Talk given to Endeavour Forum Inc. at Malvern on 24th August 1999.
Thank you very much for that kind introduction. I’ve been in Australia for about a week now on this two-week tour and certainly, when I was invited to come by Babette I felt sure that before very long I would get to see a kangaroo or two. I haven’t seen a kangaroo yet not even a wallaby. However I’m beginning to get the idea–it’s finally dawning on me that as I hop around from city to city with scarcely a moment’s rest—that I am the kangaroo, and that I will get to meet some company of like species before I leave this island. I’d just like to point out that one item on your agenda which I think needs a bit of clarification: number four—that’s me and number five—that’s Dr Robert Burton who will speak after me. Actually well we’re both professors. I am not a medical doctor and he is, just so you put this in a bit of perspective. My training and career have been entirely in research
Now I’m going to adjourn the microphone to the overheads so that you can see my material and hear me at the same time. The first aspect of the abortion breast cancer issue that might grab your attention and that is really newsworthy is that it’s hardly news. This [overhead] is the cover page of a paper from April 1957, which is the first paper that actually showed a significant link between induced abortion and breast cancer. It showed that among over 1,000 women in Japan with breast cancer and almost 2,000 women without breast cancer to whom they were compared the women with breast cancer were found to have three times as many pregnancies that had ended in what they call artificial abortion. It translates—putting it into current epidemiological mathematical terms—to-a relative risk of about a 2.6 or 160% increased risk of breast cancer among women who had had an induced abortion. Also it’s interesting that this study did not find a significant link between breast cancer and spontaneous abortion (commonly known as miscarriage) and that it is also a finding which through all the decades since has held up generally in epidemiologic studies. As you might gather, in the early years there weren’t too many studies done because abortion was not legal in many places in the world.
This [overhead] is a landmark study which is cited all the time as the study which establishes age at first term pregnancy as a risk factor for breast cancer. In other words, the older you are when you have your first term pregnancy the higher your risk for breast cancer. It’s probably a good time to say that when we talk about risks and relative risks, that you don’t sit there and figure out:‘Well if I have my baby at 27, it means I’m more likely to get breast cancer than if I had my baby at 23’ and that sort of thing. It’s important to note that most of the risk factors that we know about for breast cancer are really secondary risk factors. They have to do with levels of estrogen, the female hormone estrogen and compounds that act like estrogen. And all of them tend to contribute a little bit--twenty per cent; thirty per cent; fifty per cent. So that with induced abortion, as with most of these other reproductive risk factors for breast cancer, it can certainly be said that most women who have an abortion do not get breast cancer and most women who have breast cancer have not had an abortion. What is particularly important and even unique about induced abortion as a risk factor for breast cancer is that it is, as a matter of fact, a matter of choice—in practicality if not legally—depending upon the jurisdiction. That is to say that it is an elective surgical procedure and a woman’s exposure to the hormones of early pregnancy—if it is interrupted—is so great, that just one interrupted pregnancy is enough to make a significant difference in her risk. That makes it the most avoidable risk factor for breast cancer. It doesn’t mean it is the major one or the major cause of breast cancer.
1970 Lanmark WHO study
But in any case this study was another one that gave the hint of abortion as what you might call an independent risk factor because it found that not only is a full term pregnancy or a child birth beneficial to a woman in terms of her breast cancer risk but—contrary to what the simplest kind of theory might propose, i.e. a full term pregnancy is good so a partial pregnancy is probably good but not as good in terms of breast cancer risk—but abortion seems to go the other way and increase the risk. And what these researchers found—and this was the World Health Organisation’s seven-centre study multi-centre study for women around the world—they said "differences between cases and controls"—that means women with breast cancer and the healthy women with whom they were compared—"with respect to frequency of abortion were observed in only a few centres"—By the way, that is four out of seven—"and were in the direction which suggested increased risk associated with abortion, contrary to the reduction in risk associated with full term births." So a part- term pregnancy seemed to confer greater risk than no pregnancy at all. This is contrary to the simplest possible model and, it should be noted that, abortion was not looked at in its most precise way. That is to say this study, like many others, did not distinguish between induced abortion and spontaneous abortion which would explain why there was only a hint of a risk and why it was not observed in all their study centres, but it was enough to make note of.
This [overhead] was the first study that hit the papers in the Western world really that looked at induced abortion specifically as well as spontaneous abortion. It was done by Malcolm Pike and his colleagues at the University of Southern California in1981--scarcely a decade after abortion had been legalised in the United States. So they couldn’t work with many patients with breast cancer who may have been young enough of the age that abortions might have taken place before it was legalised. So their patient dates only went up to age 33. And they found that oral contraceptive use before first full term pregnancy—as well as a first trimester abortion—was associated with an elevated risk Both of them were associated with approximately a 2.4 relative risk or 140% increased risk. Now these data—when you actually took them to look at just induced abortion by itself—were not even statistically significant. They only had 163 patients in the study. They had very few with induced abortion or spontaneous abortion.
