In defense of being underweight

It is common wisdom to claim that underweight people are at increased risk of death and disease. Epidemiologists, doctors and nutritionists would tend to favour this position but certainly not biogerontologists (1) because weight-loss is very similar to calorie restriction (CR), which is one of the most robust life-extending interventions known. Now the question is which science best informs health policy in an aging, obese world and were do these differences in opinion come from? This review published in 2014 by Luigi Fontana, an expert in human CR, and the distinguished epidemiologist Frank Hu is worth a read as primer (1).

First of all, weight-loss, low bodyweight, low adiposity and CR are not one and the same. CR is, however, associated with initial weight-loss, low bodyweight and adiposity. Mice can be obese but still in a CR-state because they lost weight from their individual ad libitum set point. To define our working hypothesis I favour the idea that low body-weight in healthy people is suggestive of a CR-like state (i.e. mild CR). Being thin is a similar phenotype to CR and it seems like a plausible idea.

As is often the case reality is more complicated than our assumptions. Existing associations may be real but exaggerated or not as well supported as we thought, which is the case for cancer & obesity (4). We know that adiposity is harmful, but recently the controversial idea of an "obesity paradox" was suggested. Observational studies have found that being somewhat overweight could be healthy, but before making a final judgement we have to consider other study types that are in disagreement. What is more, new evidence suggests even the very observational studies are flawed.

The three strongest study designs informing our opinion on CR and leanness are epidemiology, biomarker studies and animal experiments*. One may wonder why I would mention human and animal studies in the same breath if everyone knows that human studies are superior. However, the point is that human studies are only superior if all else is equal. A well-designed, controlled mouse experiment measures healthspan across 100% of the animal's lifespan. If these experiments can be replicated in diverse non-human species the data cannot be ignored anymore. In contrast, observational studies follow human subjects only across 25% of their lifespan and usually much shorter. They are also uncontrolled which can lead to technically insurmountable biases (residual confounding, self-selection, etc).

*other type of circumstantial evidence exists e.g. the Okinawan population, see (1)

Just to drive this point home: One could claim that the CR studies and BMI epidemiology measure something completely different. A low BMI is healthy if you arrived at it per force (e.g. experimental imposition in animal experiments and elite dieters), but not if you were always thin, i.e. "self-selected" (what the observational studies measure). In reality I think there is an intersection between the two at a CR-like state. It is also important to address the BMI literature because for better or worse it is often trotted out as a counter-argument to do human CR.

Historically, animal and biomarker studies have favoured the biogerontologist's view (thinner is better, see [1]) while epidemiology did not. Therefore in this post I will mostly discuss how observational studies have recently shifted closer to the biogerontologist's view. In addition I will mention some studies not discussed in (1), including self-selected bodyweight in animals and the anorexia nervosa literature.

What does BMI measure?
We do not know if BMI reflects a phenomenon similar to CR. It certainly is not equivalent, but it stands to reason that a measure of body-weight and height would be somewhat related to the concept of CR. The largest difference is that CR describes a change from a steady-state whereas BMI describes a stable(?) steady-state at a point in time.

The low BMI group will be composed as follows: naturally thin people, elite dieters (very rare) and those with an eating disorder (orthorexia, anorexia, bulimia, etc). So in reality, we are most likely looking at those "naturally thin" people. But what does this mean? Even this group will be made up of sick-thin people and healthy-thin people.

Sick-thin people may suffer from mild conditions precluding eating, e.g. gastrointestinal disease, or severe conditions with a long latency like cancer. Also obviously smoking and drug abuse and severe stress are known to reduce bodyweight.

Whatever the reasons, "naturally thin" people could be slightly starved. On the systemic level their tissues may be somewhat calorie- and protein-restricted but this is not "noticed" by the CNS, hence they are in equilibrium. However, this stability will diminish the hypothalamic CR- and hunger-response which is essential to many benefits of CR.

