Validation of clinical and preclinical obesity across global cohorts

Medical diagnoses typically define disease on the basis of objective signs of organ dysfunction² (see https://www.merriam-webster.com/dictionary/diagnosis). By contrast, the health impact of obesity has historically been assessed using body mass index (BMI)—a proxy for body size—alongside obesity-related comorbidities^3,4,5. This approach provides an aggregate measure of ill health rather than a disease-specific diagnosis.

The absence of a clinically grounded definition has impeded recognition of obesity as a distinct disease, contributing to gaps in care, inconsistent prevention, misallocated resources and persistent stigma^6,7.

To address this, in 2025, a Lancet Commission from The Lancet Diabetes & Endocrinology proposed a new diagnostic framework that distinguishes between two categories of obesity: clinical obesity, defined by excess adiposity impairing organ function or daily activities; and preclinical obesity, characterized by increased adiposity and increased health risks but with preserved organ and whole-body function¹. The framework also recommends supplementing BMI with other anthropometric measures, such as waist circumference (WC), waist-to-height ratio (WHtR) or fat quantification with dual-energy X-ray absorptiometry, when available, to improve the detection of excess adiposity. Although consistent conceptually with diagnostic principles in medicine, the clinical implications of this framework remain uncertain. It is unknown how substantially the recommended use of additional anthropometric measures changes the classification of obesity compared with BMI alone, or whether the distinction between clinical and preclinical obesity captures meaningfully different health states. The natural history of preclinical obesity is also poorly characterized, including the extent to which it remains stable or progresses to clinical obesity over time. More broadly, whether these classifications show consistent associations with prognosis across populations and healthcare settings has yet to be established. We examined these questions using data from 844,108 adults from four large cohorts in China, the UK, the USA and Spain, aiming to evaluate the Lancet Commission’s proposed diagnostic framework in terms of obesity classification, the distribution and longitudinal course of clinical and preclinical obesity and their associations with mortality.

To assess whether incorporating other anthropometric measures (WC and WHtR) improves obesity detection beyond the use of BMI alone, we analysed data from four cohorts: the Kailuan Study (China, n = 137,795), the UK Biobank (UKB, n = 436,203), the All of Us Research Program (AoU; USA, n = 218,841) and the SIDIAP database (Spain, n = 51,269). The characteristics of the study populations are shown in Supplementary Table 1, selection of study participants from each cohort is shown in Supplementary Figs. 1–4 and the details of study design of each cohort are presented in the Methods and Supplementary Methods. Following the Lancet Commission’s proposed diagnostic criteria¹, obesity confirmation required any of the following three components: (1) BMI ≥ 40.0 kg m⁻²; (2) BMI ≥ 30.0 kg m⁻² (≥28.0 kg m⁻² for Chinese individuals) plus elevated WC or elevated WHtR; or (3) BMI 25.0–29.9 kg m⁻² (24.0–27.9 kg m⁻² for Chinese individuals) plus both elevated WC and elevated WHtR. Elevated WC was defined as ≥102 cm for male and ≥88 cm for female individuals (≥90/85 cm (male/female) for Chinese individuals)^8,9, and WHtR thresholds varied by ethnicity (non-Hispanic white: male/female ≥0.57/0.58; non-Hispanic Black: ≥0.55/0.57; Hispanic: ≥0.57/0.59; Asian: ≥0.56/0.59; others: ≥0.50)¹⁰.

Classification based on BMI alone resulted in both over- and under-identification of obesity across all four cohorts (Fig. 1 and Extended Data Table 1). The proportions classified as obese by BMI but without elevated WC or WHtR were 3.1% in the Kailuan Study, 1.9% in the UKB, 1.9% in the AoU and 1.2% in the SIDIAP, whereas 6.6%, 5.2%, 7.5% and 8.6%, respectively, of individuals with BMI-defined overweight met both WC and WHtR criteria for obesity-level adiposity. With the Lancet Commission’s recommended multi-criteria definition, hereafter referred to as the combined anthropometric definition of obesity (CAD-obesity), overall obesity prevalence increased from 17.8% to 21.3% in the Kailuan Study, 24.0% to 27.3% in the UKB, 42.5% to 48.1% in the AoU and 18.7% to 26.1% in the SIDIAP. Overall, CAD-obesity yielded a modest net increase in obesity prevalence of 3.5% in China, 3.3% in the UK, 5.6% in the USA and 7.4% in Spain.