But Malcolm Pike is a very well known and respected epidemiologist and people took him seriously. A number of people including some who have backed off from the issue since—one of them being Willard Cates writing in Science magazine, [overhead] a very prominent science publication, and writing on the public health record in 1982, a year after Pike’s study and based on Pike’s study alone, essentially because that was really the only one which was well known—he raised on his bottom line "whether abortion causes adverse effects on future desired pregnancies and whether it increases the risk of breast cancer in certain women" (those women being before they had a full-term pregnancy in this particular study). "Accurate information will help policy makers, medical practitioners and those most directly concerned, women of child-bearing age, to make rational decisions about this subject." This sounds like a very reasonable position to take, and, like most other factors which may be appearing in one or two studies—even though it’s preliminary and even if they are not very big—if they are credible researchers and credible journals, especially if it’s an elective exposure, then it’s the kind of thing which might be worth letting the public know about.
Well the other kind of reaction is the same kind of reaction that unfortunately we see whenever the abortion breast cancer connection gets major publicity. This [overhead] is a study in the same British Journal of Cancer 1982 the year after Pike’s study. Sir Richard Dole and Martin Vessey, very prominent epidemiologists, looking at a lot more women from London and Oxford with 1,176 breast cancer patients aged between sixteen and fifty years and saying "a recent publication from California has suggested prolonged oral contraceptive use and abortion before first full term pregnancy increased the risk of breast cancer in young women." But their (Vessey et al.) data on more women of a much bigger age span and all the way up to fifty years "are entirely reassuring, being in fact more compatible with protective effects than the reverse." Well, as any epidemiology study should be, it is loaded with numbers and tables and it’s all about statistics and precise numbers. And the first piece of information that you look for in a study like this—if it is showing information about induced abortion and anything—is: How many patients in the study with breast cancer, and how many women without breast cancer actually had an induced abortion? The precise number is: "only a handful". That is how many women in this study had an induced abortion. That is the most precise number you get. So you see, this study is not "entirely reassuring" with regard to the question of induced abortion and breast cancer; the study is entirely irrelevant but they put it up as a reassuring study to reassure the world that this finding that abortion may be dangerous really isn’t true.
Experimental Animal Data
Well, backing away from, or just stepping away from epidemiological studies for a while just to explain a little bit about what these studies mean, most of them are set up in such a way as to emulate the controlled experiment. You might know from scientific studies of your own or courses that you might have taken that the heart and soul of scientific research is the controlled experiment. So, for example, you take a cage of rats or several cages of rats if you are looking at different things, and control every aspect of their lives while you are doing the experiment and you change just one condition at a time. [overhead] So each cage here has only one thing different from the next and if you control the experiment that way then you can be reasonably assured that if there is a difference in the outcome at the end of the line, that that outcome is due to the change you made to the exposure of these different animals. So here we have nine rats in the cage—Sprague-Dawley rats, which can be reliably initiated to get breast tumours with the chemical carcinogen DMBA.
And you notice also in a controlled experiment one of the things that is nice about it is you can use very small numbers. You can get reliable results with only a handful of rats or mice. You don’t need hundreds or thousands of them and this particular study these groups are "pregnancy plus lactation", "pregnancy alone and not allowed to lactate", "interrupted pregnancy" and these are "virgin rats". So these are cages of rats that can be initiated to get breast cancer and they were either not allowed to get pregnant at all when they were sexually mature, or allowed to get pregnant, have their pups and breast feed them, or just have their pups and not breast feed them, or have the pregnancy surgically interrupted or aborted. And you can see now, down the line just like in the human study that I showed you from the World Health Organization: Instead of an interrupted pregnancy providing some kind of protection from breast cancer that was somewhere intermediate between a full pregnancy where you got zero out of nine or one out of nine breast cancers per cage and no pregnancy at all where you got between sixty seven and seventy one per cent incidence of breast cancer, instead it was the worst of the lot. It seemed to add, if anything, more risk and go up close to eighty per cent incidence of breast cancer. So there you have some experimental verification. And incidentally, when you are establishing a risk factor for anything you can’t just depend upon one epidemiologic study or even twenty epidemiologic studies. That is, if all you find is a statistical connection, that is really not enough. You need to find some underlying biology that makes sense. You need—or at least it’s desirable if it can be reproduced experimentally in the laboratory—to do so.
