Perhaps just as notable as the Food and Drug Administration's
announcement of its openFDA project during this week's Health Datapalooza conference was the tone signaling a courtship of the coding community. The agency has been a target of criticism from Silicon Valley-style digital innovators, perhaps best exemplified by its recent spat with 23andme, the large, Google-funded personal genomics company.
The FDA is hoping to get developers interested in using its datasets for the public good, and it's starting with adverse drug events. In this case, the data for drug events was always available—but not readily so. According to Matt Mohebbi, a programmer with Iodine—the contractor that helped the FDA with the data release—the old data was scattered and typically available in XML format, which wasn't terribly friendly for developers. The hope is that by using an open-source search program called Elasticsearch (it powers Wikipedia's search function) and building an API, interested developers will be able to build tools and critique the underlying software, driving constant improvement.
The format, said Adam Baker, Mohebbi's co-developer at Iodine, is about “extending the olive branch a little bit farther” to developers. It's supposed to be easy. They want “anyone coming to the site (to) understand what's possible.”
The effort is extending its outreach to places where only developers roam—they were proud that the story was a hit on Hacker News, an online hangout for Silicon Valley coders; they'll be taking questions and critiques on other coder forums: Stack Exchange and GitHub, to name other examples.
"These guys (at FDA) are really responsive when we e-mail them really late at night,” said Mark Silverberg, a co-founder with Social Health Insights.
Besides developers, the agency officials on hand were also keen on emphasizing benefits for academic researchers as well. The format of the site means that researchers can publish in academic journals using the openFDA protocol. “We've heard from researchers that it can be difficult to publish anything that has proprietary format, because of obvious problems for reproducibility,” Mohebbi said.
The officials and developers were eager not just to sing the benefits, but also demonstrate them. The demo featured users going to the website—www.open.fda.gov
—and scanning through the reports. They looked at the number of reports from October 2004 to June 2013 that involved a drug whose indication was for hypertension. They were able to narrow the search to reports involving a drug indicated for hypertension and specific to men older than 65.
Taha Kass-Hout, the FDA's chief health information officer, said that he hopes that “you guys can surprise us how you use it.” And the FDA hopes to open and organize other datasets in the future such as drug labeling and recalls and medical device adverse events, but those releases depend on conversations with the various centers that make up the agency.
There were also caveats. Sean Herron, a Presidential Innovation Fellow with the FDA, was quick to warn, “This is in beta. Don't build your $100 million app on this.”
When asked a question regarding the pharmaceutical industry's lack of enthusiasm for the project, Kass-Hout replied: “They've said good things, and they've expressed concerns as well” regarding the potential for the data to be misinterpreted. Herron suggested that the data was best used as driving the creation of theories, which could then be verified more rigorously, rather than coming to strong conclusions.
“We will be correcting misinterpretations,” Kass-Hout said. Follow Darius Tahir on Twitter: @dariustahir