Clinical obesity was defined as CAD-obesity accompanied by at least one of the 18 adult diagnostic criteria specified by the Lancet Commission. These criteria represent obesity-related organ dysfunction or functional impairment across several systems, including cardiovascular (for example, heart failure, atrial fibrillation, thromboembolic disease and hypertension), respiratory (obstructive sleep apnoea and chronic obstructive pulmonary disease), metabolic (combined dysglycaemia and dyslipidaemia), hepatic (MASLD with fibrosis), renal (chronic kidney disease), neurological (intracranial hypertension), musculoskeletal (hip or knee osteoarthritis), reproductive (infertility), urological (urinary incontinence) and lymphatic dysfunctions, together with limitations in daily activities. Criteria for clinical obesity were operationalized using predefined clinical, laboratory, imaging or diagnostic criteria adapted to the variables available in each cohort. Detailed definitions and cohort-specific assessment methods are provided in the Supplementary Methods and Supplementary Table 2.

However, not all organ dysfunction that coexists with CAD-obesity can be assumed to be attributable to obesity. We therefore applied an attributable fraction (AF) approach to provide a population-level estimate of attribution and to adjust subsequent estimates accordingly¹¹. The AF was calculated as: ({rm{AF}}=frac{{rm{HR}}-1}{{rm{HR}}}), where HR denotes the hazard ratio for the association between CAD-obesity and risk of incident organ dysfunction with adjustment for relevant confounders using Cox proportional hazards models. The cohort-specific HRs used to derive the AF estimates were 1.62 (95% confidence interval (CI), 1.58–1.66) in the Kailuan Study, 1.68 (1.64–1.72) in the UKB, 1.67 (1.61–1.73) in the AoU and 2.07 (1.97–2.18) in the SIDIAP (Extended Data Table 2), corresponding to AF values of 0.383, 0.405, 0.401 and 0.517, respectively.

Because the distinction between clinical and preclinical obesity depends on the presence of organ dysfunction among individuals with CAD-obesity, we first quantified the observed crude prevalence of organ dysfunction and then applied the cohort-specific AF estimates to account for the proportion plausibly attributable to obesity at the population level. The crude prevalence of organ dysfunction among people with CAD-obesity ranged from 56.7% in the AoU to 75.3% in the UKB (Fig. 2; Extended Data Table 3). AF-adjusted clinical obesity accounted for 26.3% (95% CI: 25.2–27.3) of total obesity in the Kailuan Study, 30.5% (29.4–31.6) in the UKB, 22.7% (21.5–24.0) in the AoU, and 31.3% (30.9–31.8) in the SIDIAP (Fig. 2; Extended Data Table 3). Accordingly, the proportion of individuals with CAD-obesity but without any detectable organ dysfunction—representing the minimum estimate of preclinical obesity—ranged from 24.7% in the UKB to 43.3% in the AoU, and the maximum estimate of preclinical obesity (1 minus the AF-adjusted prevalence of clinical obesity) ranged from 68.7% in the SIDIAP to 77.3% in the AoU (Extended Data Table 3).

The overall prevalence of AF-adjusted clinical obesity in the full population was 5.6% (95% CI: 5.4–5.8) in the Kailuan Study, 8.3% (8.0–8.6) in the UKB, 10.9% (10.4–11.5) in the AoU and 8.2% (8.1–8.3) in the SIDIAP (Fig. 2 and Extended Data Table 3).

Sensitivity analyses restricting to the nine organ dysfunction indicators available across all datasets yielded results that were highly consistent with the primary analyses. Prevalence estimates changed minimally under the restricted definition, indicating that additional cohort-specific indicators contributed little to classification (Supplementary Tables 3 and 4). Indicators that were not universally available were individually rare, and most participants presenting them also had at least one of the nine common indicators, supporting the robustness and comparability of the classification across cohorts (Supplementary Table 5).

To assess the progression from preclinical to clinical obesity, we performed an analysis among individuals with CAD-obesity but without any organ dysfunction at baseline. The AF-adjusted 10-year incidence of clinical obesity was 21.2% in the Kailuan Study, 8.7% in the UKB and 22.6% in the SIDIAP; because of a shorter follow-up in the AoU cohort, only an AF-adjusted 4-year incidence was calculated and an incidence of 8.5% was found (Fig. 3). Thus, approximately 70–90% of individuals with preclinical obesity remained stable over a decade. The incidence rates were similar between male and female individuals (Supplementary Fig. 5), and sensitivity analysis restricting to the nine organ dysfunction indicators yielded similar results (Kailuan Study: 21.4%; UK Biobank: 8.8%).