Cell Proliferation and Differentiation
Anyway, another advantage of the controlled experiment is that you can also remove the animals variously during the study and take actual tissue samples and cut them up and look at them under the microscope and see what changes are taking place in the tissue. This [overhead] is a composite drawing taken from that very experiment that I just showed you. The experiment was done by Jose and Irma Russo—a husband and wife team—then in Detroit Michigan now in Philadelphia—and I say a composite drawing because this, on the right, is what the tissue tends to look like with sexual maturity, but in the absence of a full term pregnancy. And here is a mature breast that is now ready to lactate at the end of a full term pregnancy. And you can see these rather primitive terminal end buds connected by rudimentary ducts: Not a very dense situation. And here you have mature, fully differentiated lobules of alveoli which secrete and eject milk. And a general feature of cells which are terminally differentiated is that their capacity for growth is turned off. This [overhead] fits into a little cartoon I drew here which kind of gives a global scheme of things. As an example you have the development of an entire body, but it also has to do with the development of the breasts during a pregnancy and many other developmental processes. You have a complementary interplay between these two global processes of proliferation and differentiation, proliferation being cellular multiplication and differentiation being modelling those cells to perform a specific function. And differentiation is characterised by a switching off of the capacity of the cells to proliferate. Now you are probably all aware that cancer in general is a disease or a series of diseases in which proliferation has gained the upper hand and gotten out of control. In fact, in any kind of pathological examination of cancer tissue from a patient, the more the cells show the capacity to proliferate and the less they show signs of differentiation, the more malignant they are described as being. But normally of course there is a proper interplay, and in the beginning of the process proliferation predominates, and towards the end of the process differentiation predominates. Now that switch in the breast tissue has recently been pinned down but it will need more research to be very precise about it and sure about it, but it looks like about the last eight weeks of pregnancy is the time in a human pregnancy when the differentiation takes place, so that at the end of a full term pregnancy a woman has fewer of those cells which are capable of proliferation and ultimately potentially becoming cancerous than she had before the pregnancy began. So full term pregnancy is protective; it lowers breast cancer risk. But if that pregnancy is cut off artificially somewhere in the middle after some weeks or months, she has far more cells in her breasts that are capable of proliferation, and that have proliferated, than she did at the beginning of the pregnancy, which translates statistically into a higher chance of getting breast cancer later in life. And the mechanism by which this is likely to occur is almost certainly the same kind of hormonal stimulus that is responsible for the action of most breast cancer risk factors. They are attributed to overexposure to some form of the female hormone, estrogen. In this case it would be estradiol or in the Commonwealth we say oestradiol and spell it with an "o". It’s the most common estrogen—the main one secreted by a woman’s ovaries. It is a natural compound. It is not a primary carcinogen. It doesn’t make cells in the breast or anywhere else become abnormal But it stimulates the cells of the breast to grow, and it is just as good at stimulating abnormal and pre-cancerous or even cancerous cells to grow. And most risk factors for breast cancer are attributable to some form of over exposure to some form of estrogen. So for example it is known and universally acknowledged that women who go into menarche; get their periods earlier in life and therefore have more menstrual cycles, or who go into menapause later in life and therefore have more menstrual cycles, or women who chronically use alcohol which inhibits the metabolism of estrogens, end up with chronically slightly higher levels of estrogen in their body, and it translates as higher risk of breast cancer.
Estradiol Levels
Now this [overhead] is a very good illustration from a paper published in 1976 by a pair of Swiss obstetricians who had a clinic for what is called threatened abortion: women who were pregnant but they had vaginal bleeding; a sign of a threatened spontaneous abortion. This has been found in other studies but this is the best graphic presentation. These women walked into the clinic anywhere from six weeks to nineteen weeks gestation. Whenever they happened to walk in they had a blood sample taken and they measured a bunch of hormones including estradiol. The open circles represent those pregnancies which were viable and went to term. The closed circles represent pregnancies which ended in spontaneous abortion. So you can see how the estradiol goes up steadily and very steeply as the pregnancy progresses, certainly after 20 weeks and beyond well into the second trimester. Whereas pregnancies that end in spontaneous abortion seem to be characterised by very low levels of estrogen, which would explain why spontaneous abortions are not usually associated with increased risk of breast cancer.
