The effects of obesity on mortality—what can we learn from weight histories?
The New Year in 2013 began with release of a major study on the health risks of obesity. For a nation grappling with rising levels of obesity, the news was comforting. The data suggested that people with overweight could expect to live longer. The results also indicated that having slight obesity conferred no excess risk of death. The results were picked up by many major news outlets, was discussed in op-eds, and on social media and sparked livid reactions from critics. The influence of the study was tremendous, which was not surprising. It was the largest study ever carried out on the subject—summarizing data on close to 3 million individuals from 97 studies. Unfortunately, the debate quickly grew acrimonious. In a rare step, Nature issued a reprimand and urged the scientific community to accept the fact that obesity’s relationship with health might be less clear cut than imagined.
But is it? This is the question I set out to answer. It seemed a futile pursuit in the face of such a large and impressive study. But a part of me wondered whether the meta-analysis just reflected systematic bias in the underlying studies.
The very first step I took was to try to replicate an earlier study by Katherine Flegal and colleagues that came to much the same conclusions as the meta-analysis. The earlier paper was based on data from the National Health and Nutrition Examination Survey (NHANES), considered the gold standard survey for monitoring the health of the American population.
The most common critique of that study, which emerged again after the 2013 meta-analysis, was that the results—a protective effect of overweight, no increased risk of mild obesity—reflected confounding by illness or reverse causality. The idea is that being slim may appear risky if some people in the sample are slim because of an illness that caused them to lose weight.
A lot of studies have tried to address this bias, but none of the approaches used have proven satisfactory. One common strategy has been to try to isolate a healthy subset of the sample by excluding people with known or suspected illnesses, people who lost weight or people in poor self-rated health. However, this approach can introduce its own biases and has been criticized for excluding large fractions of people in the sample, calling into question the generalizability of the results.
It was clear to me that to improve upon the work by Flegal and colleagues I would need to get to the heart of the reverse causality matter and find a way to address it. So what could I do differently?
The solution turned out to be really simple. Studies of obesity and mortality were almost all based on weight at a single point in time. But many people gain or lose weight during their lifetimes, especially in times of illness. I needed to find a way to incorporate weight histories.
Incorporating histories is common practice in studies on smoking. In that literature, the non-smoking population is almost always separated into never-smokers and those who smoked in the past and quit. If you didn’t separate out these two groups, you’d likely reach the conclusion that smoking is not all that harmful—it might even appear protective. But that’s only because the low risks of the never-smokers are being masked by the much higher risk of the former smokers, many of whom smoked throughout their lives.
Surprisingly, this distinction, which is clearly essential for obtaining accurate estimates of the effects of smoking on mortality, is rarely made in the obesity literature. Studies almost always lump together individuals who have never had obesity, with individuals who no longer had obesity (but once did), despite the fact that these two groups may have very different mortality risks. If it were only possible to disentangle these two groups, we might be able to obtain better estimates of the effects of obesity on mortality that are not affected by reverse causality.
So after replicating the basic approach used by Flegal and colleagues, I took the next step of incorporating weight histories to determine what people weighed over the course of their lifetime. This was enabled by a question in the survey asking subjects to recall their lifetime maximum weight.
I used this information to separate the normal weight category into two groups: those who always maintained normal weight versus those who formerly had overweight or obesity and then lost weight. I found that mortality risks in the latter group were much higher than the risks in the former group. I demonstrated that combining these two groups raises the mortality rate in the normal weight category and obscures the low risks of those who maintained normal weight throughout life. Finally, I showed that when the normal weight category is redefined to only include the always normal weight individuals, the association between excess weight and mortality strengthens dramatically. These results were published last year in Population Health Metrics.
My findings thus suggest that weight histories are an essential piece of the puzzle for understanding obesity’s effects, just as former smoking status is important in the study of the effects of smoking on mortality. Failure to take weight histories into account has likely caused widespread bias in the literature, with the effect of obscuring obesity’s true toll.
Andrew Stokes is an Assistant Professor in the Department of Global Health at Boston University. His research is focused on the causes and consequences of the global obesity epidemic and developing novel approaches to combating obesity at the population level through interventions that target aspects of the social and physical environment. You can also follow him on Twitter.