On Social Networks and Study Recruitment
Inspired by a question posted by a friend of mine on Twitter, I’ve been pondering this article and the use of social networks as a tool for public health research. Since Twitter’s character cap means subtly and nuance are never, ever going to be their strong suit what follows is a longer musing about the question here.
First a disclaimer: I have been fortunate (unfortunate?) enough to have escaped ever having to directly recruit subjects for a study. All of my research involves either already collected data being re-used for further research purposes, or mathematical modeling/simulation studies, which use entirely virtual subjects. So this is the musing of a random epidemiologist who thinks about study design on occasion, not a battle-worn veteran of the subject recruitment wars.
Online communities represent a really great way to drum up interest for something, and access concentrated groups of individuals. One only needs to consider my post on the traffic on this blog. Two posts on message boards for a relatively small hobby have driven 8,765 pageviews to this site.
Its not hard to see how patient-driven social communities online could represent a spectacular resource for studying rare diseases. It’s extremely common to over-sample individuals with a disease using other types of resources (registries, hospital records, recruiting specialists to enroll their patients), and there’s ways to handle that kind of oversampling statistically. Heck, the entire case-control design is devoted to oversampling cases. So yes, there is some awesome potential for social sites to be used for study recruitment. As a bonus, its the internet, so by and large reaching these people is at a much lower cost than say mailing recruitment information, or advertising. And in this day and age, anything that’s easier on a study budget is a good thing.
I do however have two major concerns about its use, neither one of which is inherently fatal, but do warrant some caution:
- Social sites over-sample cases, to be sure. But they also over-sample other covariates in a study – covariates like race, income, and gender. It is entirely possible to bias a study by sampling cases from a population with one demographic structure, and controls from a population with another (using non-social media recruitment methods). I’ve done a little research on the reverse problem, perfectly sampling controls and then sampling cases from a population with an unusual demographic structure (in this case, people with landline phones), and it can substantially bias your results. This kind of problem is something that should absolutely be approached with caution.
- “The Echo Chamber”. Some communities…don’t have the best relationship with the medical community. The Autism-Vaccine link people come to mind immediately. A recent survey, properly eviscerated on Respectful Insolence has generated 7,825 responses. The study is biased, terrible, and has about as much scientific rigor as a badly conceived middle school science project. But numbers of participants like that give it weight, let those looking not for empirically backed truth but a confirmation of their own beliefs say “A study with nearly *8000* people…”.
To be honest, the first problem concerns me more, as its far more subtle. The second problem…is a long-running one, and while its enabled by internet communities, its hardly going to go away by everyone being a little more careful with their methodology.
Filed under: Epidemiology | Leave a Comment