Follow me on

Why There Is So Much Confusion About Obesity And Mortality

scaleAny follower of media reports or even research papers on the relationship between obesity and mortality should be righty confused by now.

Not only are there publications suggesting that the relationship between obesity and mortality isn’t that strong after all and that perhaps the BMI levels associated with the longest survival are somewhere around 30 (and not below 25) but then there is the issue of the obesity paradox, or the finding that among people with chronic (and some acute illnesses), a higher BMI is associated with better survival than being of “normal” weight.

On the other hand, there is overwhelming evidence that higher BMI’s are associated with an increased risk of a wide range of health problems – from diabetes to cancer.

This is not to say that everyone with a higher BMI is sick – they are not! But there is no doubt that the risk of illness does increase with higher BMIs.

In our own study on the Edmonton Obesity Staging System (EOSS), which classifies individuals based on their actual health rather on their BMI, we found that while about 50% of individual in the BMI 25-30 range can be considered healthy (EOSS Stage 0 or 1), this number drops to below 15% for individuals in the BMI 40+ range.

So, if obesity is such a risk factor for disease, why do epidemiological studies struggle to consistently show an effect of obesity on mortality?

Now, a paper by Andrew Stokes and Samuel Preston, published in the Proceedings of the US National Academy of Science, suggests that it is not current weight (as used in many studies) but rather the highest lifetime weight that is most clearly associated with mortality.

Their reasoning is as follows. “Intentional” weight loss in the population is rare (very few people in the general population ever consciously manage to lose a significant amount of weight and keep it off)

In contrast, “unintentional” weight loss, when it occurs is generally a bad sign. Indeed, one of the best indicators of poor prognosis (for almost any health condition) is when someone loses weight. In many cases, this “spontaneous” weight loss can precede overt illness or death by many years.

Thus, the authors argue that most of the literature on this issue is simply confounded by the confusion caused by all the people who have unintentionally lost weight due to an underlying health problem (diagnosed or undiagnosed).

As these people would be at higher risk of death, despite measuring in at a lower weight, they muddy the waters making lower BMI levels look more dangerous (or in comparison higher BMI levels look safer) than they are.

To test their hypothesis, the researcher looked at data from the US NHANES study linked to death registers using four different approaches:

Model 1: BMI measured at the time of survey (this is the method most commonly used in epidemiological studies)

Model 2: The highest reported lifetime BMI at the time of survey

Model 3: Individuals surveyed in their current BMI class who had never been heavier compared to individuals in that BMI class who reported formerly being in a higher BMI class.

Model 4: Individuals surveyed in their current BMI class compared to people who were formerly in that BMI class but had moved to a lower BMI class by the time of the survey.


In both models 1 and 2, there was a greater risk of mortality with higher BMI class, but the relationship was stronger in model 2 (highest lifetime BMI) than in model 1 (current BMI).

In model 3, there was still an increased risk with higher BMI class but within each current BMI class, risk was higher in individuals who had previously belonged to a higher BMI class.

In model 4, mortality also rose with the highest weight achieved but was markedly higher in individuals who lost weight after achieving a particular BMI category compared to those who remained at that maximum.

These findings have important implications for our understanding of the relationship between BMI and mortality.

As the authors note,

“Confining analytic attention to survey BMI alone thus sacrifices important information provided by an individual’s maximum BMI. The poor performance of the survey-only model is especially salient because models using only BMI at survey dominate the set of findings in the literature on the relation between BMI and mortality.”

The errors in not considering highest BMI are not trivial.

For example,

“33.9% of individuals in the sample who were normal weight at survey were formerly overweight, and this group had three times the prevalence of diabetes and cardiovascular disease CVD) relative to those who were in the normal-weight category at both max and survey.”

Here is how you would interpret the data,

“Disease prevalence and mortality both rise with increases in maximum BMI and rise even further for those who reach a particular maximum BMI category and then lose weight. These patterns strongly suggest that obesity raises the risk of diabetes and CVD and that, once acquired, these diseases often precipitate weight loss….Only by using weight histories can this pattern of erasure be identified and corrected.”

