Since I wrote about screening mammography a week ago, I thought I would do a quick follow-up. A story in today’s New York Times reported on a study that “discovered” that cancers that are detected between screening mammograms are more aggressive than those detected by mammography.
I used the scare quotes because this was not a discovery; it was a confirmation of a safe prediction. There was almost no chance that was not true, even absent empirical confirmation. That was not the only such example. The NYT story included the observation, “Scientists know that both missed tumors [i.e., those that are detected in the interval between mammograms, but that should have been detected on a mammogram, when re-viewed in retrospect] and interval cancers [the study’s jargon for those that are detected in the interval and were not visible on the previous mammogram] tend to be larger and more advanced than tumors detected at regular screenings.” I suppose that the intended subtext might have been “this should be obvious”, but I have a feeling the reporter thought that it was news. I trust it is obvious that tumors that are detected because they are big enough to be felt or that are causing symptoms are more advanced, on average, than those that can only be seen with a mammogram.
The new “discovery” is not quite so obvious as that. It requires a little more thinking and there was a tiny chance it was not true. To understand that one, think of different hypothetical cancers that are growing at different speeds (a chalkboard would be useful now, but I will try to do it with prose). They will be detected by screening mammography when they get to size S and even without screening when they get to size T, with T>S. Something growing very slowly is very likely to reach size S but not yet size T when a scheduled mammogram occurs, so it will be detected by mammogram. Something growing very fast will spend very little time between S and T, so it is very unlikely a mammogram will be scheduled during this time, and so it will not be detected by a mammogram. As the growth speed varies in between those, it should be obvious that the faster the growth, the more likely it will get from S to T before the next scheduled screen. Thus, it is inevitable that the cancers that are detected between screens will be faster growing. It is possible that fast growing cancers would not be more aggressive (in the sense of spreading beyond the breast and being life-threatening) but that is rather unlikely.
This well-known phenomenon is one of the reasons that cancer screening in general is not as useful as the typical naive assessment suggests. Screening is much better at catching the cancers that are slow growing and would thus be less of a problem even if detected later without screening. So the screen-detected cancers are easier to successfully treat, making it look like screening was causing more effective treatment when it was really just selecting the cases that are easier to treat.
The new study result should have been presented as a confirmation of an obvious prediction, not a discovery. It is really too bad that people in health research and reporting do not realize that a hypothesis is not just a word that is preceded by “null” and used in statistical testing.
On the positive side, the policy recommendation by the study authors was much better than usual:
If we set aside that this was not really a “finding” and that there is no such thing as an existential unconditional “need”, we can appreciate the use of “suggests”, meaning “all else equal, knowing this (again, pretending we did not know it already) argues for moving our priorities more toward….”. And the “different approaches” could almost be seen as a heretical suggestion that the boondoggle that is the current screening mammography industry could be improved upon. It was probably already considered heretical that they did not call for even more of the current tech screening.