BOSTON -- Demonstrating the potential power of Internet search trends in measuring the impact of health policies, researchers at Children's Hospital Boston report a strong correlation between Google searches for "abortion" and local abortion policies and rates.
The researchers analyzed one year's worth of summary search statistics for the 50 U.S. states and 37 countries, representing millions of anonymous searches. They found that states and countries with lower abortion rates and less permissive abortion policies had higher search volumes for the single word "abortion." Conversely, states and countries with higher abortion rates and more permissive abortion policies had lower "abortion" search volumes.
"One would have expected that 'the more abortions, the more searches for abortion,'" said Ben Reis PhD, of the Children's Hospital Boston Informatics Program and assistant professor of pediatrics at Harvard Medical School. "Instead, these turn out to be inversely related."
The paper, published online August 25 in the open access journal BMC Public Health (http://www.biomedcentral.com/bmcpublichealth/), discusses the potential value of augmenting standard data collection with tools like Google. "We propose Internet search engine data can also be used more broadly to study the impact of public health policies at fine geographic scales and with fewer delays in data collection," write Reis and co-author John Brownstein PhD, also a faculty member of the Children's Hospital Informatics Program and assistant professor of pediatrics at Harvard Medical School.
"With traditional epidemiology, there's no way you could collect geographically comprehensive data on people's behavior in a day," Brownstein said.
Abortion search volume was significantly higher in states where fewer than 10 percent of counties had providers, and in those with a mandatory waiting period, mandatory counseling, mandatory parental notification for minors, or mandatory parental consent for minors.
Geographically, the researchers saw lower "abortion" search volumes in U.S. coastal states, where abortion rates are higher, than in non-coastal states. Similarly, South American countries with conservative abortion policies sharply contrasted with Eastern European countries having more permissive abortion policies.
"One possible explanation for these inverse relationships is that people with limited access to local abortion services are using the Internet to find providers outside their health system or outside their region, while people with more access are able to go through standard local healthcare channels to find an abortion provider," Brownstein said.
"This exploratory analysis highlights the emerging potential of search data to benefit public policy debates," said Reis, director of the Predictive Medicine Group at Children's, adding that the method needs more validation before any firm conclusions can be drawn.
The team chose abortion as their test case for proof of concept because of widespread public interest and varied regulations among U.S. states and between countries.
In the last year, Internet search trends have become a complementary tool to help scientists and health officials detect and track emerging infectious diseases, such as H1N1 and seasonal influenza. "There's no reason this kind of search data analysis cannot be applied to all sorts of policies that affect large segments of the population," said Brownstein, director of the Children's Computational Epidemiology Group and co-founder of HealthMap (http://healthmap.org), an Internet-based aggregator of infectious-disease surveillance sources (such as news stories, government reports, and individual experts).
Search data have several advantages over traditional surveillance sources, the researchers said. Search engine data are available in nearly real-time, while traditional information can take years to compile. A similar example is Google Flu Trends, which has been validated and can detect regional influenza outbreaks up to a week or two earlier than conventional surveillance by the U.S. Centers for Disease Control, which relies on local laboratory and clinical data that is first compiled at the state level.
Other advantages include efficiency, especially in countries without strong public health reporting infrastructures, and transparency, because search data are outside of traditional health reporting channels and may be less subject to government censorship and control, the authors said.
The researchers used aggregate yearly data on search volume available at Google Insights for Search (http://www.google.com/insights/search), which provides data in relative percentages and only in geographic distributions where volume is sufficiently high to maintain anonymity. Google Search fields several hundred million search requests daily and is the most-used search engine in the United States and in the world.
Reis and Brownstein are testing their search engine data analysis methods with other health policies. "Our goal is to support an informed public health policy debate by maximizing the amount of useful information available to all stakeholders," Reis said.
One limitation of search engine data is that it may not be a representative sample of the general population, especially in poorer regions where fewer people may have Internet access, also known as the digital divide, the researchers said.