Friday, February 12, 2016

Is the "Obesity Paradox" an Illusion?

Over the last two decades, multiple independent research groups have come to the surprising conclusion that people with obesity (or, more commonly, overweight) might actually be healthier than lean people in certain ways.  This finding is called the "obesity paradox".  Yet recent research using more rigorous methods is suggesting that the paradox is an illusion-- and excess body fat may be even more harmful to health than we thought.

Introduction.  What is the obesity paradox, and why does it matter?

We have an enormous scientific literature suggesting that excess body fat increases the risk of many diseases, including the ones that kill us most-- cardiovascular disease, cancer, and diabetes.  Yet around the turn of the 21st century, evidence began to emerge that the story might not be so simple.

Researchers reported that people with cardiovascular disease, kidney disease, and lung disease tended to fare better when they had obesity than when they didn't (1).  Then in 2005, a National Institutes of Health researcher named Katherine Flegal dropped a bombshell on the research community: in her analysis of Centers for Disease Control data, people who were overweight had a lower overall mortality risk than people who were lean or obese (2).  In other words, people who carry some excess fat are the least likely to die, implying that they're the healthiest overall.

Flegal followed this up with more papers in a similar vein, including a 2013 meta-analysis representing 2.88 million people that confirmed her group's previous finding: overweight people, not lean people, have the lowest mortality risk (3).

The potential implications of these findings are enormous.  Two-thirds of American adults have obesity or overweight, and we know that major weight loss is hard to achieve and sustain.  If the obesity paradox is true, then it implies that we should just stop pestering people to lose weight-- there's no problem to begin with!  This is a comforting idea for people who carry excess weight, particularly if they've struggled with weight loss.

It's also a comforting idea for the sugar water industry, including companies like Coca-Cola.  If obesity isn't as bad as we thought, then maybe sugar water isn't as bad as we thought either.  Coca-Cola has funded several researchers whose work supports the obesity paradox* (4).  I'm not questioning the integrity of these researchers, but Coca-Cola's interest in this idea certainly highlights its importance.

But is it real?

The idea of an obesity paradox was controversial from the beginning, and many people have critiqued it over the years.  This came to a head in 2013, when Walter Willett, chair of the Harvard Department of Nutrition, aggressively criticized Flegal's study, stating on NPR that it "is really a pile of rubbish, and no one should waste their time reading it".  He also organized a symposium at Harvard explicitly to criticize the concept (5).  Willett was rebuked for his aggressive tactics, but despite this, many people agreed with his scientific position.

The fundamental problem with the obesity paradox is that it's based almost entirely on observational evidence, meaning that it doesn't come from controlled experiments that are better at identifying cause-effect relationships.  And in this particular case, it's not hard to imagine ways in which observational methods could obscure the true relationship between body fatness and health.  In other words, the obesity paradox could be nothing more than an illusion of the particular research methods that were used to identify it**.

How might this work?  As an example, we know that illnesses such as diabetes and Alzheimer's disease often lead to weight loss (sometimes many years before diagnosis, meaning that you can't entirely eliminate the problem by excluding people with diagnosed disease).  So if you're currently lean and sick, researchers may associate your illness with the lean category, even if you used to be overweight, and even if that excess fat caused your illness to begin with.

A similar phenomenon happens with cigarette smoking.  Smoking causes weight loss, and it's also a fast track to illness and premature death.  It's not hard to imagine how smokers might make leanness look a lot more dangerous than it really is.  It's also not hard to imagine that some smokers don't accurately report their smoking habits, making it difficult to fully account for in studies.

Researchers have recognized these pitfalls for many years, and those who did the best job of correcting for them usually found that lean people tend to be healthiest (6, 7).  Yet a new research method developed by Andrew Stokes, assistant professor of global health at Boston University, promises to provide the clearest picture yet of the true relationship between body weight and health.