Cox proportional hazards models were used to assess associations of BMI-based obesity and CAD-obesity with all-cause mortality, adjusting for sociodemographic covariates. The AoU cohort was excluded because of limited follow-up. Median follow-up was 15.3 years in the Kailuan Study (17,642 deaths), 13.8 years in the UKB (31,812 deaths) and 11.2 years in the SIDIAP (2,338 deaths).

Compared with individuals with no overweight or obesity, participants with BMI-defined overweight or obesity but without CAD-obesity had a lower all-cause mortality risk (pooled HR 0.88 [95% CI: 0.86–0.90] and 0.93 [0.86–0.99], respectively). By contrast, individuals with CAD-obesity, regardless of whether their BMI was in the overweight or obese range, had a significantly higher, albeit modest, mortality risk (pooled HR 1.06 [1.03–1.10] and 1.07 [1.05–1.10], respectively; Fig. 4a and Extended Data Table 4).

Further cause-specific mortality analyses were done in the UK Biobank (the only cohort with available cause-specific mortality data), including cardiovascular mortality, obesity-related cancer mortality and their combination (Supplementary Table 6). After multivariable adjustment, BMI-defined overweight or obesity without CAD-obesity was not associated with increased risks of these cause-specific mortality outcomes. By contrast, CAD-obesity was associated with higher risks, including among individuals in the BMI-defined overweight range.

In further sensitivity analyses, exclusion of deaths occurring within the first 5 years of follow-up did not materially alter the associations, arguing against reverse causation as a major explanation for the findings (Supplementary Table 7). The proportion of missingness was low across all cohorts, with no variable exceeding 20% (Supplementary Table 8), and a sensitivity analysis restricted to participants without missing variables also yielded robust results (Supplementary Table 9). Stratified analyses by smoking status confirmed that smoking was a strong effect modifier; however, among never-smokers, CAD-obesity remained consistently associated with higher mortality risk, whereas BMI-defined overweight and obesity in the absence of CAD-obesity showed no or modest mortality risk after multivariable adjustment (Supplementary Table 10).

We evaluated the association between clinical obesity and mortality risk. Participants were first categorized into four baseline groups: (1) no CAD-obesity and no organ dysfunction (reference group); (2) no CAD-obesity with organ dysfunction; (3) CAD-obesity without organ dysfunction (preclinical obesity); and (4) CAD-obesity with organ dysfunction (clinical obesity). Clinical obesity was associated with a substantially higher mortality risk (pooled HR, 1.69; 95% CI, 1.64–1.73), whereas preclinical obesity showed a modest association (pooled HR, 1.19; 95% CI, 1.14–1.24; Fig. 4b and Extended Data Table 5). Cohort-specific estimates were directionally consistent.

Sensitivity analyses by excluding deaths occurring within the first 2 or 5 years of follow-up or by incorporating waist-to-hip ratio into the diagnostic indicators did not materially alter the associations (Extended Data Table 6). The association between clinical obesity and mortality was stronger in female individuals and in those aged younger than 60 years (P < 0.001 for interactions; Extended Data Fig. 1 and Supplementary Table 11). In the age-stratified analysis, sensitivity analyses by further excluding deaths occurring within the first 5 years or restricting participants without missing variables yielded similar results to those of the primary analysis (Supplementary Tables 12 and 13), providing further reassurance against reverse causation and selection biases.

To disentangle the relative contributions of BMI-defined obesity, CAD-obesity and organ dysfunction to mortality risk, participants were further classified into eight groups according to the presence or absence of each component. In groups defined by organ dysfunction and CAD-obesity, BMI-defined obesity conferred little additional risk, when compared with the BMI-defined non-obese status. On the other hand, CAD-obesity showed a consistently higher mortality risk, and the highest mortality risks were observed among participants with both CAD-obesity and organ dysfunction (that is, clinical obesity) (Extended Data Table 7).

Separate cause-specific mortality analyses of clinical and preclinical obesity in the UK Biobank dataset showed that clinical obesity was consistently associated with substantially higher risks, with effect sizes generally stronger than those observed for all-cause mortality, whereas preclinical obesity showed elevated but consistently lower risks (Supplementary Table 14).

We performed further complementary analyses to examine the appropriateness of the ethnicity-specific WHtR thresholds used to confirm obesity status. First, the relationship between BMI and WHtR was broadly consistent across cohorts (Pearson r ≥ 0.64; Extended Data Fig. 2). On the basis of fitted linear regression models, the conventional WHtR cut-off of 0.50 corresponded approximately to the BMI-defined overweight threshold, whereas WHtR values corresponding to the BMI-defined obesity threshold were generally higher than 0.55 (from 0.55 to 0.60 across the four cohorts).

