Relationship Between BMI and Disease, and Longevity

[corybroo]

As noted numerous times here, BMI is at best an approximate for body fat and says nothing about type of body fat.

One third of normal-weight individuals are obese, according to study

The disparity between the two indexes has generated a phenomenon called 'the paradox of obesity with normal weight'—higher than normal body fat percentage in normal-weight individuals.

The researchers analyzed the anthropometric data of 3,000 Israeli women and men, accumulated over several years: BMI scores; DXA scans (using X-rays to measure body composition, including fat content); and cardiometabolic blood markers.

 About one third of the participants—1,000 individuals—were found to be within the normal weight range. Of these, 38.5% of the women and 26.5% of the men were identified as "obese with normal weight"—having excess fat content despite their normal weight.

Matching body fat percentage with blood markers for each of these individuals, the study found a significant correlation between "obesity with normal weight" and high levels of sugar, fat, and cholesterol—major risk factors for a range of cardiometabolic diseases. At the same time, 30% of the men and 10% of the women identified as overweight were found to have a normal body fat percentage.

individuals, being within the norm according to the prevailing BMI index, usually pass 'under the radar.' Unlike people who are identified as overweight, they receive no treatment or instructions for changing their nutrition or lifestyle

they suggest that body fat percentage should become the prevailing standard of health, and recommend some convenient and accessible tools for this purpose: skinfold measurements that estimate body fat based on the thickness of the fat layer under the skin; and a user-friendly device measuring the body's electrical conductivity

"Our study found that obesity with normal weight is very common in Israel, much more than we had previously assumed, and that it is significantly correlated with substantial health risks. And yet, people who are 'obese with normal weight' are not identified by today's prevailing index, BMI. We also found that body fat percentage is a much more reliable indicator than BMI with regard to an individual's general health.

[IgorF]

Thanks corybroo

while nothing unexpected, I would rather say - completely normal and expected findings, with beautiful continuous distribution, it is always good to have such things from different studies to cement the foundations of understanding that constant excess energy is bad and different bodies are sinking it with different success and distribution methods to all possible places due to plethora of reasons.

For me the idea of skin conductivity sensor looks questionable, we already have a market (10years ago unimaginable) of devices with 60% accuracy that reports some health parameters that pretend to be "integral" and even selling something back to users, based on math manipulations over something that inaccurate. (I think this describes my intuitive guessing about BIA deeper https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2543039/)

But nevertheless, the quick and dirty check for triglies being done several times on a stable diet/spending timelines within 70-100 diapason seems very good test on itself (given no issues with morning glu offcourse). If a person is gravitating to 100 that is as mentioned in many other places a strong sign to revise the person's strategy because in a decade things will be fishy and potentially not reversible.

For those practicing CR these things are IMHO not applicable, too many moving parts and packing them into a cohort is not ok from the statistical tooling perspective and mechanical understandig of energy flows in bodies. But we all have others who asks for advice sometimes or relies on care, so understanding the numbering / convincing with data is a nice to have tool.

 

Br,

Igor

 

 

Edited July 16, 2023 by IgorF
added link for BIA

[Saul]

18 hours ago, corybroo said:

One third of normal-weight individuals are obese, according to study

The disparity between the two indexes has generated a phenomenon called 'the paradox of obesity with normal weight'—higher than normal body fat percentage in normal-weight individuals.

The researchers analyzed the anthropometric data of 3,000 Israeli women and men, accumulated over several years: BMI scores; DXA scans (using X-rays to measure body composition, including fat content); and cardiometabolic blood markers.

 About one third of the participants—1,000 individuals—were found to be within the normal weight range. Of these, 38.5% of the women and 26.5% of the men were identified as "obese with normal weight"—having excess fat content despite their normal weight.

Matching body fat percentage with blood markers for each of these individuals, the study found a significant correlation between "obesity with normal weight" and high levels of sugar, fat, and cholesterol—major risk factors for a range of cardiometabolic diseases. At the same time, 30% of the men and 10% of the women identified as overweight were found to have a normal body fat percentage.

individuals, being within the norm according to the prevailing BMI index, usually pass 'under the radar.' Unlike people who are identified as overweight, they receive no treatment or instructions for changing their nutrition or lifestyle

they suggest that body fat percentage should become the prevailing standard of health, and recommend some convenient and accessible tools for this purpose: skinfold measurements that estimate body fat based on the thickness of the fat layer under the skin; and a user-friendly device measuring the body's electrical conductivity

"Our study found that obesity with normal weight is very common in Israel, much more than we had previously assumed, and that it is significantly correlated with substantial health risks. And yet, people who are 'obese with normal weight' are not identified by today's prevailing index, BMI. We also found that body fat percentage is a much more reliable indicator than BMI with regard to an individual's general health.