Naturally thin people may tend towards a less anabolic diet low in "high quality" protein, methionine and high-glycemic carbohydrates (vegan, vegetarian-like diets, picky eaters, orthorexia)

It is difficult to speculate why a low BMI would induce a CR-like state, but it seems like an intuitive working hypothesis. Within reason, a low BMI index could reflect reduced anabolic signalling, and especially GH action. Somewhat surprisingly, however, adiposity is generally associated with reduced stimulated and basal GH secretion, but this may well be a response to increase anabolic signalling downstream of the receptor (i.e. negative feedback).

A reduced BMI may also reflect early life decreases in growth hormone signalling, which is at least partly true in mice.

Given the evidence from CR and CR-like states we should nevertheless favour the working hypothesis that low BMI is protective of disease. How does this assertion hold up in practise?

BMI and mortality
A meta-analysis of over 200 studies including 30 million participants was recently published. This is no small feat and enables a lot of room for statistical adjustment (3). Since disease is associated with weight-loss statistical correction is essential for accurate results. And what do you know? The better we control for confounding the healthier a low BMI appears. Overall we still see a U-shape of the mortality curve suggesting that a low BMI is unhealthy, but not by much.

For the sake of comparison let us estimate the point at which the excess mortality is the same for low BMI and high BMI. In healthy never smokers: a BMI of 15 is as dangerous as 35, but 17.5 only as bad as 27.5. All in all, the excess risk at a BMI of 15 is only 1.48 (1.32 to 1.65)!

Furthermore the authors say:
In the non-linear dose-response analysis restricted to studies with ≥20 or ≥25 years of follow-up, there was no increased risk at the low BMI range down to a BMI of 20, while risk increased slightly even within the high normal range (BMI of 24-<25) but was more pronounced in the overweight, obese, and severely obese BMI ranges...

What about extreme underweight? These poor skeletons must be dying? Well, no. If we exclude studies with follow-up shorter than 20 years and only look at never smokers to reduce bias we see a symmetry at BMIs of approx. 16 and the deadly 25!

Most impressively the risk in the low BMI category (15-20) continues to decrease when follow-up increases from 20-25 years to >25 years.

Controlled studies of "self-selected" bodymass
These are impossible in humans but necessary. We have seen that the mortality risk in the low BMI category (15-20) continues to decrease when follow-up increases from 20-25 years to >25 years (20-30% of lifespan). What if we extend follow-up to 100%?

Selective breeding for slower early life growth rates also produces lines of small mice that exhibit exceptional longevity. These findings prompted an examination of the hypothesis that small body weight at early ages might be associated with increased life expectancy in a segregating population of mice derived from four common laboratory inbred strains. Our results show that low body weight, as early as 2 months of age, is a predictor of longer life span....Our data support the hypothesis that genetic alleles that reduce early life growth rates or body dimensions may, as a secondary effect, also retard the aging process in mammals. The most dramatic illustration of this tendency comes from studies of dog breeds, among which breed-specific body weight predicts 56% of the variation among breeds in mean longevity (Miller, 1999).

While it may be different in humans, in mice it takes 400-600 days before the survival curves start to separate. That is a follow-up of 30-50% of their natural lifespan! These studies are far from perfect or comprehensive, but certainly suggestive even though they do not measure BMI per se. Similar experiments in non-human primates could rival the rigour of good observational studies and are direly needed. To my knowledge, there is no such data.

Studies of adolescents
Apparently there is a benefit to mandatory military service after all: the medical examination preceding it. In this study (5) around 2.3 million Israeli adolescents were evaluated for long-term mortality in relation to BMI. Overall death was lowest at below-average BMI:
The rates of death per person-year were generally lowest in the group that had BMI values during adolescence in the 25th to 49th percentiles, with higher rates observed among those below the 5th percentile.

The excess mortality at low levels was equal to excess mortality for the 85th percentile. Since participants were young non-CHD deaths were the most important at 8 per 10'000 vs approx. 0.5 per 10'000 for coronary heart disease deaths. What about CHD? Here being underweight is an advantage and at worst it is neutral.

