Health Policy$ense

Explaining the Obesity Paradox

This month’s issue of Mayo Clinic Proceedings features two new articles on the obesity paradox—the finding that overweight/obesity confers a survival advantage in individuals that have been diagnosed with a medical condition. These paradoxical findings have also been found in the general population. Indeed, the largest meta-analysis to date on weight and mortality in the general population summarized 97 studies consisting of close to 3 million observations and found that being overweight is associated with lower mortality relative to normal weight and that being slightly obese confers no excess risks.


Andrew Stokes is an Assistant Professor in the Department of Global Health at Boston University and recent graduate of the University of Pennsylvania, where he earned doctorate degrees in Demography and Sociology.

The latest meta-analysis investigates the relationship of body mass index (BMI) with total mortality, cardiovascular (CV) mortality and myocardial infarction (MI) after coronary revascularization procedures. Across 36 studies, it finds that the risk of total mortality, CV mortality and MI was highest among underweight patients, and CV mortality was lowest among overweight patients.

The second study examines BMI and attributes of body composition [lean mass index (LMI) and body fat (BF)] in a population of patients referred for echocardiography. Once again, the authors find that higher BMI is associated with lower mortality. But when they separate the attributes of body composition, they find that a higher lean mass index (i.e., muscle) eliminates the protective effect of higher body fat, and is itself associated with lower mortality in both obese and non-obese people.

Reverse causality
These studies seem to confirm the obesity paradox. But do these findings reflect true causal effects of obesity on mortality or confounding? Some scholars suggest that excess weight may help protect against frailty and wasting diseases. On the other hand, skeptics argue that confounding masks the true causal relationship between weight and mortality. A key culprit cited is reverse causality, which refers to the fact that an individual’s weight may be a reflection of illness. Are we mixing up cause and effect?

I have recently published work that attempts to address reverse causality and to explain why findings of weak or inverse associations between excess weight and mortality are so prevalent in the literature. My approach uses lifetime maximum weight to determine an individual’s weight before any potential onset of illness. Use of maximum weight makes it possible to identify normal weight people who have always belonged to the normal weight category versus those who were formerly overweight or obese and subsequently lost weight. Previous studies have not been able (or even attempted) to disentangle these two groups; however, this turns out to be quite consequential to understanding the obesity paradox.

Using data from the National Health and Nutrition Examination Surveys, I compared associations between excess weight and mortality using BMI at survey and BMI at lifetime maximum. I studied US adults ages 50-84 who never smoked (since smoking is a strong confounder of the relationship between excess weight and mortality) and used multivariate regression models that adjust for confounding by social and demographic factors.

As shown below, using max BMI strengthens the associations between excess weight and mortality. Excess risk associated with overweight relative to the normal weight category is -2% using BMI at survey compared to 28% using max BMI; excess risk for obese class 1 is 18% vs. 67%; and excess risk for obese class 2 is 31% vs. 115%. The associations are statistically significant for obese class 1 and obese class 2 using maximum BMI, whereas in the case of BMI at survey, none of the associations are significant.

chart 1

Using maximum BMI strengthens these associations for two reasons. First, when using BMI at time of survey, the normal weight category is a mix of low-risk stable weight individuals and high-risk individuals who have experienced weight loss. These groups have very different mortality rates, as shown below. Specifically, the overall mortality rate in the normal weight at survey group of 10.4 deaths per 1,000 person-years is the weighted average of the mortality rate for the stable normal-weight group (7.2) and the much higher mortality rates of those formerly overweight, obese class 1 or obese class 2 (14.2, 16.5 and 66.6, respectively). In contrast, using max BMI, the normal weight category only includes individuals who were consistently normal weight throughout their lives.

chart 2

A second reason the associations strengthen using maximum BMI is that the overweight and obese groups contain not only the relatively healthy stable-overweight (or stable-obese) individuals but also the higher-risk individuals who experienced weight loss, making the relative comparisons more pronounced.

These findings suggest that reverse causality is a fundamental source of bias in studies of the mortality risks of obesity. As a result of this bias, many studies in the literature find weak or even inverse associations between excess weight and mortality. The extent of the problem depends on the particular subpopulation being investigated—the bias is expected to be largest in studies focused on the elderly and clinical populations. Not surprisingly, these are the subpopulations in which the obesity paradox is most commonly reported.

Identifying the optimal weight category for mortality risk is important for shaping clinical and public health guidelines. Many studies find that being overweight or moderately obese may not be deleterious to health. But if the tenuous and inverse associations identified in the prior literature are a result of bias—and my analysis suggests they are—then we are at risk of misinforming both patients and the public about what they can do to improve to their chances of living long and healthy lives.