Two Things That Get on My Nerves, Part II

Saturday, June 28, 2008
Confusing Correlation and Causation

Recently, a paper was published that examined the association between sleep duration and the risk of death. Ferrie et al. showed that in their study population, subjects who slept either more or less than 7 hours a night had an increased overall risk of death. Here's how it was reported in Medical News Today:

Too Little Or Too Much Sleep Increases Risk Of Death
And here's a gem of a quote from one of the study's authors (excerpt from the article above):
In terms of prevention, our findings indicate that consistently sleeping around 7 hours per night is optimal for health and a sustained reduction may predispose to ill-health.
There's only one small problem: the study indicated no such thing. What the study showed is that people who sleep more or less than 7 hours tend to die more often than people who don't, not that the lack or excess of sleep caused the increased mortality. Have you ever noticed that you sleep more when you're not feeling well? Have you ever noticed that you sleep less when you're stressed? Could the increased mortality and sleep disturbances both be caused by some other factor(s), rather than one causing the other? We don't know, because the nature of the study doesn't allow us to answer that question!

The message the public ends up hearing is that no matter what feels right for your body, 7 hours of sleep is the optimum for health. Even though you'll have to go to work with bags under your eyes, feeling like crap, it's healthy. Even though you have the flu, you'd better not sleep more because it might give you a heart attack. I find that conclusion difficult to swallow.

The only way we could say that 7 hours of sleep is the healthiest amount (for the "average" person), would be to do an "intervention study", in which the subjects are manipulated rather than simply observed. Here's how it would work: we would take a large group of people and randomly assign them to either 5, 7 or 9 hours of sleep a night. We would then look at mortality over the course of the next few years, and see who dies more.

Intervention studies are the only way to establish causality, rather than simple association. At the end of our study, we could rightfully say that X amount of sleep causes an increase or decrease in mortality. Obviously, these types of studies are challenging and expensive to conduct, so it's tempting to over-interpret observational studies like the one I mentioned initially.  These studies are useful, but should be taken with a grain of salt.

This has to be one of the gravest, most frequent mistakes in the realm of health research and reporting. So many of the health recommendations we get from the media, the government, and even scientists are entirely based on associations!