Mortality analyses showed a similar pattern. Restricted cubic spline models indicated that mortality risk began to increase at WHtR values around 0.55 (Extended Data Fig. 3). Consistently, participants with a WHtR between 0.50 and the ethnicity-specific thresholds had no excess mortality risk (pooled HR, 0.99; 95% CI, 0.97–1.02), whereas those with a WHtR above the ethnicity-specific threshold had a higher risk (pooled HR, 1.21; 95% CI, 1.18–1.24; Supplementary Table 15).

Using 0.50 instead of the ethnicity-specific thresholds substantially increased the proportion of participants defined as having obesity by meeting the WHtR criterion across all four cohorts (Kailuan, 59.2% versus 18.3%; UKB, 66.7% versus 27.2%; AoU, 74.3% versus 48.8%; SIDIAP, 71.1% versus 33.3%). When these thresholds were incorporated into the definition of confirmed obesity, the same risk pattern was observed: participants classified only by the 0.50 threshold showed no excess mortality (pooled HR, 1.00; 95% CI, 0.96–1.03), whereas those meeting the ethnicity-specific thresholds had a higher mortality risk (pooled HR, 1.20; 95% CI, 1.18–1.22; Supplementary Table 16).

To our knowledge, this is the first large-scale, multinational study to comprehensively evaluate the Lancet Commission diagnostic framework¹ against the widely used BMI-based classification. Previous studies have examined individual components of the framework—such as expanded anthropometric criteria or prevalence estimates of clinical obesity—but have lacked integrated assessment or longitudinal follow-up^{12,13,14,15,16}. This study evaluated several interrelated elements of the new diagnostic framework: the implications of using combined anthropometric criteria to confirm obesity; their effects on prevalence and risk stratification compared with BMI alone; the importance of selecting appropriate anthropometric cut-offs; and the prognostic implications of distinguishing clinical from preclinical obesity.

With regard to anthropometric classification, our findings show that BMI alone both over- and under-identifies obesity and does not reliably distinguish individuals who have an increased mortality risk. These results support the Commission framework’s use of complementary anthropometric measures alongside BMI to improve obesity classification and risk stratification. Although WC and WHtR are, like BMI, indirect measures of adiposity, increased values for these measures are (for anatomical and physiological reasons) more likely to reflect fat-mass expansion, whereas a higher BMI might also reflect greater lean mass. Their correlation with direct body-composition measurements¹⁰ supports their use as a pragmatic means of detecting excess adiposity where direct measurement is impractical. Future studies should determine whether direct measurement of fat adds meaningful precision and could ultimately replace combined anthropometry if reliable bedside technologies become widely available.

We observed substantial discordance between BMI-based classification and CAD-obesity, with BMI overestimating obesity in 1.2–3.1% and underestimating it in 5.2–8.6% of participants across cohorts. When considered against national obesity estimates from the 2025 World Obesity Atlas¹⁷, these proportions suggest that BMI misclassification affects tens of millions of individuals in countries with large populations, underscoring the potential public-health implications of seemingly modest differences in classification.

However, the effects of using a combined anthropometric definition of obesity depend, crucially, on how the Commission’s recommendations are operationalized. Unlike our findings, previous studies in US cohorts reported that applying the Lancet Commission framework substantially increased overall obesity prevalence^{12,13,14,15,16}. These discrepancies mainly reflect methodological differences, particularly the choice of anthropometric cut-offs and incomplete adherence to the framework’s recommendations. In one study¹⁴, higher prevalence estimates were driven by the use of a single low WHtR cut-off of 0.50 across populations, potentially classifying individuals with milder increases in adiposity and associated risk as having obesity. Another study¹⁶ reported an overall obesity prevalence of 60% in the AoU cohort, compared with 48.1% in our analysis. In addition to using a single WHtR cut-off of 0.50, the study applied the cut-off across the full BMI range, including individuals with a normal BMI. This approach diverges from the Commission’s recommendation that anthropometric confirmation be used mainly in individuals with BMI values below—but near —the BMI obesity threshold¹. Applying anthropometric criteria outside this context risks classifying individuals with normal body weight and localized abdominal adiposity—such as those with lipodystrophy or related conditions, which are biologically and clinically distinct—as having obesity. For this reason, the Commission retained BMI as the initial screening tool¹, and we limited the assessment of obesity to participants within the conventional BMI-defined overweight and obesity ranges.