Thanks Corybroo!

 

Absolutely no surprise.  And I think that Igor is correct, in deprecating electrical conductivity testing.  

Testing subcutaneous fat with skin calipers is a good idea.  Much better (but harder to do) would be a measurement for visceral fat. 

Perhaps the ratio between waist circumference and hip circumference might be a rough test for visceral fat -- but the problem is that some people are wide, and others narrow, at the hips, due to differing bone structure.

But BMI is certainly a very poor measure of both total body fat percentage, and of visceral fat percentage.

  --  Saul 

[kpfleger]

DXA (aka DEXA) scans from cosumer-facing outfits like BodySpec are inexpensive and don't generate very much radiation so reasonable to do annually or even a few times per year and while they are disparaged to some extent by doctors who declare that prescription-ordered DXA is much better (I think primarily for bone health assessment) the consumer-facing machines are probably much better than calipers for measuring body fat %, lean %, & specifically visceral fat.

[Ron Put]

On 7/19/2023 at 4:25 PM, kpfleger said:

the consumer-facing machines are probably much better than calipers for measuring body fat %, lean %, & specifically visceral fat.

Absolutely agree.

I have started doing a DEXA scan yearly, and find it really useful.

It also validated the fat, muscle and bone mass data I get daily from my Withings Body Cardia scale (the scale is very consistent at underestimating my body fat by 2%, so when I see 9% on the scale, it corresponds to 11% based on the DEXA scan.

I think DEXA is also good enough for general bone density tracking.

[amelia1917]

Hey all, first post, please pardon formatting errors.

I haven't seen discussion of this recent paper on the forum, and it has shaped how I think about the relationship between BMI and disease risk, and I thought I'd briefly summarize it and ask for your perspective.

 

Ryan K. Masters (2023) Sources and severity of bias in estimates of the BMI–mortality association, Population Studies, 77:1, 35-53, DOI:10.1080/00324728.2023.2168035

 

I have seen discussion here of papers that adjust BMI analyses for health-status (e.g. number of risk factors or healthy lifestyle factors, here and the lancet analysis referenced earlier in this thread, here).

 

We have also seen discussions of having a longer follow-up time, to eliminate those whose weight was due to an active disease status, further reducing the J shaped relationship (here)

Colloquially, body fat % as a marker for adiposity, quantity of visceral fat approximated via body shape indices (e.g. waist-to-hip ratio, waist circumference, etc) are relevant as well (I can cite later if you'd like, I'm in a bit of a rush). Further, the time-dependence of BMI (thinking of time-integrated obesity, similar to cigarette 'pack-years') is shown to be relevant as well, with many of these studies giving just a single measurement.

Masters claims that leaving any one of these factors out leads to substantial bias in the data. He wrote this flowchart where, assuming we have a *perfect* metric for someone's level of obesity/overweight (which we do not), what measurement biases arise in typical analyses.


Based on this, he did the following (long, tl;dr below)
 

Quote

Aims
This study explores three sources of bias in estimates of the BMI–mortality association: (1) confounding bias from within-BMI-group variation in body shape; (2) positive survival bias among high-BMI samples due to recent weight gain; and (3) negative survival bias among low-BMI samples due to illness-related weight loss (i.e. reverse causation). First, because variation in body shape not captured in standard BMI categories likely confounds the BMI–mortality association, I document variation in body shape within BMI groups and examine the extent to which this variation is associated with poor health, biomarkers of cardiometabolic diseases, and mortality risk. Second, I examine how recent weight gain and recent weight loss affect the composition of BMI groups at time of survey. Obesity rates among the US adult population have increased since the 1980s due to rising exposure to an obesogenic environment (Wang and Beydoun 2007; Reither et al. 2009, 2015; Ljungvall and Zimmerman 2012; Masters, Reither et al. 2013). Rising prevalence likely means that large subsets of overweight and obese samples have spent relatively short durations of time at these BMI levels. Also, long durations of time spent obese can increase risk of disease (Abdullah et al. 2012; Reis et al. 2013; NASEM 2021) and may lead to illness-related weight loss, resulting in reverse causation bias whereby samples with low BMI are composed partly of respondents whose BMI levels were at one time overweight and/or obese (Lawlor et al. 2006; Preston et al. 2013; Cao 2015; Stokes and Preston 2016b). Thus, on the one hand, health profiles of samples with low BMI are likely biased downwards by a subset of respondents whose BMI is low due to recent weight loss. On the other hand, health profiles of samples with over- weight and obese BMI levels are likely biased upwards by a subset of respondents with high BMI from recent weight gain. After documenting the sources of these biases in NHANES 1988–2006, I examine the severity of these biases on estimates
of the US BMI–mortality association. I compare estimates of BMI differences in all-cause mortality risks among US adults aged [45–85) from two different models. I fit the first model using baseline BMI measures to indicate five categorical levels of BMI (<18.5, [18.5–25.0), overweight, class 1 obesity, and class 2+ obesity), and I fit the second model using these five categorical levels of BMI while adjusting for all three sources of bias. I test for age-based differences in the BMI–mortality associations by refitting both models separately to ages [45–70) vs [70–85), and I identify the functional form of the BMI–mortality association by refitting both models using nine categories of BMI. Finally, I use estimates from both models are to calculate the percentage of US adult deaths attributable to overweight and obesity.

