12 NOV 2026 — A definition of obesity that adds waist measurements to height and weight sorted people by their risk of dying better than body mass index (BMI) on its own, according to an analysis of 844,108 adults in China, the UK, the US and Spain, published in Nature on 11 November. The cohorts are observational, so the study shows which labels line up with later deaths, not what caused them.
The definitions being tested came from a 2025 Lancet Commission report. Clinical obesity is excess body fat with organ dysfunction or limits on daily activities; preclinical obesity is excess fat while the organs still work normally.
BMI stays as the first screen. The authors counted obesity at a BMI of 40 or more, at 30 or more plus a large waist or a high waist-to-height ratio, or at 25 to 29.9 plus both, with lower BMI cut-offs of 28 and 24 for Chinese participants. Clinical obesity then needs at least one of the Commission's 18 adult criteria, from heart failure and sleep apnoea to hip or knee osteoarthritis.
How often does BMI get it wrong?
In every cohort, and in both directions. BMI labelled 1.2% to 3.1% of participants obese although neither waist measure was raised. It missed 5.2% to 8.6% whose BMI sat in the overweight range but who met both waist criteria. Overall prevalence rose only modestly under the new rule. The Kailuan Study, which follows employees of China's Kailuan Group, went from 17.8% to 21.3%, and the US All of Us cohort from 42.5% to 48.1%.
Why adjust the clinical obesity count?
Someone with excess fat and heart failure may have heart failure for other reasons. So the authors used an attributable-fraction adjustment, which counts only the share of organ problems that the extra risk seen in people with obesity would account for. In Kailuan the hazard ratio linking obesity to a new organ problem was 1.62, and (1.62 − 1) ÷ 1.62 gives the 0.383 applied there. The paper says the method "cannot determine attribution at the individual level". After adjustment, clinical obesity made up 22.7% to 31.3% of all obesity, or 5.6% of the whole Kailuan sample and 10.9% in All of Us.
Does preclinical obesity usually progress?
Mostly not, over a decade. Among people with obesity and no organ dysfunction at the start, the share who went on to clinical obesity within 10 years, after adjustment, was 21.2% in Kailuan, 8.7% in UK Biobank and 22.6% in SIDIAP, a primary-care records database in Catalonia. All of Us, with shorter follow-up, showed 8.5% over four years. Across the three ten-year figures, that leaves the abstract's 77.4% to 91.3% who did not progress. By our arithmetic, SIDIAP sets the low end (100 − 22.6) and UK Biobank the high end (100 − 8.7). The authors suggest Kailuan's higher rate may partly reflect its health checks every two years.
Is the new label better at predicting death?
Three cohorts had enough follow-up, with 51,792 deaths between them by our sum: 17,642 in Kailuan over a median 15.3 years, 31,812 in UK Biobank over 13.8 and 2,338 in SIDIAP over 11.2. Against people with neither obesity nor organ dysfunction, clinical obesity carried a pooled hazard ratio for death of 1.69 (95% confidence interval 1.64 to 1.73) and preclinical obesity 1.19 (1.14 to 1.24).
The more telling finding is for people BMI alone labels obese. Where their waist measures were not raised, their pooled hazard ratio was 0.93, slightly below people with no overweight or obesity at all. People in the overweight BMI range who did meet the waist criteria came in at 1.06.
Where does Asia fit?
Kailuan is the only Asian cohort, and there the cut-offs matter. The study used waist-to-height thresholds that vary by ethnicity, 0.56 for Asian men and 0.59 for Asian women, in place of the familiar 0.50. Using 0.50 would have lifted the share meeting the waist-to-height criterion in Kailuan from 18.3% to 59.2%. Across the cohorts, people who met only the lower waist-to-height cut-off showed no excess mortality, with a pooled hazard ratio of 1.00. The authors want more prospective work to refine population-specific thresholds.
Who is behind the study?
The corresponding authors are An Pan of Huazhong University of Science and Technology in Wuhan and Francesco Rubino of King's College London. The paper says Rubino conceived the study. Nature's news team reported that he led the effort to create the framework itself. "We did not know until now whether or not this new framework had implications in the real world and a prognostic value," he told the journal. No industry funding was received, the paper says, though four of its 16 authors, Rubino among them, declare consulting fees, grants or honoraria from companies including Novo Nordisk and Eli Lilly. The analysis code is published as an open release.