19 August 2011

Sprey-Cockburn Report

Did Fukushima Meltdown Increase Infant Death Rates in the United States?

Attached below is an extremely important report written by my friends Alexander Cockburn and Pierre Sprey.  They describe on the ramifications of the still unfolding Fukushima nuclear disaster in Japan — specifically, its impact on infant death rates in the United States, and in so doing they also discovered the disastrous state of radiation monitoring by the Environmental Protection Agency.  

Alexander needs no introduction, nor should Pierre, but for those who do now know of him, Pierre is a brilliant statistician/engineer.  He was a key member of the design teams that shaped the enormously successful F-16 and A-10 combat aircraft and he is founder and key sound engineer of the innovative Mapleshade recording studio, renowned for the quality of music among audiophiles.  Pierre approaches statistics in the exploratory empirical tradition that allows the universe of data observations help to shape the hypothesis under test.  Moreover, he does this in a rigorously unbiased way that ends up milking the maximum real (as opposed to spurious) information out of the data.  His analyses are not only elegant and brilliantly simple, they can be awe-inspiring to the analytically inclined. [Pierre's technique is very similar the Exploratory Data Analysis philosophy pioneered by the renowned statistician John Tukey, an accessible description of which can be found here.]

In this case, Pierre constructed a simple statistical test (the Mann-Whitney U Test — explained in the technical addendum I have appended to the end of the Sprey-Cockburn Report) to determine if and how infant death rates in the United States changed after the Fukushima nuclear disaster in Japan.

I urge readers to study this very important report carefully, especially those of you who are infatuated with the idea of building a carbon-free economy, because a carbon-free economy means many more nuclear power plants.

The Sprey-Cockburn Report appeared in the subscriber edition of Counterpunch.  The editors kindly gave me permission to distribute it.  While the free Counterpunch website is a source of much useful information on an incredibly wide range of subjects, the juiciest morsels of investigative journalism are usually reserved for paid subscribers … I strongly recommend that readers subscribe, it will be worthwhile, even if you are interested in only one article out of every six.

Chuck Spinney
Nice France

Where did the Radiation Plumes from Japan Touch Down?
The Fukushima Disaster and Infant Death Rates in the U.S.
· At Least 19 Cities in Harm's Way
· Why EPA Monitoring is a Joke
By Pierre Sprey with Alexander Cockburn
CounterPunch, Volume 18 no 13, July 1-31, 2011 
[Subscriber edition — distributed with permission of editors]

