Wednesday, April 2, 2014

Uncovering the True Health Costs of Excess Weight

Is excess weight hazardous to health, or can it actually be protective?  This question has provoked intense debate in the academic community, in some cases even leading researchers to angrily denounce the work of others (1).  There is good evidence to suggest that excess body fat increases the risk of specific diseases, including many of our major killers: diabetes, heart attack, stroke, heart failure, cancer, and kidney failure (2).  Yet strangely, the studies relating excess weight to the total risk of dying-- an overall measure of health that's hard to argue with-- are inconsistent.  Why?

These studies typically rely on a measure of body fatness called the body mass index (BMI), which essentially captures weight-for-height.  A BMI of less than 18.5 is considered underweight, 18.5-25 is normal weight, 25-30 is overweight, 30-35 is obese, and greater than 35 is very obese.  Most studies agree that underweight and obese people have an elevated risk of death.  Yet when we look in the overweight range, the picture is less clear: some studies suggest that lean people are healthiest (3, 4), while others suggest that overweight people are actually healthiest, particularly at older ages (5).  The latter conclusion contradicts the message we often hear from public health authorities and in the doctor's office, so it seems like an important question to resolve.

The studies involved in this debate are all observational, meaning researchers simply observed people of different BMIs and attempted to determine the average death risk of each BMI category.  But as we know, observational studies come with important caveats.  A big one is that it's difficult to be certain that a person's BMI actually caused their health status, rather than BMI being associated with something else that caused their health status (a "confounding factor").  This is one of the main reasons observational studies are considered less convincing than controlled trials, where people are randomized into treatment groups.

In the case of BMI and mortality, we have at least two obvious confounding factors: smoking and illness-induced weight loss (6).  People who smoke cigarettes tend to be leaner than nonsmokers, and they also tend to die more because tobacco smoke is toxic.  People who are ill tend to lose weight, and they also tend to die more frequently than people who aren't ill.  Both of these confounding factors would tend to make leanness appear more hazardous than it really is, and by comparison, make overweight and obesity appear healthier.  Studies have attempted to correct for these major confounds, and those that do the best job of it tend to support the idea that lean people are healthiest (7, 8).  However, the corrections made for illness-induced weight loss tend to be crude because they can only correct for known illness.

Recently, a University of Pennsylvania graduate student named Andrew Stokes published a paper that examined this question in a groundbreaking way (9).  Instead of simply relating current BMI to mortality risk, he related maximum BMI to mortality risk.  In other words, what was the highest BMI for each individual between 1988 and 2004, and how does that relate to the risk of dying over the same time period?  The advantage of this measure is that it's resistant to confounding by illness-induced weight loss, because the maximum recorded BMI is likely to have occurred at a time prior to the onset of illness.  The limitation in this case is that maximum BMI was self-reported rather than objectively measured by investigators.

He also excluded people who had ever smoked regularly, reducing the risk of confounding due to smoking-induced weight loss.

Here are two graphs I created to illustrate his main findings.  Similar to previous studies, in Stokes' data set, current BMI was only weakly related to mortality risk (OW=overweight; OB=obese; VOB=very obese).  This is the type of result that led Dr. Arya Sharma to refer to BMI as a "meaningless integer".

Yet when he examined maximum BMI, a different picture emerged (same scale in both graphs):

This analysis uncovered a much stronger relationship between BMI and death risk, suggesting that previous observational studies may have been seriously confounded, and as a result, may have greatly underestimated the health impact of excess body fat.  This is the main finding of the paper.

Another fascinating tidbit is his estimate of the percentage of total US deaths that can be attributed to excess BMI.  When using current BMI, only 5 percent of deaths in this group of people can be attributed to excess BMI.  When using maximum BMI, 33 percent of deaths are attributable to excess BMI.  Another way of putting this is that excess weight may be linked to one out of three deaths among US adults age 50-84.  This could resolve the apparent paradox that excess weight promotes deadly diseases, yet doesn't seem to be strongly associated with the risk of dying.

Presumably, the explanation for the paradox is that people often lose weight before dying, so their death is associated with a lower BMI group, despite the fact that their former excess weight contributed to their demise.  This is consistent with Stokes' additional finding that people who had lost the most weight were at the highest risk of death.