Although this study suggests elevated non-CVD mortality at low BMIs, which we should not discount, it is important to note that these were likely not-age related conditions given the young age of the subjects (in their 40s). In the end, we see that a low BMI was associated with the prevention of an age-related disease at the cost of increased non-age-related early life mortality, consistent with animal studies. I wonder if there are more such studies and if they were included in the large meta-analysis already discussed? Probably not because this is not a true prospective, controlled observational study. Because of this the authors were unable to perform the adjustments that further improved the results in the above meta-analysis ("we were unable to account for important cardiovascular lifestyle risk factors that may confound the BMI association [incl smoking!]").

Waist circumference and waist to hip ratio: size 0 is almost good enough
The BMJ paper (3) reminds us of the following:
A pooled analysis and a meta-analysis and the European Prospective Investigation into Cancer and Nutrition study reported linear increases in risk of mortality with greater waist circumference and waist to hip ratio, even among older people and within the normal BMI range...

As per the meta-analysis published in 2013 (2):
Waist circumference as low as 65cm is beneficial in women showing a J-shaped relationship with mortality. Waist to hip ratio shows even more of a linear trend and BMI is the only exception, which is truly odd.

Using a quick definition of the dreaded size 0 from wikipedia we get "56-81 cm" up to "64-89" for stomach and hip measurements which yields a waist to hip ratio of 0.69 to 0.72. Well, maybe it is not true that the fashion industry sends "dangerous and unrealistic messages". Maybe the messages are just unrealistic.

Better than expected survival in anorexia nervosa (AN)
The long-term survival of mild cases is surprisingly good despite excess mortality attributable to AN/starvation, hypercholesterolemia, mental disease, suicide, substance abuse and overt malnutrition. There are trends for reduced CVD and cancer in some studies (Korndörfer et al.).

The mortality rate, particularly for AN, was considerably lower for those studies that had a long follow-up period, such as that by Korndörfer et al,9 which showed an SMR for AN of 0.71 and had 27.1 years of follow-up, or the study by Crow et al,21 with an SMR of 1.7 and a follow-up of 18.13 years...Twelve studies described deaths from suicide in patients with AN, and analysis showed that the weighted annual mortality due to suicide in AN was 1.39, which means that 1 in 5 individuals with AN who died had committed suicide.

And another meta-analysis:
although multiple articles have asserted that AN has one of the highest premature mortality rates of any disorder, [2,4,5,68] after re-examining the source articles cited, we found the evidence to be ambiguous when comparisons were based on the SMR, which standardizes mortality against expected rates in a common population (and thus allows for a valid comparison).

Korndörfer et al.

Given these findings I am curious to see the results of this ongoing meta-analysis and whether they will be able to control for alcoholism and smoking:
Catalá-López, Ferrán, et al. "Anorexia nervosa and cancer: a protocol for a systematic review and meta-analysis of observational studies." Systematic Reviews 6.1 (2017): 137.

Presently the data is rather mixed, for example (Mellemkjaer et al):
In total, 366 cases of cancer (excluding non-melanoma skin cancer) were seen among women with anorexia nervosa, and the IRR for all cancer sites was 0.97 (95% CI = 0.87-1.08) adjusted for age, parity and age at first child. There were 76 breast cancers corresponding to an adjusted IRR of 0.61 (95% CI = 0.49-0.77). Significantly increased IRRs were observed for esophageal, lung, and liver cancer.

It will be quite interesting to find out if esophagal cancer is due to purging and smoking, lung cancer due to smoking and liver cancer due to alcohol abuse resulting in an overall decrease of non-AN related cancers.

Whether AN is a reasonable model of CR to begin with is another question. The disease certainly replicates some benefits of CR on bio-markers but not all. AN is associated with a pro-inflammatory state and increased circulating TNF-α, IL-6 and IL1-β as per meta-analysis (Solmi et al,), but not all parameters show these trends (Omodei et al.)

Conclusion and outlook

We can say with some pride that the biogerontologic approach is "validated" or even vindicated by epidemiology. Evolutionary reasoning and animal experiments appear to predict the BMI epidemiology based on the concept of CR, but we will have to see how the story unfolds over the coming years to be sure.

There are also some real world implications. Incorrect understanding of the epidemiology can bias ethics committees and make human CR studies more difficult and expensive. It wastes the time of doctors who are taught to harass thin people from outdated textbooks.