To minimize the pitfalls of traditional methods, Stokes looks not only at current body weight, but at weight history.  He asks a simple question: how does a person's maximum attained weight associate with health outcomes?  Weight is compared using the body mass index (BMI) scale, which corrects for the effects of height on weight.

Maximum weight should do a better job of capturing the effect of excess body fat on health, because traditional methods don't consider the possibility that a person might have previously carried more fat, and may have been exposed to its damaging (or protective) effects for many years.  If this is true-- which seems fairly obvious to me-- then Stokes's method is a better test of the obesity paradox hypothesis than previous methods.

In 2014, I wrote about Stokes's first study using this method, which suggests that lean people have lower mortality risk than people who are overweight or obese (8).  Recently, Stokes and his mentor Samuel Preston published two more papers on the subject, and they are much more detailed (9).

Using the maximum weight method, Stokes confirmed his previous finding that people who remain lean throughout life have the lowest risk of dying.  Consistent with this, they also have the lowest risk of developing diabetes and cardiovascular disease.  Both associations were substantial.  Furthermore, people who already have diagnosed cardiovascular disease have a higher risk of dying if they're also obese, not lower.  There is no paradox in these data, which is a good sign that we finally understand what's going on.  It's also reassuring that his results align well with what we know from experimental studies in animals and humans.

His paper goes on to provide valuable insight into why previous studies often went astray.  With the weight history method, he was able to show that people who lost weight and went down a BMI category were inflating the risk level of that category, confirming that this was warping the results of previous studies.  The most likely explanation is that these people were losing weight due to illness (perhaps subclinical) or smoking (perhaps unreported), making leanness appear less healthy than it is.

As a reminder, there is no evidence that voluntary fat loss is unhealthy, and quite a bit of evidence that it's beneficial (10).

Conclusion

The obesity paradox is probably an illusion of the research methods that are used to investigate the relationship between body weight and health.  We can never be sure that the results of observational studies are free of confounding, but there are good reasons to believe that Stokes's maximum attained weight method is more accurate than methods that only consider a weight snapshot.  I hope these findings will stimulate other researchers to adopt his methods.


* To my knowledge, Flegal has never received funding from Coca-Cola.  But in any case, I don't think it's ethical to assume a researcher is biased by a funding source without evidence to support that accusation.  It's logical to think in a general sense that industry funding has the potential to bias research, but specific accusations demand specific evidence.  I don't find it particularly rational to dismiss research findings out of hand just because of a funding source.  In fact, ironically, this is one of the most common debate tactics used by people who are ideologically biased themselves.

** We can add this to a long list of incorrect conclusions that have resulted from confounding in observational studies.  A few examples off the top of my head: the relationship between calorie intake and body weight, the effect of vitamin E on cardiovascular health, and the effect of hormone replacement therapy on cardiovascular health.

19 comments:

  1. As for the funding I think it's important to note that it's possible for funding to bias the field without actually biasing any individual researchers.

    Consider two researchers, Alice and Jenny. For various reasons Alice believes the Obesity Paradox is real while Jenny thinks it's an illusion. Coca-Cola likes Alice's research better so it gives her a big grant to fund it, Jenny meanwhile doesn't have the same funding so she studies other topics or produces less influential publications.

    The result is that because of Coca-Cola's money the published research now strongly supports the idea of the Obesity Paradox, even though Alice and Jenny are still perfectly ethical and unbiased.

    People often seem to miss this point, particularly as it applies to money in politics as well.

    ReplyDelete
  2. I remember reading some decades ago that insurance actuarial tables reported that people who gained a small amount of weight every decade after 30 lived longer. It's been long enough that I don't remember the source.

    ReplyDelete
  3. I think this country would be a better place if we stopped saying "we have an obesity epidemic" and started saying "we have a poor health epidemic" (if that's actually true).

    Then we could talk about what you should be eating MORE of instead of telling people 'just eat less' or 'move more'... which leads to increased stress and other health disasters.

    ReplyDelete
  4. Hi Aluchko,

    I totally agree. I think publication bias is potentially one of the biggest pitfalls in research, and a difficult problem to detect.

