An Oregon hazelnut packer recently refused to reveal his list of nut suppliers because, he said, health officials had not proved that hazelnuts were responsible for an outbreak of toxic E. coli.

Midwest health officials had used traditional interviews to determine that seven people had been sickened by hazelnuts, but they had not yet found a contaminated hazelnut. One Oregon packer warned that it would be “disastrous” to reveal nut suppliers in the absence of “conclusive” evidence that nuts were the culprit.

Health officials understand the packers’ reluctance. They’ve seen similar responses from producers of suspect foods ranging from beef to alfalfa sprouts to raw milk. 

In the last few weeks, cantaloupes suspected as the cause of Salmonella infections and beef bologna tied to an E. coli outbreak prompted product recalls, although the pathogens had not been confirmed in cantaloupe or bologna samples.

But skeptics who doubt the association ignore the fact that traditional epidemiology is statistically as significant as a gee-whiz genetic fingerprint match.

“The food industry has a hard time understanding how epidemiology works,” says Dr. Kirk Smith, foodborne illness supervisor for the Minnesota Department of Health. “People don’t grasp how powerful it is.”

Genetic “fingerprinting,”  the popular term for pulsed field gel electrophoresis, or PFGE is a relatively new technology that has profoundly changed the science of epidemiology.  By analyzing the genetic pattern of E. coli O157:H7 or other toxins in stool specimens, it can establish with a high degree of probability that patients were, or were not, sickened by the same food.

PFGE can also be used to establish links between sick people and a specific source, be it hamburger or hazelnuts.  But often there is no sample of the suspect food. By the time an outbreak is detected, that food has often been consumed or discarded, so epidemiologists fall back on traditional techniques.

 “It would be great if we could just buy the product, take it to the lab and find Salmonella.  That’s something anybody can understand,” says Dr. Bill Keene, senior epidemiologist at the Oregon Health Department. “But when you offer up P values and probabilities, people want to say: ‘That’s just statistical mumbo jumbo.'”

Take the hazelnut outbreak, for example.  In mid-February, Minnesota health officials detected a possible “cluster” of E.coli O157:H7 infection — seven sick people across the upper Midwest. PFGE analysis of stool samples showed they were sickened by a common source.  But what source?  They lived hundreds of miles apart, and had not eaten at the same restaurants.

Each of the victims was questioned at length about what they had eaten over the past few weeks, looking for common denominators.  Each, for example, had eaten ground beef.  But two had eaten privately slaughtered meat, which meant they did not get their meat from a common source.

But all seven had eaten nuts, and this got officials attention.

It’s basic statistics, Kirk explains.  “Seven out of seven of our cases ate hazelnuts, and the PFGE says they almost certainly were sickened by the same source.”

The next step is to establish the “background rate,” he explains.  What is the likelihood that people will eat nuts in a given week?  There is no such data for hazelnuts, but there is data that says about 34 percent of Americans will eat some kind of nuts in a given week.

Overlay those two figures – all seven cases and 34 percent background – and you get a probability of about .05% — five one-hundredths of one percent.  “So the odds of hazelnuts not being the cause of this outbreak were very, very small,” Smith says.  “Possible, but highly unlikely.”

From a statistical standpoint, those results are as persuasive as a PFGE match between sick people and a specific source.  But the old-fashioned interview approach may not impress a business owner facing the enormous costs of recalling a contaminated product.

“Everybody loves lab results,” Keene says.  “Me too.  Nobody sends more stuff to the lab than I do.”

Genetic fingerprinting continues to bring profound changes to epidemiology.  A generation ago, most outbreaks of foodborne illness were local affairs traced to a bad batch of chicken at a restaurant or a church potluck. Today, PFGE enables officials to establish links between illnesses scattered across the landscape and traceable to mass-marketed products.

“Twenty years ago, we would have had those seven E. coli cases, and no way to link them to a common source,” Smith says. “Without PFGE it would have been a lot harder to implicate hazelnuts.”

The technology, however, isn’t perfect.  Epidemiologists are haunted by the experience of a huge Salmonella  outbreak in 2008, which was originally blamed on tomatoes before further investigation swung the onus to jalapeno peppers.

But PFGE remains just one tool in the epidemiologists toolbox, which is why they still rely heavily on traditional questionnaires, Smith says. That’s not likely to change any time soon.

And even the reluctant hazelnut packer soon had to agree that traditional strategies are credible. The very day he refused to hand over his list of nut growers, the Minnesota Department of Agriculture announced that PFGE analysis had identified the outbreak strain of E. coli on the exterior of a hazelnut found in the home of one of Minnesota’s E. coli patients.  In that case, case closed.