Observational Studies, Part 2

We've been handed a teachable moment by the study reported in The Lancet finding that exposure to CT scans in childhood is associated with an increased risk of cancer later in life.
This is called a retrospective cohort study. It would be impossible to do this particular study in the United States, because of our fragmented health care system. In the UK however, they do indeed have socialized medicine. While this means that Britain must be a totalitarian dungeon, it also means that they have universal access, much lower costs, and better outcomes than we do. Also the people are more satisfied with their health care. I know, The Horror, The Horror!

It also means that it is possible to link events that befall specific people at one time in their lives with their later fate. Since everybody gets their health care from the National Health Service, the records are available for every single child who had one or more CT scans. The study (by Mark S. Pearce and colleagues) included every person who did not have a prior cancer diagnosis, who had one or more CT scans while they were younger than 22 years from the years 1985 through 2002. The UK also has a central cancer registry, so it was possible to identify all of these individuals who were subsequently diagnosed with cancer in the years 1985-2008. Again, this could not be done in the United States. Finally, they developed models for total radiation exposure to the brain and bone marrow based on which body parts were scanned and the age at which it happened. This enabled them to calculate a dose-response relationship between exposure to these tissues, and subsequent blood (leukemia) and brain cancers. Because the radiation dose to bone marrow and brain from scans of the extremities is at or near zero, this gave them what amounts to a non-exposed comparison group.

The net result is a study that is in many respects as good as a Randomized Controlled Trial (info to follow in a few days). The investigators excluded leukemias that appeared within 2 years of the scan, and brain cancers that appeared within 6 years, so as to eliminate the possibility that the scan was done because of suspicion of cancer. That means that as far as we can tell, the exposures were random with respect to cancer risk. (They would typically have been done because of a traumatic injury, or abdominal pain.) The diagnoses of cancer were made by physicians who had no idea this study would ever happen and most likely did not even know that the scans had ever taken place, so it qualifies as blinded; and in any case diagnoses of leukemia or brain cancer are pretty definitive and unlikely to be much influenced by ascertainment bias.
The results are additionally compelling because there is a very pretty, linear dose-response relationship for both categories of cancer; and because the results are similar to what has been found from other observational studies such as observation of Hiroshima survivors. So this gives very strong support for causal inference.

Caveats? A couple. I said "almost as good as an RCT" because it is possible to imagine some confounders. Actually I don't think the authors even mentioned this, but head trauma is an indication for a head scan, obviously, and could plausibly be related to later risk of brain cancer. Similarly, one can imagine that there is some as yet unrecognized relationship between causes of abdominal pain or intestinal symptoms in children and later leukemia. One could also imagine that there are certain socio-economic or environmental circumstances associated with the probability of receiving a CT scan in childhood and cancer risk. However, the linear dose-response relationship would seem to argue against these possibilities.

The good news, for those of you who are now paralyzed in fear because of that abdomen scan they gave you when you had suspected appendicitis, is that the relative risk reported in the corporate media -- 3 to 7 times for the highest doses, for leukemia and brain cancer respectively -- actually represent very low absolute risk. By 10 years after exposure, you would expect one case of leukemia and one brain tumor per every 10,000 patients. Unfortunately the elevated risk, as far as we know, continues after that, probably for life, but it still adds up very slowly. One more takeaway -- an I'm volunteering this one -- this study does support the widely used assumption that there is no safe threshold for radiation exposure, that the risk for cancer is linear and that the line goes right through zero. But -- the risk at low levels is extremely small.

Finally, obviously, this is one more cautionary tale about overutilization of medical procedures. CT scans are still well worth doing under some circumstances. They are actually much safer than, say, riding in a car every day. And wouldn't you want to know if your child had a cerebral hemorrhage that might kill her? Say, because she'd been in a car crash? But, make sure there really is a darn good reason.