    Hi Tina,

    I think it's possible to recognize that many of our foods are fattening/unhealthy, and advise eating less of those foods, without simply telling people to "eat less, move more". I don't think we're going to get very far by eating more healthy foods without reducing our intake of unhealthy foods.

    ReplyDelete
  5. Let me see if I get this straight.

    Stokes et al. look at the obesity paradox and think "Hey, I bet if we rehashed the data differently [using maximum BMI instead of current BMI], we wouldn't see the paradox anymore".

    So they go and test their idea on a data set *where there was no obesity paradox in the first place*. They rehash the data and, guess what, there is still no obesity paradox.

    And that proves the obesity paradox never existed???

    ReplyDelete
  6. Thinking about the actuarial tables, weight (or BMI) is not a factor in the Framingham CVD risk score. The important factors that a person can control are HDL and systolic blood pressure. While these factors can be related being overweight, they are better related to level of activity. "Move more" has more effect than "eat less".

    ReplyDelete
  7. Another great example of being led by observational rather than controlled studies. As a physician it is galling to me that this is not stressed more in training. We are easily fooled by incorrect assumptions. Keep up the great analysis!

    ReplyDelete
  8. Hi Valerie,

    I think I understand what you're saying. Two counterpoints:

    1) In his initial paper, the weight history method did eliminate an obesity paradox (or at least an overweight paradox) that was present in the snapshot data. This was also the case for his paper on CVD that I cited in this post.

    2) Not all snapshot analyses show the paradox to begin with, and as you pointed out, the one he used for his recent PNAS paper didn't. However, consistent with the other studies, the weight history method strengthened the association between weight and mortality and made leanness look significantly healthier than overweight. And, just as importantly, it offered compelling evidence that reverse causality is a major confounding factor in snapshot analyses.

    So your point is valid, but on balance I still think the conclusion is well supported.

    Hi thhq,

    The Framingham risk score was derived using the types of analyses that Stokes's research suggests are confounded. So it's likely that weight should be part of the picture. That doesn't undermine the importance of physical activity of course-- it's important no matter what a person weighs.

    ReplyDelete
  9. @Stephan Guyenet said:
    " I think publication bias is potentially one of the biggest pitfalls in research,"

    Excellent point!

    To the extent that there is useless research (what I call grant seeking behavior - simply support the current trends and maintain money flows) that gets published with glaring flaws in logic and deceitful headlines means we are all harmed - and limited research funds are wasted. What is ever stranger is the history of blocked research - I believe there are people in the agricultural industry that have blocked important studies over the years - limiting of trans-fats was delayed perhaps a decade? ( Remember the heart healthy logo on coco-poofs? a type of breakfast candy? Do you think the money coming in from a certain grain company might have also influenced which research got funded? )

    There was an interesting paper on study size vs conclusions - the small - low 'n' papers that supported the PC theories were published, larger studies finally turned the science had a hard time finding a publisher.

    The Framingham paper with the graph that had bogus intervals in order to support the 'Cholesterol is bad' meme comes to mind.

    ,.,.

    Someone that goes into the hospital with a serious illness - 10 days in ICU - can lose a LOT of weight - someone with a bit of fat has an advantage.

    There is also the example of a weight lifter that can squats twice his body weight - his BMI is likely to look bad - but his health is great. One of the key points to notice is that people with good muscle mass are less likely to lose BG control - strength is a much better predictor of mortality than BMI or FRS(Frammington Risk Score) or other scores.

    Here is the con-founder - people that carry an extra 30LB 24-7 end up getting stronger - more muscle. As long as they keep moving.