CounterPunch has established that in the eight weeks after the nuclear disaster at the Fukushima complex in Japan on March 11, infant mortality in 19 U.S. cities increased by 35 per cent.
   In the course of this review, conducted by CounterPunch's statistical consultant, Pierre Sprey, it also became clear that the Environmental Protection Agency's monitoring system, known as RadNet, is hopelessly inadequate to assess the effect on U.S. public health of a nuclear accident either overseas or here in the Homeland. EPA's routine sampling is laughable, with sampling frequency and geographic coverage that are hopeless for tracking radiation exposures of concern to public health. EPA's extra sampling following disasters like Three Mile Island or Fukushima can, at best, identify only a tiny fraction of the significant touchdowns of the concentrated radiation plumes from an accident site.
   This past June, to check on a Sherman and Mangano piece on the CounterPunch website showing elevated infant deaths in eight cities in the Pacific Northwest post-Fukushima, we asked CounterPunch's statistical advisor Pierre Sprey to review data available from the Center for Disease Control (CDC) in the form of weekly deaths of infants of one year or less in 122 reporting cities across the U.S.A. This is the only available database where one can get numbers bearing on very recent mortality trends within a week or so after the deaths occurred. Most other mortality databases are not published within a year or more of the events covered.
   In June, Sprey reviewed data from all eight cities mentioned in the Sherman/Mangano article, as well as the three remaining Pacific area CDC-reporting cities to the north of these eight: San Jose, Santa Cruz, Fresno, Berkeley, San Francisco, Sacramento, Portland, Seattle, Tacoma, Spokane and Boisie. Sprey found that the four northernmost Pacific Northwest cities of these eleven- Portland, Tacoma, Seattle and Spokane - showed remarkably significant results.
   During the ten weeks before March 11 the four cities suffered 55 deaths among infants less than one year old. In the ten weeks after Fukushima 78 infants died - a 42 per cent increase, one that is statistically significant. To confirm once again that these results were not due to seasonality, Sprey compared the infant deaths in the ten weeks after Fukushima to the deaths in the equivalent ten weeks a year earlier. The results were almost identical with the ten weeks before Fukushima in 2011. Within the equivalent ten weeks of 2010, 53 infants died in these four cities.
   The post-Fukushima deaths are 47 per cent higher than they were in the same period a year before - once again, statistically significant. If you add the equally far north city of Boise, Idaho, to the four-city sample, the results remain almost unchanged.
Such results are not as surprising as they seem: Chernobyl was associated with similar spikes in infant mortality at various distant locations in Europe. Even though our northern Pacific coast is 4,500 miles from Fukushima, significant localized concentrations of radioisotopes are to be expected because the meltdown's radiation plumes carried by the Pacific's westerly winds, much like pollution plumes and ash plumes elsewhere, did not disperse uniformly with distance - contrary to the equations used by all atmospheric computer modelers. In fact, actual observations of radiation dispersal after Chernobyl or volcanic ash dispersal after any notable eruption, including the recent Icelandic eruptions, always show that the particles disperse in unpredictable and surprisingly concentrated plumes, which touch down occasionally and with high concentrations at great distances from the source.
   In his June review on our website, Sprey pointed out that an important line of inquiry would be to correlate the sampled cities' infant death results with contemporaneous measurements of radiation levels in the drinking water, and possibly the milk supplies in nearby areas. In that review, Sprey selected the four Northwest city samples because he was trying to analyze the data from all CDC-reporting cities within a geographically consistent area that might have been exposed to Japanese radiation. 
   Now Sprey, with the help of CounterPunch researcher Jed Bickman, has widened his purview and sharpened his selection criteria to include more cities. He looked for cities specifically defined by elevated radiation levels near them, using EPA iodine-131 measurements as the indicator of a nearby plume touchdown. Locations in the EPA RadNet database showing significantly raised iodine-131 in any air, rain or drinking water sample (or significantly raised strontium-89 in a milk sample) within the 20 days following Fukushima were selected. If any of the 122 cities reporting weekly mortality to the CDC was nearby or within 100 miles or so downwind of one of these RadNet locations, this was taken as an indicator that the reporting city could have been exposed to a radioactive plume touchdown. Note that, as discussed at greater length below, the EPA RadNet samples are so sparse in time and space - days or weeks apart and often hundreds and hundreds of miles between monitoring sites - that the vast majority of actual plume touchdowns across the country almost certainly remained undetected. 
   Using this approach, Sprey selected 19 cities showing evidence of being near a touchdown within 20 days of the Fukushima disaster. Five proved to be Portland, Seattle, Spokane, Tacoma and Boise, the cities already examined in the June analysis. Five more from the Sherman/Mangano study also met the criteria: Santa Cruz, Sacramento, San Francisco, San Jose and Berkeley. New cities added were Long Beach, Las Vegas, Ogden, Salt Lake City, Colorado Springs, Denver and, surprisingly, three cities in Florida - St. Petersburg, Tampa and Jacksonville. 
   Sprey took infant deaths during the eight weeks after the Fukushima disaster on March 11 (weeks 11 through 20 of 2011) and compared them to two control samples. One control sample was the deaths during the identical eight-week period from a year earlier (weeks 11 through 20 of 2010), and the second was the eight-week period in 2011 just before Fukushima. 
   Sprey found that, when compared to 2010, infant mortality in the 19-city sample increased by a statistically significant 35 per cent. The raw numbers are 305 infant deaths in the eight weeks after Fukushima and 226 deaths for the same eight-week period in 2010. Comparing the 305 post-Fukushima deaths to the 259 infant deaths in the eight weeks just before the meltdown yielded a statistically significant 18 per cent increase.
The percentages are not quite as large as the percentage increases for the four-city sample of the June review, but statistical reliability has increased considerably, because the sample almost doubled the number of cities and quadrupled the number of deaths included.
