The obesity paradox encompasses the fact that a body mass index (BMI) in the overweight category appears to correlate with the lowest death rate. The optimal BMI in studies that have found this ranges from about 25 to 27, which is solidly within the overweight – but not obese – range. For a review and (credulous?) discussion, see “Obesity Paradox Does Exist”.(1)
A longstanding criticism of the obesity paradox is that BMI is measured one time only, at the time of the survey. Then the subjects are followed to see whether they develop a disease or die. This process skips the problem of reverse causation.
The idea behind studying health outcomes as they relate to body mass index is, if it isn’t completely obvious, to see how BMI affects health. But what if health affects BMI?
If someone were in the overweight or obese category and developed an illness, let’s say a subclinical tumor, he might very well lose weight. So, he does so, then takes part in a survey studying BMI, registers as normal, then a few years later his cancer is found and diagnosed. He would register in the study as a normal weight person who got cancer, and his case would raise the average risk for the normal BMI category, which is 18.5 to 24.9.
But if he was in the overweight or obese category to begin with, his BMI could very well have contributed to his cancer.
In this illustrative case, an illness caused weight loss, and he was getting unhealthier. That’s reverse causation.
The way to get around this, or one way, is to use maximum BMI instead of BMI at the time of the survey. I wrote about this recently.
The author of the study I discussed, Andrew Stokes, is making a nice career out of his unique measurement process, and has just published a new paper: “Revealing the burden of obesity using weight histories”.(2) From the paper:
There is substantial uncertainty about the association between obesity and mortality. A major issue is the treatment of reverse causation, a phrase referring to the loss of weight among people who become ill. Weight histories are vital to addressing reverse causality, but few studies incorporate them. Here we introduce nationally representative data on lifetime maximum weight to distinguish individuals who were never obese from those who were formerly obese and lost weight. We formally investigate the performance of various models, finding that models that incorporate history perform better than the conventional approach based on a single observation of weight at the time of survey. We conclude that the burden of obesity is likely to be greater than is commonly appreciated.
Using maximum lifetime weight, Stokes found, for example, that of those who had a normal BMI at the time of survey, only 61.5% had always had a normal weight; 34% had at one time been overweight, 3.3% had been obese, and 1.3% had been morbidly obese (BMI >35). See here.
Clearly, if having a lifetime maximum BMI in the overweight or obese category is detrimental to health, and those people are then registered as normal weight at the time of the survey, that increases the perceived risk of being normal and decreases the perceived risk of overweight/obesity.
This is exactly what Stokes found.
The chart shows mortality rate for all individuals who had normal weight at the time of the survey, divided into whether they were always normal, or at one time overweight or obese. Those with a maximum BMI greater than normal had much higher death rates than those who had always been normal, exaggerating the risk of being normal, and underestimating the risk of overweight/obesity.
In my previous post on this topic, I noted that Stokes estimated that, using maximum lifetime BMI, death due to overweight or obesity in never-smoking Americans, aged 50-84, was about 33%. This compares to about 5% using BMI at the time of the survey.
Hence, there is no obesity paradox.