Now you might say in that graph there is an alternative explanation: Maybe in those women who walked in with threatened abortion and had a blood sample taken and their estrogen was very low, maybe the baby had already died and that is why the estrogen is so low. So to pin that down there is some good evidence coming in from one of the rare instances where you can do essentially a controlled experiment on human beings. Here [overhead] is an artificial insemination clinic data set from California in 1993. And of course in artificial insemination it’s necessary to measure hormones on a daily basis to see where women are in their cycles so you know when to attempt the artificial insemination procedure. So here we have the open circles representing the menstrual cycles of about a dozen women, but these (filled circles) would be the non-conceptive cycles; where conception did not take place. You see the typical pre-ovulatory peak of estradiol and a small secondary rise in the luteal phase towards menstruation. But in a viable pregnancy—of course the pre-ovulatory peak being the same—you can see almost immediately—and in a dozen women it was statistically significant (that is what the stars mean), five days after conception—a clear difference in the amount of estradiol between the non-pregnant and the pregnant state. And in the case of pregnancies which began, but ended in spontaneous abortion in the first trimester, [overlay next overhead] you can see in this artificial insemination data (the squares representing pregnancies like that), the rise is lower and slower and by the time of the missed period it is almost flat. So that estradiol doesn’t even seem to get as high as in the pre-ovulatory peak value. In fact if you look at some summary data from yet another source, (and none of these sources disagree; it’s standard textbook material), by seven or eight weeks gestation you can see, in a normal pregnancy, estradiol is more than double where it was in the pre-ovulatory peak. Whereas (and this is the level of conception where it comes down from the peak) whereas in a spontaneous abortion, or in most first trimester spontaneous abortions, the level doesn’t even get as high as the pre-ovulatory peak. So most spontaneous abortions don’t have normally very high levels of estradiol. Now the reason for this is also fairly straightforward, in that the reason why—or at least the proximal reason why—spontaneous abortions occur is because there is not enough of the hormone progesterone to maintain the pregnancy. Well, estradiol is made from progesterone and they rise and fall in parallel in early pregnancy, and so when the progesterone is low, so is the estrogen. The estrogen is not necessary for maintaining the pregnancy. Its job is to prepare the breast to secrete enough milk to feed the baby.
1994 Janet Daling Study
The next time the issue of abortion and breast cancer hit the news [overhead]—and that is a long time between 1981 and 1994, although a number of other studies had been done rather quietly around the world—but-it hit the news again in 1994 with the work of Janet Daling of the Fred Hutchinson Cancer Research Institute in Seattle Washington, a very prominent cancer research institute. Her case-control study was of approximately 1800 women: comparing 900 women with breast cancer with a like number of control women drawn from the same population. You see when emulating a controlled experiment you can’t control all the factors, but you try to match your control population that doesn’t have the disease as closely as possible to your population that does have the disease. And then, using highly trained nurse interviewers, you go out and interview that population and find out things about their reproductive history, their family history and every other variable you think might have some effect on breast cancer risk. I might point out that a lot of criticisms have been raised about a potential weakness of such studies—that you might get a difference in the accuracy of response between the patient population and the control population, which would introduce something called the response bias and give you what appears to be an answer or an effect of the variable you are testing that is really not due to the variable you are testing. However, there is a very precise, exact science that has been developed about doing these interviews. They use very highly trained nurse interviewers. They interview the study subjects blind. That is to say that when the interviewer goes out to interview study subject number 197, she doesn’t know if subject number 197 is a breast cancer patient or a control subject. You can readily imagine that if the interviewer knew the difference there would be some subtle differences in the way questions might be asked which would show up as a biased finding on the final result. So there are a lot of tremendously good data that are drawn from so called retrospective interview-based case-control studies. It is really the bread and butter of epidemiology, as opposed to what is called the cohort study—where you start out recording who had an abortion and exactly when, and everything else about the medical history when it happened, and then you look down the line to see who got breast cancer or whatever other outcome you are looking at. Cohort studies may take many years or decades to complete whereas within the space of several months or years you can complete a case control study. So it’s certainly very valuable and shows up a lot of things. And by the way, this study showed up about a fifty per cent increased risk of breast cancer in women who had reported a history of induced abortion.
What I’ll also tell you is that there was another finding in the study that was kind of buried—that is to say it didn’t make the news; wasn’t quoted. And that is an apparent synergy between induced abortion and a family history of breast cancer. So for example—in this passage that I’ve highlighted—here, in women with no family history the overall size of the increased risk associated with induced abortion was 1.4; women with no family history had a forty per cent increased risk of breast cancer. But women with a positive family history: sister, mother, aunt or grandmother with breast cancer, the overall risk was 1.8. That is, women with family history and abortion, compared with women with family history and no abortion. And instead of there also being a forty per cent increase, it was eighty per cent so it seemed to be a synergistic—a greater risk increase than adding the two together alone; the whole greater than the sum of its parts. But it was particularly strong for a first abortion that occurred prior to age 18 years. Twelve case patients and zero control patients; relative risk: infinity. In other words, in their whole population of 900 patients with breast cancer by the age of 45, and 900 patients who didn’t get breast cancer of the same age group, they found twelve study subjects who had had an abortion before age 18 and also had family history, and none of them turned up in the non-cancer group. They all got breast cancer by age 45. Now no other study is going to find that sort of absolute association. However other studies also report an apparent synergy, a much greater increased risk in women who have a positive family history of breast cancer. So for example a study in France published in the same year in 1994 showed that women in France who reported two or more induced abortions had about a 600 per cent breast cancer risk increase.
