The obesity paradox paradox (crossposted at Crooked Timber)

I see lots of stories made up of handwringing over the “obesity paradox”, normally presented as saying that even though obesity is a risk factor for all kinds of diseases, obese people appear to have lower mortality than others. A typical finding is the one reported here

being overweight or slightly obese was linked to about a 6 percent lower risk of dying, compared to people considered “normal weight. Being severely obese, however, was still tied to an almost 30 percent higher risk of death.

People are tying themselves in knots over this, but it doesn’t seem to me that there is any paradox to be explained. The obvious reading of the data is that the Body Mass Index[^1] ranges used for the various categories (20-25 Normal, 25-30 Overweight etc) were set a bit too low when they were originally estimated, or rather, guessed. From my quick look at the data, if you bumped the ranges up by a couple of points, the paradox would disappear. People at the bottom of the current normal range, who tend to have high mortality, would be classed as underweight, while those currently classed as slightly overweight would be reclassified as normal.

Am I missing something?

[^1] This point is logically separate from the general problems of the BMI, regarding muscle mass and so on.

26 thoughts on “The obesity paradox paradox (crossposted at Crooked Timber)

  1. According to the conventional BMI index, AFL athletes Adam Goodes, Lance Franklin and Nic Naitanui are all overweight, and quite a few of the participants in the current State of Origin series are obese.

  2. A more accurate measure might be of body fat ie BF%.

    Studies into diabetes and mortality in aboriginal communities indicate that weight gain on a generally lighter frame results in a greater build up of abdominal fat, something that may not picked up using BMI.

  3. But, to restate the point in footnote 1, the difficulties with BMI aren’t the problem here. The paradox arises in people who aren’t athletes, and whose height is mostly in the normal range, so that BMI is a pretty good proxy for BF%.

  4. I think you are right J.Q. And like many good insights it is surprisingly simple and even obvious once someone has pointed it out. Being a little above the top end of the current healthy BMI range, I am quite happy to believe it needs to be bumped up a bit. It’s always easier to believe what suits us!

  5. If I were giving odds–I’m not, but it’s the way to bet–the shift in “acceptable” BMI (form what, 30 to 32 or 33?) corresponds to medical advances in the treatment of the illnesses to which one is more prone at the higher level. Which also probably includes earlier detection and treatment.

    More diabetics-in-training living longer–and take better care of themselves earlier, therefore avoiding or curtailing some (bad) behaviors that otherwise would have abided for an extra few years.

    And since you have the “developed” world with a higher cohort of people whose weight goes higher than underweight (speaking from my 25-30 pounds too heavy position), you research their issues for the same reason Willie Sutton robbed banks.

    Result is a better outcome at the margin on the high side, and no improvement on the low side, so it looks better for the normal-bodied BMI 30-35 cohort.

    The parallel would be the post-war progress in deaths from heart illness, most especially since ca. 1965 and transplant technology. If 10% of people don’t die from heart attacks and live ten years longer, average life expectancy rises one year. Similar effect here.

    If I were betting. Or putting together a research proposal.

  6. Your hypothesis JQ is simpler, more explanatory and better empirically supported than its competitors. It’ll do me. I hadn’t thought about it much until now since as noted the BMI measure is well-known to be a poor measure. I think the simpler ratio of waist to height should be more useful. Last month’s SMH reprinted a good London Telegraph story on that subject:

  7. I agree with @Ken Houghton , surely the answer to this ‘paradox’ is likely that being slightly overweight is correlated with affluence in USA and therefore with a greater probability of access to half-decent health care?

    And regarding BMI, despite it being touted as a pretty good proxy for per cent body fat, surely per cent body fat would be the best measure to use, especially for actual medical research? I used to do a lot of this in a previous career, it only takes a few minutes to calculate, tho I guess that is a long time compared to a few seconds for BMI.

  8. @Ken Houghton

    My BMI is about 27. It is clear from looking sideways in the mirror I am overweight. I have a significant pot belly, man cans and a layer of unecessary fat over my whole torso. Apologies if this is too much information. Without the excess weight I am a natural ectomorph so it is no surprise I look overweight at 27 BMI.

    However, I find it hard to believe that anyone is “normal-bodied” in the BMI 30 to 35 cohort. All of this cohort would have very high fat ratios except perhaps some squat-built and very heavily muscled athletes. A well muscled mesomorph at my BMI of 27 might be a near perfect speciman but at 27 BMI I am already nearly a whale.

  9. The epidemiological debate currently in this area is as much to do with entrenched positions as it is with the evidence.
    The evidence for many decades indicated that a BMI of about 21 to 27 was pretty good, with a minimum mortality risk at about 23, but really the higher risk of 27 or 21 as compared to 23 wasn’t great. This evidence was simplified into advice to the general public that 20 to 25 was good, 25 to 30 was not so good, and 30+ was bad. There never was strong evidence that the overweight range of 25 to 30 was all that bad, and the main reason for trying to avoid that range was that it put you on the road to the obese category (30+) which is bad.
    Now the evidence has shifted somewhat, with many studies indicating that the minimum mortality risk BMI at present is higher than 23. There is still a lot of uncertainty as to what the minimum risk BMI is exactly, but the cat is among the pigeons, because the dominant paradigm is being challenged.