The use of historical data in determining risk would not be a new concept,

“The introduction of historical data in the analysis of smoking occurred more than a half century ago, when studies began to distinguish among current-, former-, and never-smokers.”

Similarly, in the context of obesity one would need to differentiate between people who currently have obesity, people who previously had obesity, and people who never had obesity.

All of this only works, because in these type of epidemiological studies, “intentional” weight loss, be it through behaviour change, medication or surgery, is so rare as to be non-existent. Virtually all weight loss seen at a population level in “unintentional” and probably related to underlying health issues.

Thus, one should not interpret these findings to mean that someone intentionally losing weight through behavioural, medical or surgical treatments is at a higher risk for mortality – the intervention studies we have on that (this cannot be studied in population studies as there are so few cases of “treated” obesity), suggest otherwise.

For clinicians, these data point to the importance of noting the highest BMI and not just current BMI – if the patient has lost weight (especially if this is not explained by obesity treatment), then this may be a high-risk patient.

Edmonton, AB



  1. I clicked through to the abstract and tried to click on the “Free Full Report” but that link is broken, so I’m commenting based only on the abstract and your analysis. I appreciate that you, by the way, are a careful reader, but you lack an important skepticism that I plan to provide you now. First of all, I’d like to challenge a conclusion summarized in a sentence in the abstract:

    “These distortions make overweight and obesity appear less harmful by obscuring the benefits of remaining never obese.” NO. What the study finds is that weight fluctuation is harmful. They are demonizing obesity when clearly it’s the fluctuation — whether by YOYO weight cycling or bariatric surgery — that is dangerous.

    The latter factor is emphasized by the time period studied. The researchers are consulting NHANES surveys as far back as 1988. As you know, bariatric surgery has gone through a revolution since then. The early surgeries were much more dangerous and had higher post-op mortality than now. (And, of course, all surgery carries higher risk of mortality.) Regarding bariatric surgery, however, if someone had attained a higher BMI in, say, 1985, went through Bariatric Surgery and had a lower BMI at the 1988 survey, their risk of dying in the next ten years was markedly increased by merely having had that surgery.

    I will also attack an assumption. This time I’m quoting you: “Their reasoning is as follows. “Intentional” weight loss in the population is rare.” and “Virtually all weight loss seen at a population level in “unintentional” and probably related to underlying health issues.” Is that their assumption? That’s crazy. You parenthetically qualify the first statement by saying that maintenance is rare. That I agree with, but I think that’s you talking, not them.

    As you know, weight loss is going on all the time. Depending on when you’re surveying people, a quarter or more of the population may be on a “serious” diet and may be losing or have lost a significant amount from their highest established BMI. (Of course, they will regain it.) Between NHANES surveys, some people might lose and regain over and over. Their weight may be anywhere in a cycle, whether moving upward or downward. (They’ll also be more enthused to complete a questionnaire if they are near the lower end of their yoyo weight cycle.) That is probably what is dangerous, and I think that the results from models 3 and 4 support this, but, again, I can’t see the original study. I’m basing it on your summary.

    I invite you to read this study through my eyes and to rehash it in a way that doesn’t demonize obesity (and obese people by extension).

    Post a Reply
  2. Not to be a pest, but their survey period also includes that awful time in which Fen-Phen and Redux were introduced, killed people, then yanked from the market in 1997. The danger people faced who took those drugs was NOT their highest BMI, but the desperate treatment that they accepted to try to relieve themselves of it.

    Post a Reply
  3. According to your interpretation, more than one third of normal-weight people are living on borrowed time (they are lean onky because they are very sick).

    That number is huge! Does it really seem realistic to you?

    Like Debra pointed out, I think you are leaving out all the people who are right about to regain the weight they just lost. They might eve form the majority of that 33.9% of normal weight people.

    How do you distinguish people who lost weight from sickness, from people who (temporarily) lost weight intentionally?

    Post a Reply

Submit a Comment

Your email address will not be published. Required fields are marked *