This is no empty threat. This bias already ruined a good CR study and wasted taxpayer money (see also: "Set up to fail: are scientists stupid?").

In general our society is getting more accepting of mildly overweight people while underweight people are more and more marginalized. Yet we should not forget the problem of obesity is much larger than the problem of anorexia. Given the evidence, regulations banning models below a BMI of 18.5 seem questionable. We could with the same confidence say that models with a BMI higher than 24-25 are promoting dangerously unhealthy obesogenic ideals, because that's the point of symmetry on the BMI-mortality graph. Of course one has to keep in mind that models may be at elevated risk for anorexia and argue that this is the reason for the cut-off. However, most people claim that modelling promotes a body image that is unhealthy. This is unscientific and wrong and body-shaming.

In fact, we can go a step further. We know from animal experiments that you may have to suffer a little hunger (maybe not much) for the benefits of CR. So if anyone out there benefits from CR it might be young women obsessing and hungering over skinny bodies and healthy diets. This is an extremely serious topic that needs to be investigated: is the grey zone between anorexia and normal weight a CR-like state in humans? How can we safely promote life on the edge?

Researchers should try to evaluate "elite dieters" (e.g. young health conscious women and women in modelling jobs) as well as people who meet the criteria for mild anorexia and orthorexia with extreme caution about extrinsic risk factors like mental health. Given the extreme dieting we see in elite runway models, which is often accompanied with some form of orthorexia, it is surprising that no one investigated this population.

If we become more tolerant of thin people this might help us fight the obesity pandemic.

Thank's to Whole Health Source for inspiration and calling my attention to (5) and (3).

"A high amount of body fat can lead to weight-related diseases and other health issues and being underweight can also put one at risk for health issues. "

1. Fontana, Luigi, and Frank B. Hu. "Optimal body weight for health and longevity: bridging basic, clinical, and population research." Aging cell 13.3 (2014): 391-400.

2. Carmienke, S., et al. "General and abdominal obesity parameters and their combination in relation to mortality: a systematic review and meta-regression analysis." European journal of clinical nutrition 67.6 (2013): 573.

3. Aune, Dagfinn, et al. "BMI and all cause mortality: systematic review and non-linear dose-response meta-analysis of 230 cohort studies with 3.74 million deaths among 30.3 million participants." bmj 353 (2016): i2156.

4. BMJ. 2017 Feb 28;356:j477. doi: 10.1136/bmj.j477. Adiposity and cancer at major anatomical sites: umbrella review of the literature. Kyrgiou et al.

5. Twig, Gilad, et al. "Body-mass index in 2.3 million adolescents and cardiovascular death in adulthood." New England Journal of Medicine 374.25 (2016): 2430-2440.

Obesity & GH:
Obesity and Growth Hormone Secretion
Takara L. Stanley

Bodyweight in animals
Miller, Richard A., et al. "Big mice die young: early life body weight predicts longevity in genetically heterogeneous mice." Aging cell 1.1 (2002): 22-29.

Anorexia nervosa:

Korndörfer, Sergio R., et al. "Long-term survival of patients with anorexia nervosa: a population-based study in Rochester, Minn." Mayo Clinic Proceedings. Vol. 78. No. 3. Elsevier, 2003.

Arcelus, Jon, et al. "Mortality rates in patients with anorexia nervosa and other eating disorders: a meta-analysis of 36 studies." Archives of general psychiatry 68.7 (2011): 724-731.

Mellemkjaer, Lene, et al. "Cancer incidence among patients with anorexia nervosa from Sweden, Denmark and Finland." PloS one 10.5 (2015): e0128018.

Omodei, Daniela, et al. "Immune-metabolic profiling of anorexic patients reveals an anti-oxidant and anti-inflammatory phenotype." Metabolism 64.3 (2015): 396-405.


Keshaviah, Aparna, et al. "Re-examining premature mortality in anorexia nervosa: a meta-analysis redux." Comprehensive psychiatry 55.8 (2014): 1773-1784.

Solmi, Marco, et al. "Inflammatory cytokines and anorexia nervosa: A meta-analysis of cross-sectional and longitudinal studies." Psychoneuroendocrinology 51 (2015): 237-252.