Will Internet Searches Become Health Data?

“Big data” has become a big buzzword in health care, usually invoking the idea that large sets of data combined with computer-assisted analysis can reveal new patterns and correlations. While big data is often associated with genomic medicine or machine learning across vast sets of medical records and journal articles, scientists are exploring a simpler source of health data: internet searches.

The most recent example is a study published this summer in The Journal of Oncology Practice, which was conducted by Microsoft researchers Eric Horvitz and Ryen White, with Columbia University graduate student John Paparrizos. The team analyzed large samples of internet queries in an attempt to improve early detection of pancreatic cancer.

Using data from the Microsoft search engine Bing, the study began from queries suggesting a user’s recent diagnosis, such as “Just diagnosed with pancreatic cancer.” They then backtracked to view earlier searches by the same user (all anonymized) that might suggest preliminary symptoms of the disease.

Those early queries, researchers believe, can act as warning flags. As Dr. Eric Horvitz told The New York Times, “We asked ourselves, ‘If we heard the whispers of people online, would it provide strong evidence or a clue that something’s going on?’”

Remarkably, the researchers claimed they could identify between 5 and 15 percent of pancreatic cases based on search queries, with false positive rates of as low as one in 100,000. They acknowledged that false positives could lead to higher medical costs or anxiety for people who later found out they were not sick.

While five-year survival rates for pancreatic cancer are extremely low, early detection can prolong life in a small percentage of cases. The study suggests that early screening could increase the five-year survival rate to 5 to 7 percent, compared to just 3 percent for most pancreatic patients today.

These findings open up exciting possibilities for using search engine data to improve population health, but experts emphasize that we should be cautious to celebrate any kind of breakthrough.

The Microsoft study is “exciting but preliminary,” says Stephen Mooney, an epidemiologist at Columbia University who has studied the use of big data in public health. "The potential benefit is huge, but it would be easy to naively assume we know more about this than we do."

In particular, many epidemiologists worry that trends extracted from internet searches risk confusing correlation and causation.

Mooney points to Google Flu Trends as an early attempt to use Web searches as health data, which reviewers later proclaimed to have “failed spectacularly.”

Google Flu Trends attempted to track flu activity in real time based in users’ Web searches. While the project worked for a while - and produced some attention-grabbing headlines - the algorithm infamously missed the peak of the 2013 flu season by 140 percent.

Failure of these projects doesn’t negate their usefulness, it just emphasizes that big data is most appropriately deployed alongside other types of information streams. Against this backdrop, we can expect web searches to become another potential tool to inform decision-making by traditional health professionals.