The agency’s solution was PREDICT, or Predictive Risk-based Evaluation for Dynamic Import Compliance Targeting, which is now used at 70 percent of U.S. import operations covering land, sea and air ports of entry.
PREDICT is an artificial intelligence program that scans information on every food and drug import, searching for high-risk shipments that should be physically examined before entering the country.
Foods susceptible to contamination, such as seafood and fresh produce, are assigned a higher risk score. The program is designed to take into account whether shipments contain food products under recall as well as a company’s history of violations and previous test results for its products.
Investigators who monitor the flow of imports look first at shipments with the highest risk scores, flagging those they think should be held for physical inspection.
The system, said Denise Williams, a supervisor in the FDA’s Division of Import Operations in Los Angeles, is “the greatest thing since sliced bread.”
Jessica Bentz, an FDA investigator who trains others on using PREDICT, said she has noticed an increase in “good calls” by the computer program, meaning investigators can use their time more efficiently by concentrating on the riskiest shipments and letting the rest pass into the marketplace more quickly.
The system doesn’t have access to millions of documents, including companies’ violation histories, which are stored mostly on paper at FDA locations across the country, he said.
Nielsen said PREDICT “can only do what the data allows it to do” and the data is “not in a usable format to be applied at the border for decision-making.”
“There are some instances,” Bentz said, “where the [firm’s] history is not as advanced as we’d like it to be … because it’s still on paper.”
The system will improve as investigators input more data, she said.
Another issue that has plagued the FDA and U.S. Customs and Border Protection for years is what’s known as FDA Establishment Identifier numbers. Many foreign firms have more than one identifier – a code that tells the FDA who they are.
According to a U.S. Government Accountability Office report released in May 2010, the FDA has an average of three unique identifiers per firm, causing delays in investigating a firm’s history.
The agency is addressing the problem, Bentz said, by having investigators manually merge firms’ identifiers when they find more than one.
“Nothing ever works 100 percent the way you want it when it starts,” she said.
The May 2010 report also concluded that although PREDICT nearly doubled the percentage of field examinations resulting in violations during its pilot phase, further work is needed to ensure the system’s effectiveness. This includes eliminating gaps in collaboration between front-line agencies such as Customs and Border Protection and the U.S. Department of Agriculture’s Food Safety and Inspection Service.
Yet investigators who work with PREDICT on a daily basis maintain that it has improved their performance tremendously, Williams said.
“I love it,” he said. “I love PREDICT.”