    But much more important that the weight is BG - the idea that a high-carb and particularly high sugar diet is harmless for someone with poor BG control is insane. IMO the targets for A1c, fasting-BG,postprandial-BG recommended by the ADA are much too high. If you look at young/healthy children - levels in people that are not exposed to industrial foods - fasting is typically below 90 and postprandial below 110. The idea that one shouldn't change one's diet until it hits the ADA numbers is simply wrong in MO. Once fasting BG start rising - a lot of damage has already been done. People that have excellent BG control, end up with much lower cancer rates, lower diseases of aging in most organs, lower BP - better quality of life, particularly in the later decades of life.

    The stress that stimulates muscle mass - (this comes from training - not low impact exercise - they are NOT one and the same) - also increases bone density (less likely to spiral from a hip fracture) - and stronger tendons. This requires a certain level of protein - protein that is missing from many high-carb diets.

    This paper , (correlative evidence does not show causation) begs careful research to find the 'WHY'. ( my hunch is BG is part of the story).

    Look - I'm not a low-carb fanatic - but if I am in the 40% of the public that has lost good BG control - I'm going to change my diet. I'm going to do my best to keep my post-pradials below 110 and my A1C below 5.

    ReplyDelete
  10. @karl @steven I brought up Framingham because it is used by actuaries and because doesn't use weight as a factor. It puzzles me that BMI or weight are not factors, because this is information that would have been collected. Maybe adding weight as a factor doesn't improve the correlation when systolic blood pressure and HDL are being used?

    @karl please provide a link to your Framingham paper. I only use the CVD risk tool (http://cvdrisk.nhlbi.nih.gov), which from what I can see does not have a strong "cholesterol is bad" bias. I've spent a lot of time testing "what-ifs" with the tool. I have found that the predicted CVD risk is fairly insensitive to increasing total cholesterol, but is strongly reduced by raising HDL (though almost flat above 65 mg/dL). Hindsight would tell you to test different things, but it'll take another 50 years of mortality data to improve upon it.

    I made myself obese and diabetic eating mass quantities of Raisin Bran and Frosted Mini Wheats. I used to get cases of them for free at work. Every time I ate five boxes I got a free DVD for sending in coupons on the back, which incentivized me to empty the boxes more rapidly. They had the heart healthy logo (dry cereals are the base of the good old Food Pyramid, along with those heart-healthy pop tarts), and all those vitamins and minerals. What could possibly be wrong with eating more? As I related earlier I've done a lot of stupid N=1 health experiments, but this one had by far the worst outcome. The only positive was the excellent free DVD's, which represent three cases of cereal eaten.

    ReplyDelete
  11. Framingham also has a model which uses BMI instead of lipids:

    https://www.framinghamheartstudy.org/risk-functions/cardiovascular-disease/general-cvd-risk-prediction-using-bmi.php#

    ReplyDelete
  12. You raise the need for the research to be based on "weight history" and to account for morbidities. It's been done. Read the longitudinal study by Zheng, Tumin, and Qian (2013), Obesity and mortality risk: new findings from body mass index trajectories. Approx. 10k Americans over age 51 were interviewed at two year intervals for up to 26 years. The subjects were sorted into six sizes by BMI (two BMI trajectories each from the three BMI ranges of normal, overweight, and obese). Kaplan-Meier type curves were plotted for survival, and hazard ratios were calculated. Among the major findings: the poorest survival occurred among the two extreme sizes; for a full 10 years (ages 61-71), the survival curve of the thin subjects coincided with that of the very obese; the two "overweight" sizes showed the highest survival; the fully adjusted hazard ratio of the thinnest was head and shoulders worse than the four better survivors. As for the thin subjects (and at BMI of 21.5 to 22.0, they were not extremely thin), even after accounting for every kind of morbidity, including behavorial and psychosocial (being lonely or indigent), their *fully adjusted* hazard ratio for death was 1.64 -- and the 95% CI of this hazard ratio did not even overlap with the 95% CI of the higher survivors! Also, the thinnest cohort was the only one not to gain weight after age 51 (ie, the only cohort with a downward body size trajectory). The skinniest and the fattest broke away from the pack well before age 60, and at age 65 their mortality was ca. 10% higher than the reference cohort's. Multiple studies dating back to at least 2001 concur with Flegal (2013) and Zheng (2013). Except for about 4 out of 5 cancer types (by site), most chronic morbidities have their median age of diagnosis before age 60 (for diabetes, for example, it's around 54). And the median age at diagnosis for almost all cancer types is below 70. All is to say that Zheng et al. did not miss emergent disease. You refute Flegal (2013) citing de Gonzalez (2010) (your footnote 7); the former study cited the latter and refuted it. This raises doubt on your assertion that "those who did the best job of correcting for [cited methodological pitfalls] usually found that lean people tend to be healthiest".