Masters combined all of these potential confounders, used the NHANES cohort (both the '88-'94 and '99-'06 data), and performed principal component analysis (PCA) on them. This produced a new model labeled 'adjusted' in the following plots. Note: he fit the model separately for the groups [45-70) and [70-85), providing some supporting evidence for discussions here about the appropriate body weight for those considered elderly. A caveat is he doesn't show an age-specific analysis for BMI within the normal range, only between normal/overweight/obese/obese+. When he does show more fine-grained BMI metrics, it's for the whole population.

First, headline results

The range of 18.5-20.0 being the lowest risk is likely a relief to those here, but that depends on how you're doing in terms of the adjusted model. What did he control for?

 

Quote

The Baseline model shown in this paper includes five categories measured from base- line BMI at time of survey: underweight, overweight, class 1 obesity, and class 2+ obesity, with BMI [18.5– 25.0) as the reference category. The Adjusted model includes 10-factor indicators of the PCs of body shape and includes underweight, overweight, class 1 obesity, and class 2+ obesity measured from weight 10 years prior to survey, with BMI [18.5–25.0) 10 years prior to survey as the reference category.

Unfortunately, when actually describing his principal components, he was a bit vague. (Edit: I just found the supplementary materials, and there's too much there to share here but it's a good read. Brief summary - the PCs are the same between sexes, the PCs are independent of each other, the body shape indices he used were all highly correlated with each other so only one came out in the PCs, ABSI has the lowest correlation to BMI).

Quote

Three principal components are estimated to account for about 90 per cent of variation in US adult body shape. PC1 accounts for about 51 percent of variation and is negately associated with indicators of general adiposity (e.g. −0.40 loadings for weight and skinfold measures). PC2 accounts for about 25 per cent of variation and is associated with indicators of body shape and central adiposity (e.g. 0.60 loading for ABSI and −0.30 loading for thigh circumference) (Reis et al. 2009; Maessen et al. 2014). PC3 accounts for about 13 per cent of variation and indicates stature (e.g. 0.90 loading for height).

Where, from the "marginal effects of PC1/2" below, the PC based on general adiposity is highly inversely correlated to BMI and the lower values it has the worse health status you have. PC2 which grows with ABSI ("A Body Shape Index", (ABSI) = waist(cm) / (BMI^{2/3} × height(cm)^{1/2}) and shrinks with thigh circumference. As PC2 grows, you have worse health, and it is generally much less correlated with BMI. Therefore, this PC which is mostly explained by having a smaller ABSI and larger thigh circumference, explains the majority of in-BMI variation in health status. That is, if I'm interpreting this right!

More information on the PCs I found in the supplementary material. Waist to thigh ratio also quite good! The thigh circumference having a two-way relationship (recall, PC2 we want to be small and PC1 we want to be big) comes from adiposity vs muscle-mass, obviously. For PC3 it is unclear if larger or smaller values are good, but it's almost entirely defined by height and weight, so I suspect that lower values are more desirable but he should have made this more clear.


He then shows that recent changes in BMI can lead to bias as well. Note: he only has 2 measurements, 10 years apart. If he had more frequent measurements, it's likely that the magnitude of this bias could be even greater. Either way, the results are significant!
Those who lose weight to the normal range have a worse health status than those who stayed, and those who gain weight to a heavier status have better health than those who were stable. This is predictable but the results are perhaps surprising; those who have lost weight into the normal range have the highest proportion of the poorest health status of anyone in the table! Tied for the highest level of inactivity, very high CRP, etc etc. Those who gained weight from normal to overweight in those 10 years have a surprisingly good health status as well. Based on the ratios in the NHANES database shown in this plot
These losers/gainers are a significant proportion of the database! Big confounder!

