Medical Examiner

Trend Spotting

Even Google can’t cure the flu.

Google Flu Trends.

This flu season, if your head starts aching, your nose gets clogged, and your body gets the chills, there’s a decent chance you’ll turn to the Internet before heading to the doctor. You won’t be the only one. Last month, to harness the powers of our collective Web searching for a cure, Google announced a cool new system called Flu Trends that tracks outbreaks by monitoring when users type search terms such as “flu symptoms,” “influenza,” or “the flu.” Google’s team even published its findings in the prestigious scientific journal Nature. They reported that Flu Trends tracks almost perfectly with data on influenzalike illnesses that the CDC obtains from doctors’ offices. And as an added bonus, Flu Trends detects outbreaks up to two weeks earlier, when people are still sitting at home sneezing into their keyboards.

“Syndromic surveillance,” or using data that track trends in patient symptoms, has been a hot topic since the Sept. 11, 2001, and the anthrax attacks. The initial thought behind the now gazillions of dollars dumped into syndromic surveillance was to use sensitive monitoring systems to identify clusters of E.R. and doctor visits by patients exhibiting worrisome symptoms, like fevers and rashes, and to create an early-warning system for bioterrorist attacks. But as researchers from the Rand Corp. point out, syndromic surveillance is not all that when it comes to detecting bioterrorism. Anthrax symptoms like fever and cough mimic the flu, making it tricky to differentiate an attack from an influenza outbreak. But what syndromic surveillance is good at is detecting trends in the flu itself. And Flu Trends, which is really a new type of syndromic surveillance, reportedly has benefits over other types of surveillance.

Flu Trends is certainly better than, where users report their symptoms. When sick, people are more worried about getting better (and using Google to figure out how) than reporting symptoms to a random Web site. Flu Trends is similar to a system in which doctors’ offices report data on patients with flu symptoms to the CDC, which publishes the results each week. But we don’t how Flu Trends compares with other types of surveillance systems, such as data from drugstores (to track purchases of cold medicine), telephone medical hot lines (to monitor patients who call in with flulike symptoms before a doctors’ visit), school absences (to know when kids are not showing up at school), E.R.s (to identify diagnoses of actual flulike illnesses), laboratory data (to confirm the infection and know the viral strain), and hospital discharge data (for information on severity of illness and intensive-care use). But Flu Trends has one clear benefit over these systems: They are costly to maintain, while Flu Trends can get similar data on the cheap.

But sometimes you have to spend a little to get a little. The problem with Flu Trends is that it doesn’t give detailed info on who is sick. Apart from knowing where their Internet connections are located, Flu Trends tells us nothing about the users. Other systems come with additional information, such as demographics (age, gender, etc.) and give more detailed information on where the sick people actually live. This “situational awareness” helps epidemiologists know how the disease is spreading in particular populations and regions.

But if officials monitored only Flu Trends, it would be difficult to sort the signal from the noise—in addition to losing critical details on who is sick. Things besides an actual flu outbreak can cause people to search the Internet for flu information. We would imagine that Flu Trends would spike on the release date for a flu-related movie—maybe Outbreak 2: Electric Booga-Flu. And what happens if a pandemic flu scare hits the nightly news? Flu Trends’ ability to detect when the real pandemic hits will be obliterated when people, including those without symptoms, start to search the Internet. Monitoring drugstore sales has the same issue: A jump in cold-medicine sales may mean a flu outbreak, but it could also mean that CVS is running a sale or that flu fear is causing people to stock their medicine cabinets.

The real question here is: So what? Why is so important to predict a couple of days ahead of time which states have lots of people with flu symptoms? Some say this information may spark health officials or doctors to produce special news bulletins, intensify hand-washing campaigns and flu vaccination efforts, let hospitals know to up their staffing levels, or even prescribe preventive antiviral meds, like Tamiflu, to high-risk people. Those sorts of prevention efforts do actually reduce the flu spread. But it is unclear what good a one-week heads-up can do when it is already clear that the flu is coming, anyway.

If Flu Trends is not the perfect surveillance system, then what is? The ideal system would let health officials know where influenza hits, how serious it is, and what the viral strain is. It would give us advanced warning to help prevent spread. Problem is, no data source has all this info—each has limitations. The flu has a predictable pattern: You get infected, you spread it to your family and friends, you start feeling awful: fever, cough, aches, headache. Then you complain to your wife or mother (whom you have already probably infected), telephone a nurse hot line, or buy some cold medicine. Ideally, you call in sick to work or avoid school, and, finally, you type “flu symptoms” into your Google browser. There is surveillance for all these activities—except for the whining, of course. Most folks ride it out at home and stay in bed for a couple of days, but some go to doctors’ offices or to the E.R, and some of those doctors will order influenza tests. But by the time we have the hospital data, the flu has not just arrived, but has infected lots of people, and interventions to reduce its spread are less effective.

Probably the best answer is for officials to use multiple data sources for syndromic surveillance simultaneously and perform graded interventions with each new bit of information. When the earliest warning system goes off (like Flu Trends or drugstore purchases), health officials might announce “Flu is here” and urge people to get more rigorous with hand-washing and get vaccinated if they haven’t already. Then, when the hospital data warn that the flu has landed in a particular location and is having an extreme effect on a population (like children or the elderly), targeted interventions can be implemented at the regional level, as with Tamiflu prophylaxis, or even considered for more drastic measures, such as temporarily shutting day care centers.

Another future trend in syndromic surveillance is sharing information between jurisdictions that now adhere to political boundaries such as states, counties, and cities. Diseases often spread across state lines without asking permission from health officials. So regionalization or nationalization (like Biosense) of surveillance may help us identify outbreaks that are not confined to one particular area.

But despite all this, it is important to keep a grounded perspective. The way to prevent yourself from getting the flu is to get your shot, wash your hands, and stay away from coughing people. When you do get it, stay at home, for goodness’ sake. And along with your daily searches for Britney Spears and government bailouts, keep on Googling for flu remedies. By doing so, you’re not just informing yourself, you’re also helping the health department. But no matter how accurate Flu Trends is, we can be certain about three things this winter: 1) Flu will come; 2) many people will get sick, and some will die from the infection; and 3) Google Flu Trends will prevent neither 1) nor 2).