    ReplyDelete
  13. Stephan. May I suggest you post on health effects of ectopic fat? This seems to be a neglected area, with the exception of abdominal fat. My experience is that small ectopic fat deposits have some characteristics that may be used to characterize its effects and correlate as a leading health indicator. Examples: occurs in body areas where some damage has already occurred such as face, kidneys (love handles), pubic area, upper under arms, intestines; does not correlate with overall weight loss. I am old and my concern is not with weight (low level chronic infection). I have never been obese and only mildly overweight in younger days but I have experienced ectopic fat deposits changes over time. I am lean and so it is easy to observe changes in these deposits. I observe these effects in the general population. I have experienced useful correlations with the ectopic fat abundance.

    ReplyDelete
  14. Hello cgnty69,

    Thanks for your comment. I looked at the study you cited (Zheng et al), and it does not support the strong conclusions you draw from it. Here's why:

    1) The higher mortality risk for the "normal weight downward" category is consistent with Stokes's finding that people who trend downward in weight have higher mortality. This is precisely the confounding factor that has limited snapshot analyses of the weight-health relationship.

    2) What we need to consider as the "healthy lean" category in Zheng is actually the "normal weight upward" category, which increased from 23.1 to 23.6 kg/m2-- this group started at a healthy BMI and remained largely weight stable.

    3) When we compare the "normal weight upward" category to the "overweight stable" category, the hazard ratios are statistically indistinguishable. The confidence intervals overlap one another broadly. Statistically speaking, the HRs are the same.

    4) Similarly, in table 3 that considers only the healthiest subset of people, the HRs are virtually identical.

    5) The older the age of the sample population, the weaker the relationship between BMI and mortality risk. Zheng et al is an older sample.

    I don't see how this study poses a challenge to Stokes's findings, or the idea that the obesity paradox is an illusion.

    ReplyDelete
  15. Oh, one more thing I forgot to mention: the "overweight stable" group that had the lowest mortality HR was barely overweight (BMI 25.8 -> 26.9). About 3 BMI points above the "normal weight upward" group.

    ReplyDelete
  16. (Prologue on terminology and categories. Although mathematically, the concept of "overweight" encompasses the concept of "obese", in scientific and clinical jargon, the two words are interpreted as being mutually exclusive. I suspect that many discussions of the health impacts of being "overweight" conflate the obese and "subobese" categories. This thread opened with, ". . . people with obesity (or, more commonly, overweight)". I beg pardon in case of misunderstanding which range of BMI values other commenters are addressing. The objections I am raising do not extend to very obese BMI values.)

    Two of your responses bolster the argument that "overweight" (BMI between 25.0 and 29.9) is not in itself to be taken as a concern for mortality.

    (1) As for the HR's of Zheng 2013's two "subobese" groups being statistically insignificant from the HR of the group whose BMI trajectory was 23.x: but that is the contrarian claim. "Overweight" (ie, in the sense of what I call "subobesity") does not increase mortality, because in Zheng 2013 the overweight people did not die sooner to a statistically significant extent. In fact, they didn't die sooner, period (the longevity increases were on the order of half a year). Moreover, the size category consisting of BMI 23.x was outlived by both the higher overweight subjects (~6 BMI units higher) and the lower overweight subjects (respectively: the "overweight obese", those whose 340 month BMI trajectory rose from ~29.0 to ~31.5; and the "overweight stable", those whose trajectory rose from ~26.0 to ~27.3). The contrarian view is that the moderately chubby differ in mortality only trivially from the "normal" sized, and that possibly even the increases in their morbidity are milder than rumored. The contrarian view does not need to claim that these people live the longest.

