Back to Home
ABC in the courts: Dramatic ABC testimony
in Florida’s parental notification appeal
By Joel Brind, Ph.D.
Startling admissions—and blunders—by one of ABC link’s biggest foes.
Readers may remember Boston University epidemiologist Lynn Rosenberg as the author of the
1994 editorial8 which accompanied the publication of the ABC study by Daling et al. 9 in the
Journal of the National Cancer Institute. Rosenberg was also the expert witness for the
Southeast Pennsylvania Transit Authority, which government agency was in Federal Court in 1996
to defend—unsuccessfully—its removal of privately funded ads warning the public about the ABC
link (see ABCQ Update, Summer 1997, Summer 1998).
This November, Dr. Rosenberg was in court again, this time as the ABC expert witness for a
group of Florida abortion practitioners, challenging the state’s new law requiring parental
notification within 48 hours before a minor girl could obtain an abortion.
While staunchly maintaining that "there is a consensus of opinion that a link between
induced abortion and breast cancer has not been established", Dr. Rosenberg was forced
to make some dramatic admissions under cross-examination:
"Q: So in other words, a woman who finds herself pregnant at age 15 will have a higher breast cancer risk if she chooses to abort that pregnancy than
if she carries the pregnancy to term, correct?
A: Probably, yes.
Q: Looking at that another way, let’s compare two women.
Let’s say both got pregnant at age 15—one terminates the
pregnancy, but the other carries the pregnancy to term.
And both women go on to get married and have two children, say, at age 30 and age 35. Is the risk of breast cancer
higher for the woman who had an abortion at age 15 or the woman who had a baby at age 15, all other things being
A: It’ s probably higher for the one who had an abortion
at age 15."
While this appears to be a striking contradiction to the idea that the ABC link has not been
established, it actually refers to one aspect of the ABC link that has been universally accepted for
many years10 and that is the following: a full-term pregnancy early in a woman’s life (before age
30) lowers her future risk of breast cancer. Aborting the pregnancy abrogates this protective
effect, and so a woman who has an abortion rather than a baby at a young age is at higher risk, as
Dr. Rosenberg readily admitted. But she did so only when—under oath—the question was
phrased in an unequivocal way, for the detractors of the ABC link are masters of equivocation on
this point. For the most recent example, Melbye et al., in a recent letter published in
Epidemiology11, perform the following gymnastic maneuver:
A related misunderstanding is to argue that induced abortion enhances the risk of breast
cancer because primigravida (i.e., first time pregnant) women having an induced
abortion postpone the first birth they would otherwise have had. Nevertheless, this effect
is again a biological effect of the first birth and not of the induced abortion."
In our response to the Mclbye et al. letter (slated for publication in March12), Dr. Chinchilli and I
just talk straight about "abortion’s nonspecific effect in raising breast cancer risk by
delaying the first birth." Hence, what is ground-breaking about the Rosenberg testimony is that
she was forced to make the plain language admission.
It is, then, a separate but key issue as to whether abortion also raises breast cancer risk
independently of its effect in delaying the first full-term pregnancy. This effect is attributable to the
effect of the huge quantities of growth promoting estrogen flooding the breasts during the
pregnancy before it is aborted. This independent effect is what we documented in our 1996
"Comprehensive review and meta analysis"7. The basis of all the arguments by those such as
Rosenberg, who claim that this link has not been established, is a different interpretation of the
body of evidence of the independent effect.
Under the skillful cross-examination of Mr. Ken Sukhia of Fowler White, the Tallahassee law firm
working for the state of Florida in the case, Dr. Rosenberg admitted a great deal more than
abortion’s non-specific effect in delaying first full-term pregnancy. As an expert in the field, she
showed an astonishing—and convenient—lack of knowledge about the the medical record which
has documented the ABC link.. Hence a detailed analysis of the major misstatements she made in
her videotaped testimony for Florida Circuit Judge Terry Lewis is well worth the read:
One major bone of contention concerned the first published study to document the ABC link.