I spoke with Stokes about his paper on Monday.  He is cautiously optimistic about the finding, and he feels that its primary limitation is the fact that maximum BMI was self-reported.  BMI measurement error has the potential to either strengthen or weaken the BMI-mortality association, depending on whether people tend to under-report or over-report their maximum weight.  He would like to repeat his analysis using existing data sets in which maximum weight was objectively measured by investigators rather than self-reported.  If confirmed, his findings may suggest that excess weight is the single most important public health concern in the United States.

Stokes recently accepted a faculty position at Boston University in the Department of Global Health.  I hope he'll be able to continue his work on this important question.


Beth@WeightMaven said...

How does Stokes' study address other potential confounds? E.g., a nutrient-poor, energy-dense Western diet that leads to weight gain and also presumably contribution to disease through inflammation etc. Or the implications of weight stigma on delaying health care and/or inappropriate health care (given the link between excess BMI and income).

Wouldn't it be more correct to say "excess weight is linked to one out of three deaths among US adults age 50-84" rather than "excess weight may cause one out of three deaths among US adults age 50-84"?

rif said...

Fascinating. How can this deal with people who change their food habits and lower their BMI? Their maximal BMI remains the same. A measure like this seems like it could never support the idea of lowering your BMI since the metric wouldn't change.

elbatrofmoc said...

Thanks a lot for reporting this and summarizing the findings. Why hasn't it made headlines?? I know more studies are needed to confirm these results, but still, considering how many weak observational studies make the news, it's seems bizarre to me. This is huge, in my honest opinion.

Stephan Guyenet said...

Hi Beth,

What you said makes sense. Perhaps a more accurate way of putting it is to say that elevated BMI-- and the diet/lifestyle that typically accompany it-- are strongly associated with elevated mortality. I struggled with the wording on that part of the post.

Hi rif,

That's another potential confounder, but it should also be accounted for by this method. BTW, intentional weight loss has not been found to increase health/mortality risks in other studies.

Hi elbatrofmoc,

Perhaps because he chose to publish it in a lower-impact journal. He explained the reasoning behind that decision when we spoke, but I can't share it without his permission. In any case, it may end up having a major impact anyway, but just take longer to do so, as the method becomes more widely adopted and published.

Gretchen said...

I hope this study *doesn't* reach the popular press. If it does, a lot of overweight people might stop trying to lose weight. Why bother if it won't affect mortality.

BTW, I think we're all going to die. What time frame does "risk of death"
refer to.

rif said...

How would people who change their BMI deliberately be accounted for by this method? It seems that somebody who had a high BMI and managed to lower their high BMI by modifying food intake, somebody who had a BMI and lowered it due to illness, and somebody who had a high BMI and kept it til they died would all be treated the same by this metric?

Did you mean that intentional weight loss has not been found to increase mortality, as you wrote, or not been found to decrease?

Phil GB said...

Seems that some of the people assigned to the "healthy overweight" group could be attributed to the fact that the BMI only measures height/weight ratio. BMI doesn't take into account a larger muscle mass, and thus, a healthier lifestyle than some "normal weight" people.

Reijo said...

This is potentially very important finding.

Another way to tackle the confounding caused by illness-induced weight loss might be to look peri-midlife BMI and its effect on mortality or morbidity.

There is data from meta-analyses showing pre-midlife BMI is strong risk factor for CHD and midlife BMI for dementia.

And there is at least this US cohort for CVD mortality

RD Catherine said...

Well written review of the issues of BMI and morbidity/mortality. As a RD I think there is a potential vast difference between an overweight individual who has achieved that via high fat/ fast food diet combined with lack of exercise, versus an individual of similar weight who exercises regularly, drinks alcohol modestly, and eats plenty of fruit and veg as part of a (calorie excessive) Mediterranean-style lifestyle.

This weeks epidemiology paper on fruit and vegetable intake in England ( threw up some interesting data. Across the quintiles of fruit and vegetable intake, an overweight BMI was consistently around 56% in all intake groups, similarly obese classification was relatively static at over 22% across all intakes of fruit and veg. Yet with increasing plant food intake, relative risk of mortality from cardiovascular disease and cancer was reduced - whatever the BMI. Perhaps we need to look objectively at what is first defined as a 'healthful diet' as a more acceptable approach to population dietary changes. I'm not advocating a HAES approach with the clinically obese, but perhaps allowing some 'wiggle room' around the healthy overweight concept may empower people to engage more with dietary messages

Matt Lentzner said...

