Induced Abortion and the Risk of Breast Cancer
ABSTRACT & COMMENTARY
Melbye et al from Denmark established a cohort of 1.53 million women, born 1935-1978, by linking data from the Civil Registration System, the National Registry for Induced Abortion, and the Danish Cancer Registry. From this cohort, 280,965 women with 370,715 induced abortions and 10,246 women with breast cancer were identified. After adjusting for age, parity, age at delivery of first child, and other risk factors, the risk of breast cancer was identical in women with and without a history of induced abortion (even with more than one induced abortion). No effect was found with time interval between induced abortion and diagnosis, age of the women at time of diagnosis, or calendar period (year) at diagnosis or induced abortion. A trend was found for an increasing relative risk with increasing gestational age at induced abortion; however, the analysis of late abortions was limited by small numbers (6 at 13-14 weeks, 17 at 15-18 weeks, and 14 at > 18 weeks). (Melbye M, et al. N Engl J Med 1997;336:81-85.)
COMMENT BY LEON SPEROFF, MD
Conflicting results (increased risk of breast cancer, decreased risk, and no effect) have been reported in over 20 studies of the relationship between induced abortion and the risk of breast cancer. Concern for an adverse effect has been based on the theoretical suggestion that a full-term pregnancy protects against breast cancer by invoking complete differentiation of breast cells, but that abortion increases the risk by allowing breast cell proliferation in the first trimester of pregnancy without allowing the full differentiation that occurs in later pregnancy (Am J Pathol 1980;100:497-512). This controversy was highlighted in the media by a meta-analysis published in 1996 concluding that there was a 30% increased risk of breast cancer in women with a history of induced abortion (J Epidemiol Community Health 1996;50:481-496). This added further fuel to the debate between "pro-choice" and "pro-life" advocates.
This whole experience is an excellent example of how a meta-analysis of observational studies (case-control and cohort studies) cannot correct for biases present in the individual studies. For this subject, there is a major problem of recall bias. Women who develop breast cancer are more likely to truthfully reveal their history of induced abortion than healthy women. How can this bias be avoided?
In the Danish study, recall bias was avoided by obtaining all data from national population registries, avoiding personal interviews. In a country where reporting of induced abortion and cancer is mandatory, computer linkage and follow-up are effective and complete. Furthermore, the impressive size of the study strengthens the credibility of the results. The number of women with breast cancer who had had induced abortions was 4-6 times larger than case numbers in previous case-control studies.
The finding of increased risk with later abortions should be viewed very cautiously because of the small numbers.
The underreporting of induced abortion is a well-recognized phenomenon (Demography 1992;29:113-126). A recent Dutch case-control study on induced abortion and risk of breast cancer emphasized the importance of recall (reporting) bias (J Natl Cancer Inst 1996;88:1759-1764). The Dutch study evaluated induced or spontaneous abortion in two geographic areas that differed in prevalence of and attitude toward induced abortion. There was no evidence for an increased risk of breast cancer associated with spontaneous abortion, and an increased risk with induced abortion was attributable to underreporting by the healthy subjects in the more conservative regions of the country with a predominantly Roman Catholic population. Supporting the recall bias explanation was the finding that in these same regions the patients underreported their use of oral contraceptives in comparison with physician reports. The failure to find an increased risk of breast cancer associated with spontaneous abortion also supports the recall bias explanation.
The current epidemiologic conclusion regarding induced abortion and risk of breast cancer is undeniable: there is no solid evidence for an increased risk. In addition, there is an important lesson to be gained from the more than 20 studies, the very recent epidemiologic data, and the meta-analysis. Clinicians must not assume that the results of a meta-analysis are reliable and without problems. The technique of meta-analysis was developed to bring together results from multiple small, randomized clinical trials in order to gain statistical power that is lacking with small studies. A meta-analysis of randomized trials is very reliable and powerful (e.g., the analysis of breast cancer treatment data in Lancet 1992;339:1). However, a meta-analysis of observational studies (case-control and cohort) cannot correct for the biases and errors inherent in the individual studies, and the conclusion of the meta-analysis may magnify the problems. The meta-analysis of induced abortion and breast cancer risk is an excellent example of this epidemiologic difficulty. I strongly urge clinicians to view a meta-analysis of observational studies critically and with skepticism.