In the comments to the recent post on BMI, commenter Colst pointed to another study of mortality and BMI that found significantly higher risks for overweight people. Today, I see that Kevin Beck at Dr. Bushwell’s Chimpanzee refuge has a post describing what I think is the same study, with the title Risk of death much higher in overweight and obese. Which is true, if you look at the data in the right way.
Kevin posted a bunch of graphs from the study, and I’ll excerpt two of them to keep things readable. The first is the relative risk of death for all the men in the study, as a function of BMI:
That seems to put the minimum risk of death at a BMI of around 26, slightly into the “overweight” category. So where does the “much higher” come from? (Below the fold.)
The first graph is just a compilation of all the data for all the men in the survey. That by itself doesn’t mean all that much for researchers, so they break it down by a lot of different factors. Kevin shows a graph of the risk for different age groups, which doesn’t make much of a difference, but the interesting one is the graph showing people sorted by smoking status:
(This is the only graph to show any real separation between different groups.)
This is really interesting, and kind of weird. The minimum risk of death is around a BMI of 26 or 27 for anybody who ever smoked (whether they smoke now or not), but just under 25 for men who never smoked. What’s sort of strange here is that there’s really no difference at all between current and former smokers.
If you cut the data even further, a graph I’m not going to bother to show, and look at the relative risk to men who never smoked as a function of their BMI at age 50, the minimum shifts over a little more, to between 23 and 24 (still on the high end of “normal”). The striking thing about this graph is that the risk increases much more rapidly for this group, and indeed, is almost double the minimum for people at the border between “overweight” and “obese.” Hence, “much higher risk.”
(Kevin also posts graphs for women, which are generally similar, though all the curves shift to the left a bit.)
Is this sort of data slicing legitimate? Well, it’s pretty typical of the analysis of medical studies– Bob Park has a great riff on the studies that claim to show a connection between power lines and cancer. The numbers in the sample are large enough that you wouldn’t expect it to cause any really weird problems, and age and smoking status are well-known risk factors that ought to be controlled for. So there’s nothing really inappropriate about doing these kind of cuts.
Of course, making really sweeping claims based on these cuts is very slightly dodgy. I mean, the relative risk of death really is much higher for male non-smokers who were classified as “obese” at age 50 than their thinner counterparts. But then, the difference between the high end of “normal” and “obese” for a former smoker is much lower– maybe 30%. So, if you plan to be fat when you’re fifty, start smoking, and then quit. Or something.
(Actually, don’t– the increase in mortality from smoking is probably significantly greater than the effect of weight. These graphs plot relative risk only, and if you look at Kevin’s graphs, you’ll see that there were close to ten times as many dead smokers as non-smokers…)
In the end, there’s probably something for everyone in this study. No matter how you cut the data, the risk of death is definitely higher for people in the “obese” range, so Kevin’s happy. And even the most impressive cut through the data shows that the minimum risk is to the high end of “normal,” which fits with my contention that the target range is a little lower than it should be.
So, um, cookies for everyone!
Damnit, this means I didn’t have to quit a decade ago, since apparently it doesn’t make much difference.
A real and possibly unavoidable problem is that data that serve populations well often have little application to individuals, and this is especially true with the BMI. Obviously choosing 25 as the cut-off for overweight is as arbitrary as choosing any other number; were I to pack on 35 pounds of muscle and fat — which would be difficult to do given my genetic make-up, but still possible with some sturdy work — I’d go from what most would consider very skinny (BMI of right around 20.0) to “overweight.” And many of my fellow runners, who could outpace your average skinny sedentary person of the same age and sex in virtually any physical task of your choosing, would also be considered overweight by BMI standards.
Still, the majority of Americans considered obese are not built like NFL fullbacks; pointing out hypermuscular “obese” outliers is a bit of a strawman, akin to pointing out the 100-pack-year smokers with no manifest health problems. It seems obvious that very obese people, regardless of where one plunks the arbitrary cut-off BMI value, suffer all sorts of health woes disproportionately. This is much less clear in the high-20s BMI range and, taken together, studies relying on this parameter are certainly contradictory enough so that lifestyle recommendations and public policy decisions founded on the stat should be avoided.
So how is the data corrected for age. If it is relative risk, does that mean that risk of death for 50 yr old male is compared over these weight classifications, and that for 40 yr old male and so on. Clearly we know that weight and BMI increase with age in the US male population. We also know that risk of death increases with age.
So to reach my ideal weight I either have to gain a couple pounds and start smoking or lose 10 and stay a nonsmoker 😉
When God hands you still bottoms, sell asphalt.
http://english.people.com.cn/200608/24/eng20060824_296181.html
The best the Irish could do was wakes.
A real and possibly unavoidable problem is that data that serve populations well often have little application to individuals, and this is especially true with the BMI. Obviously choosing 25 as the cut-off for overweight is as arbitrary as choosing any other number;
Sure, if you just pick numbers.
My point is that there is, in principle, a way to make this non-arbitrary. Namely, doing studies like the one cited here, and setting the “normal” range based on where the health problems are minimized. For graph “A” above, you’d want it to be centered on something like 26, plus or minus a reasonable amount. A “normal” range of 24-29 would pretty much cover the region below the horizontal line, for example.
The choice of “normal” as 20-25 is fairly obviously an arbitrary choice based on the fact that people like multiples of five. But it doesn’t need to be that arbitrary– you could do it scientifically, and the only arbitrary element would be the increase in risk used to set the thresholds.
OK. But just to put the discussion in perspective.
According to the data from the CDC Vital Statistics System about 550 out of 100000 males aged 45-55 died in the year 2003.
Of those, about 105 died either by accident, cirrhosis of the liver or suicide. I seriously doubt that any of these causes will be reduced by lowering the national BMI average.