The U-shaped curve: BMI and healthy life expectancy

BMI is the most criticised measurement in popular health writing, and much of the criticism is fair. It's a two-hundred-year-old ratio that can't distinguish muscle from fat, ignores where fat sits, and was derived from a narrow European population. It also happens to predict population-level health outcomes with enough reliability that essentially every large cohort study uses it. Both of those things are true simultaneously, and the tension between them is worth sitting with.

Why use BMI at all

Three reasons, none of them about BMI being a good measure of individual body composition.

First, availability. Weight and height are collected in nearly every health survey and cohort study ever run. That means BMI is the one body-size variable with decades of longitudinal outcome data attached to it across millions of people. A better measure with no historical data attached is less useful for building a model.

Second, population-level performance. BMI's failures are mostly at the individual level — the muscular athlete misclassified as overweight is the standard example. Across a large population, those cases are a small minority, and the correlation between BMI and adiposity is strong enough that the aggregate signal survives.

Third, reproducibility. You can measure it yourself, at home, consistently. Waist-to-hip ratio and body-fat percentage are better measures that most people measure badly or not at all.

So BMI earns its place in the model as a practical proxy, not as a truth. If you have access to a body composition scan, that tells you more about your own body than this factor does.

The shape of the curve

The relationship between BMI and mortality is not a straight line. Large pooled analyses — most notably the Global BMI Mortality Collaboration's 2016 analysis in The Lancet, covering several million participants across four continents — consistently find a U-shaped or J-shaped curve: elevated risk at low BMI, a broad low-risk trough, and steeply rising risk at high BMI.

The rise on the high side is well explained. Excess adiposity, particularly visceral fat, drives insulin resistance, chronic low-grade inflammation, dyslipidaemia, and hypertension — the cluster underlying type 2 diabetes and cardiovascular disease. It also imposes mechanical load, contributing to osteoarthritis and sleep-disordered breathing. For healthspan specifically, that last group matters a great deal, because joint disease and sleep apnoea are strongly disabling conditions that rarely appear in mortality statistics.

The rise on the low side is more complicated, and we'll come back to it.

The flat bottom, and why it matters

The most practically useful feature of the curve is that its trough is wide and flat. Across large analyses the lowest-risk band typically sits somewhere between about 20 and 25, and the difference in risk within that band is small — well within the noise of any individual's other risk factors.

This has a direct implication that runs against a lot of health messaging: there is no meaningful evidence that moving from a BMI of 24 to a BMI of 22 buys you healthy years. The gains are concentrated in moving into the trough from outside it, not in optimising your position within it.

Reverse causation and the obesity paradox

Two related complications make this literature harder to read than it first appears, and any honest treatment has to address both.

Reverse causation at the low end. A significant share of the elevated risk observed at low BMI is not caused by being thin. It's caused by illness that makes people thin. Undiagnosed cancer, COPD, cardiac cachexia, frailty, and heavy smoking all reduce body weight, and in observational data they show up as "low BMI" with poor outcomes. Studies that exclude deaths in the first several years of follow-up, or that exclude smokers, generally find the left arm of the curve flattens considerably — though usually not entirely, since genuine underweight also carries risks from reduced muscle mass, bone density, and immune reserve.

The obesity paradox. In certain specific patient groups — people with established heart failure, those on dialysis, some elderly cohorts — higher BMI has been associated with better survival. This finding is real but frequently misapplied. The leading explanations involve selection effects, the protective value of metabolic reserve during acute illness, and the same reverse-causation problem operating in the opposite direction. What it does not support is the general claim that carrying excess weight is protective for healthy adults, and it is regularly quoted as though it did.

What BMI misses

Four things, each of which can materially change what a given BMI means:

• Body composition. Muscle is denser than fat. Two people at BMI 27 with substantially different lean mass have substantially different risk profiles, and BMI treats them identically.
• Fat distribution. Visceral fat around the organs is far more metabolically harmful than subcutaneous fat on the hips and thighs. Two people at the same BMI with different distributions differ meaningfully in risk.
• Ethnicity. Cardiometabolic risk rises at lower BMI thresholds in South Asian and East Asian populations, which is why several countries use lower cutoffs — commonly around 23 for overweight rather than 25. The standard thresholds were derived from European-ancestry populations.
• Age and sarcopenia. In older adults, BMI can stay stable while muscle is replaced by fat. A steady BMI in your seventies is not necessarily a steady risk profile, and low muscle mass is independently predictive of functional decline.

Why waist circumference often does better

Because fat distribution matters and BMI can't see it, waist circumference and waist-to-height ratio frequently outperform BMI as predictors of cardiometabolic outcomes — sometimes substantially. Waist-to-height ratio in particular has an appealingly simple heuristic behind it: keep your waist under half your height. It requires one tape measure and no arithmetic.

We haven't made it an input to the calculator, and the reason is practical rather than principled: self-measured waist circumference is unreliable — people measure at inconsistent anatomical points, over clothing, and while holding their breath. Height and weight are self-reported with far less error. It's a deliberate trade of theoretical accuracy for data quality, and it's a trade worth knowing about when you read your result.

How the model scores it

The calculator computes BMI from your height and weight and applies a banded modifier:

Note that the whole 18.5–25 band scores identically. That's deliberate, and it follows directly from the flat-bottom finding — the model declines to imply a precision the evidence doesn't support. Note also that the penalty at the top of the range (−7) is close to the smoking penalty (−8), which reflects how steep the high end of the curve gets in the pooled data.

One last framing point. BMI is the factor people most often approach with shame, and shame is a poor motivator over the decades-long timescales that matter here. The evidence on sustained weight change is fairly consistent that gradual approaches maintained over years outperform aggressive short-term interventions that don't stick. If this factor is where your leverage is, that's worth more than any specific dietary approach.