Then he produced the headline plots I linked at the top (though I wish there were more of them), and proposed a new analysis of

Conclusions I have taken: BMI of 18.5-20.0 is likely optimal, though the HR from 20-22.5 or 22.5-25 is small enough that I'm comfortable with it. Within that BMI range, maximizing thigh circumference (leg muscle mass, we've known this is significant) and minimizing ABSI = waist(cm) / (BMI^{2/3} × height(cm)^{1/2}), i.e. minimizing waist circumference, is ideal (though measuring ABSI on its own and reading the paper that it came from might have some value). Maintain that body weight, and stay active. Stuff we already knew, but this paper points out that much of the epidemiology of BMI is biased due to those losing weight for reasons of disease, a population-wide weight gain like we've seen over the last 50 years, and body shape.

One other thing I'll point out is, much like in the original ABSI paper, hip measurements did not come out of this (e.g. waist to hip ratio did not give independent value). The hip measurement always seemed odd to me, as it seems quite genetically dependent on bone structure. Thigh circumference as a marker that we have a lot more agency over.

Recall, also, that this is not a population with optimal or even reasonable nutrition. This paper did not look at that whatsoever. I suspect that the relative effect would only grow with healthy diet, but that's my bias talking.

What do y'all think? Interesting paper? Illuminating? Or are the author's own biases showing too strongly?

Edited September 27, 2023 by amelia1917
mild typo

[drewab]

Hi Amelia, 

Welcome to the CR Forums! I appreciate the comprehensiveness and thoroughness of your post. What stood out to me most was your conclusion, which I would generally agree with. I come at it from a different angle though: I ask what behaviours (diet, exercise, stress reduction, sleep, social connectedness, etc.) contributes to the biomarkers (bloodwork, BP, RHR, imaging, etc.) that will help me increase health span and perhaps lifespan. I engage in these behaviours to optimize my biomarkers and my BMI happens to land pretty consistently at 20.5 to 21.5 at most. When I deviate above or below this threshold I personally start to experience challenges. For example, when my BMI dips below 20 I find that I am too weak and don't have the energy level I desire and insomnia kicks in significantly. When my BMI hits 22+ my blood pressure increases from a stellar 105/62 to more like 115/70... some people might be fine with this, but I feel that lower is better. 

In short, I don't actively seek out a BMI, though I am sure many people do. I engage in behaviours that I believe maximize health/life span and my BMI simply lands where it lands...

 

 

[mccoy]

On 9/25/2023 at 2:19 PM, amelia1917 said:

I can cite later if you'd like, I'm in a bit of a rush

I liked the above. I can just imagine if you had ample time at your disposal! 

Apparently, Masters did a good job of trying to clear up some confusion and confounders in the previous BMI models, which appeared to carry not a little conceptual discordance.

More work has probably to be done, but we are in the right direction.

Basically, though, I would reject all those curves, the previous inverted J and the inverted L proposed by Masters.

For example, my present BMI is 23.4, with an estimated adiposity of around 10-12%. Am I in the same hazard level of people with my same BMI but an adiposity of 40%? Probably not.

This issue is discussed above but there is not a solution, which might have been providing different hazard curves for different subgroups of people with different adiposity, measured by a more suitable metric.

Unless the solutions are present in the article or the supplemental material.

Edited September 27, 2023 by mccoy

[Ron Put]

I am kind of with mccoy on the curve.

Based on what I read, I believe it makes most sense to consider total body mass. It appears to correlate with longevity; thus, short and stocky would be similar to tall and skinny in terms of the likelihood of longevity.

See this, for example:

Longevity, mortality and body weight - PubMed (nih.gov)

The purpose of this study was to analyze the relation of total body weight to longevity and mortality. The MEDLINE database was searched for data that allow analysis of the relationship between absolute body weight and longevity or mortality. Additional data were used involving US veterans and baseball players. Trend lines of age at death versus body weight are presented. Findings show absolute body size is negatively related to longevity and life expectancy and positively to mortality. Trend lines show an average age at death versus weight slope of -0.4 years/kg. We also found that gender differences in longevity may be due to differences in body size. Animal research is consistent with the findings presented. Biological mechanisms are also presented to explain why increased body mass may reduce longevity. Life expectancy has increased dramatically through improved public health measures and medical care and reduced malnutrition. However, overnourishment and increased body size have promoted an epidemic of chronic disease and reduced our potential longevity. In addition, both excess lean body mass and fat mass may promote chronic disease.