But as I said, that didn’t make the papers, and the reason is pretty obvious. When an epidemiological study comes out in a medical journal, it may come out with an editorial. There are usually one or two editorials about what the editor thinks are the most important papers in that issue of the journal. And usually—or almost all the time—the editorial will tell you why the paper is important, and why you should take it seriously. But in this particular case, mirroring the kind of reaction that Pike’s study got in 1981, the editorial was there to tell you "However the overall result as well as the particulars are far from conclusive and it is difficult to see how they will be informative to the public." Right! An elective procedure that is exceedingly common seems to show an increased risk of getting a life-threatening disease that is also exceedingly common and the editorial writer can’t figure out how that might be informative to the public! That statement strikes me as disingenuous to say the least. Well, what was the main criticism of the methodology and of the results of the study by the editorialist, Lynn Rosenberg? "A major concern", she said, "especially because the observed effect was small"—and that is true: a fifty per cent increased risk overall is epidemiologically relatively small, kind of near to the borderline of what can be accurately measured—"is the possibility of reporting bias."
Reporting Bias
Reporting bias as I told you before was a difference in the accuracy of responses. In this particular case, by the way, you would guess from an editorial like that, the paper at hand—the Daling study—did not deal with the question of bias or certainly didn’t deal with it adequately. The bias in particular that she is talking about is based on a hypothesis that was put forth by a team of Swedish researchers headed by Olaf Meirik of the World Health Organisation in Geneva. They had done a couple of studies on reproductive risk factors in breast cancer. One of them was of the same kind of retrospective interview-based case-control type, and another one was based upon prospective records; a computerized registry of abortions from Sweden where the records were generated at the time of the abortion. And, since everybody in Sweden has a number, and has these records, they were actually able to obtain the computerized records from everybody in the case control study, and therefore, compare how accurate were the responses of these women. Well they claim to have found a statistically significant difference—that is what this P007 means—between "underreporting of previously induced abortions among controls..." (That is what they have hypothesized; that healthy women would be more likely to lie about their abortions, but that breast cancer patients, considering their life-threatening condition, would be more likely to be honest about reporting their abortions.) They found a difference between "underreporting of previous induced abortions among controls relative to over reporting among cases." In other words, that women with breast cancer would make up abortions that didn’t happen. That is really the sole basis of their statistically significant finding: that seven breast cancer, seven Swedish breast cancer patients in their study and one healthy Swedish woman reported having had an abortion of which the computer had no record. Sorry lady: the computer says you didn’t have one; you didn’t have one. In other words, their statistically significant conclusion that response bias applied was absolutely dependent upon the assumption that the computer record was right, that women did not have an abortion unless the computer said they did; even if the woman said she had one but the computer said she didn’t, she didn’t. Well over reporting, I think, is a pretty preposterous assumption, and I’ve used that word in describing it in publications. Janet Daling, in her study, was much more diplomatic. However she wrote "We believe it is reasonable to assume that virtually no women who truly did not have an abortion would claim to have had one". I think that is reasonable too. In fact, that evidence, that these Swedish women who claimed to have had an abortion that the computer had no record of represents over reporting, has since been retracted in March of 1998, and that group—that Olaf Meirik, group in their subsequent research, do not mention reporting bias as an explanation for the connection of abortion and breast cancer any more.
Well another thing that is interesting about the bias connection is that subsequent to the Daling Study—about five months later [overhead]—a study came out by Lipworth and colleagues, which showed actually the overall identical result of a fifty one per cent increase in breast cancer risk in Greek women. But they treated response bias by looking, by doing a literature review in Greece, noting "even before legalization, induced abortions were practiced in Greece with widespread social acceptance.
This can be interpreted as indicating that healthy women then in Greece report reliably their history of induced abortion." So they claim their finding was not attributable to response bias. Interesting that this study was submitted for publication on October 20th 1994, exactly one week before October 27th 1994, when Harvard epidemiologist Karen Michaels told Dr Lawrence Altman, epidemiologist reporter of the New York Times, that "that is a flaw in the design because women who have breast cancer are more likely to disclose an abortion than women who did not develop breast cancer". You see it’s a fact.
1996 Brind Meta-analysis
Everybody knows it. Who should that Karen Michaels be but [overlay overheads] the same one who is on the by-line of the study in Greece: Karen B Michaels! What a difference a week makes in one’s interpretation of whether it’s response bias or not.
