Researchers looking for previously undiscovered drug side effects are turning to web searches for answers.
When drugs are approved by the Food and Drug Administration (FDA), they are vetted for potential side effects, and drug makers are required to divulge these on their products’ labels. But not all side effects emerge in the short term studies that manufacturers conduct, so many only come to light when hundreds of thousands, and even millions of people start using a medication. And what better way, a group of researchers from the Stanford University School of Medicine and Microsoft Research decided, to suss out some of these unexpected side effects than to turn to where people are most likely to report, share and ask about them — the internet.
The researchers combed through a year of web search history from 6 million Internet user volunteers. Using automated tools, the scientists were able to to mine anonymous data from 82 million drug-symptom and condition searchers made by the participants who agreed to let the users install a Microsoft plug-in to monitor their history.
The team was rewarded with a previously unreported interaction between two commonly prescribed drugs — paroxetine (Paxil), an anti-depressant, and pravastatin (Pravachol), a statin that lowers cholesterol. About 5% of the people who searched for paroxetine also searched for a descriptor of hyperglycemia, or high blood sugar, which can trigger symptoms of blurry vision and dehydration. Less than 4% of people searching for pravastatin also searched for hyperglycemia symptoms. However, among those who searched for both drugs—which suggested they were taking both—10% typed in hyperglycemia-related searches. That hinted that users of both medications were experiencing a drug-drug interaction that previous studies had not identified, since drug approvals don’t include exhaustive studies of every potential combination of drugs that patients may take.
But the web searches could serve as an early harbinger of potential problems when certain drugs are combined, as well as reveal new side effects of individual medications that did not emerge during drug development. While the FDA currently tracks this information, as part of its Adverse Event Reporting System, the system relies on voluntary reports from doctors, pharmacists, patients and drug companies, and these reports often lag behind initial reports which in many cases need immediate medical attention. (After mining reports sent in to the FDA, the study authors discovered hyperglycemia was a side effect of the joint use of paroxetine and pravastatin, confirming their search-based findings.)
To further validate their results, the research team also studied searchers surrounding 31 drug interactions known to cause hyperglycemia, and 31 safe drug interactions. They did find more hyperglycemia-related searches for people using the drugs known to cause problems, but also a high rate of hyperglycemia searches among patients using drugs with no known related side effects. The scientists believe that in order to decrease the amount of false-positive results, they need to combine data from other sources such as social media, medical records and patient support forums.
The idea of exploiting the vast amount of information embedded in internet searches has already been used in improve flu tracking. The Centers for Disease Control (CDC) has created a FluView Influenza-Like Illness Activity app that shows geographic flu-like activity and trends based on information reported to the agency from health care providers. Google Flu Trends uses search terms the search engine has determined to be indicators of flu activity, such as fever, influenza and flu, and aggregates that search data to estimate flu activity by state. Similarly, the healthcare product company Help Remedies launched it’s lighthearted Facebook app called “Help I Have The Flu,” which scans through users’ Facebook accounts to identify which of their friends is most likely to give them the flu.
TIME covered the app during the peak of flu-season:
The app looks for friends’ comments and statuses relating to flu activity, like “coughing” and “sneezing” and even tracks trends such as late night posts, since lack of sleep is considered a risk factor for getting sick. The result? A list of risky individuals from whom you can steer clear. Or, perhaps just as important, blame if you come down with a bug. After all, nothing is more comforting than being empowered to point fingers.
Of course, the scientists acknowledge that an individual’s search data doesn’t always reflect an accurate picture. A person may search symptoms based on a news story they read or because a family member is having issues with their medications. Still, the authors write that they see “a potential public health beneﬁt in listening to such signals, and integrating them with other sources of information.” Like the traditional canary in the coal mine, these indicators could be the first signs of potentially harmful side effects or drug interactions that could be avoided if detected sooner rather than later.
The study is published in the Journal of American Medical Informatics Association.