Yesterday's health news delivered another paper slamming mammography. A report out of Denmark uses statistics to show that overdiagnosis is incredibly frequent. An accompanying editorial, by Dr. Otis Brawley of the American Cancer Society, is a call for acceptance by doctors of the overdiagnosis phenomenon.
You might say it’s all about “overdiagnosis awareness.” So let’s consider that.
Overdiagnosis is a concept: the idea of finding medical information that does not benefit the person, and which may cause harm, especially if it leads to overtreatment. However, the concept is misapplied when it comes to finding invasive breast cancer in women who are otherwise likely to live. Cancer science dictates that over time in most cases, unchecked tumors gain mutations and spread. There is no pathology report I have ever seen that shows a breast tumor to have vanished.
Meanwhile, breast cancer is a leading killer of women. Bayes’ theorem supports screening of populations at significant risk for a disease that can and should be treated. Breast cancer is a good, if not excellent, example of that. What worries me about so many publications on “breast cancer overdiagnosis” isn’t so much that women won’t choose to get screened, which is a genuine concern, but that women may begin to think, mistakenly, that breast cancer is no big deal.
Is there such a thing as statistical gaslighting?
Limitations of the current paper, published in the Annals of Internal Medicine, include the omission of breast ultrasound and other imaging methods. Breast cancer screening is much better than many people think, in part because it's not just "mammography," as considered in this article. As I have written, breast ultrasound is crucial for detecting tumors in premenopausal and some older women who have dense breasts. Others would point out that 3D mammography is much better than 2D, or non-digital before that, and that MRIs and molecular breast imaging can be useful. Yet these other methods are kept out of the story.
Another issue is the cross-sectional nature of this far-from-randomized study. A statistician might inform you that you can’t compare outcomes between distinct groups of women, with age brackets chosen by the analyzers, without accounting for known and potential differences between them. It's tricky. Yet that’s what the authors did.
Relative to some champions of “evidence,” I’m fairly open to learning from population-based analyses, if and when strong and clear patterns emerge from data. However, this new study provides neither a strong nor clear pattern of anything; there are no findings that leap out from the graphs or tables or unadjusted supplementary data.
Some vocal mammography critics dismiss non-randomized studies except for those on mammography. You might wonder why they do so: What is the agenda behind their arguments? Are they doctors competing for other resources, such as grants to support their research? Maybe they’ve published articles or books against breast cancer screening and this report, whether or not it’s a good analysis, bolsters their reputation.
It’s simply not valid to cherry-pick findings of non-randomized studies to support one’s views.
Those same critics might at least weigh the benefits of all women having access to better-quality screening—with ultrasound and 3D mammography—or at least argue for careful trials of those, rather than conflating the methods or blurring the facts about breast cancer's potential lethality. Show me a single case of invasive breast cancer in a 50-year-old woman that’s not worth finding early, and I’ll pay more attention to this overdiagnosis hype.