We finally came out with our study: [overhead] "Induced Abortion as an Independent Factor for Breast Cancer – A Comprehensive Review and Meta-analysis" in the Journal of Epidemiology and Community Health in October of 1996. That is a British Medical Association publication. It was no accident that I submitted the paper to an English journal—and this journal in particular—because I felt strongly that we would get fair treatment. I didn’t want, for example, to publish the paper in the Journal of the National Cancer Institute and be sabotaged by an editorial like Daling’s study was. PS: two months later the Journal of the National Cancer Institute published an editorial directly attacking our research anyway, but at least we had a little bit of lead time. Well this study was rather wordy: we analyzed every study that had been done and published and also stacked them all up in a meta-analysis. That is, we lined up all the studies and ended up—through a statistical compilation method called the weighted average—using a couple of different models, and the most conservative estimate gave us a thirty per cent increased risk on average overall.
This [overhead] is an updated meta-analysis. We had at the time 23 studies. Now there are 31, and 25 out of the 31 show data with a point estimate to the right side of this line of unity, i.e., increased risk, with 18 out of the 25 statistically significant on their own. That is where this whole confidence interval is to the right side of the line—doesn’t cross the line—and studies on the left side would be negative, or studies showing a negative association, or that abortion would be a protective effect. This is what you find with just about any, even well-acknowledged risk factors. You always find a couple of studies that go the other way but the overwhelming predominance of these studies is to the right side of the line. Now the study designs are very different. In a lot of cases the point estimates are very different. These studies may be described as being rather heterogeneous which makes it a little bit unreliable to say 30 per cent. Maybe it’s fifty per cent, eighty per cent? It’s best to say that there is a range of increased risk and the only thing you can say really safely though is that there is certainly going to be an overall positive association when you have such an overwhelming predominance of the data looking that way. Incidentally not all of these studies are of the retrospective case-control type. For example this study here in 1989 by Howe and colleagues is based entirely on prospective data—death certificates filed at the time of abortion—and they found a ninety per cent risk increase. So you find positive associations whether it is a case control study or other kind of study. Prospective, retrospective; it seems to be coming up as a risk factor in the overwhelming majority of studies.
Controversy
Well not surprisingly we were greeted somewhat controversially. [overhead] That was the headline in the Wall Street Journal. Incidentally, Janet Daling, who had published that study in 1994 (and by the way, in terms of the whole abortion debate, most of this work has been done by pro-choice researchers); Janet Daling, describing herself as very strongly pro-choice, said our paper was "very objective and statistically beyond reproach." Incidentally, though just to give you a flavour of international views of how the media treat things, the British Medical Journal thought that even that article was very biased even though it was one of the fairest articles describing our work. [overhead] They did something very scientific in analyzing the press coverage on that particular article and concluded that "all in all more column inches were devoted to the paper’s critics than to the research itself." Right! they just added up how much of the article criticized the research and how much of the article was interested in describing the research. So they concluded it was biased.
"File Drawer" Effect
Meanwhile, in the Journal of Epidemiology and Community Health, our article did not come out with an editorial in the same issue. But the kind of treatment we received when it was published so troubled the editor-in-chief, Dr Stuart Donnan from the University of Manchester, that he decided to write an editorial in the next issue which came out in December. [overhead] In pertinent part he says "In the light of recent unease about appropriate but open communications of risks associated with oral contraceptive pills it will surely be agreed that open discussion of risks is vital and must include the people—in this case the women—concerned." He goes on: "I believe that if you take a view—as I do—which is often called pro-choice, you need at the same time to have view which might be called ‘pro-information’, without excessive paternalistic censorship (or interpretation) of the data". So that was his take on the reaction to the study.
Meanwhile, it is not to say that there isn’t valid criticism of a meta-analysis. One I just mentioned to you: If the studies are heterogeneous, it’s hard to rely on a particular number like thirty per cent and say that is the real average. But another criticism which is probably the most valid in general of meta-analyses of this type is something called the file drawer effect. That is to say we based our meta-analysis only on data which had been published. Those are really the only data which are available to us. And we thought it was a fair thing to do. However the argument would go something like this:
"Maybe you found 23 studies and 18 out of the 23 show increased risk of breast cancer and you got an overall significant association. But suppose there were really 123 studies and 100 of them didn’t show anything at all and we all know—and this is a well known fact about research in general—studies which are negative, that is, they don’t show anything significant, are likely not to be published at all. We argued that because of the contentious nature of abortion and the obvious reluctance of even researchers who documented the link in their own papers to report on it, that there was probably a reverse effect going on, that there was a reverse file-drawer effect; that the data which did show a connection, instead of data which didn’t show a connection, were probably suppressed. We didn’t have any direct evidence of it but direct evidence did appear about a month after we submitted our meta-analysis for publication.