Published in 1957 in the journal GANN (The word ‘gann’ means ‘cancer’ in Japanese, and the
journal’s name is an acronym for Japanese Journal of Cancer Research). Since this study by
Segi et al.2 was published before current epidemiological formulations were standardised it
reported data in terms numbers of pregnancy outcomes (e.g., full-term birth, induced abortion,
spontaneous abortion) rather than numbers of women with pregnancy outcomes, among women
with v, without breast cancer.
Rosenberg tried to argue that it was invalid to include the Segi study in our meta-analysis, because
we needed to use data from other Japanese studies to translate the Segi data into modern, relative
risk format. (The argument was intended to serve a twofold purpose: 1., to eliminate the 160%
increased risk the Segi study showed from the both of evidence of the ABC link, and 2., to
undermine the credibility of the meta analysis and particularly its first author—me——, since I was
the ABC expert testifying for the State of Florida in the case.)
But when cross-examined in regard to the Segi study, Rosenberg said two things—on the
record—which were simply and totally false:
"Q: I want to show you that study. (the Segi study) Have you reviewed that study?
A: No, I have never seen this study. It was published in a Japanese journal, in Japanese. This is the first time I’ve ever seen it
Sorry, Dr. Rosenberg. GANN is an English language journal. Secondly, although it might be true
that she had never seen a copy of the study, she certainly had no excuse for this being the case.
That’s because she advanced essentially the same arguments against the Segi study when she
testified in the CBM v. SEPTA case in 1996 (see articles in Spring and Summer 1997 and
Summer 1998 ABCQ Update) where the very same English Segi reprint was provided to her
(though she may never have even looked at it)!
We move on:
"Q: You had indicated—I believe you used the word that you felt it was an
A: Yes, I do think it’s an invalid study if it doesn’t give information on the
numbers of induced abortions in the study."
Now that’s a very big "if" indeed, since the Segi study does report precise data on the number of
Of course, one need not take what might be termed an unduly cynical approach here. Perhaps it is
too much to expect an expert to place much credence in an old, somewhat obscure study that
predated modern epidemiological methodology.
But then, how does that explain—or excuse—Dr. Rosenberg’s lack of familiarity with what is
arguably the best published research done on American women? In fact, the 1989 study of Howe
et al.13 is the only one based entirely on prospective data, so that even the possibility of response
bias can be absolutely excluded. (The Howe study also reported a statistically significant, 90%
increased risk of breast cancer for women who had had an induced abortion, a fact which was
also acknowledged by Dr. Rosenberg in her testimony.) But Dr. Rosenberg’s memory was
demonstrably—and conveniently—flawed in this regard:
"Q: Now, wasn’t that study one that was based on prospective data?
"A: Let me check. Yes, it was.
Q: So there couldn’t poasibly be—at least with regard to that study-any
response bias or reporting bias—I’m sorry, as you put it—in that study since it was based on prospective data, i e., fetal death certificates?
A: Just let me check please. I’m really not familiar with that study. I haven’t looked at it recently.
Q: So the reports of abortion were obtained from fetal death
A: Yes, I believe they were, based on my review. Right. The problem with this—and I must disagree with you that there wasn’t reporting bias—is
that unlike the registry in Denmark, for example, where by law the abortions are reported to the registry, in this study the reports of
the induced and spontaneous abortions came from the women themselves. So once again, we have a problem of reporting, potential problem of
"Q: Well isn’t that (i.e., abortion data in the Howe study) then limited
to those which were reported on the fetal death certificate by the physician?
"A: By the woman. The physician gets the information from the woman.
Q: Well, let me ask you to read from page 301 at the top, the first sentence there.
A: Oh, by the physician, you’re correct.
"Q: Okay. So would that not exclude what you’re saying; that is, reporting or a response bias?
"Q: So this was a study that does not suffer from what is the suspected reporting bias that you testified about?
