Induced Abortion and Risk of Breast Cancer

To the Editor:

In their recent reply to Sharpe, ¹ Melbye et al² defend the methods employed in their earlier study on induced abortion and breast cancer risk, ³ and reiterate their claim of “no biological effect of induced abortion.” We concur with Melbye et al that it is reasonable to ascertain the effect of induced abortion, independent of abortion’s nonspecific effect in raising breast cancer risk by delaying the first birth. We are surprised, however, that Melbye et al stand firm by their null result, inasmuch as their methodology contained serious flaws ⁴ that they have since been corrected in their recent study on the effect of preterm delivery on breast cancer risk. ⁵

Both studies used the same overall cohort of Danish women born between 1935 and 1978. In the former study ³ all women in the cohort were automatically entered into the analysis and followed for breast cancer from age 12, whereas women were only entered into the analysis in the latter study ⁵ if and when they gave birth in 1978 (or later), the year in which the gestational age registry began. This restriction eliminated the obvious misclassification of a large portion of the study cohort, had the authors assumed that women giving birth before 1978 had no preterm births. It was precisely this misclassification—of 60,000 of the oldest members of the study cohort—which did occur in the former study, ³ because 80,000 legal abortions, which occurred among the study cohort before the inception of the computerized registry, ⁶ were ignored. Yet in the latter study, ⁵ Melbye et al explicitly took pains to avoid any misclassification even of women, included in the analysis, who may also have given birth before 1978, by statistically estimating the number of preterm, pre-1978 deliveries.

A second fundamental error in the former study ³ was the inclusion of breast cancers diagnosed since 1968, despite the fact that the recording of the exposure variable—induced abortion—did not begin until over 5 years later in 1973. This error necessarily caused an underestimation of the relative risk. The error was corrected in the latter study ⁵ by the inclusion of breast cancer cases diagnosed only since the inception of the preterm birth registry in 1978.

These methodologic corrections (and the exclusion of nulliparae) reduced the number of cases included in the analysis from 10,246 to only 1,363. The result is a meaningful contribution to knowledge of the effect of pregnancies of different length on breast cancer risk, specifically, that pregnancy shifts from being a risk-increaser to a risk-decreaser at approximately 32 weeks gestation.

We have previously suggested ⁴ that Melbye et al reexamine their induced abortion data by comparing birth cohort-matched subjects with vs without induced abortions. They rejected this suggestion, claiming that their cohort design took birth cohort differences into account “more efficiently.” Nevertheless, regardless of the statistical model used, there is simply no way to assess accurately an association between two variables when the exposure (induced abortion) and the outcome (breast cancer) happened in two largely separate populations (ie, the younger vs the older members of an overall cohort that spans over four decades). Nor can a valid result be obtained by ignoring the demonstrated misclassification of tens of thousands of subjects.

We challenge Melbye et al again to reexamine their induced abortion-breast cancer study, this time with the more modest suggestion that they apply their own, corrected statistical methodology, and with the additional caveat of omitting the adjustment of the data for birth cohort. Because the birth cohort effect on breast cancer incidence in Denmark ⁷ —whose cause is unknown—closely parallels induced abortion exposure by birth cohort, ⁶ any contribution of induced abortion to breast cancer risk would necessarily be underestimated by adjusting it for birth cohort.

Joel Brind

Vernon M. Chinchilli