What BMI Actually Measures, and What It Misses

Body mass index is the most widely used number in health, and the most widely misread. It appears on insurance forms, in clinic notes, on gym intake questionnaires and at the top of most weight-loss apps. It is also a ratio of two numbers that says nothing whatsoever about what your body is made of.

That does not make it useless. It makes it a specific instrument with a specific job, and the confusion comes almost entirely from asking it to do a different one. What follows is an account of where the index came from, what the large mortality studies actually show, the two opposite ways it misclassifies people, and what to measure alongside it.

A nineteenth-century average, repurposed

The ratio behind BMI predates the concept of obesity medicine. Adolphe Quetelet, a Belgian astronomer and statistician working in the 1830s, wanted to describe how human physical traits were distributed around a population mean. He found that among adults, weight scaled roughly with the square of height. The ratio he published was a descriptive tool for populations, and he made no claim about individual health.

It acquired its modern name and purpose more than a century later. Ancel Keys and colleagues, writing in the Journal of Chronic Diseases (1972), compared several simple height-weight ratios against densitometry, the underwater weighing that was then the reference standard for measuring body fat. Across five cohorts they concluded that weight divided by height squared correlated best with measured fatness while being least distorted by height, and they gave it the name body mass index.

Keys was direct about the limits. The index was, in his phrasing, satisfactory for population studies. The leap from that to a number printed on an individual patient's chart happened afterwards, and largely without argument.

Where the index genuinely holds up

The case for BMI rests on scale. The Prospective Studies Collaboration pooled fifty-seven studies covering close to 900,000 adults in the Lancet (2009) and found all-cause mortality was lowest between roughly 22.5 and 25.0, with each five-unit increase above that band associated with about 30% higher mortality. The Global BMI Mortality Collaboration repeated the exercise in the Lancet (2016) across more than ten million participants on four continents and reached a consistent conclusion, with the association strengthening once smokers and people with existing disease were excluded from the analysis.

No cheaper measurement produces a relationship that robust. BMI needs a scale, a tape and no training. It is reproducible between observers, it does not drift with equipment calibration, and it can be collected on millions of people at effectively no cost. For public health surveillance, for tracking a population across decades, or for triaging who warrants a closer look, that combination is hard to beat.

The mistake is treating a curve fitted across a million people as a statement about the one person standing on the scale. A population-level association tells you how risk shifts on average as a group's weight distribution moves. It does not tell you which individuals within that group carry the risk.

The muscle problem

BMI cannot distinguish tissue. Muscle is denser than fat, so a body carrying a lot of it registers as heavy for its height, and the index reports that heaviness as excess.

The effect is not marginal. Ode and colleagues, in Medicine and Science in Sports and Exercise (2007), compared BMI classification against measured body fat in college athletes and found the standard overweight threshold flagged a substantial share of male athletes whose body fat percentage sat comfortably in the healthy range. Sensitivity for detecting genuine excess fat was poor in exactly the group whose training makes them heavy. Anyone who has watched a rower, a rugby forward or a competitive powerlifter receive an obese classification has seen this failure in real time.

The practical reach of this problem is narrower than the internet suggests, because most people are not athletes. But it still matters clinically. An index that misfires on muscular bodies will also misfire on anyone partway through a serious resistance-training programme, whose scale weight may hold steady or even rise while their body composition improves substantially.

The opposite error, and it is more common

The failure that gets less attention runs the other way. A person can sit squarely inside the healthy BMI range and still carry a fat mass and fat distribution associated with elevated cardiometabolic risk.

Romero-Corral and colleagues described this pattern in the European Heart Journal (2010) under the name normal-weight obesity: a BMI in the normal band combined with body fat in the highest tertile, which was associated with metabolic syndrome and, in women, with cardiovascular mortality. Tomiyama and colleagues, analysing NHANES data in the International Journal of Obesity (2016), measured the mismatch from both directions. Close to half of adults classified as overweight, and about a quarter of those classified as obese, were metabolically healthy on standard markers. Meanwhile roughly a third of adults sitting inside the normal BMI band were not.

Two groups are especially poorly served. Older adults lose lean mass steadily with age, so an unchanged BMI across three decades can conceal a slow exchange of muscle for fat. And people with a genetic tendency to store fat viscerally rather than subcutaneously can reach a metabolically meaningful visceral load at a body weight no screening threshold would ever flag.

The threshold is not the same everywhere

The familiar cutoffs were derived largely from European-ancestry populations, and the risk relationship does not transfer cleanly. The WHO Expert Consultation reported in the Lancet (2004) that people of Asian descent tend to show higher body fat percentage and greater cardiometabolic risk at any given BMI than European populations, and it recommended additional public health action points at roughly 23.0 and 27.5 rather than the standard boundaries.

This is not a rounding adjustment. It means a South Asian adult can carry the metabolic risk profile associated with the overweight category while their BMI reads as comfortably normal on a Western chart. Several national bodies have since issued lower thresholds for South Asian, Chinese and other Asian populations. Our own BMI calculator reports both sets of thresholds side by side for this reason: showing only one of them would be presenting a regional convention as a universal fact.

What to measure alongside it

The most useful addition costs about the same as BMI and captures what BMI cannot: where the fat sits.

Waist circumference and waist-to-height ratio

Visceral fat, the kind packed around the abdominal organs, is metabolically active in a way that subcutaneous fat is not, and it tracks disease risk far more closely. Yusuf and colleagues, reporting the INTERHEART study across fifty-two countries in the Lancet (2005), found waist-to-hip ratio a substantially stronger predictor of myocardial infarction than BMI, with the association holding across every region studied. Pischon and colleagues, publishing EPIC cohort data in the New England Journal of Medicine (2008), found abdominal measures predicted mortality even among participants whose BMI was in the normal range.

Waist-to-height ratio is the simplest version of this and the easiest to remember: keep your waist under half your height. Ashwell, Gunn and Gibson's meta-analysis in Obesity Reviews (2012) found this ratio outperformed both BMI and waist circumference alone for detecting cardiometabolic risk, and the 0.5 boundary works across sexes and most ethnic groups without adjustment. NICE adopted it in UK guidance for the same reason.

An estimate of body composition

If you want to know how much of your weight is fat, you have to estimate fat directly. Tape-based circumference methods carry roughly three to four percentage points of error against DEXA and were validated on military populations, so treat a single reading as a rough figure rather than a measurement. Their value is in the trend: measured the same way, by the same person, at the same time of day, the direction of change over several months is far more informative than any one number. The body fat calculator runs several published equations at once so you can see how much they disagree.

Using the number well

BMI works as a first screen and nothing more. If it flags something, that is a prompt to look closer, not a diagnosis. If it flags nothing, that is not clearance.

Three habits make it more useful. First, read it alongside a waist measurement, because the pair answers a question neither answers alone. Second, watch its direction rather than its absolute value, since a stable BMI over five years means something quite different from one that has climbed four units. Third, treat any classification near a boundary as noise; the difference between the top of one band and the bottom of the next is well within the measurement error of a bathroom scale and a morning weigh-in.

What none of this changes is the underlying evidence. The mortality curves are real, and the association between sustained excess adiposity and cardiometabolic disease is one of the better-replicated findings in epidemiology. The argument against over-reading BMI is not an argument that body composition does not matter. It is an argument that a ratio of weight to height, on its own, is too blunt an instrument to tell you where you personally stand.