Abortions and breast cancer: Record-based case-control study

Abortions and breast cancer: Record-based case-control study

It has been suggested that abortions leave the breast epithelium in a proliferative state with an increased susceptibility to carcinogenesis. Results from previous studies of induced or spontaneous abortions and risk of subsequent breast cancer are contradictory, probably due to methodological considerations. We investigated the relationship between abortions and subsequent breast cancer risk in a case-control study using prospectively recorded exposure information. The study population comprised women recorded in the population-based Swedish Medical Birth Register between 1973–91. Cases were defined by linkage of the birth register to the Swedish Cancer Register and controls were randomly selected from the birth register. From the subjects' antenatal care records we abstracted prospectively collected information on induced and spontaneous abortions, as well as a number of potential confounding factors. Relative risk of breast cancer was estimated by odds ratios (OR) with 95% confidence intervals (95% CI). A reduced risk of breast cancer was observed for women with a history of at least 1 compared to no abortions (adjusted OR = 0.84, 95% CI = 0.72–0.99). The adjusted OR decreases step-wise with number of abortions to 0.59 (95% CI = 0.34–1.03) for 3 or more compared to no abortions. The patterns are similar for induced and spontaneous abortions. In conclusion, neither a history of induced nor spontaneous abortions is associated with an increased risk of breast cancer. Our data suggest a protective effect of pregnancies regardless of outcome.

Hormonal and reproductive factors have been identified as major determinants of breast cancer risk.1 Both the number of pregnancies and early age at first pregnancy are inversely associated with breast cancer risk, possibly due to a pregnancy-induced terminal differentiation of the mammary epithelial cells.2 Animal studies suggest, however, that this protective effect is confined to term pregnancies, and that an abortion might leave the breast epithelium in a susceptible intermediate stage, where proliferation has begun but full differentiation is not reached.3

Results from previous studies on abortions and subsequent breast cancer risk are conflicting. Spontaneous abortions have generally not been found to influence the risk of breast cancer.4 A positive association between induced abortions and subsequent breast cancer risk has been reported in several epidemiological studies,5 but a large Danish record-based cohort study6 and 3 recent record-based case-control studies7, 8, 9 have failed to reproduce this association. A recent review concluded that the data presented to present do not justify a warning to women when counseling about abortion, but neither exclude the presence of an association.10 One suggested reason for the conflicting results is that reporting bias has produced false associations in interview-based case-control studies on induced abortions.4 If women with breast cancer were more likely to report accurately previous induced abortions than community controls, this would create a spurious association between abortions and breast cancer.

If abortions influence breast cancer risk, the impact on public health could be substantial. Breast cancer is the one of the largest contributors both to cancer mortality and morbidity among women in the western world.11 Abortions constitute a common exposure among women in fertile ages. In Sweden 350 induced abortions are carried out per 1,000 live births, and this number is comparable to that of other western countries.12, 13, 14

We studied induced and spontaneous abortions and subsequent breast cancer risk in a case-control study using information on abortions and potential confounders collected before the onset of breast cancer.

MATERIAL AND METHODS

Data sources

Data from 2 population-based registers, the Swedish Medical Birth Register and the Swedish Cancer Register, and from medical records were used for the study. The registers were used to identify cases and controls, whereas all information on abortion and potential confounders was collected from medical records.

The Swedish Medical Birth Register has recorded information on close to 100% of all births in Sweden since 1973.15 In the Swedish Cancer Register, information on all incident cancers has been recorded since 1958. The notification to the Swedish Cancer Register is mandatory both for clinicians and pathologists, resulting in an estimated completeness of more than 98%.16 Data on individuals recorded by the registries and in medical records can be linked using the unique 10-digit identification number assigned to all Swedish residents.

Study population

The study compared the number of reported abortions at antenatal care in women with a diagnosis of breast cancer (cases) with the number in women without breast cancer (controls).

Eligible cases for the study were identified through linkage between the birth register and the cancer register. These included all women who had given birth in Sweden between January 1st, 1973 and December 31st, 1991 with a subsequent diagnosis of primary breast cancer during the same time period. Women with a diagnosis of breast cancer before giving birth were excluded.

Two potential controls per case were selected and individually matched to cases on year of birth (within a 3-year interval from the case) as well as hospital and year of giving birth. Controls were randomly selected from the birth register and included women who had given birth after January 1st, 1973. Controls were alive and free from breast cancer and had not emigrated by the date of diagnosis of the paired case. For each case, only 1 control was used. If exposure information could not be assessed for the first control, we attempted to retrieve information on the second. If exposure information was unavailable for either the case or for both controls, the case-control pair was excluded from the analysis.

For logistical reasons, the study was restricted to 23 Swedish hospitals. In all, 1988 breast cancer patients were included in the study, which represents 44% of all eligible patients in Sweden during the period. All exposure information was collected from antenatal care records, standardized throughout Sweden on January 1st, 1973. The record is based on a physical examination and an interview carried out by the midwife at the first antenatal care visit, typically at 8 weeks of gestation. We abstracted information on all previous abortions and births and on potential confounding factors. Data on height and weight were based on the physical examination, whereas all other data was self-reported by the woman to the midwife. As all data were collected prospectively, information on abortions and confounding variables refer to the time before this interview. Information on use of oral contraceptives before pregnancy was recorded, but there were no data on dose or duration of use. Information on smoking referred to the time of interview and was noted as daily smoker or not. Information on abortions included calendar year, week of pregnancy and whether the abortion was induced legally or spontaneous.

For 175 cases (8.3%) we were not able to retrieve the antenatal record, or the case was excluded due to incomplete data. For 123 cases (6.4%) data was missing for both controls, and for 99 cases (5.1%) data was missing for the first control but not the second. After exclusion due to incomplete data, information on 1,759 pairs of cases and controls (3,718 individuals) was available, corresponding to 84% of the potential sample. The main reason for incomplete data was that the subject gave birth during the first months of 1973, and the interview thus took place in 1972, before the implementation of the standardized record.

Statistical analysis

Relative risks were estimated by calculation of odds ratios (OR) with 95% confidence intervals (95% CI) using conditional logistic regression. Within the 3-year matching interval, there were age differences between cases and controls when the exposure information was collected. The older in each pair would be more likely to have had 1 or more abortions. To overcome this potential source of bias we excluded abortions taking place at an older age than the age at which the exposure information was collected for the youngest in the pair. Smoking, use of oral contraceptives, height, pre-pregnancy weight, number of births and age at first birth were initially included in the model as potential confounding factors. In the final analyses, adjustments were made for all variables found to influence the estimates notably. Women's year of birth was a matching criterion and effects within the matching interval were analysed with women's year of birth as a continuous variable. All other variables were modeled as series of dummy variables, thus making no assumptions about linearity. Age at first birth was analysed in 5-year intervals (≤15, 16–20, 21–25, 26–30, 31–35, ≥36) and height in strata defined by quintiles (≤161 cm, 162–164, 165–167, 168–170, ≥171). A number of subjects (12%) lacked information on height and were in the adjusted analysis assumed to have median height (165–167 cm). Sub-analyses were carried out on induced and spontaneous abortions separately and on abortions before first birth. In the adjusted analyses of induced and spontaneous abortions, the 2 lowest strata of age at first birth were collapsed together due to small numbers. All calculations were made using SAS statistical software (SAS Institute, Cary, NC).

Дата добавления: 2015-11-04