In closer alignment with the Commission’s guidance, we applied obesity-level, ethnicity-specific thresholds for WC and WHtR. In our data, the conventional WHtR cut-off of 0.50 corresponded approximately to the BMI-defined overweight threshold, whereas WHtR values corresponding to the BMI-defined obesity threshold and increasing mortality risk were generally observed around or above 0.55. This convergence between anthropometric and mortality analyses supports the use of higher WHtR thresholds for obesity classification and is consistent with the previously proposed cut-off of 0.55–0.59 (ref. ¹⁰). Further prospective evaluation will be required to refine population-specific thresholds.

The application of obesity-level thresholds for WC and WHtR further illustrates the limitations of BMI-based classification. Although these measures remain indirect proxies of body fat, the reclassified groups had distinct mortality profiles, showing that conventional BMI categories encompass individuals with heterogeneous risk and that additional measures provide clinically relevant stratification beyond BMI alone. This is particularly consequential for the overweight category. The Lancet Commission framework intentionally does not retain overweight as a separate intermediate-risk category; instead, it extends the definition of obesity to individuals with a BMI classified as overweight who meet additional anthropometric criteria indicating obesity-level adiposity. Our data support this approach: individuals with a BMI in the overweight range who met CAD-obesity criteria had a mortality risk comparable with that of individuals with a BMI in the obesity range who also met CAD-obesity criteria. By contrast, those with a BMI classified as overweight who did not meet CAD-obesity criteria had no excess risk and, in several analyses, had a lower risk than did individuals with a normal BMI. The CAD-obesity criteria therefore identify adiposity-related risk that would otherwise remain obscured within the conventional overweight range.

The implications are substantial. An estimated 1.5 billion adults worldwide are classified as overweight by BMI¹⁷, yet our findings indicate that individuals in this population should not all be regarded as being at increased risk. Our data suggest that combining BMI with WC and appropriately calibrated WHtR cut-offs, such as those evaluated here, can identify higher-risk individuals without substantially increasing overall obesity prevalence. This selective reclassification provides a clinically meaningful basis to inform individual care, health policy and resource allocation.

Anthropometric confirmation improves the identification of mortality risk associated with excess adiposity. However, because clinical obesity denotes an active disease state, its diagnosis requires a further step: establishing the presence of organ dysfunction that can plausibly be attributed to that adiposity. This requirement reflects a fundamental principle of disease definition: because similar alterations in organ function can arise from different disease processes, causality cannot be inferred from coexistence alone. For example, not all joint inflammation is caused by rheumatoid arthritis, nor is all heart failure attributable to the same underlying heart disease. In clinical practice, differential diagnosis addresses this uncertainty by considering alternative causes and, where possible, excluding them; it is therefore essential to diagnostic specificity. For this reason, organ dysfunction coexisting with obesity cannot automatically be considered obesity-related; doing so could overattribute dysfunction to obesity and substantially inflate estimates of clinical-obesity prevalence and incidence. Most previous applications of the framework classified all coexisting organ dysfunction as obesity-related without accounting for alternative causes^13,14,15,16, and might therefore have overestimated the prevalence of clinical obesity.

To mitigate this problem, we applied the AF method at the population level to estimate the proportion of organ dysfunction that is statistically attributable to CAD-obesity, and adjusted the prevalence and incidence estimates accordingly. Although this adjustment cannot establish causality in individual participants, it mitigates the risk of attributing organ dysfunction to obesity when other causes could be responsible, thereby reducing systematic overattribution and improving population-level estimates of the prevalence and incidence of clinical obesity. Notably, an earlier US study¹² using a different adjustment approach estimated the clinical-obesity prevalence as 9.7%, close to the 10.9% that was observed in the AoU cohort after AF adjustment. This concordance between studies using different population-level methods to account for plausible attribution provides some reassurance about the robustness of our estimates.

Our longitudinal and mortality analyses provide evidence supporting the distinction between clinical and preclinical obesity. Despite variation across cohorts, 70–90% of individuals with preclinical obesity did not progress to clinical obesity over 10 years. Clinical obesity was also associated with substantially higher all-cause and obesity-related mortality, whereas preclinical obesity showed only a modest excess mortality risk. The observed patterns of progression and differences in mortality risk show that these health states differ not only in their defining clinical characteristics but also in their prognostic implications. The distinct clinical and prognostic profiles of clinical and preclinical obesity provide a foundation for personalized, proportionate care. Preclinical obesity warrants a mostly preventive approach aimed at reducing future risk, with the nature and intensity of intervention guided by the individual’s risk of progression and the presence of relevant comorbidities. By contrast, established functional impairment and higher mortality risk in clinical obesity justify greater urgency and, where appropriate, more intensive treatment and prioritized access. The distinction can therefore guide the purpose, timing and intensity of individual care and support evidence-based decisions on treatment prioritization, coverage and resource allocation.

