General information, not medical advice. BMI is one input among many; discuss your own health with a clinician rather than with a formula.
On 17 June 1998, the National Heart, Lung and Blood Institute announced new clinical guidelines on overweight and obesity. The United States would now define overweight as a body mass index of 25 to 29.9, and obesity as 30 or above.
Before that day, the American thresholds were 27.8 for men and 27.3 for women — figures derived from the 85th percentile of an earlier national survey.
Nobody’s body changed. The number did.
The American Medical Association, reviewing BMI in June 2023, put the consequence in its own policy language: the 1998 change moved millions of Americans from “normal weight” to “overweight.” No official count of the reclassified was ever published, which is why you should distrust the precise figures that circulate.
Where the number came from
The arithmetic is older than the clinic. Adolphe Quetelet, a Belgian astronomer and statistician, studied in the 1830s how human weight scales with height across populations. That work sat inside a project he called social physics, built around l’homme moyen — the average man. The 1842 English translation of his treatise contains a chapter titled “Properties of the average man.”
Quetelet was describing populations, not diagnosing individuals, and he did not call anything a body mass index. That name arrived in 1972.
Ancel Keys and colleagues published “Indices of relative weight and obesity” in the Journal of Chronic Diseases, comparing candidate height-weight formulas across 7,424 healthy men in twelve cohorts across five countries. They found the ponderal index performed worst, and that weight divided by height squared — which they termed the body mass index — correlated least with height and best with measured fatness. They preferred it for its simplicity and its applicability, in their phrase, to all populations at all times.
It is worth being precise here, because the story is often told with a line Keys never wrote. He did not declare BMI unfit for individuals. He compared formulas for epidemiological work and picked the best one. The explicit restriction to population-level use is a modern statement, set out most clearly by the Lancet Commission on clinical obesity in January 2025: BMI should be used as a surrogate for health risk at population level rather than as an individual measure of health.
What it cannot see
BMI has no way to distinguish fat from muscle, or to locate either. The NHLBI’s own 1998 guidelines said so in the same document that set the new cutoffs, noting that BMI overestimates body fat in very muscular people and that its limitations must be recognised.
The popular version of this caveat is the heavily built athlete misclassified as overweight. That case is real but rare. The common error runs the other way.
Romero-Corral and colleagues compared BMI against measured body composition in 13,601 adults from a national survey. Using a BMI of 30 or more to identify high body fat, they found sensitivity of 36% in men and 49% in women, against specificity of 95% and 99%. In plain terms: if BMI calls you obese it is almost always right, and if it does not, that tells you considerably less. The study used bioelectrical impedance rather than DXA scanning, which the authors note would have been preferable.
Janet Tomiyama’s group looked at the same gap from the clinical side, across 40,420 American adults. Among people classed as overweight, 47% were metabolically healthy on standard cardiometabolic measures. Among those in the normal-weight band, 31% were metabolically unhealthy.
The cutoffs are not universal, and the WHO said so
A WHO expert consultation published in The Lancet in 2004 examined whether Asian populations needed different thresholds. It found that the BMI at which observed risk begins varies between about 22 and 25 across Asian populations, and the high-risk point between 26 and 31.
What it did next is the part usually reported backwards. The consultation declined to set separate cut-points, concluding that the WHO international classification should be retained, and offered additional public-health action points instead. The same number continues to mean different risks in different populations.
What BMI still does well
None of this makes it useless, and the strongest counter-evidence deserves stating fairly.
The Global BMI Mortality Collaboration pooled 239 studies and 10.6 million participants, restricting its main analysis to 3.95 million never-smokers free of chronic disease and discarding the first five years of follow-up. Mortality was lowest between BMI 20 and 25, and rose steeply above 30 — hazard ratios of 1.45 at grade 1 obesity, 1.94 at grade 2 and 2.76 at grade 3.
That is a large, replicated, population-level signal, and it is why neither the AMA nor the Lancet Commission proposes discarding BMI. Both propose constraining it: the AMA’s 2023 policy directs that it be used alongside visceral fat, body composition, waist circumference and relative fat mass.
One genuinely open dispute
The so-called obesity paradox is unresolved and should not be presented as settled either way.
Katherine Flegal’s 2013 meta-analysis in JAMA, covering 97 studies and more than 2.88 million people, found a hazard ratio of 0.94 for the overweight band and 0.95 for grade 1 obesity — no excess mortality, and possibly less. The Lancet 2016 collaboration, using stricter handling of smoking and pre-existing illness, found 1.07 and 1.45 for comparable bands.
The disagreement is methodological: how much of the apparent protection at moderate overweight is reverse causation, with illness causing weight loss rather than thinness causing death. It has not been resolved, and anyone citing one side without the other is selling you something.
The honest summary
BMI is a nineteenth-century population statistic that acquired a clinical job it was not built for, and a threshold that was moved by committee. It is a decent instrument for comparing groups and a crude one for describing a person. It is specific but insensitive for excess fat, blind to where that fat sits, and calibrated on populations that may not include yours.
That is a real limitation, not a reason to ignore the number entirely. It is a reason to treat it as the beginning of a conversation with a clinician rather than the end of one.
Sources
- Keys A, Fidanza F, Karvonen MJ, Kimura N, Taylor HL. Indices of relative weight and obesity. Journal of Chronic Diseases 1972;25(6):329-343 (reprinted International Journal of Epidemiology 2014).
- National Heart, Lung and Blood Institute. Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults, 1998.
- CDC. Update: Prevalence of Overweight Among Children, Adolescents, and Adults — United States, 1988-1994. MMWR 1997;46(9):199.
- WHO Expert Consultation. Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies. The Lancet 2004;363(9403):157-163.
- American Medical Association. AMA adopts new policy clarifying role of BMI as a measure in medicine, 14 June 2023.
- Romero-Corral A, et al. Accuracy of body mass index in diagnosing obesity in the adult general population. International Journal of Obesity 2008;32:959-966.
- Tomiyama AJ, et al. Misclassification of cardiometabolic health when using body mass index categories. International Journal of Obesity 2016.
- Global BMI Mortality Collaboration. Body-mass index and all-cause mortality. The Lancet 2016;388(10046):776-786.
- Flegal KM, Kit BK, Orpana H, Graubard BI. Association of all-cause mortality with overweight and obesity using standard BMI categories. JAMA 2013;309(1):71-82.
- Rubino F, et al. Definition and diagnostic criteria of clinical obesity. The Lancet Diabetes & Endocrinology 2025.