    (2) As for the point (5) regarding age:
    (a) I draw the opposite interpretation from the Kaplan-Meier curves in Zheng 2013, figure 2, in which divergences in longevity were greatest in the final 100 months of the study period.
    (b) Regarding people under 50, the correlation between "overweight" (BMI 25.0–29.9) and mortality is inherently a nonissue. Indolent, mild to moderate morbidities don't cause more than a relative handful of deaths before age 50.

    I don't see how the argument, that strong morbidities (eg, smoking) may be confounders because they depress weight, undermines Zheng 2013. First off, this study did adjust for smoking. Moreover, the thin subjects, older people with BMI <21, who fared so poorly in this study make up only 10% of Americans. 75% of Americans over 45 have a BMI above 23, while 50% are between 23 and 28 (halls.md, graph compiled from NHANES III). In Zheng 2013 the 3 body size groups within the BMI values 23–28 were bunched together in their survival/mortality throughout the 340 month study period (ages 51–79). This study found that being slender (BMI <21) is a morbidity.

    We need to distinguish morbidity from mortality: although mildly to moderately overweight persons as a group live several months longer, some of them are somewhat sicker for decades. On the other hand, I think it's likely that morbidity (eg, Type 2 diabetes) doesn't rise appreciably until BMI of 27 or 28. Research into the correlation between morbidities and BMI in the range, 25.0 to 29.9 should be more fine grained, down to the BMI unit. The science might support moving the BMI threshold of "overweight" to 26.0.

    ReplyDelete
  17. @cgnty @stephan @karl As karl points out re: bodybuilders BMI is confounded by the effect of exercise. Loss of musculature is a serious problem for the elderly. If falling BMI is accompanied by reduced thigh circumference and grip strength increased mortality would be expected.

    I ran "what ifs" on the Framingham BMI model and am disappointed. First, the risk factors are much higher compared to the lipid model. Second, lower BMI predicts lower risk. BMI of 11 (at which point most people are dead) is much lower risk than the optimal BMI of 22-26. The BMI factor appears to be a crude linear data crunch.

    ReplyDelete
  18. @cgnty mortality is my main interest. Overweight is certainly a factor but it is't the only one. And once obesity is no longer an issue - no matter which method you use to treat it - diet is no longer a major mortality factor IMO.

    Obituaries don't lie about this. My casual study of N=1's shows that whatever their diets were, Ancel Keys, Jack Lalanne, Julia Child, John Wesley, Lawrence Ferlinghetti and Harlan Sanders lived remarkably long lives. Whatever they ate and did is of paramount importance as a personal model for a long and healthy life. In contrast, Pritikin and Atkins are of much lesser interest. Their highly attenuated diets may not account for their relatively low longevity, but IMO their lives were not lengthened or improved by their lifetimes of dietary rigor.

    And Ferlinghetti is still alive. Apparently poetry matters.

    I recently got some information from Henry Blackburn (also a nonagenarian) about his years working with Ancel Keys. His first comment "Why does anyone live to be 100?" was followed helpful information about his diet, family life, and high level of physical and mental activity. Keys was a little overweight, probably in the 25-27 BMI range.

    ReplyDelete
  19. https://www.ncbi.nlm.nih.gov/pubmed/26765423
    Medicine (Baltimore). 2016 Jan;95(2):e2424
    "In conclusion, overweight adults have a lower mortality risk than normal weight adults. Our findings do not support [Stokes' contention] that the lower mortality in overweight adults is due to confounding effects of smoking and preexisting diseases."

    ReplyDelete

Comment moderation is in effect. The core criteria for publication are that a comment must add value, and must be respectful. For more information, see this post.