Analyzing queries made to Google, Bing, and other search engines can reveal the potentially dangerous consequences of mixing prescriptions before they are known to the Food and Drug Administration (FDA), according to a new study. Such data mining could even expose medical risks that slip through clinical trials undetected.
Pharmaceuticals often have side effects that go unnoticed until they're already available to the public. This is especially true of side effects that emerge when two drugs interact, largely because drug trials try to pinpoint the effects of one drug at a time. Physicians have a few ways to hunt for these hidden risks, such as reports to FDA from doctors, nurses, and patients. One study, in 2011, data-mined those FDA reports and uncovered a hidden drug interaction: When taken together, the antidepressant paroxetine and the cholesterol suppressant pravastatin cause hyperglycemia, or high blood sugar. After verifying that finding with experiments, the researchers behind the study wondered what other information sources were left untapped.
Enter search engines. Much like Google Flu Trends reveals influenza outbreaks by tracking flu-related search terms, search queries about drug combinations and possible side effects—say, "paroxetine," "pravastatin," and "hyperglycemia"—might enable researchers to identify unanticipated downsides to medications, says bioinformatics researcher Nigam Shah of Stanford University in Palo Alto, California. "If a lot of people are concerned about a symptom, that in itself is valuable information."
Although many bad reactions to drugs never get reported to doctors, people talk about what's bothering them all the time on a casual basis to their friends or online, notes computational biologist Nicholas Tatonetti of Columbia University, who was also involved with the study. "They don't really know," he says. "They're just reporting on their symptoms, which is just a normal thing that humans love to do."
So the researchers turned to Microsoft, which gets permission from many Internet Explorer users to collect Google, Bing, and Yahoo! search queries for research. Microsoft provided a database containing 82 million search engine queries from 6 million unique users from 2010. The research team looked for users who queried "paroxetine" and "pravastatin," or each drug alone, to determine if the same users also looked for "hyperglycemia" or other terms describing hyperglycemic symptoms. For example, they may have looked for "dry mouth" or its technical term, "xerostomia."
Before anyone knew about the drug combination's side effects, one out of 10 people searching for both drugs also looked for terms related to hyperglycemia—about twice as often as did people looking up paroxetine or pravastatin alone, the team reports today in the Journal of the American Medical Informatics Association. The researchers then looked for 62 other drug pairings, half known to cause hyperglycemia and the other half known not to. They found that the data-mining procedure correctly predicted whether the drug combo did or did not cause hyperglycemia about 81% of the time.
The findings add yet another resource for scientists to find clues to drug risks, Shah says, and this one can be monitored in real time. One hope, he notes, is that search engine companies could analyze the data and ship the results to FDA, which would then screen the information and alert doctors of new potential risks. "This is information, we have access to it. As a society we're sitting on it," he says. "We could use it to assist the FDA, which currently relies on the reported sources."
The new study is "exciting" because the field is so new, says biomedical informatics researcher Hojjat Salmasian of Columbia University, who was not involved with the work. "We don't know how much potential it has," he says. "But based on the results that have been published so far in this study and other similar studies, it seems like this is a very important field to explore."