Last week a paper in JAMA received wide attention from the media, including some decent coverage from Gina Kolata from NYT science, who wrote three separate pieces to discuss breast cancer treatment (the best of which is here).
Unfortunately, for the average reader who isn't into biostats, the coverage of what should have been the point of the study (the natural history of Ductal Carcinoma in Situ), has been presented as a terribly difficult topic to sort, with breaking research contributing equivocal findings to a problematic treatment decision.

That is simply not the case and the JAMA study (Breast Cancer Mortality After a Diagnosis of Ductal Carcinoma in Situ) confirms what many public health minded doctors have long understood: those are not the breast cancers that kill. 

Overall 20 year risk of death from breast cancer after being diagnosed with DCIS and undergoing mastectomy, or lumpectomy (with and without radiation) turned out to be 3.3%. I could not find the rate in the comparison group of the general population that they used, but they report the standardized mortality ratios (SMR) between the two groups as 1.8 overall. This means that a woman in the comparison group has a risk breast cancer death of 1.8% (not related to that other 1.8, it's just the number that is 1.8 times smaller than 3.3).

Gina Kolata's pieces for the times have the science of breast cancer right, but like most reporting, they are very light on the numbers and don't realize that this study is about risk. She misunderstood the most important finding of the paper above and accidentally reports that both groups had a 3.3% risk, which is not accurate. Nevertheless, she gets the lead author, Steven Narod, in interview, to state the obvious: the best treatment for DCIS may be no treatment.

As I read the coverage and the commentaries, before reading the actual paper, I kept thinking, can somebody call Gil Welch and get him to explain this for us? And to give her credit again, Gina Kolata did just that, but gave him only two sentences at the end of the article I linked above:

"Welcome to the world of dealing with low-probability events," said Dr. H. Gilbert Welch, a professor of medicine at Dartmouth and author of "Less Medicine, More Health."
"I think it is a classic example of what is and will only increasingly become a recurrent problem in medicine," he added. "The questions about what to do - if anything - are fundamentally difficult."

I am willing to bet my relatively expensive NYT subscription that Dr. Welch said some really interesting things that would ACTUALLY EXPLAIN THIS MESS right after that, but a nitwit editor, not the columnist, cut it off due to length and the fact that it contained difficult words like "ratio."

Referring to Dr. Welch as the author of "Less Medicine, More Health" is a bit like calling T.S. Eliot the guy that wrote "Cats." Dr. Welch has published a host of wonderful papers on absolute vs. relative risk, how to understand statistics and authored a number of books for laypeople on stats and risk, including, most importantly, "Should I Be Tested For Cancer? Maybe Not, and Here's Why" in 2006. In this concise, clear, rational, numbers-filled book, he explains the biggest problem with screening healthy people for cancer:

Deadly cancers grow and spread quickly (that's the whole problem with cancer), while indolent, precursor, or undefined lesions just sit around. When you screen people every few years, it would be simple luck if you happen to catch a deadly, fast growing cancer, while you inevitably find lots of indolent disease that likely doesn't deserve to be called cancer.

His most compelling argument is around prostate cancer, wherein the vast majority of treatment is for a type of growth completely unrelated to risk of death and carries very serious side effects. But he is equally clear about how we treat DCIS as a risk for breast cancer death: there is no relationship. 

Getting back to the JAMA study, what the authors actually report (whether they know it or not and I'm coming to that) is the women who had DCIS found on mammogram and subsequently treated with surgery and/or radiation died of a different breast cancer lesion.


The risk of the breast cancer lesion that did ultimately kill 3.3% of the women who had DCIS was only 1.8 times the risk in women generally. For perspective, the increased risk of lung cancer if you smoke is 35 times. There is certainly a little more risk, but it's not much. The increase in risk almost certainly has to do with something in the patients themselves, regarding how prone they are, not the DCIS.

And in the study, they found that the 3.3% had no relationship to the treatment chosen (in fact the mastectomy group did a little worse, but it's likely not significant), which tells us one thing clearly:
Treating DCIS found on screening mammograms is not improving breast cancer mortality.

Said more clearly:

Stop cutting on women who don't need to be cut!

So why do I say that the authors themselves seem to have missed the point of their own study? Well, they could/should have reported as the headline number (like, in the title) that women age 30-35 had 17 times the risk of death as the older women, or that African American women were 2.5 times more likely to die than non-hispanic whites. But the most important finding was really, that DCIS is overtreated. Or rather, treated unnecessarily in the vast majority of cases.

Instead, the article strangely comes to the conclusion that, since the women with DCIS seemed to die of breast cancer at a similar rate to the control group, despite treatment, DCIS must somehow be metastasizing or causing recurrent cancers. That it is actually MORE dangerous than suspected:

"It is often stated that DCIS is a pre-invasive neoplastic lesion that is not lethal in itself (my note: that is actually the definition of "in situ"). The results of the present study suggest that this interpretation should be revisited....Some cases of DCIS have an inherent potential for distant metastatic spread. It is therefore appropriate to consider these as de facto breast cancers and not as pre-invasive markers."

But WE ARE treating them as breast cancers. We are cutting them out. It doesn't change risk. The study showed that. It confirmed what Dr. Welch showed in his 2006 book using actual biostatistics, that deadly cancers continue to remain deadly and the cancers we are treating are not the right lesions to focus on. The authors' insistence that we would benefit patients by paying more attention to these lesions is simply incomprehensible.

The JAMA paper is free on the link above.