  10. so the old saying

    “it’s better to be a bit overweight and fit than skinny and unfit”

    would be about right?

    how much has the advertising “creatives” success in making what is a really-hard-to-maintain body size,the ideal?

  11. @Ikonoclast Belly fat is worse than other fat. Using a body fat index measure applied to the Jackson & Pollock chart I come in at “ideal” which means I can indulge without guilt.

  12. “it’s better to be a bit overweight and fit than skinny and unfit”

    May is correct . If you are a bit over you can still be quite fit but the obese cannot .

    Belly fat is worse than other fat

    Yes but fat in your core around the internal organs is the worst of all. Some are predisposed to get it there .

    Older people who have some body fat are known to live longer as when they get ill they have some reserves to help them thru.

    PrQ can we have a post on ‘the Aboriginal problem ‘ soon ?. (they tend to put on core fat )

  13. There isn’t really a paradox, because studies that make a finer gradation in BMI than the standard <18.5, 18.5-25, 25-30, etc, and do a decent job trying to account for reverse causation tend to find that mortality is lowest in the 22 to 25 range (there are studies that suggest higher optimal BMIs, but they usually fail on either one or both of the above accounts, as does the research that is reported on in the link in the post). So there isn’t really a case for raising the over-weight cut-off from 25. There may be a case for increasing the under-weight cut-off, but the question still remains as to the direction of causation: a BMI even moderately below 22 is still associated with increased mortality after accounting for smoking, disease history and recent weight loss, but it could still be the result of diseases which cause gradual weight loss over a long period of time and remain undetected at the time of study. And even if it’s low BMI that’s causal, it may be that low body mass in general, or low lean body mass in particular, is the problem, so it still doesn’t necessarily amount to an “obesity paradox”. What’s more, even if older people should gain fat to protect against certain illnesses, it doesn’t mean it’s a good idea for the rest of the population, and age-specific BMI recommendations may be desirable.

    With regard to the BMIs of football players, often reported in tabloid newspapers to allow the average couch potato to convince himself that “it’s mostly muscle”: the fat-free component of BMI, called the “fat free mass index” or FFMI, averages about 19 for males in the 18-54 age range, with a standard deviation of 1.3 to 1.4. An FFMI of 25 is roughly the limit of natural muscle development (i.e. unaided by steroids), and football players frequently have FFMIs in this range (they are, after all, physical elites). But this is a massive four to five times the standard deviation above the mean: only a tiny percentage of men are this muscly. With an FFMI of 25, 16.7% body fat is enough to place someone in the obese category, with a BMI of 30. But the average male with a BMI of 30, even allowing for a very strong positive relationship between lean body mass and fatness and guessing an FFMI of 21.7 (i.e. a very generous two standard deviations above the overall mean), is probably at least 28% fat, well above the 25% cut-off for obesity.

  14. -BMI calculated at death will put people suffering from a number of diseases to a lower than normal category.
    -Surveys often excluded hospitalised people.
    -Age related results (people over 65) won’t include people who die earlier, including due to unhealthy habits.
    -Moderately overweight may generally be wealthier and have health care.
    -BMI is a poor proxy for fat, even in “normal” people.

    Also see –

  15. @rog

    I’d be prepared to cut average build people some slack to 26 BMI. Beyond that one is badly overweight and should not kid oneself. I’ve stopped kidding myself but I haven’t lost any dang weight yet this year (well maybe one kilo). I found myself almost wishing for a severe but survivable flu this winter to fine me down. A severe case of gastro is good for weight loss too. It’s worked in the past. Not exactly a fun way to lose weight though.

  16. I am incompetent to judge the technical arguments, but I can easily imagine the social pressure there would be against moving the normal range upwards.

  17. The great positive of the BMI is that it is dead easy to measure height and weight then compute using multiplication and division. Almost anyone can calculate it and to the extent that it modifies behaviour it is a useful institution. OTOH, BMI has been demonstrated to be problematic for a number of reasons including that the integral indices of height and weight don’t represent human morphology very well, and that the BMI is an inaccurate measure of the body fat fraction, which is thought to be what we actually want to measure. Other more complex measurements like hip to waist ratios, skin folds, etc, do a better job of assessing body fat but require special equipment and/or knowledge to measure and compute reliably.

    Even allowing for these problems, the relationship between BMI and health is complex and depend on what population being measured. This may be part of the story with the BMI “paradox” – different populations. Figure 1 of the paper below – from a large population study – shows hazard ratios v. BMI split for males and females, and for healthy people and the general population (which includes a chunk of sick people.) The curves are quite different. At the low end, healthy people can have quite low low BMI with little risk, but if you are sick and thin you are likely to fit the medical acronym CTD (circling the drain).

