Monday, March 29, 2004

Abortion And Breast Cancer

The current issue of The Lancet - the premier British medical journal - includes a re-analysis of 53 studies of the putative link between abortion and breast cancer. (Link requires free registration.) By combining these studies, they wound up with a total sample of 83,000 women. That makes their conclusion that abortion does not increase the risk of breast cancer pretty damned definitive.

You're going to be hearing a lot about this study, and some of what you hear is going to attack its methodological quality. Here's the part the talking points are going to be based on: the researchers separated studies which collected retrospective information about abortion from studies which collected prospective information. "Prospective" means that they followed women who had abortions and then counted how many of them developed breast cancer. "Retrospective" means that they took women who developed breast cancer and asked them whether or not they had ever had an abortion.

The evidence for the supposed link between abortion and breast cancer all came from retrospective studies. Prospective studies tend not to show a link. What's the difference? Reporting biases. If you ask women whether they've ever had an abortion, a certain percentage of those who did are going to forget (yes, it's possible), and a certain percentage of them are going to lie. There's good reason to believe that women who have breast cancer are less likely to forget or lie than women who don't have cancer. It all comes down to the pervasive human tendency to search for meaning. When something bad, like cancer, happens, we tend to scrutinize every inch of our medical history and environment looking for possible answers. That goes double if there's a potential causal factor that you feel guilty about - like having an abortion. So cancer patients are more likely to remember their abortions, if they've had one, and are more likely to "confess."

The archetypal research showing the dangers of retrospective reporting comes from the field of ADHD, back when it was still called "minimal brain dysfunction." (No, I'm not that old, but my Descriptive Psychopathology professor was.) A whole raft of studies showed that mothers of children with ADHD symptoms reported more adverse events during pregnancy, delivery, and early infancy: falls, bleeding, abnormal test results, delivery complications, high fevers in infancy. So researchers concluded that ADHD symptoms were caused by minor brain damage early in development. Then someone went back and studied actual medical charts for families with and without ADHD kids. There was no difference whatsoever in the number or kind of adverse events experienced in pregnancy, delivery, or infancy. What was going on in the earlier studies? It turns out that mothers of children with behavior problems rack their brains looking for something, anything in their history that might explain what they're going through. Mothers of healthy children don't do that.

Prospective data is much, much cleaner. You document abortions when they happen, and you document breast cancer when it happens. You don't have to rely on anyone's memory. You don't have to rely on anyone's honesty. So the authors of the Lancet study had excellent reasons to base their conclusions on the prospective studies, and their statement that abortion does not increase breast cancer risk is perfectly justified.

(Incidentally, you wouldn't know that from the Washington Post's headline, "Abortion's Link to Breast Cancer Discounted." "Discredited" or "disproved" would have been the appropriate verb choices there - "discounted" doesn't automatically convey the information that the "link" under discussion doesn't exist. Instead, the Post headline gives the impression that the link is there, but some people don't think it's very important. It's very subtle editorial positioning by word choice. Thanks, liberal media. But I digress.)