Search results beat FDA in finding drug combo side effects
Sifting through the search queries of 6 million people turns out to be a better way to discover drug-to-drug interactions than the current gold standard, the Adverse Event Reporting System.
When it comes to scientific research, size matters -- and yes, bigger is better.
So it may come as no surprise that scientists at Stanford, Columbia, and Microsoft have used Internet search data to uncover prescription drug side effects faster than the FDA's current gold standard, the Adverse Event Reporting System. After all, the data miners had the activity of some 6 million Internet users at their disposal, whereas the FDA relies on physicians to notice and report problems.
Reporting today in the Journal of the American Medical Informatics Association, the researchers write that by analyzing Google, Microsoft, and Yahoo search engine queries, they were able to find a link between the use of the anti-depressant paroxetine and the cholesterol-lowering drug pravastatin. (The link? Using both increases a user's risk of developing high blood sugar.)
The team relied on work done at the lab of Russ B. Altman, a bioengineering and computer science professor at Stanford. Microsoft provided anonymized data from a browser toolbar users allowed to be installed to monitor their search history.
After looking at 82 million searches for drug, symptom, and condition information performed in 2010, the researchers pinpointed searches for the term paroxetine, the term pravastatin, and also searches for both terms simultaneously.
Next they calculated how likely it was for those users to search for the term hypoglycemia or some 80 symptoms, including "dehydration" and "high blood sugar."
They learned that users who searched for both drugs were twice as likely to search for one or more of those terms related to hyperglycemia than those who searched for just one of the drugs.
"There is a potential public health benefit in listening to such signals," they write in their paper.
While this is a first of its kind, the approach is similar to Google Flu Trends, and it may soon look primitive if the researchers are able to incorporate an even wider range of data drawn from additional (and massive) sources such as Twitter and Facebook. The challenge there, the researchers told The New York Times, is .