From a news story on a recent paper: Pramil N. Singh, Ella Haddad, Serena Tonstad, Gary E. Fraser. Does Excess Body Fat Maintained After the Seventh Decade Decrease Life Expectancy? Journal of the American Geriatrics Society, 2011.
Several points:
1) The authors are using a parametric model. This can drive results at the boundary in ways that do not describe the data. So I'd not take the flattening out of the relative risk cure for low male BMI that seriously, unless they have more data than the graph suggests for BMI < 21 for men.
2) You almost never see a minimum relative risk for BMIs this low, either for men or women, in US (or other country) observational studies. Most studies show a minimum mortality risk for BMIs around 25-27, i.e. light overweight.
3) There are indications from older European data that once upon a time, i.e. before 1970, the minimum mortality BMI for men was under 25 and much closer to 22, especially for men who remained at this BMI throughout their adulthood.
4) These high minimum mortality BMIs don't line up with our prior beliefs. The most common hypothesis, which is also mine, is that in the modern US population lots of low BMI person are only low BMI due to health issues and have gotten there after having been non-low BMI and then developing a disease that brings down their BMI and then, for some of them, kills them. In the past this fraction of the low BMI population wasn't large enough to drive up the minimum mortality BMI, but now, weight the greater mean BMI of the population, this fraction of the low BMI population is starting to drive results.
5) But there is little empirical work that investigates this intuition. This paper looks at, so far I can tell without having access to the paper itself, only people who never lost (much) weight (for a sustained period -- hm, this may just mean they were similar in weight at two points 17 years apart, i.e. what would be 150-160 lbs for me). I don't know they handle people who keep gaining weight though...what's the X variable for somebody who has a BMI of 21 at 20 years, 25 at 35 years and 30 at 45 years? In any case, I'd like more research on the life course of weight and mortality, ideally with some indication if weight is targeted by the individual or not (i.e. is weight loss intentional?).
I'm currently around 153-154 lbs., not getting under 152 lbs., so around a BMI of 21.75, most likely due to my running being slightly under 20 miles/week right now, though I am now trying to ramp up slowly (as well as acquiring a bike).