AN OBLIQUE BUT STARTLING CONCESSION.
Faced with the evidence, most denials, if you will, go around the research. They assert the claim that these studies results are unreliable due to a hypothetical entity referred to variously as "response bias," "reporting bias," or "recall bias."
We shall detail this alleged failure later in the article but the gist of the argument is that the way certain categories of women report (or fail to report) their abortions unfairly leaves the impression that women who've aborted have a higher incidence of breast cancer.
Quietly and obliquely, in a letter to the editor published in this month's issue of the British Medical Association's Journal of Epidemiology and Community Health (JECH), the "evidence" almost universally relied upon to back up this argument was abandoned by the very group which first presented it! Let's look at the background.
In 1991, a Swedish team, including the prominent researcher Olav Meirik of the World Health Organization, published the only paper to date which claimed to show direct, statistically significant evidence of response bias. Appearing in the American Journal of Epidemiology, the paper compared data the team had gathered previously, by two different methods, on the same population of Swedish women.
In papers appearing in the British publication Lancet and the British Medical Journal in 1986 and 1990, respectively, the group published combined data on Swedish and Norwegian women, data which were derived from interviewing a patient group and a healthy group (the "controls") about their reproductive history.
In 1989, the same group published a study in the British Medical Journal based on computerized abortion registry records of the same Swedish women. Such prospective records, i.e., records collected long before anyone in the study got breast cancer, are by nature not subject to inaccuracies due to recall which may occur in an interview-based, after-the-fact study (retrospective).
According to the Swedish research team, the crux of the response bias argument is this hypothesis: "a woman who had recently been given a diagnosis of a malignant disease, contemplating causes of her illness, would remember and report an abortion more consistently than would a healthy control" (a woman without a malignant disease). The result would be an apparent but misleading elevation in breast cancer risk since it would seem that breast cancer patients have had more abortions than healthy women: response bias.
This seems plausible, but the evidence presented to back it up--something called "overreporting"-- is dubious indeed. A look at the data tables in the 1991 paper reveals that seven Swedish breast cancer patients reported having had an induced abortion for which the computer had no record compared to only one healthy (control) subject with no computer record.
To support their hypothesis, the authors assumed that the computer record had to be right. Therefore, these eight women must have imagined abortions that never really took place: they had "overreported" them!
Common sense to the contrary notwithstanding, this dubious assumption provided the statistical significance for the otherwise insignificant finding of "underreporting" by healthy women. It was good enough to rely on by those bent on discrediting the ABC link.
Consider, for example, this wild exaggeration by Douglas Weed and Barnett Kramer of the National Cancer Institute, in a December 1996 editorial in the institute's own journal (JNCI). They wrote that the Swedish study "show(ed) that healthy women consistently and widely underreport their history of induced abortion."
Alternate interpretations of this "evidence" of response bias went relatively unnoticed. For example, Janet Daling et al., in their 1994 abortion-breast cancer study published in the JNCI, expressed a more logical view: "we believe it is reasonable to assume that virtually no women who truly did not have an abortion would claim to have had one."
That brings us up to the present, and the current issue of the JECH, where we now find Meirik et al. admitting that since the process for obtaining a legal abortion in Sweden before 1975 was "time-consuming and perceived by many as stigmatizing and paternalistic.... [s]ome women therefore had induced abortions abroad or unrecorded terminations of pregnancy." They add, "We are not
surprised to find some Swedish women confidentially reporting having had induced abortions during the period 19661974 which are not recorded (i.e., in the computerized registry) as legally induced abortions."
This admission is as oblique as it is striking. The authors use the term "overreporting" only once in the whole text of the letter, and in quotes, as if the notion were someone else's and not theirs in the first place. Even more peculiarly, having just acknowledged there is a lack of any credible justification, they reaffirm the idea of recall bias in the very next sentence!
The Meirik et al. letter was written as a criticism of the "Comprehensive review and meta-analysis" of the ABC link which I had written with co-authors from the Pennsylvania State University College of Medicine and published in the October 1996 issue of the JECH. That gave us the right to reply.
In our response, referring to the above admission by Meirik et al., we state: "This interpretation marks an about-face; an acknowledgement that the computer registry may not be the `gold standard' for assessing the occurrence of induced abortion. Thus, based on discrepancies between the interview and computer registry data, the claim of `overreporting' is acknowledged by Meirik et al. to be unfounded, and with that, significant evidence of response bias evaporates."
However, this is only half the story, for in preparing our response to the Meirik et al. letter, we had need to revisit this Swedish group's earlier studies. In doing so, we uncovered evidence that rigorously proves that they had ignored evidence of the ABC link in Norwegian women.
Specifically, the 1990 paper, which included data on induced abortion, presented combined data for Swedish and Norwegian women. Since the 1991 paper compared interview-based data to computerized data only for the Swedish women, it made it possible -- by subtracting data from the Swedish women from the combined data for the Swedish and Norwegian women -- to obtain separate data for the Norwegian subjects.
The result was startling: Though the Swedish breast cancer patients outnumbered the Norwegian patients by more than 3 to 1 (317 to 105), only 26 of the 73 patients who reported having had an induced abortion were Swedish; 47 were Norwegian. In other words, of the 317 Swedish breast cancer patients, about one in 12 (8.2%) had had an induced abortion, but almost one in two (45%) of the Norwegian breast cancer patients had had an induced abortion!
The effect of the wholly inappropriate statistical maneuver of combining the Swedish and Norwegian data was to bury a definitely positive association between induced abortion and breast cancer (i.e., increased risk) in Norwegian women.
Numerically, we were able to prove that, in contrast to the 11% decreased risk reported in the 1990 paper for the combined Swedish and Norwegian study population, the data actually demonstrated an increased breast cancer risk for Norwegian women somewhere between 12% and 123%. We were also able to determine that, if both Swedish and Norwegian women in the study were affected to the same extent by induced abortion, it would be just over 30% increased risk--the same 30% increased risk which we had calculated as a worldwide average in our 1996 meta-analysis.
It remains to be seen just how much of an impact these revelations will have on the shapers of public policy at the World Health Organization, the National Cancer Institute, and other bodies who have bent over backwards to find excuses to deny the reality of the ABC link. Now they have two fewer.
This article was adapted from an article appearing in the current issue of the Abortion-Breast Cancer Quarterly Update., which Dr. Brind publishes. Subscription information may be obtained by phone or fax at (914) 463-3728, by mail at P.O. Box 3127, Poughkeepsie, NY 12603, or from Dr. Brind's website: www.abortioncancer.com.