Patients' Internet search patterns can help alert medical officials to possible drug interactions, according to a paper published at the Journal of the American Medical Informatics Association.
Researchers from Microsoft, Stanford and Columbia University conducted a large-scale study of Web search log data gathered during 2010, paying particular attention to patients searching for information about the antidepressant paroxetine and the cholesterol-lowering drug pravastatin.
The pairing later was reported to cause hyperglycemia, but the researchers were looking for clues beforehand that might have suggested such a problem based on patient queries entered into Google, Microsoft and Yahoo search engines.
The researchers were surprised at the strength of the "signal" they found in the search data, and say it could be valuable to the Food and Drug Administration in tracking adverse drug events, reports The New York Times.
"There is a potential public health benefit in listening to such signals and integrating them with other sources of information," the authors wrote.
The automated data-mining techniques used were similar to that used for Google Flu Trends, which provides a heads-up to health officials about patterns of illness. The new research was based on 82 million individual searches for drug and health information among people who agreed to have their search histories studied.
First, the researchers looked for the words paroxetine and pravastatin, then searches for both together, then words reflecting hyperglycemia including about 80 symptoms such as "high blood sugar" or "blurry vision." They found that people who searched for both drugs were more likely to search for symptoms of hyperglycemia than those who merely searched for information on one drug--and they would do so in fairly short order.
The researchers now want to combine that information with other sources of information, such as behavioral data or posts on social media. Patient privacy, however, remains a concern in aggregating data in this way.
Researchers continue to search for better ways to identify drug interactions, including a computer algorithm created at Stanford that allows doctors to differentiate between drug-related adverse events in patients and adverse events from another illness. Other researchers are looking to Twitter and Facebook for real-time information about the spread of disease and how to contain it.
Meanwhile, researchers from Columbia University and the National Center for Atmospheric Research have developed a model combining Google Flu Trends and weather forecasting techniques that can predict the peak of flu outbreaks in specific areas more than seven weeks in advance.