Monday, November 26, 2007

Fat facts which nobody can deny

The Hoofnagles are jolly good fellows, debunking those cranky guys who deny that HIV causes Aids and that vaccines save lives and the like, uncovering the tactics they use but they happen to use one of such tactics themselves, cherry picking.

Skepticism is a good thing and vital for science and health, the problem is that those cranks, HIV denialists and antivaxers boast to be the real skeptics. The Hoofnagles try their best debunking those debunkers. Mark Hoofnagle has criticized me for having accepted Sandy Szwarc to Skeptics' Circle. I did so because I share her skepticism in the obesity issue which is an important one in these times. I did so because I think this is sound skepticism, based on facts. And this does not mean that I agree with all statements on her blog.

Mark Hoofnagle has attacked Sandy for a post suggesting that case numbers in various diseases have been increased just by altering the diagnostic criteria. Again, I am not going to comment on non-obesity issues over there because I have not (yet) studied the details. The only point I am going to discuss here is the definition of body mass index (BMI) 25 as general upper limit of normal weight and the claim that every weight beyond that limit is to be called "overweight" and is bad for health.

Such obesity skepticism, in Mark's view, is cranky, and he cites two large studies to back his view. The first one, by Calle and co-workers, has been published in the New England Journal of Medicine 1999. The problem with such old data is that they may be outdated meanwhile because, as has been shown by Flegal and co-workers, the link between obesity and mortality has gotten weaker over time. Therefore I am dealing with the newer one, by Adams and co-workers, published 2006 in the same journal. This study is really impressive; more than half a million men and women have been assessed and followed up to ten years. There are many beautiful cherries to pick from this study. Mark has picked this one.

BMI and mortality picked I must admit this is one of the most beautiful J-curves I have ever seen. It shows the relative risk of death for men aged fifty who never have smoked, plotted against BMI. Mark shows this curve to support BMI 25 as an upper border to bad health. Does it really show this? What I see here is a nadir (lowest value) at BMI 23.5 which could be called the ideal weight for this selected group, only 18 percent of the whole cohort. The upper border of the normal weight range must be defined by the mortality value. For BMI 25 we get a relative risk of less than 1.1 in this curve. Symmetry must be postulated, and we look at the lower BMI value with a relative risk of less than 1.1 which is the case at around BMI 22. In other words, with a postulated upper BMI limit of 25, the corresponding lower limit is 22, leading to a normal weight span of only three BMI points. Such a narrow definition of normal weight cannot make any sense. Rather I would accept a lower limit of normal weight at BMI 20 (according to WHO it is even lower, at 18.5). Our curve shows a relative risk of 1.25 which is again found at BMI 27 at the upper end. Thus, our curve suggests a normal weight range of BMI 20-27 for this specially selected group, but not for women, not for younger or older men, not for those who smoke or who have smoked earlier in their lives. We see, not even this best cherry to be picked is able to support the view that bad health begins at BMI 25.

BMI, mortality and smoking status Now let me pick my own cherries, nota bene from the same study. This one shows the curves for men, based on more than three hundred thousand cases, according to their smoking status. For never smokers, we see an ideal weight of BMI 25 and a "normal" weight range of 21 to 30. For former and current smokers, these values have to be shifted upwards by two BMI points. For a former smoker, a BMI 32 appears to be at the upper end of normal weight, rather than "obesity".

BMI, mortality and age This one is even better, showing roughly the same three hundred thousand men, this time sorted by age classes. Looking at the relative risk at the lower end, I am no longer ready to accept a BMI 20 as lower limit of normal weight. The point of ideal weight again has shifted upwards, now to 26, and the range of normal weight (I admit this is my personal view) is between 22.5 and 32.5 for men in their fifties, expanding gradually to 22 and 36 for men in retirement age.

Conclusions: One of the biggest recent studies on weight and mortality does NOT support a general upper limit of BMI 25 but suggests a re-definition of the term "obesity" which should be adapted to age, sex and smoking status. And even claims of so-called cranks must be checked seriously, their grain of truth may be bigger than it seems at first glance.

Open e-mail to Mark Hoofnagle

Dear skeptic colleague

You do a great job with your blog but I cannot agree with your recent post when it comes to overweight and obesity. In particular, I cannot see how you conclude a BMI of 25 as a general upper limit of good health from the study by Adams et al, NEJM 2006, that you have cited. Plus I would like to learn, besides the graph you have commented, how you interpret the other two graphs from the same study, displayed above here.

Looking forward to find your clarifications on your blog soon.

Christian Bachmann
Med Journal Watch

Photo credit:

No comments: