Self-Reported Syndromic Surveillance

I've put off linking to because it got pickup on Boingboing and elsewhere. But I thought again after I realized that no one I'd read had applied a little Epidemiology 101 to the site. So it's a cool idea: using a Google Maps mashup, let the hoi polloi post when they're feeling ill, checking off a few symptoms and other info, and plot it by location. Get enough contributors, and you have an interesting surveillance map of possible infectious outbreaks. Add a little more Web 2.0 magic - tagging, analytics - and you've got perfect blog fodder. The idea is basically a self-reported version of what's called 'syndromic surveillance,' which is basically the effort to systematically track disease outbreaks on a geospatial level, by pulling from a host of data inputs - hospital admissions, drug store purchases, school illnesses, and so forth. It's been used for infectious disease for years, but it's gotten attention more recently because of it's utility in possible terrorist or bioterrorist events. That, plus the fear of avian flu, make it quite a trendy concept.

So on the face of it, is especially enticing, because it combines that thinking with a dash of Web 2.0 special sauce. But I knew there was something a bit, well, imprecise about it. And that comes down to bias. Quick refresher: bias - the notion that systematic error can be built into a study design - is one of three fundamental mistakes in epidemiological studies, along with confounding (the unmeasured influence of an extraneous factor from the study) and random chance (the likelihood that the study result was due to chance rather than any true association). Basically, succumbs to one of the most common forms of epidemiological bias - reporting bias. This bias holds that those who are aware of being ill are sometimes more (and sometimes less) likely to pipe up and participate in a study than those who aren't ill - therefore giving a disproportionate measure of the true level of illness. Unless you're measuring the true rate of illness against a true rate of non-illness in a population, you're going to have a biased result.

So suffers from the same non-statistically valid flaw that those self-selected polls on do - there's no way to determine whether the results are in any way an accurate reflection of the actual population in question.

There's also a host of other problems with the site - starting with its categories of illness (stomach ache and muscle ache are hardly medically exact categories). So while I'm inclined to give it props as an experiment and a fun thing to have around, I sure hope it doesn't get overblown.

Thomas GoetzComment