Cancer clusters
This past Thursday I wrote a post that generated quite a bit of back and forth in the comments section, about the supposed vaccine/autism connection. An issue that’s somewhat related to that discussion is how to decide whether an observed increase in a disease or syndrome is just a coincidence, or a meaningful pattern—and, if the latter, how to determine the cause of the uptick.
These questions are also vital in the case of the phenomenon known as the cancer cluster. One was recently in the news, a dreadful grouping of cancer cases in children residing in the smallish town of Clyde, Ohio. One family even had two children with leukemia; what are the chances of that, and what could be the cause?
The answer comes not in the mind and heart gripped with understandable fear, but in the math which says every that, every now and then, statistically speaking, there will be such an increased incidence, a cluster occurring by sheer chance. It is very difficult not to act emotionally, from the gut, and insist that where there’s smoke, there must be fire.
Cancer clusters such as the one in a 12-mile radius around Clyde are reported and investigated, as well they should be. Cancer most definitely can have an environmental cause, and if that’s the case it needs to be uncovered. But the vast majority of cancer clusters are statistical glitches, although the people involved can find that very difficult to accept. Human beings want and almost need to find a cause for horrible events (or even good ones); it’s just our nature.
But true cancer clusters that have an environmental cause have certain defining characteristics. One of the best articles I’ve ever read on the issue appeared over ten years ago in the New Yorker. Written by Atul Gawande, and entitled “The Cancer Cluster Myth,” it offers this sobering quote:
“The reality is that [cancer cluster investigations are] an absolute, total, and complete waste of taxpayer dollars,” says Alan Bender, an epidemiologist with the Minnesota Department of Health, which investigated more than a thousand cancer clusters in the state between 1984 and 1995. The problem of perception and politics, however, remains. If you’re a public health official, try explaining why a dozen children with cancer in one neighborhood doesn’t warrant investigation.
Note that statistic: more than a thousand cancer clusters in Minnesota alone in a single decade, none of them found to be anything more than statistical bumps. No doubt many people would conclude that wasn’t because some environmental problem isn’t causing them, it must be because the investigations were inadequate in some way and/or even corrupt and compromised. But I would not be among those making that claim.
What are the characteristics of meaningful cancer clusters? Red flags are rare cancers rather than common ones, the same exact type of cancer rather than ones in a generally similar category, a particular cancer in an age group or other population that doesn’t usually exhibit it, or an extraordinarily high rate (for example, one environmentally caused cluster in Turkey featured a rate 7,000 times higher than usual, although the rate needn’t be quite that high to be significant). Most neighborhood clusters almost never feature any of these characteristics.
Note that in the situation in Clyde, the involved children have many different kinds of cancer, and mostly leukemia, which is (unfortunately) the most commonly-occurring childhood cancer. And even then, the two children suffering in a single family have two different types of leukemia. The childhood cancer cluster in Clyde also involves brain tumors and lymphoma, two other of the most common forms of childhood cancer. This is highly unlikely to be a meaningful cancer cluster—although it’s completely understandable that the parents there are frightened out of their wits. But exhaustive investigations have produced no smoking gun.
That’s because there probably is none. But tell that to the parents of Clyde.
Cancer cluster = news story.
Cancer void = zilch.
This is one of the cases where common sense expectation (intuition) and mathematics tell us two different stories. Just make a random distribution of dots on a square, by using random number generator for X and Y coordinates. You will immediately see some clasters, that is, places where density of dots is higher then average. Our perception works remarkably well in detecting any regularities or anomalies. Moreover, if you do not see any clasters this means that dots distribution is, in fact, non-random: there should be some anti-correlation, repulsion of dots from each other. Only test on statistical significance can tell us whether these apparent clasters are simply random density fluctuations, inevitable for true random distribution. And yes, this is always hard to explain to people without mathematical training.
Repeat over and over until it internalizes: Correlation does not imply causation. Really, if you simply remember that *one* sentence from a statistics course you will note that a VERY large number of “studies” are bogus.
South Park had a funny episode called “Medicinal Fried Chicken”. Colorado passed a law barring fast food from low income areas (apparently they were actually discussing doing this for our improved health) and a medicinal marijuana moved into the old KFC that was removed because of the law. A number of the residents figured out if they gave themselves cancer they could get legal marijuana and they all did so.
In the end a study found that where the KFC’s were replaced by medicinal marijuana places that the testicular cancer rate (the one chosen by the male residents) shot through the roof. Therefore one could only conclude that fried chicken has a miraculous medicinal property to prevent testicular cancer.
A great deal of these studies are just as good as that one, it is just they play on fears the public has so few think they are stupid, yet they truly have the same amount of religious methodology. Small sample size, correlation means causation, and an unwillingness to explore anything other than the outcome the researchers wanted couple to make a horrid report and equally horrid policies based on it.
When my son was diagnosed with leukemia, we were pretty quickly made aware of various other cancer diagnoses in the neighborhood. My wife and I had many discussions about possible environmental causes, because as parents the need to find a REASON was overwhelming. In the end we decided that it didn’t make sense. After much study we concluded that we simply couldn’t pick a cause or source for the disease. Now many years later and more than a few stats courses, I still think that’s the right answer. No clear causality, which does not generate a comforting narrative, but I prefer that to a fairy tale.
This is similar to how global warming has people pointing out wild and whacky weather. Strictly a problem in understanding average is not the same as normal. Averages are compiled out of whacky which is still in the category of normal. What would be abnormal is if whacky no longer occured.
Lots of adult cancer patients about to embark on treatment asked me about causality. It was always an early question, pre-empting the discussion of treatment options, toxicities, outcomes, etc. My reply started along the lines of, “When your house is on fire, do you ask the firemen as they show up, ‘What caused it? How did it start?’ ”
If we could assign a cause to each case of cancer, how would that give comfort? To tell a young woman, non-smoker, non-drinker, with oral cancer that it is most probably viral-induced will comfort or help her how, as she faces possibly deforming surgery and radiation therapy?
Are we hung up on cause and effect in other realms? Are Obama’s actions foolish or the result of knavery? I get no comfort or satisfaction one way or the other. Like cancer, I don’t care about the cause; I want to stop him any way we can.
Cancer is not a single disease, but hundreds of different diseases with very different ethiology. Some are quite common, some are exceptionally rare. Sarcoma Kaposhi, for example, was so rare that most GP never seen it in their practice, untill epidemic of it occured in San-Francisko gay community; that is how HIV virus was discovered, by a sudden claster of a very rare cancer.
People hate to think they have no control. Randomnity sucks.
Having control requires power over the issue, which first means knowing a cause. Then the cause can be addressed, or at least blamed.
And, while correlation does not equal causation, it does require explainin’
All same-same vaccination nuttiness.
I’m impressed that the New Yorker published the Atul Gawande article. After all, it put the powerline/cancer myth on the map with an article by Paul Brodeur in 1990. I wonder if they ever published a retraction? Two magazines in one, as James Taranto is fond of writing.