Ah, the “disease of the week” syndrome. A very interesting, if unintended, side-effect of the popularization of epidemiology. This unpleasant condition can be avoided by keeping a few basic principles in mind.

1. Even a study in the most prestigious peer-reviewed journal may have design problems.

There are more forms of bias than you can shake a stick at. Epi folks are pretty adept at avoiding them as best as they can, but it can be pretty difficult to dodge every single one of them, especially on the budgets that most studies are operating under.

Confounding, considered by some a form of bias, is another can of worms. It means that two factors may appear associated, but the apparent relationship is actually caused by a third variable that is related to both of them. To use an epi-prof stand-by, initial studies 50 years ago found that drinkers had a higher incidence of lung cancer, compared to non-drinkers. Once they looked into that, they found that most of those drinkers also smoked, of course. Drinking was a confounder obscuring the relationship between smoking and cancer. Epidemiologists can control for confounding during analysis, but there is a possibility that confounding variables may be missed.

Small study size also falls under this heading. Studies with few participants can be helpful as a pilot or when operating on a shoestring, but they often won’t have enough power to show statistical significance—even if one does, you should look at a study that has an ‘n’ in the double digits pretty critically.

2. Correlation is not the same thing as causation.

The number of pirates has decreased over the past two centuries. At the same time, global temperatures have risen. Oh noes! Lack of pirates causes climate change! (Can I get a “Ramen“?)

Just because two things are associated does not mean that one causes the other. A risk factor and an outcome may be correlated without the risk factor actually causing the outcome.

Wikipedia has a fab article on all of that here.

3. One of the key criteria for causation is repeatability of findings.

One study doesn’t mean all that much. Even beyond the whole design problem issue, chance can and does interfere with getting nice and neat study results.

Of course, if the study is a follow-up on drug licensing and shows that 50% of folks taking Drug X have a stroke, that drug will be off the shelf before you can blink. When you’re dealing with life and death, it pays to be a bit more conservative.

For our purposes on this blog, though, don’t start sweating until you see several nice, peer-reviewed studies showing the same kind of association.

*Reading rec: for a good, short overview of Epi that will do wonders for helping make you a smart consumer of health info, check out PDQ Epidemiology by Streiner and Norman.