Monday, August 16, 2010

Science in the Herald

From the “Healthy Living” section:
People who sleep more or fewer than seven hours a day, including naps, are increasing their risk for cardiovascular disease, the leading cause of death in the United States, a study says.
“More or fewer than seven” hours of sleep a day – that just about covers us all, doesn’t it. The Stratford Theory of Numbers strikes again. Let’s look more closely.

The study was published in Vol 33 issue 8 of the journal Sleep and is based on research by Anoop Shankar, an associate professor in the West Virginia University’s department of community medicine. The abstract of the article “Sleep Duration And Cardiovascular Disease” says: 

Previous studies have shown that both short and long sleep durations are related to increased likelihood of diabetes and hypertension. However, the relation between sleep duration and cardiovascular disease (CVD) is not clear. We examined the hypothesis that compared with sleep duration of 7 hours, shorter and longer sleep durations are independently related to CVD. [. . . ]
Conclusion: Compared with sleep duration of 7 h, there was a positive association between both shorter and longer sleep durations and CVD in a representative sample of US adults. These results suggest that sleep duration may be an important marker of CVD.
But they are not talking about five or 10 minutes too short or too long. No, they are talking about a couple of hours. As WVU’s website puts it: 
If you’re sleeping less than five hours or more than nine hours, you could be putting yourself at an increased risk for heart disease . . .
The rest of the article – which was syndicated from AFP – is fine but that first paragraph is not just innumerate, it is downright misleading. I suppose they just whacked it in without reading it. Or perhaps one of the Herald’s sub-editors “improved” it. The headline is a doozy too: 
Seven hours the magic number in the sleep game
Nope. Could be six for older people, could be eight for younger. Seven is the median, the middle of what seems to be the normal healthy range. Nothing magic about it.

The Waikato Times, which ran the same story days later in its “Wellbeing” section, used this headline: 
Snoozing way to health
Nope. “Snoozing” is not mentioned in the story at all. And of course nowhere is the idea canvassed that maybe people sleep too little or too long because they are, you know, unwell. With something like, perhaps, cardiovascular disease or similar. That whole correlation-not-being-causation thing.


Max said...

The Herald is, of course, a terrible newspaper and it's true that "most journalists are innumerate" but it might be worth reading abstracts - or, just to be shocking, the whole article (!) - a bit more carefully to avoid the same criticism in future. The abstract says:

"the multivariate odds ratio (95% confidence interval) of CVD was 2.20 (1.78, 2.71), 1.33 (1.13, 1.57), 1.23 (1.06, 1.41), and 1.57 (1.31, 1.89) for sleep duration ≤ 5 h, 6 h, 8 h, and ≥ 9 h"

- that is, that there's a materially greater level of cardiovascular disease in those sleeping 6 or 8 hours, contrary to what is said here; and, further:

"association persisted in subgroup analyses by gender, race-ethnicity, and body mass index categories. Also, similar associations were observed when we examined myocardial infarction and stroke separately"

- that is, the association was common across the group, even differentiated for personal characteristics, and again contrary to what is suggested here.

Stephen Stratford said...

Fair enough - I was taking a shortcut so as not to get to statsy (and it's a long time since I have dealt with confidence intervals so am not so confident with them these days).

But the WVU summary says: "Shankar said that those who sleep less than five hours are twice as likely to develop heart disease. On the opposite end of the spectrum, those who sleep nine hours or more a day are one-and-a-half times more likely to develop heart disease than those who sleep seven hours a day."

That is, the real risks, or rather correlations, are at the extremes of the range studied, not around the middle.