As Sprey reviewed the data available, he was astounded at the sampling inadequacy of EPA's RadNet, all that the United States has available to monitor the exposure and health risks of large masses of people during a nuclear accident - most importantly during a domestic power plant accident where far more Americans would be at risk than from Fukushima. 
Radiation does not disperse according to any model. Plumes move unpredictably. Thus, the only way to monitor is to have a network which is geographically dense enough, and to sample often enough that it doesn't miss a lot of plume touchdowns. And the sampling frequency needs to be adequate for each of the ways in which the public could be exposed to harmful radiation: through drinking water, milk, rain, and airborne particles. 
As stated at the outset, EPA's RadNet is hopelessly inadequate. For an example, Sprey looked up what the RadNet database had collected for the decade 2001 to 2010 in the most populous state in the union, California.
Air: first, consider airborne isotope results across 10 years for the entire state of California, measuring six isotopes of concern for public health: iodine-131 and five others (in EPA lingo, these are called air filter samples). Across the entire decade, there were only 11 readings, all of them conducted on one day, December 31, 2009 - and only one sample for each of 11 cities on that day. 
Milk: in a decade, EPA took only six readings for all of California, one in Los Angeles, and five samples in San Francisco.
Rain: EPA measured only gross gamma radiation and a short list of isotopes, which doesn't include iodine-131 or strontium-89, both of prime interest for public health after a nuclear power plant accident. Over the decade, rain sample readings were taken only once a month, and only at one site in all of California: Berkeley till March 2004, then Richmond from March 2004 to June 2010. After June there are no readings, because either EPA failed to update the database or lost interest. 
   Drinking water: this is a major health concern because a city's drinking water exposes citizens to radioactive particles washed into the system from across an entire watershed. Only three cities received any drinking water readings at all. In Los Angeles, EPA took one isotope reading a year but only for iodine-131, cesium-137 and tritium. Tritium measurements of interest only to detect if a thermonuclear explosion has taken place, are useless, and iodine has an eight-day half-life, so it disappears long before the next one-a-year sample is taken. In Richmond, EPA measured drinking water once a year, but only for seven years of the decade. Berkeley made do with three years. 
   In New York State the situation is just as bad. In all of New York, air filter isotope readings were taken on just one day, December 31, 2009. For drinking water isotopes, EPA measured one sample per year in only three cities: Albany, Niagara Falls, and Syracuse. New York City had no drinking water readings for the entire decade.
For rain, there were monthly readings at two places in New York State: Yaphank and Albany. New York City's rain remained unmonitored. 
For milk: over the decade, three milk samples were measured for Buffalo and two for Syracuse; New York City milk drinkers were left to fend for themselves.
Amid these appalling deficiencies, EPA thumps its chest proudly for its small network of about 125 "near-real time," continuously monitoring stations across the U.S.A. - stations that measure gamma radiation as continuous graphs (at infrequent intervals, these stations also send in air sample filters for the conventional laboratory air-isotope readings discussed above). Typical continuous gamma graphs from these stations provide little or nothing of public health interest, because they consist only of very small perturbations above a steady background level. EPA claims, however, that slightly atypical perturbations can alert EPA scientists that something unusual is happening. Then, by comparing the recorded gamma particle counts in each of ten bands of energy levels, trained technicians are said to get good indications of what radioactive isotope may be the source. 
  The catch? EPA's continuous gamma monitoring database doesn't disclose when scientists have determined that an unusual event has occurred, what isotope was identified, or what action was taken. In other words, EPA's years and years of stored gamma graphs yield nothing of interest to anyone outside EPA, neither public health officials nor the public. CP
Pierre Sprey has been a consulting statistician for EPA studies in air and water quality monitoring and related health effects and a principal member of the Pentagon's concept design teams for the F-16 and A-10. He is now running Mapleshade, his record label that sets new standards for sound quality and manufactures pioneering cables, vibration control devices and other upgrades for perfectionist audio systems. 
Technical Addendum
CS Note: I asked Pierre Sprey for a technical description of the statistical test used to determine the significance of his results.  For the statistically inclined, attached below is Pierre’s description of the Mann-Whitney U test.  Used properly, it is an extremely powerful non-parametric or distribution free test [a good textbook explaining the power and use of these kinds of statistical tests is James Bradley's classic Distribution-Free Statistical Tests ]:
“The appropriate  distribution-free test is the Mann-Whitney U test for comparing the medians of two different samples (the samples can be of different size and don't even have to have the same underlying distribution). It tests the hypothesis that the median week's death count for the 19 cities during the 10 weeks after meltdown is to the median week for the 10 weeks before (or, alternatively, the same 10 weeks one year earlier). The alternate hypothesis is that the "after" sample is greater.

The test works by rank ordering all 20 weeks taken together, then looking at T, the sum of the rank positions only in the "after" sample of weeks. From this you calculate U as follows:

 U = n1 n2 + {n1 (n1 + 1)/2} – T , where n1 and n2 are the sizes of the first and second samples respectively.

In this case, the two sample sizes to be compared are equal (i.e., 10). The distribution of U is known (as long as you know the two sample sizes) so you can look it up in a table of the U-statistic to get the significance level. The reason the distribution of ranks (properly normalized) is known and invariant is because ranks are a distribution-free measure. Think of it this way: the chance of the second sample value you draw having the 5th largest rank doesn't in any way depend on the distribution of the actual sample value itself.”