[IgorF]

Few more cents on reliability of bmi approach on centenarians example:

 

Review
Nutrition in centenarians
Dorothy B. Hausman ∗, Joan G. Fischer, Mary Ann Johnson
Department of Foods and Nutrition, University of Georgia, 280 Dawson Hall, Athens, GA, USA

doi:10.1016/j.maturitas.2011.01.003

(full paper available via scihub)

Quote

Body mass index is a simple index of weight-for-height that
is frequently used in the assessment of nutritional status. It is
defined as weight in kilograms divided by the square of height in
meters (kg/m2) and allows for the classification of individuals as
underweight (BMI < 18.5), normal weight (18.5–24.99), overweight
(25.0–29.99) or obese (≥30.0) [14]. A low BMI is often associated
with an increased risk of mortality in seriously ill or hospitalized
older adults [15,16], whereas a high BMI or overweight/obesity is
more closely associated with chronic health conditions and not nec-
essarily with increased mortality at advanced ages [17]. There are
some problems, however, associated with the use of BMI as an indi-
cator of health and nutritional status in older adults including the
difficulty of adjusting for loss of height with age and inability to get
a true measure of height due to inability to stand or amputations
[15]. Further, other anthropometric indicators such as mid-arm cir-
cumference [18] and measures of weight change over time [19] may
be better nutritional predictors of morbidity and mortality in older
populations.

Despite the problems associated with its use, BMI has been
included as an indicator of nutritional status in several cente-
narian studies from around the world (Table 2). Typically, BMI
of centenarians is lower than that of their older adult controls
[6,8,20,21] and appears to further decrease as centenarians age
beyond 100 yrs [22]. In addition to nutrition, BMI of centenari-
ans also appears to be dependent on the ethnicity and overall
stature of the representative population group. Thus, average BMI
is less in Italian and Japanese centenarians of overall smaller
stature (BMI ∼ 19) [6,20] as compared to those of larger stature
from northern Europe (BMI ∼ 23–26) [23,24]. BMI was also higher
in African American as compared to Caucasian participants from
the Georgia (US) Centenarian Study, 24.3 vs. 22.2, respectively
[25]. In addition to African American race, a higher BMI in the
Georgia centenarians was also associated with a low intake of
fruits and vegetables and was an independent risk factor for both diabetes and high systolic blood pressure [26]. Given the
many adverse consequences of both under- and overweight, efforts
are needed to maintain a healthy weight, even in the oldest
old.

bold is mine, it is obvious but usually dissolves in statistic aggregates completely

also the older the person is getting - the less meaningful is bmi at all, thus aggregated data studies will be built on a non-linear variable

Br,

Igor

Edited September 28, 2023 by IgorF

[amelia1917]

I'm going to offer the perspective that this analysis is quite strong, even in the face of some of the relationships (which are true!) that you all have pointed out.

Essentially, he has found the impact of time spent at a BMI entirely independent of body shape, for a wide set of ages (including for a population that was between 70-85, which is firmly in the elderly bracket), for the broader American demographic. This does not say anything about centenarians, but does offer quite a diverse demographic, and the conclusions are similar at least for those of different ages.

But what does it mean for BMI to be independent of body shape, if it is a literal measure of body shape? PCA found other factors that explained in-BMI variance in 20 year followup of mortality.

To mccoy's point, you are certainly not in the same hazard level as someone with significant adiposity, but Masters' analysis has controlled for that already. In fact, given that you are well muscled, you are likely at significantly less risk than someone with an 'ideal' BMI who is even somewhat less muscular. But if you maintained the same degree of relative muscularity (depending on its distribution, I guess), but had a BMI that was slightly lower, you might expect to be slightly (that is 20% at most, note that BMIs between <18.5 and 25 all are statistically insignificantly differently associated from the mortality associated with the "ideal" range in his paper) better off.

To Ron's point, the 3rd principal component in the paper, which explained roughly 10% of the variance within-BMI, is explained purely by height and weight, essentially stature. It isn't clear which direction the relationship is, but it's reasonable to suspect that it's negative; that is, 10% of the difference in within-BMI mortality is explained by height, which is relatively minor.

 

To Igor's point, this is a really reasonable critique; this says nothing about centenarians and I would expect that avoiding frailty is the dominant concern. But this paper does point out that even in a quite aged population (70-85) the relationship holds (though only for the major ranges of normal/overweight/obese; I would be biased towards the upper range of normal for the quite elderly).

For people in this forum, who are likely American and under 85 on average, I think it's quite a powerful and relevant result 🙂

[kpfleger]

Great first post. I agree that this paper is good. I've been linking it on Twitter since it was featured in a news story by Lifespan.io. As you say, the difference between 18.5-20 vs 20-25 isn't huge, but this should be a good counter-argument to those studies implying that overweight (25-30) is healthier than "normal weight".