The perils of not being fat.

[uglybeech]

Thought this might interest some of you. I certainly found it entertaining. The Journal of the American Medical Association came out with a 12 question longevity test you can use if you’re over 50 to predict whether you’ll drop dead in the next 4 years. Turns out if you are not overweight (if your BMI is less than 25) you are **twice ** as likely to die than if your BMI is over 25. (Hazard ratio = 2.0). A BMI of 25 is the cutoff between normal weight and overweight, btw.

Here’s the AP story on it with a link to the test. And here’s the original JAMA article.

Have at it.

[uglybeech]

Sorry - I should be clear. JAMA didn’t come out with the test. It was reported this week in JAMA I should have said.

[cerberus]

Well, it’s not at all clear that being grossly overweight, say having a BMI > 35 isn’t a problem.

And there may be a survivor effect: significantly overweight people may have been “culled” in their late forties and early fifties by Heart Disease, CVA, Diabetes Mellitis or Cancer.

[uglybeech]

@cerberus

Well, it’s not at all clear that being grossly overweight, say having a BMI > 35 isn’t a problem.

And there may be a survivor effect: significantly overweight people may have been “culled” in their late forties and early fifties by Heart Disease, CVA, Diabetes Mellitis or Cancer

Yeah, I wondered about the extremes too. Here’s what the JAMA article has to say why they settled on having just two categories (above 25 and below 25):

We dichotomized BMI because our multivariate analysis showed that when other variables were considered, more extreme values of BMI did not improve the performance of our model.

I read that as saying all other things being equal, a very high or very low BMI didn’t change the results much. But make of it what you will. In any event it sounds to me like they did analyze the extremes separately and didn’t find it a useful predictor (at least not on it’s own). The fact that they analyzed the extremes separately would also render any culling effect more or less moot, as it would only be an issue if they never bothered to break them out.

[Lightray]

@uglybeech

I read that as saying all other things being equal, a very high or very low BMI didn’t change the results much. But make of it what you will. In any event it sounds to me like they did analyze the extremes separately and didn’t find it a useful predictor (at least not on it’s own). The fact that they analyzed the extremes separately would also render any culling effect more or less moot, as it would only be an issue if they never bothered to break them out.

I would read that as meaning that the responses attributed to other variables so overwhelmed the response from BMI that the additional predicitive accuracy for including BMI was negligable.

I’d suspect, since they did multivariate analysis, that BMI didn’t turn up as contributing to any of the significant principal components.

I’d also suspect that their data set may not have included a sufficiently large sample of BMI extremes for the analysis to have flagged it as significant. Could be that many of the BMI extremes had already been eliminated from the over-50 populations they studied.

[uglybeech]

@Lightray

I would read that as meaning that the responses attributed to other variables so overwhelmed the response from BMI that the additional predicitive accuracy for including BMI was negligable.

I’d suspect, since they did multivariate analysis, that BMI didn’t turn up as contributing to any of the significant principal components.

No - BMI did turn up as highly independently predictive (hazard ratio of 2.0) for the normal or underweight. BMI extremes didn’t turn up as independently predictive.

[uglybeech]

Also, I figure if they didn’t have enough people at the extremes of weight, I’d expect they’d have said so. Not that they had analyzed the effect of extremes and found it didn’t help their model.