Prescription drugs that kill: The challenge of identifying deaths in government data

Mike Stucka at Data Driven Journalism: “An editor at The Palm Beach Post printed out hundreds of pages of reports and asked a simple question that turned out to be weirdly complex: How many people were being killed by a prescription drug?

That question relied on version of a report that was soon discontinued by the U.S. Food and Drug Administration. Instead, the agency built a new web site that doesn’t allow exports or the ability to see substantial chunks of the data. So, I went to raw data files that were horribly formatted — and, before the project was over, the FDA had reissued some of those data files and taken most of them offline.

But I didn’t give up hope. Behind the data — known as FAERS, or FDA Adverse Event Reporting System — are more than a decade of data for suspected drug complications of nearly every kind. With multiple drugs in many reports, and multiple versions of many reports, the list of drugs alone comes to some 35 million reports. And it’s a potential gold mine.

How much of a gold mine? For one relatively rare drug, meant only for the worst kind of cancer pain, we found records tying the drug to more than 900 deaths. A salesman had hired a former exotic dancer and a former Playboy model to help sell the drug known as Subsys. He then pushed salesmen to up the dosage, John Pacenti and Holly Baltz found in their package, “Pay To Prescribe? The Fentanyl Scandal.”

FAERS has some serious limitations, but some serious benefits. The data can tell you why a drug was prescribed; it can tell you if a person was hospitalized because of a drug reaction, or killed, or permanently disabled. It can tell you what country the report came from. It’s got the patient age. It’s got the date of reporting. It’s got other drugs involved. Dosage. There’s a ton of useful information.

Now the bad stuff: There may be multiple reports for each actual case, as well as multiple versions of a single “case” ID….(More)”