Well well well, Agronomist, you raise very interesting points about follow-up time and my "having it both ways". Indeed. In the standard case-control setup, one identifies a group of cancer patients and compares them to healthy women of the same age. The follow-up time issue is taken care of by making sure that abortion was available for long enough for women to develop breast cancer, testing the hypothesis that the latter was caused by the former. That's why the first US study--Pike et al in 1981 (OR = 2.4)--was restricted to women under age 33, abortion having only been legal about a decade. (Also why the Howe et al study of 1989, based on prospective data from fetal death certificates, was restricted to women under age 40. BTW, most of these prospective studies--echoed by some commenters here--deny the existence of the Howe study, since it found a signicant OR of 1.9; an inconvenient truth.) But in Brauner (2013) healthy older women (age >50) only were recruited for the prospective study and followed for an average of 12 years. Essentially, it was like following up a control group for a standard case-control study, rather than being a standard case-control study.

In contrast, the slew of prospective studies between 1997 and 2008 (Danish study of Melbye 1997, Harvard Nurses study, CA teachers study, EPIC study) all share the flaw (among many) of having as little as ZERO follow-up time. So in Melbye, for example, a 15-year-old who had an abortion on 31 Dec. 1992 (when the ascertainment period for the study ended), was in the study as having an abortion, but no breast cancer! In fact, in Melbye, fully 358,000 of the 1.5 million women (25%) were under age 25 on that date. They represented over 100,000 abortions, but only 8 cases of breast cancer!

So it is the studies that deny the link that want to have it both ways: Studies that have inadequate follow-up time and those that allow too much time to elapse (thus eliminating those afflicted with breast cancer in the interim), thus making the link conveniently disappear.

Joel Brind, you are wrong about the Howe et al. study. It was not a prospective study, but a case-control study. It did not use interviews and so doesn’t suffer the problem of recall bias that most case-control studies have, but it likely had some issues around data matching.

Leaving that aside, I see you are going for a version of the Golden Mean. An effect will only be observed if the follow up period is exactly right. Too short and no effects; start too late and no effects. When you have to create very specific circumstances like this in order to find an effect, it is likely that the effect has limited or no applicability

Imagine yourself a pregnant woman now? In the light of this debate how safe would you feel about an abortion now. The link is not definite but it is a strong possibility. Wouldn't you think that the risk is there. Wouldn't you think that it's too big a risk. The studies that go against the link have serious flaws. Of course you could take a risk on it but the consequences could be fatal.

Gadfly42,

If I was a pregnant woman and the scans showed abnormality I would immediately progress with termination.

Just to give one example to show that people are quite capable of managing risk themselves.

I would definitely put any political party at the bottom of a long preference list if they even looked like changing that and trying to make up my mind for me.

Agronomist, I like your rational thinking.
Brind you have been outplayed here, and your obvious bias is very blatant.
Why not use your scientific talent to find out the real reasons for breast cancer occurring, reasons that aren't attached to your pro-life stance?

Gadfly, you have absolutely no idea of what goes on in a woman's mind when she is considering having an abortion.
She would already know that the possible consequences of having an abortion include puncture of the uterine wall, hysterectomy, or infertility, and more.
If they are still needing to have the abortion, then do you honestly think a tiny chance of this breast cancer link will make a blind bit of difference?

The choice will always be the woman's right in this country, so you need to move on...

OK, Agronomist, you are correct in that Howe et al was not a typical prospective study, like a cohort study in which the following of subjects over time is built into the original study. Rather, it may be classified as an historical prospective study, in which the data have all been entered prior to any disease determination; hence, there is not even the possibility of recall bias. (BTW, the recall bias hypothesis has actually been disproved many times in ABC link research. The original "evidence", presented in a 1991 study by Lindefors-Harris et al, was that at interview, fewer controls admitted to abortions that were on the computerized medical record than did breast cancer patients. But the underreporting by controls only showed up relative to "overreporting" of abortions among the patients, i.e., that the abortions they reported to the interviewer had not taken place, because the computer had no record of them! Of course, this is nonsensical on its face, and the original purveyors of the "overreporting " subsequently retracted the idea in published correspondence in the Journal of Epidemiology and Community Health in 1998. There has never been any credible evidence of response bias in ABC link research, yet it still gets trotted out like fact, even by some commenters on this forum. Meanwhile, even the idea of response bias fades from credibility in the face of the magnitude of the ORs reported recently from Asia, as response bias only becomes a possibility when the association is weak.)

But I digress: The Howe 1989 study (a case-control study nested in a prospective data base) is really the gold standard type of epidemiological study, because it avoids the weaknesses of cohort studies (like lack of adequate follow-up time), while having all the strengths of a classical case-control study, where one may pair-match cases for different potential confounders: much better than statistical adjustment which has inherently much more uncertainty.