45th gestation weeks. For each mother, the gestation week she gave birth was determined by using the date of estimated timing of delivery. Women who did not fulfill these criteria were excluded (Fig. 1). In addition, women with multiple births were excluded because their weight gain is not comparable to that of mothers of singletons. Mothers with pre-eclampsia or eclampsia were excluded because they accumulate weight as fluid retention during pregnancy, and have been reported to have a reduced breast cancer risk.[11,12]
Study population and exclusions
Pregnancy weight gain was first calculated as the difference between the first and last visit to maternity center. However, this window varied considerably among included mothers (range 3-295 days). The time-period of calculated weight gain was therefore adjusted by extrapolating a line for each mother to reflect her weight increase during pregnancy. The calculations are described in detail in Additional File 1. After the unstable period of early pregnancy, a mother’s weight increases linearly.[24] Mothers usually begin to gain weight after the first trimester (e.g [25]). We extrapolated the line separately for 0-15th (Line A) and 15-40th gestation weeks (Line B) for each mother. Weight gain was extrapolated to continue until 40th gestation week for all mothers, although 22.2% of mothers delivered at 39th gestation week or before.
For mothers for whom both Line A (n = 2,143) and Line B (n = 2,184) were available, total pregnancy weight gain was calculated by adding the extrapolated weight gains from both periods. Thus, total pregnancy weight gain could be extrapolated only for 2,089 women. Cases and controls (66.5% vs 65.0%) did not differ concerning the number of available weight measurements. For the rest of the women, either the first weight measurement was later than 24th gestation week, the last weight measurement was before 30th gestation week, or information on pre-pregnancy weight, weight at the first or the last visit, or timing of the visits was not available. As indicated above, these subjects were excluded from the analyses.
We compared the characteristics of the mothers who were excluded (n = 2,001) to the characteristics of the mothers in the final study population for whom total pregnancy weight gain could be determined (n = 2,089). The two groups were similar in regard to breast cancer incidence, age at menarche, height and the frequency of visits to a private doctor. However, the excluded mothers were older (mean: 27.1 years vs. 26.5 years, p < 0.001), heavier (58.7 kg vs. 57.3 kg, p < 0.001; body mass index, BMI: 22.3 kg/m2 vs. 21.8 kg/m2, p < 0.001), older at first birth (25.2 years vs. 24.7 years, p = 0.016), had more children during index pregnancy (1.92 vs. 1.81 at index birth, p < 0.001), and were more often exposed to estrogen or progestin drugs (50.3% vs. 48.6%, p = 0.021). Their children were shorter (mean 49.3 cm vs. 50.3 cm, p < 0.001) and weighed less (mean 3,310 g vs. 3,472 g, p < 0.001), suggesting that excluded mothers' pregnancy weight gain might have been lower. It is probable that exclusion of these women had no major effect on the findings.
The Case-control Study
A nested case-control study was performed to determine whether later weight development confounded the association between pregnancy weight gain and breast cancer risk. A sample of women was chosen from the final cohort (n = 2,089) that included all breast cancer cases with data on pregnancy weight gain (n = 123). For each case, we chose seven randomly selected controls (n = 856) who were born in the same year as the case. These 979 women were linked to the Hospital Inpatient Registry to obtain information on the women's stays in hospitals. 117 cases were identified with a hospital visit in average 0.4 months after breast cancer diagnosis (median 0.0, range from -16.3 to 17.2), and of these cases information on body weight and height was available for 65 (53% of 123 cases). Among the controls, 699 had been a patient in a hospital at a similar age than their respective cases (maximum difference ±5 years), and 431 of them had weight and height available in the hospital archives (50% of 856 controls, 6.6 controls/case).
The breast cancer cases with no information on later body weight did not differ from the cases used for the nested case-control analysis. The controls with no information on later body weight were approximately 1.5 years older at the time of the hospital visit than the controls included to the study (p = 0.027).
Statistical Analysis
Statistical significance of possible differences in baseline characteristics of the study population, pregnancy weight gain and postpartum weight loss and weight retention by tertiles of pregnancy weight gain was tested by using analysis of variance for continuous variables and ?2-test for proportions. The incidence of breast cancer per 100,000 person years was counted by groups of 5 kg pregnancy weight gain, tertiles of pregnancy weight gain and tertiles of postpartum weight retention. Person years were calculated from the delivery to the diagnosis of breast cancer or other endpoint including death, emigration or end of the study.
The association between pregnancy weight gain and breast cancer risk was further examined using a Cox regression model. Total pregnancy weight gain was included as a categorical covariate (tertiles) in the model. Age at menarche, age at first birth, age at index pregnancy, BMI before pregnancy, and parity (at index birth) were all used as continuous covariates in the model. Postpartum weight retention 51 days after delivery (mean) was later added to the model.
The incidence of breast cancer was counted and the Cox regression model was carried out also separately for pre- and postmenopausal breast cancers. Information on the age at menopause was not available. Therefore all women were expected to have menopause at the age of 50 years.
In the case-control study, weight and BMI change between pre-pregnancy and at the time of later hospital visit were compared between the tertiles of pregnancy weight gain (analysis of variance). A Cox regression model that included later BMI was also used to analyze the data.