Rohan Study Adelaide
These are data from [overhead—with two rightmost bars obscured] the only study I’m aware of in Australia on reproductive risk factors in breast cancer. The study was conducted in the early 1980s on women in Adelaide and it was mainly focused on dietary risk factors. In fact they found a slightly protective effect of beta carotene in the diet, but they also had to look at all the other factors which are known to effect breast cancer risk to see how they affected dietary factors and also so show that the population was typical. So yes indeed, benign breast disease somewhat increased risk and obesity. Fat cells make estrogens so obesity is associated with an overall slight increase in breast cancer risk. Older age at menarche decreased risk (fewer menstrual cycles).
Older age at first birth increased risk, no births at all increased risk, older age at menopause (more menstrual cycles, more estrogen exposure) increased risk, both ovaries removed ("surgical menopause"—many fewer menstrual cycles, much less estrogen exposure), decreased risk, family history somewhat increased risk. But nothing about abortion. In the methods section it seemed that they collected the data about abortion but they reported nothing when the study was published in the American Journal of Epidemiology in 1988. Every other variable, but not abortion. It was 1995—right after we submitted our meta-analysis for publication—when a small meta-analysis came out in France by Nadine Andrieu and colleagues, where she put together data from other studies, from six studies looking at the synergistic effect—which they found some evidence of—between family history and induced abortion. And they used data that wasn’t all published. Some of the data had not been published and just became published by virtue of being in this meta-analysis, including all the data in the study by Rohan et al on the women of Adelaide. and for the first time [unveil two rightmost bars on overhead] the data on spontaneous abortion—nothing significant—and induced abortion saw the light of day: an overall 160% increased risk of breast cancer among the women of Adelaide Australia.
It was the strongest risk factor they found. It was the only one that was clearly statistically significant. And this you don’t do. This is not what you see in scientific research, ever. I’ve never seen it before, where the most significant finding in a study is specifically left out of a research paper. So this was direct evidence of this, what we would call the reverse file drawer effect, where real evidence or positive evidence of a connection between abortion and breast cancer ended up stuck in the file drawer. And we hypothesize that there is more of it. This is one case where it came out of the file drawer—quietly, seven years later.
Melbye Study
Well three months after the publication of our meta-analysis came a real salvo, [overhead] a real attempt to shoot it down and to convince the world that "induced abortions have no overall effect on the risk of breast cancer". Period. Also a disturbing participation of the US National Cancer Institute in a lot of this covering up of what is going on, including the editorial that came out with the Daling study the editorial that attacked us and this editorial in the New England Journal of Medicine in January 1997 [overhead], which championed the findings of this study by Melbye and colleagues from Denmark. Patricia Hartge of the US NCI saying "In short a woman need not worry about breast cancer when facing the difficult decision of whether to terminate a pregnancy." A very odd position for the National Cancer Institute of a country to take, of a country in which twelve studies had looked at the issue and eleven out of twelve of them had found increased risk in women who had had induced abortion, increased risk of breast cancer. Eight out of eleven of them were statistically significant most of them funded by or even done by the very National Cancer Institute. Including that Howe Study here which found an effect—based entirely on prospective data—that could not possibly be subject to any response bias. You see every odds ratio here is above one except the one that is just one exactly. So the overwhelming—and even more overwhelming than the world wide studies—were the American studies that showed the link. So why would the National Cancer Institute and others including the Department of Defense from the US—which funded the study from Denmark—why would they say that this study was definitive? Well this was a very big study. This is a million and a half women. This is every woman born in the state of Denmark between 1935 and 1978. This is over 400,000 abortions. This is over 10,000 cases of breast cancer. If any study is going to be definitive this would be it, one would think. A couple of things however are a bit odd about their methodology. [overhead] First of all, they started logging abortions in 1973, saying that "the legal right to an induced abortion through 12 weeks gestation was established for women with residence in Denmark". So it’s just like the US Roe v. Wade decision. One would think it was legalized in 1973, and therefore any abortions before that year were very few and illegal and probably wouldn’t matter. Well something that you can see right in the paper that is odd is starting to log abortions in 1973.