Another confounding factor is that not all BMI is created equal. "Lean" people can have a fairly high body fat if they lack muscle mass. Likewise a very fit person can easily be overweight or even creep into obese if they are heavily muscled.

BMI seems too crude to be useful, but maybe there aren't enough of these well-developed people to matter from total population aspect.

For an average height male I would consider 185 to be close to an ideal lean weight. That's a little over 10 pounds into the overweight category according to BMI. My opinion, but the average American is not only fat, but also under-muscled compared to the genetic ideal.

Sanjeev said...

>explanation for the paradox is that people often lose weight before dying, so their death is associated with a lower BMI group
I've been reading this proposal since the early 80s - the earliest I can pinpoint my memory was in discussions of some of Reuben Andres's work and am surprised no one's found ways to tease out the contributory effects

raphi said...

Hi Stephan,

I applaud Andrew Stokes' decision to publish in an Open Access Journal. "Impact Factor" is a primitive method of judging the scientific 'track record' of a journal and is highly susceptible to political/corporate opportunism.
You want to know what is what? Publish all data & make it available to everyone - that's it!

Stephan, the study doesn't seem to specify - did the subjects have their BMI officially recorded and then went on to report their highest BMI they can recall with that reference measurement in mind? Or did these people simply report what they thought was their highest BMI, simply based on a their understanding of what BMI is and where they might stand relative to others?

The difference is small but may be disproportionately meaningful.


PS: the 'innovation' of how to make the best use of poor data is at the heart of good science and an interesting example you provided here.

RLL said...

Former smoking eventually drops a person's risk somewhat close to baseline over IIRC 15 years. But it never returns totally to baseline. I would suspect something similar with weight loss. Rob

Stephan Guyenet said...

Hi Reijo,

Thanks for the links. One other way to analyze the data is to look at the total duration of overweight or obesity. There are some interesting studies supporting the idea that the duration of obesity is more strongly related to health risks than current obesity. That way of looking at it makes good sense to me.

Hi Phil and Matt,

BMI is a crude measure on an individual basis, but it works fairly well in studies that look at averages of groups of individuals. That's because muscular people tend to be balanced by under-muscled people in each category. Getting more sophisticated measures of body comp is a lot harder than BMI, and they may not be that much better at identifying health risks.

Hi raphi,

Maximum weight was self-reported, and objectively recorded height was used to calculate BMI from it. Other BMI points were from objectively collected data.

IcedCoffee said...

Woops just refreshed. I find it a little difficult to imagine that undermuscled overweight people is a serious confounding variable. LBM almost invariably goes up as a result of increasing body weight. I suppose it might equalize, but I'd be skeptical until I saw some data.

Another thought, I would imagine health care provision would be greater in the increasingly obese, whereas unexpected deaths would be more likely in people who otherwise appear healthy. (The skinny fat, so to speak.)

Keith Hansen said...

How would using average BMI over a period compare to maximum BMI?

glib said...

Almost certainly the right variables some combination of time and BMI. Someone who was obese for 20 years will die sooner than someone who was obese for 10. The ground breaking part of this is the elimination of confounding effects, and although it is not proven, it is very reasonable to assume that the least time one spends in the obese zone, the worse the outcome. Yes, to prove this rigorously would be a bitch, following people for 30 years or so.

Chris said...

Hi Stephan,

I can see the issue with self-reported BMI being by far, the largest issue.

There's still a lot of questions:

1. Maximum BMI: Does losing the excess weight cause the mortality to reverse? In the case of tobacco uses, after quitting for about ~20 years, it drops back to almost normal for risks of mortality.

For Maximum BMI, one question worth asking is, does the risk drop back to normal for those who lose the excess weight? If the answer is, only partially, or an outright no, that could have huge implications.

Perhaps more research is needed. It would be an expensive study I propose. Follow a large group of people, divide them into: people who did not lose weight, people who lost weight but regained it, and people who lost weight and managed to keep it off (small group I know). Test for mortality.

2. As noted the reported BMI is an imperfect method. First because people are less likely to report their correct mass.

The other issue is BMI accuracy. Perhaps a better measure may be visceral fat? But how? East Asians in particular tend to accumulate such fat at lower BMIs than other races.

Hmm ... the reason why BMI was adopted was because it was quick and easy to measure. Maybe measuring the circumference of one's waist and adjusting for height is the best way? It's quick and easy too.