Notably, using this distinction to guide proportionate care does not imply withholding treatment from people with preclinical obesity. One criticism of the Lancet Commission framework is that distinguishing preclinical from clinical obesity could encourage ‘watchful waiting’ or restrict access to treatment^18,19,20. However, this concern seems to conflate diagnostic classification with treatment eligibility. Like any diagnosis, the distinction between clinical and preclinical obesity provides a snapshot of an individual’s current health status; it does not prescribe inaction or determine eligibility for treatment by itself. Analogous distinctions—such as those between prediabetes and diabetes or adenomatous polyps and carcinoma—guide the purpose, urgency and intensity of intervention without precluding care for prediabetes or adenomatous polyps²¹. Accordingly, preclinical obesity might warrant active preventive treatment, including lifestyle, pharmacological or, in selected cases, surgical interventions, depending on progression risk and comorbidities that could benefit from weight reduction. Treatment decisions should therefore be informed by diagnosis but not reduced to diagnostic category alone.

Several limitations should be acknowledged. The retrospective design of our study and its reliance on electronic health record (EHR) data could limit diagnostic precision and introduce inter-cohort variability. Organ dysfunction was identified using administrative codes rather than standardized clinical assessments, and not all 18 dysfunction indicators were available in every cohort, potentially leading to the underestimation or misclassification of specific organ disease subtypes. This limitation is highlighted by the Kailuan cohort, in which the higher observed incidence of clinical obesity might reflect more frequent (biennial) clinical evaluations. However, sensitivity analyses showed that the additional indicators contributed minimally to classification, supporting the conclusion that differences in the number of available organ-dysfunction indicators are unlikely to be a major driver of the inter-cohort variability observed in the primary analyses. Finally, the AF method cannot determine attribution at the individual level, and future prospective studies using clinically adjudicated diagnoses of clinical and preclinical obesity will be essential to refine estimates of disease burden and progression. Nonetheless, the consistency of findings across geographically and demographically diverse cohorts strengthens the robustness, generalizability and clinical relevance of the proposed diagnostic framework.

In summary, BMI alone misclassifies obesity in a substantial proportion of individuals, both overestimating and underestimating the clinical relevance of excess adiposity, and does not meaningfully discriminate higher mortality risk. Integrating other anthropometric measures (such as WHtR and WC) and clinical criteria reflecting organ function into a multi-criteria framework improves obesity detection and aligns diagnosis more closely with true disease status. Clinical and preclinical obesity represent distinct and prevalent subgroups, with markedly different mortality risks. This study demonstrates the clinical utility and prognostic importance of distinguishing between clinical and preclinical obesity, and supports the global applicability of this classification scheme for guiding care and informing health policy.

Study population

This population-based multinational study analysed data from four large and well-characterized adult cohorts: the Kailuan Study (China), the UK Biobank (UKB; UK), the All of Us Research Program (AoU; USA), and the Information System for the Development of Research in Primary Care (SIDIAP, Spain). The Kailuan Study is an ongoing prospective cohort initiated between 2006 and 2007 among more than 100,000 employees of the Kailuan Group, with biennial follow-up examinations thereafter²². The UK Biobank enrolled around 500,000 individuals aged 37–73 years from 2006 to 2010 with questionnaires, physical examinations and biomarker assessments²³. The All of Us Research Program is a US precision-medicine initiative aimed at building a diverse biomedical database to study the biological, clinical, social and environmental determinants of health²⁴. The SIDIAP database includes pseudonymized EHRs from more than eight million individuals in Catalonia since 2006, with approximately 75% of the regional population represented²⁵.

The Kailuan Study was approved by the ethics committee of the Kailuan Medical Group. The UK Biobank received ethical approval from the North West Multi-Centre Research Ethics Committee. The All of Us Research Program is overseen by the All of Us institutional review board. The SIDIAP analysis was approved by the IDIAP Jordi Gol ethics committee (20/204-P). All studies followed the Declaration of Helsinki, and all participants gave their written informed consent. No additional ethical approval was required for the current study because no new data were collected and all analyses were performed using previously collected data.