    On the “fat side,” hazard ratios increase more strongly with BMI for healthy people than for the general population. This is what we would expect, if you are sick you are more likely to die but it is less likely to be your BMI that gets you.

    The net effect is that the good BMI range ends up being significantly higher in the mixed populations (like 23-29) than in the healthy population (like 19-25). The data would suggest that if you are sick with a BMI of 21 you should actually fatten up a few units but this prescription might be mixing chickens and eggs. Studies that aim to determine optimum BMI typically exclude some categories of sick people so they don’t skew the data but this means that the result will be at odds with the data for the general population.

  18. My dad is in his 80’s, and has always been skinny. But at some point skinny seems to turn into frail and weak, and then it doesn’t seem to be a good thing.

    The Health Report on Radio National had a story recently where some guy in America reckoned that all studies connecting morbidity to being overweight failed to properly take exercise into account, and if they did, it turned out that exercise was more important than having a normal weight. See here .

    I’m overweight and half fit, and the extra weight loves to stick to my gut. Can anyone tell me how to selectively remove fat from my gut?

  19. @John Brookes: This is the big change that happens as people age. According to US data, the average lean body mass for males aged 60-79 is 59.7 kg (which is not much reduced from the 20-59 age group), but for those who are 80 and over it is 52.3kg (of course, part of the difference could be a cohort effect). Meanwhile, body fat percentage for the two groups is similar: 30.8% and 30.7% respectively. Some argue that obesity in older people might make them more robust to falls, but surely for the purposes of avoiding and withstanding falls it’s more beneficial to stay strong, and given that lean body mass and fat mass are positively correlated, it may be that it’s the extra strength of heavier old people, rather than their extra fat, which helps. But you obviously can’t tell from BMI studies.

    With respect to exercise, some studies certainly do try to take exercise into account, but perhaps not to the satisfaction of that person. For example, the study that Jim Birch linked to took account of overall physical activity, categorised as low, medium or high. But including physical activity in the model had very little effect on the results.

    With regard to losing weight, you can’t spot reduce, but there is evidence that visceral fat is preferentially lost when you start losing weight (

  20. As far as longevity goes calorie restriction is still the gold standard . Its effects have been known and well tested for 40 years .You get less of just about everything that way (cancer too). In the lab animals live longest when fed 50-70% of what that species would eat if food was always there to be eaten whenever they like (a situation most of us are in). Its not fully understood why this is but its not because of weight gain issues . Constant digestion makes lots free radicals ,and there are repair and maintenance functions that work best, or only at all, when our system is empty . Also brain function is maintained better as we age if we eat less -that has big effects on longevity (less falls etc).

    Seventh Day Adventists life expectancy is 90 because they are vegetarian ,dont smoke or drink ,have community interaction, and advocate exercise. To that I would recommend strength training ,stretching, occasional fasting and dont keep your body too clean on the outside or the inside (especially important for children and babies with developing immune systems). Good sleep patterns are also very important .One study showed health outcomes for workers who rotate morning afternoon and night shifts to be as bad as those who smoke a packet a day .

  21. John Brookes :My dad is in his 80?s, and has always been skinny. But at some point skinny seems to turn into frail and weak, and then it doesn’t seem to be a good thing.
    The Health Report on Radio National had a story recently where some guy in America reckoned that all studies connecting morbidity to being overweight failed to properly take exercise into account, and if they did, it turned out that exercise was more important than having a normal weight. See here .
    I’m overweight and half fit, and the extra weight loves to stick to my gut. Can anyone tell me how to selectively remove fat from my gut?


    worked for me.

    six months,three times a week.

    i didn’t change the way i ate and dropped ten kilos.

    it was bloody hard yakka and the first three weeks i couldn’t rollover in bed but that passed.
    it’s a bit expensive though.

  22. oh,neglected to say the weight came off the area between the bottom of the ribs to the top of the legs.

  23. To the extent the paradox is a non-linear relationship between weight and incidence then the reclassification does nothing to change this as far as I can see. They will still be trying to understand why incidence rises again as weight falls further. Surely, no serious medical research on disease incidence is based on bmi, in any event.

  24. The big problem with the BMI is that it works 2 dimensionally whereas the body is in three dimensions, which essentially means as you get taller you must get thinner to stay within the BMI range. For example someone 1.5m tall with a weight of 56kg has a BMI of 25, now if you were to scale that person up to 2m leaving all the same proportions they would weigh in at 133kg (they’re 1.3 times taller, 1.3 times wider and 1.3 times thicker, or 2.37 times the volume)which puts them into the obese category at a BMI of 33.3. If they wanted to maintain a BMI of 25 they’d have to lose 33kg.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s