Pioneering Ideas

Do you flu Yahoo!?

Google.org may have grabbed headlines last week with the announcement that search term activity on its engine may forecast real-life flu activity, but grantees Phil Polgreen and Forrest Nelson are releasing the first peer-reviewed journal article that documents this trend.  Together with partners at Yahoo! Research and Harvard, they published a study in Clinical Infectious Diseases that finds that frequency of Web searches on flu and influenza (excluding searches related to avian flu, vaccines and other outlier terms) predicted increases in cultures positive for influenza one to three weeks in advance.  Their study, and a quote from Phil, were cited in the New York Times cover story on the Google flu trends service.

Given the ultra-real-time nature of Web information, and the fact that 8 million people search for health information every day, it’s perhaps not surprising that this is a potentially rich avenue for exploring whether people’s hunt for information online signals their health concerns, experiences, conditions, behaviors, expectations and even outcomes.  This has sparked privacy concerns -- also not surprising.

Polgreen, Nelson and colleagues examined the relationship between influenza culture data and Yahoo! searches at the national level, breaking searches down in to 9 Census regions and tracking activity over time.  They reported a statistically significant relationship between intensity of flu-related queries and levels of flu cases and even deaths.  The methodology has some limitations due to the fact that data only go back four years and other reasons, but it’s another potentially valuable tool in public health professionals’ toolbox.  And given the reality that current CDC surveillance activities identify disease activity only as or after it occurs, any advance jump on an outbreak seems like a good thing. 

This is what drives Polgreen’s and Nelson’s other Pioneer-funded work to test the use of electronic prediction markets to forecast domestic and avian flu activity.  The premise being that more knowledge, gained more quickly, fuels wiser policy and resource allocation decisions, better prevention and treatment actions, and hopefully less incidence of disease with less harmful effects. 

Search term surveillance may yield trendspotting clues for public health officials fighting emerging and reemerging infectious diseases, changes in phenomena tied to chronic illnesses or trends in STD infections ahead of official reports of disease activity.  It’s fascinating to think of the potential that exists in mining aggregate data from the ways we digitally engage in the world in the course of our everyday lives, and how that collective information may be applied to improve public health and health care practices.  This is an area that Pioneer might explore more down the road – we’d like to hear what you think about this, and how it might be applied.

How am I feeling...next Wednesday?

What if you knew what was going to happen in the next two days? I would bet on the lottery number that I knew was coming up and, perhaps, avoid some of the mistakes I almost certainly would make.

 

This might sound a bit like science fiction but researchers at the Center for Future Health at the University of Rochester are beginning to demonstrate that people with heart disease will be able to know their health status future and be able to take action to change it. Under that scenario, I would be able to take action on Thursday so that I’d be able to dance at my daughter’s wedding Saturday night.

 

This is not as far fetched as it might sound. Engineers have been using sensors to monitor the “health” of machines so that they can predict and avoid failures, and they have teamed with medical professionals and computer scientists to see if they can develop comfortably wearable devices that can measure and interpret signals that can define a health index that adapts as it learns more about the individual who is wearing it. Such a system can translate those learnings into actions that can help the end user get better or alert others that other actions need to be taken.

 

Early results are promising. Working with a small sample of patients and healthy people, they’ve found that they can use sensors that are easy to wear, collect appropriate data and can transfer readings to a smart phone. They can distinguish between healthy and sick people and detect changes that seem to presage future events. They’ve even been able to build a wearable prototype as a proof of concept. The patients (from varied backgrounds) that have volunteered to be part of this project are all engaged and enthusiastic.

 

However, there are a huge number of questions that need answers. Data are being collected in a way that they haven’t before so interpretation is not yet validated or refined. It’s still unclear what needs to be collected and how good a job one can do in predicting future events. Does the methodology generalize to other conditions? How will the health profession be engaged and participate? There are certainly more questions than answers at this point.

 

But we have learned that the field of machine monitoring and prediction can teach us something about health management. And maybe, someday, I’ll discovery the winning lottery number a day before it’s picked.