Participants aged 18–74 years who were free from stroke, myocardial infarction and cancer at baseline and for whom anthropometric (BMI, WC and WHtR) and organ-dysfunction data were available were included. The final numbers included in the prevalence estimates were 137,795 from the Kailuan Study, 436,203 from the UKB, 218,841 from the AoU and 51,269 from the SIDIAP. The progression from preclinical obesity to clinical obesity was estimated among those with CAD-confirmed obesity but without any organ dysfunction, and the sample size was 8,009, 29,338, 54,535 and 5,273, respectively. For the mortality analysis, the AoU study was excluded owing to its relatively short follow-up duration; for the other three cohorts, the sample size was generally the same as that in the prevalence analysis, except for the Kailuan Study, in which three participants without valid mortality data were further excluded. Details of the study design for each cohort can be found in the Supplementary Methods, and the selection of study participants is in Supplementary Figs. 1–4.

Definitions of clinical and preclinical obesity

Following criteria from the Lancet Commission¹, clinical obesity was defined as excess adiposity plus at least one sign of obesity-induced organ dysfunction. Preclinical obesity was defined as excess adiposity without such dysfunction.

Excess adiposity was defined by any of the following criteria, to be consistent with the proposed framework: (1) BMI ≥ 40 kg m⁻²; (2) BMI ≥ 30 kg m⁻² (≥28 for Chinese individuals) plus elevated WC or WHtR; or (3) BMI 25–30 kg m⁻² (24–28 for Chinese individuals) plus both elevated WC and WHtR. The thresholds for population-specific WC-defined adiposity are ≥102 cm for male and ≥88 cm for female individuals in the Western population, and ≥90 cm for male and ≥85 cm for female individuals in the Chinese population^8,9. The thresholds for race- and sex-specific WHtR-defined adiposity are 0.57 (male) and 0.58 (female) for non-Hispanic white individuals; 0.55 (male) and 0.57 (female) for non-Hispanic Black individuals; 0.57 (male) and 0.59 (female) for Hispanic individuals; 0.56 (male) and 0.59 (female) for Asian individuals; and 0.50 for others¹⁰. To illustrate that the commonly used WHtR cut-off of 0.50 (ref. ²⁶) is not an appropriate measure for central obesity, the bivariate scatter plots between BMI and WHtR were generated with fitted linear regression lines. The optimal cut-off values for WHtR that correspond to established BMI thresholds of obesity (28 kg m⁻² for Chinese individuals and 30 for others) were derived from the linear regression lines given the very high Pearson correlation coefficients in the four cohorts (r ≥ 0.64).

Organ dysfunction was defined as the presence of at least one of 18 clinical indicators listed in the Commission report. Notably, in line with the Commission’s recommendations, the metabolic cluster criterion for clinical obesity (hyperglycaemia together with low high-density lipoprotein (HDL) and high triglycerides) was considered to be met only when diabetes and dyslipidaemia occurred together, and not when either condition occurred alone. Data availability about organ dysfunction ranged from 9 (SIDIAP) to 18 (UKB) indicators. Case ascertainment was based on medical records, self-reports and health assessments (Supplementary Table 2).

Covariates

Socio-demographic and lifestyle data were collected for each cohort; details can be found in the Supplementary Methods.

Outcomes

Organ dysfunction and death events were identified through self-reports, clinical exams, medical records and provincial death certificates in the Kailuan Study. In the UKB, events were ascertained through linkages with hospital records and death registries. Incident organ dysfunctions were obtained through EHRs in the AoU study, with measurements recorded at different time points and clinic visits. Outcomes in the SIDIAP were identified through linkage to primary care and hospital EHRs.

Measures of prevalence and incidence

Given the challenges in establishing causality between obesity and coexisting organ dysfunction at the individual level, we applied the AF approach at the population level to quantify the proportion of organ dysfunction cases attributable to obesity among individuals with obesity¹¹. The AF was calculated as: ({rm{AF}}=frac{{rm{HR}}-1}{{rm{HR}}}), where HR denotes the hazard ratio for the association between obesity and risk of incident organ dysfunction with adjustment for relevant confounders using Cox proportional hazards models. Details are provided in the Supplementary Methods.

Statistical analysis

Descriptive statistics were reported as mean (s.d.), median (interquartile range) or percentages, as appropriate. Multiple imputations using chained equations (linear and logistic regression) were used to impute the missing data (mice package in R, v.3.19.0). Prevalence estimates and 95% CIs were calculated using the epi.conf function from the epiR package in R (v.2.0.95). To quantify the misclassification of obesity by BMI alone, we estimated the prevalence of abdominal obesity (based on WC and/or WHtR) in each BMI category. We further estimated the prevalence of clinical obesity among people with obesity, as well as among the total population, with adjustment using the AF method. The AF-adjusted prevalence of clinical obesity was estimated by multiplying the crude prevalence of clinical obesity and the corresponding AF values in each cohort. The unadjusted prevalence (that is, the observed prevalence) was considered as the maximum of the prevalence of clinical obesity.

