The nation’s leading search engine has quietly begun giving researchers access to its data troves to develop analytical models for tracking infectious diseases in real time or close to it. UIC is one of at least four academic institutions that have received access so far, along with the U.S. Centers for Disease Control and Prevention.
Researchers can mine Google data to identify searched phrases that spiked during previous upticks in a particular disease. Then, they measure the frequency of those searches in real time to estimate the number of emerging cases. For instance, a jump in gonorrhea might coincide with more people searching “painful urination” or other symptoms.
“If this works, it could revolutionize STD surveillance,” said Supriya Mehta, an associate professor of epidemiology at the UIC School of Public Health.
Search trends can be broken down by city and state, weighted according to their significance and combined with other data sources to give a snapshot of where disease is spreading well before public health agencies report the number of verified cases.
“We’re hoping for a bit of creativity to flourish around this,” Christian Stefansen, Google disease trends senior engineer, said during a visit to UIC last month, where he spoke to about 100 people about lessons Google learned in its attempts to mine data for public health. “There’s no shortage of communicable diseases, sadly.”
Sexually transmitted diseases are a growing threat, worsened by the progress of antibiotic-resistant strains, according to the CDC. The agency reported in November that STDs, including chlamydia, gonorrhea and syphilis, all increased in 2014, with chlamydia reaching a record of more than 1.4 million new cases. Diagnoses are highest in 15- to 24-year olds, an age group where technology use also is high.
Public health advocates have long salivated over the idea of using Internet searches to track all sorts of diseases but were limited to the publicly available Google Trends tool. It restricts the number of phrases that can be tracked and does not report searches that fall below certain undisclosed volume thresholds.
Google invited infectious disease researchers to apply for unrestricted access to search data in August as it disbanded its own real-time tracking tool, Flu Trends. Launched in 2008, Flu Trends broke ground but presistently overpredicted cases, and Google came under fire from some researchers for not disclosing its methodology. According to a paper published in Science by some independent researchers, Flu Trends stumbled because it used search terms that correlated with flu season but not actual cases of the flu and failed to adjust after Google introduced “search suggest” and other features to guide users to information.
Google is the most commonly used search engine in the U.S., with a 63.9 percent market share in October, according to comScore, a Reston, Va.-based analytics company.
Google searches can be tracked by city, providing more refined data than the national and multi-state data reported by the CDC. “It’s a phenomenal data feed to work with, and there’s a lot that can be done with it from a research standpoint,” said Jeffrey Shaman, an associate professor in environmental health sciences at Columbia University’s Mailman School of Public Health, which was given access to the data.
But no matter how great it is, some researchers say they can’t rely on Google alone. Take flu, which is furthest along of any real-time disease-tracking effort, with at least nine teams working with the CDC on 12 forecasting models for the current season. This fall Boston Children’s Hospital and Harvard Medical School launched HealthMap FluCast, a tool that gives one- and two-week predictions by incorporating Google searches with the CDC’s weekly surveillance reports; electronic medical records from athenahealth; and Flu Near You, a website of patient-reported data.