Reporting bias should be familiar tot the reader as the main justification for those—like
Rosenberg-—in the denial camp of the ABC link to dismiss the findings of increased risk in the
vast majority of ABC studies. Since these studies generally rely upon data given by the
participants in the study about their own reproductive history, it is hypothesized that women with
breast cancer will be more forthcoming and accurate about their history of abortions than healthy
women. Were this true, the observation of increased risk could simply be due to this bias in
reporting ‘by the women, resulting in apparently more abortions among the cancer patients
compared to the healthy controls. (See Spring 1997 and Summer 1999 ABCQ Update.)
The only direct. evidence of such bias was reported. in a 1991 Swedish study14—interestingly,
published in Dr. Rosenberg’s American Journal of Epidemiology—which compared
computerized, prospective medical records of abortion with interview-based data on the same
women. Based on the fact that seven Swedish breast cancer patients— compared to only one
healthy woman—had reported an abortion of which the computer had no record, the authors
claimed to have observed. "overreporting". In other words, the computer record was presumed to
be true, and if it showed no abortion, the woman was presumed to have made it up! Not only
was this preposterous notion of "overreporting" roundly criticised by Daling et al. in their 1994
study9, but it was actually retracted by the very authors who proposed it, in a 1998 letter
published in the Journal of Epidemiology and Community Health. Amazingly, Dr. Rosenberg
admitted all this under cross-examination without hesitation:
"A: Yes, I think they did talk about overreporting.
"Q: In other words, that persons who never had had an abortion, according to their study, actually responded or told interviewers that,
yes, they’d had an abortion?
"A: They suggested that that night be going on, yes.
"Q: Since then are you aware that they have acknowledged that actually--in that particular country where that study was
taken—actually, as it turns out those woman had abortions--or as best as they were able to determine--in other countries?
"Q: So they were not,indeed, overreporting?
It is also noteworthy that in her direct testimony about the Lindefors-Harris et al. study14, in the
context of its showing evidence of reporting bias Dr. Rosenberg left out any mention of
"overreporting", saying only: "What she showed was that the controls—there was more
under-reporting compared to that registry for the controls than for the cases."
Wherefore, then, the evidence of reporting bias? Not surprisingly, Dr. Rosenberg relied heavily on
the 1997 Danish study by Melbye1, as the best example of a study which showed no risk increase
but which—since it relied on prospective data-—was necessarily free of reporting bias. Said
Rosenberg in direct testimony:
"The strengths (of the Melbye study) are, first of all, that it relied on abortion records that are legally required in this country, in Denmark.
So that was the first strength :it relied on records, so there was no reporting bias, Reporting bias is absent from this study "
The manifold flaws in the Melbye study are well known to ABCQ Update readers (see sunimary
in Spring 1999 issue), and they will not be summarized here. Two of those flaws are relevant to
the Rosenberg testimony in Florida: 1) the misclassification of tens of thousands of women as not
having had an abortion, even though they did have legal abortions in the records of vital statistics
(but not in the computerized registiy that Melbye et al. used for their study), and 2) the use of
breast cancer incidence (i.e., the outcome variable) statistics going back to 1968, while using
abortion (the exposure variable) statistics starting only in 1973.
With respect to the first of these flaws, Dr. Rosenberg went only so far as to admit: "It might
possibly chage the results, yes."
With respect to the second flaw, Dr. Rosenberg insisted—on three separate occasions during her
testimony—that it would have no effect on the result:
"would the cases (of breast cancer) identified between 1968 and 1973 affect
the relative risk estimate because they had been misclassified? And the answer is, no, they contribute nothing."
Her rationale is essentially that, although every case of breast cancer diagnosed between 1968
and 1973 that occurred in a woman who did have a pre-1973 (i.e., unrecorded) abortion would
be misclassified as not having had an abortion, ". . . so would all the women who didn't
have breast cancer be classified as not having an abortion.. It contributes zero
No, Dr. Rosenberg: not exactly. It doesn’t contribute "zero information"; rather it contributes
(mis)information that the risk increase is zero (because both breast cancer patients and healthy
women from that age group would show up as having the same number of abortions; zero). And
that is entirely different. And that is one main reason why the Melbye study came up with a zero
overall risk increase, and why that result is invalid.