Melbye Misclassification
Why on earth are they starting to log breast cancer cases from April 1st 1968? That means that there were five and a half years that they were just logging cases of breast cancer—and we calculated there were over 300 cases of breast cancer—that occurred in women who were guaranteed not to have had an induced abortion, because they didn’t even start collecting records before 1973. So you see we have the statistical cart before the statistical horse. Entirely incorrect, totally invalid methodology. Then you go back and you check a little more with other sources and you look at this 1973 legalization of abortion and you find in 1973 the abortion laws were liberalized for the third time, liberalized before that in 1970 and before that in 1956. But abortion was legalized for reasons other than medical necessity not in 1973, but in 1939 and [overhead] you can get out your handy dandy Befolkningens bevægelser—right?--which means vital statistics in Danish. Doesn’t everyone have one of those? And, well I have one of those. I took some pains to get one. Fortunately it has English subtitles, and you have lots of tables in it including this one on induced abortions—legal induced abortions. And you see 1940, 1941 all the way up through the current statistics, at the time when I got this in 1994. (This is the 1996 or 1997 version). And you can add these up and calculate how many women in Melbye’s cohort born since 1935, and you can come out with 80,000 abortions, representing 60,000 women who had abortions and who are in that study as not having had an induced abortion. They did have a legal induced abortion on the record but not the records that were used in that study. Well you’d think that might wipe out the increased risk of breast cancer with induced abortion but they still managed to find a relative risk of 1.44—a 44 per cent increased risk of breast cancer with induced abortion, and they did a bunch of statistical adjustments including something called the cohort adjustment.
This [overhead] is the cohort effect on breast cancer in Denmark. And what that means is that a birth cohort of women born in about the same year are compared for breast cancer incidence. This is just overall breast cancer incidence by birth year. So the benchmark year is women born in 1868 compared to those, women born in 1918 have approximately double in terms of the incidence of breast cancer; women born in 1948 have approximately triple the average incidence of breast cancer, and thankfully it seems to be actually going down in Denmark. So this is something interesting to look at from the theoretical point of view. They adjust it for this effect. In other words the theory goes: you can’t just compare women who were born say in 1958 with women who were born in 1935 (with earlier ones in the study). You have to adjust for this cohort effect and most of the breast cancer cases in their study were women who would have been born in this area here—the oldest women in the cohort, born in the late 1930s or early 1940s. So you adjust for this cohort effect.
Cohort Effect
You sort of flatten out the curve statistically so that you can compare these women directly. Well the thing is, abortion is going up through most of the 20th century through these years and therefore if your hypothesis is correct, that induced abortion does increase the risk of breast cancer, then it necessarily is part of this rising pattern. If you correct for it and flatten it out you are guaranteed to get a null result. And to see exactly how they compared, I did something very simple and plotted the exposure to induced abortion on the same scale. [overhead] Fortunately the abortion incidence in Denmark is a very mathematically regular thing. That is to say the age distribution of women who get abortions—most of them are between the ages of 20 and 35, and the average age is right in the middle at 27. So you just take a 15-year running average and it’s abortion year minus 15 and you get birth year so you can plot them on the same curve, and you find that they rise and fall together. And in fact one of the things that we’re going to do in the next year or so, in re-analyzing more carefully this whole Danish cohort study, is to take a look and see whether we can in fact predict the frequency of breast cancer in women of varying ages in Denmark in years to come, based upon their exposure to induced abortion alone. In other words it looks like in Denmark—as one earlier study by Ewertz and Duffy in 1988—suggested. Induced abortion in Denmark is a particularly strong risk factor for breast cancer somewhere between double and triple the risk. It may be one of the major risk factors for breast cancer in Denmark. As I said with world wide data, when you add it all together it isn’t that strong a risk factor. But for certain people —maybe for Denmark in general, and certainly for people who have family history—it seems to be a more important risk factor. Meanwhile National Cancer Institute [overhead] on its web page (which is the same as its fact sheet) continues to say things like: "Although it has been the subject of extensive research there is no convincing evidence of a direct relationship between breast cancer and either induced or spontaneous abortion". Notice the high bar for evidence "convincing evidence of a direct relationship", and then it is between induced or spontaneous abortion kind of mixing the two together. But certainly the last line is an outright lie. "The scientific rationale for an association between abortion and breast cancer is based on limited experimental data in rats and is not consistent with human data." It is consistent with human data and it is not just based upon limited data and rats. There is all the other biological evidence of what happens during pregnancy what it is that makes breast cancer cells grow and what is the difference between a spontaneous and induced abortion. In other words the whole biological story is consistent. [overhead] We suggested—or I suggested in a long letter to the Wall Street Journal in 1997 that "the NCI (that is, the US National Cancer Institute and its journal would do better to protect American women" and by extension, women in the rest of the world as well) "by warning them about abortion; what most evidence indicates is the single most avoidable risk factor for breast cancer, rather than protecting the abortion industry by invoking flawed analyses from Sweden, the Netherlands and Denmark." And I also suggested "The commerce department could also help, by banning the importation of red herring from the North Sea." I think I’ll leave it there. Thanks for your long attention.