The web is now used for all kinds of illness tracking, notably during flu season. By analyzing when people search for, or tweet about, symptoms, it’s possible to build maps of how people are affected during epidemics. That can help professionals size up the problem, and prepare for the worst.
But flu-mapping is likely just a forerunner of the Internet’s health sensing potential. Now, researchers have shown its usefulness for something even more helpful, and previously unknown: drug side effects. In time, their techniques could give earlier warnings of when drugs, or combinations of drugs, are causing problems, and alert regulators to action.
Researchers at Microsoft, Columbia, and Stanford had a hunch that all the searching people do for health information online could throw up some insights. In particular, they were interested in patients who take both an antidepressant called paroxetine, and a cholesterol medicine called pravastatin, and whether the mixture caused an increase in the rate of hyperglycemia (high blood sugar).
To find out, they mined 82 million searches across Google, Bing, and Yahoo. First, they identified searches for paroxetine and pravastatin, and then how likely people were to search for "hyperglycemia," or one of its symptoms. They found significantly higher likelihood. People were roughly twice as likely to make hyperglycemia-related searches if they searched for both drugs, than if they’d looked for just one.
"Current methods for tracking side effects and interactions among medications in the post-marketing phase typically rely on a slow process of the manual reporting of symptoms," he says. "While the log-based methods may be noisier, they are inexpensive and fast, and there is the promise of developing tools that can skim over large numbers of combinations of medications."
Web-mining won’t replace traditional methods completely, White says. But it could help regulators to know where to look, by "guiding efforts to confirm or rule out potential interactions".