Over the years, the New York Times and writers associated with it have done more than most of us to debunk scares over purported cancer clusters. “When multiple cases of cancer occur in a community, especially among children, it is only human to fear a common cause,” wrote George Johnson in a 2015 Times piece. “Most often these cancer clusters turn out to be statistical illusions, the result of what epidemiologists call the Texas sharpshooter fallacy. (Blast the side of a barn with a random spray of buckshot and then draw a circle around one of the clusters: It’s a bull’s-eye.)”
Johnson’s writings in publications other than the Times have cogently analyzed dubious claims of cancer clusters in Toms River, N.J. and Love Canal, N.Y. Times science reporter Gina Kolata was a pioneer in questioning claimed incidence patterns, and a Times editorial helped in dispelling one of the most famous cancer cluster theories, that of breast cancer on Long Island.
Dozens of refuted cancer cluster scares later, are we more cautious when new ones are put forward? Or has nothing changed?
On Jan. 2 the Times published, to predictably sensational reaction, a piece by Hiroko Tabuchi profiling a claimed childhood cancer cluster in Johnson County, Indiana. Local campaigners have collected cases of childhood cancers — diverse kinds of it, not all one type — associated with the county since 2008, which would imply a rate of six cases per year. The piece, unfortunately, omits to mention the county’s population; it’s 139,654. It does concede, somewhat backhandedly, that the county is within a broadly normal range on its numbers, by noting that it has an incidence of childhood cancer slightly above the average for American counties, placing it in the 80th percentile of all such counties, which in this kind of statistical distribution means not really any great outlier at all.
Cluster alarms call for a culprit, and the local campaigners have settled on a now-shuttered industrial plant that used TCE (trichloroethylene), a solvent familiar from dry cleaning and used at many thousands of sites. They suspect it may have spread through the water and subsequent evaporation, which would explain — or would it? — which some sick children lived in homes distant from the plant and why current water testing is at best inconclusive.
I’ve written a fair bit about cluster controversies in the past, including the one in Hinkley, Calif. made famous by Erin Brockovich, and the Woburn, Mass. story captured in the book and movie “A Civil Action.”
7 Comments
The childhood cancer cluster was reported by residents in the city of Franklin, Indiana and NOT throughout Johnson County. By using the population of Franklin, and not the whole county, you’ll see there is a high rate of cluster.
I have not examined the data for Franklin, Indiana in detail, but having studied statistical analysis, I can attest to the fact that. as Mark Twain said quoting Disraeli, there are liars, damned liars and statistics. Without denying the conclusion implied in your assertion, there are other reasons that might contribute to such a cluster, such as genotype propensity, family habits and presence of other carcinogenic factors. In addition, when
When confronted in court with the death of children, the habit of people to assert a single cause to any sad event becomes overwhelming. OVERLAWYERED being the sort of site it is, regular readers tend to be skeptical of such claims. I find it troubling that one of the sections of the TIMES stories is headed “One House, Two Families”. I would be interested in a close examination of the house and the habits of its occupants before attributing Trichloroethane as the cause of those particular cancers.
That’s the trouble with statistics: moving statistical conclusions to individual cases and vice versa is a process fraught with leaps of logic.
Bob
There are always skeptics that want to say these rare cancers are caused by something else. I hope these women continue to push forward in their fight. My father died of AML cause by benzene exposure in a GM plant. His death certificate was signed, cause of death, AML caused by benzene exposure. See if one has all the facts, a lot of these rare leukemia’s and cancers are only caused from exposure to the chemical or prior treatment from chemotherapy.
Corporate Greed
There’s another type of greed, Sherry. Lawyers looking for somebody to sue. I find it ironic that researchers don’t know what causes cancer, but lawyers and politicians do.
That’s not irony, Jim.
Bob
A fun little experiment one can run in Excel in regards to cancer clusters.
Set up a macro to randomly generate two numbers between 1 and 10. These are X and Y coordinates.
Run this 100 times and plot the frequency of the results in a 10 x 10 grid, where the X coordinate is left / right and the Y coordinate is up / down. So, if the random numbers are 3,9, go over to 3rd column and up to the 9th row and add a tally mark.
I can guarantee you that it will NEVER (ok, perhaps not literally never, but never in the number of times you care to run the experiment) ever turn out as 1 instance in each square. What I can guarantee is that nearly every time, there will be one cell with several hits – well above average.
Do it again with a 21×21 grid and random number generators for 1 to 21. That results in 441 combinations (435 house seats, approximately equally apportioned by population). Run it about 2205 times, or 5 per square. There will be multiple squares with 0 hits…and others with 8, 10, 12 hits…..your so called cancer clusters. Even running it only 882 times, or two per square, you’re likely to have a cell or two with significantly more hits than 2.
These are easy experiments for an individual with a bit of spread sheet skill to run to see what “clusters” are in fact “normal” artifacts of random distribution. Just because there are only 882 cases per year nationally, 2 cases of a particular cancer per year for each 748,000 people (the average congressional district), doesn’t mean that one of those districts may have many times that simply as a result of “normal” random distribution.
That’s why sample size is important in determining the likelihood that you can actually draw conclusions from a set of statistics. The larger the samples, the more likely your statistics can “mean” something. Although what they “mean” may not be what you think they mean.
For example, establishing substance x and effect y and attempting to show a relationship, how many other substances did you not test? For example, i’m sure all those children were exposed to oxygen, nitrogen, co2, etc, etc, etc… So you can never isolate x as the cause since what if it is just a risk factor, not a definitive causal relationship? After all, apparently 200 times the normal consumption of root beer is carcinogenic…