But Dr. Rosenberg seems not to appreciate the difference. In fact, she—like Melbye and
certain other epidemiologists—are convinced that no matter how disparate the patients and the
healthy women to whom they are compared are, a statistical adjustment can always be made to
account for the difference, as long as there is any degree of overlap. They further believe that such
statistical adjustment is quite equivalent to matching the cases and controls for confounding factors
such as age.
Mr. Sukhia pressed Dr. Rosenberg hard on this point, because it involves a very fundamental
understanding of epidemiological principles:
"Q: Are you familiar with Mantel and Haenzel16?
Q: And are they people or is that a publication which you think is authoritative in the field?
Q: Would you then take issue with their statement that a sensible rule is to match for those factors such as age and sex?
A: I think what they meant was to control for them. They were the individuals who developed the statistical method for actually taking
those factors into account.
Q: Okay. And would you take issue with this statement, ‘In devising rules
for selecting controls, those factors known or strongly suspected to be related to disease occurrence should be taken into account. If unbiased
or more precise test of the significance of the factors under investigation are desired, a sensible rule is to match for those factors
such as age and sex’16? ...But you would think it would be not a sensible approach to match for those factors?
A: It is not only not sensible, it is impossible."
Sorry again, Dr. Rosenberg: It is not only not impossible to follow the "sensible rule" of the very
authors you have recognized as authoritative, but it is done in the very best studies. For example,
the 1989 Howe study13 (see above) was a study in which every one of the 1,451 breast cancer
patients was matched to a healthy woman of the same age, and who lived in the same zip code!
Yet even as she defended her own picking and choosing among the rules of epidemiology to fit
her conclusions (This could be called ‘outcome based science’.), she attacked our methods and
conclusions with vigor. Consider this outrageous example from her direct testimony:
"Q: The final error that he made in his (Brind’s) meta-analysis’ is to
equate statistical significance with cause and effect. He basically says since we have found a statistically significant assocaition, this means
that induced abortion is a cause of breast cancer."
Sorry again, Dr. Rosenberg: Nothing could be further from the truth. Here is what we said in the
meta-analysis7 about the statistically significant effect we observed:
"Furthermore, this consistent statistical association is fully compatible with existing
knowledge of human biology, oncology and reproductive endocrinology, and supported
by a coherent (albeit incomplete) body of laboratory data as well as epidemlological data
on other risk factors involving estrogen excess, all of which together point to a plausible
and likely mechanism by which the surging estradiol of the first trimester of any normal
pregnancy, if it is aborted, may add significantly to a woman’s breast cancer risk"
There are yet other examples in Dr. Rosenberg’s testimony—too numerous to mention all of them
here—of state ments contrary to the published record, or even to what she had just said a
moment before. Consider this exchange with her own (i.e., Plaintiff’s) attorney near the end of her
"Q: Specifically, do you know if 25 studies have shown a relative risk greater than 1.0?
A: I don’t know. I haven’t counted them, but that’s not the way one makes
an inference of cause and effect, by counting studies.
Q: And how would one make such an inference?
A: Well, there are a lot of factors. . Results need to be replicated over and over."
As to my own testimony in the case, all that would be new to report here concerns the overall
relevance of the ABC link to a parental notification law. The basic thrust of my testimony was that
the law, by involving parents— many of whom would be college educated and internet-literate—would increase the likelihood that a’ teenager seeking abortion would find out
about the ABC link before making her decision. This would be particularly important in cases
where the girl has a family history of breast cancer, of which, in many cases, she would be
unaware, but her parents would know.
In this context it is also striking that Dr. Rosenberg admitted—under cross-examination—that her
own recent ABC study17 also reported a particularly strong (though not statistically significant)
increased risk of 220% for nulliparous (childless) women with family history who also had had an
Perhaps Judge Lewis will exercise some good judgement and do what Judge Marguerite Simon of
the Superior Court of New Jersey just did on December 13 (in a case in which I filed an ABC
affidavit in support of an amicus brief), and rule that the parental notification law shall go into