Using longitudinal data from all four cohorts, we estimated the cumulative incidence of incident organ dysfunction among individuals with preclinical obesity. Follow-up time was calculated from baseline until the first occurrence of incident organ dysfunction, death or end of follow-up. Cumulative incidence curves were estimated using Kaplan–Meier methods implemented in the survival package in R (v.3.8.6), with cumulative incidence calculated as 1 minus the survival probability. The AF-adjusted cumulative incidence of clinical obesity was estimated by multiplying the observed cumulative incidence of organ dysfunction by the corresponding AF estimates in each cohort.

For mortality risk analysis, participants were categorized into four groups: (1) no confirmed obesity and no organ dysfunction (reference group); (2) no confirmed obesity but with organ dysfunction; (3) confirmed obesity without organ dysfunction (preclinical obesity); or (4) confirmed obesity with organ dysfunction (clinical obesity). To evaluate whether the composite anthropometric criteria could detect high-risk individuals misclassified by BMI alone, we conducted the analysis of the association between obesity status and mortality in the following five groups: (1) no overweight or obesity (BMI < 24/25; reference group); (2) BMI-based overweight without CAD-obesity; (3) BMI-based overweight with CAD-obesity (that is, both elevated WC and elevated WHtR; this group can be viewed as the underdiagnosis of obesity by BMI alone); (4) BMI-based obesity without CAD-obesity (this group can be viewed as the overdiagnosis of obesity by BMI alone); and (5) BMI-based obesity with CAD-obesity. To further evaluate the effects of composite anthropometric criteria for excessive adiposity, participants were categorized into eight groups based on BMI-defined obesity status (yes, no), CAD-obesity status (yes, no) and presence of organ dysfunction (yes, no) at baseline.

The association between obesity subtypes and mortality was estimated using the Cox proportional hazards models implemented in the survival package in R (v.3.8-6). Study-specific HRs and 95% CIs were pooled using fixed-effects meta-analysis implemented in the meta package in R (v.8.5.0). Time-at-risk extended from baseline to death or administrative censoring (31 December 2023 for the Kailuan Study, 19 December 2022 for the UKB and 31 December 2021 for the SIDIAP). Proportional hazards assumptions were verified using log–log survival plots. We performed several robustness checks, such as excluding deaths occurring within the first 2 or 5 years after baseline or including waist-to-hip ratio (≥0.90 for male and ≥0.85 for female individuals) as an additional criterion for central obesity. Stratified analyses by age groups (<60 and ≥60 years old), sex (male and female) and smoking status (never, former and current) were further performed to test the robustness of the results across subgroups of the participants.

All analyses were performed using SAS v.9.4 and R v.4.4.2. Statistical significance was defined as P < 0.05 (two-sided).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

The code used in this study is available on Zenodo at https://doi.org/10.5281/zenodo.20256910 (ref. ²⁷).

1. Rubino, F. et al. Definition and diagnostic criteria of clinical obesity. Lancet Diabetes Endocrinol. 13, 221–262 (2025).

   Article 
   PubMed 
   PubMed Central 
   
   Google Scholar 

2. Elstein, A. S. & Schwartz, A. Clinical problem solving and diagnostic decision making: selective review of the cognitive literature. BMJ 324, 729–732 (2002).

   Article 
   PubMed 
   PubMed Central 
   
   Google Scholar 

3. Hubbard, V. S. & Hall, W. H. Gastrointestinal surgery for severe obesity. Obes. Surg. 1, 257–265 (1991).

   Article 
   CAS 
   PubMed 
   
   Google Scholar 

4. Sharma, A. M. & Kushner, R. F. A proposed clinical staging system for obesity. Int. J. Obes. 33, 289–295 (2009).

   Article 
   CAS 
   
   Google Scholar 

5. Whyte, M. B., Velusamy, S. & Aylwin, S. J. B. Disease severity and staging of obesity: a rational approach to patient selection. Curr. Atheroscler. Rep. 16, 456 (2014).

   Article 
   CAS 
   PubMed 
   
   Google Scholar 

6. Rathbone, J. A. et al. How conceptualizing obesity as a disease affects beliefs about weight, and associated weight stigma and clinical decision-making in health care. Br. J. Health Psychol. 28, 291–305 (2023).

   Article 
   PubMed 
   
   Google Scholar 

7. Yanovski, S. Z. & Yanovski, J. A. Approach to obesity treatment in primary care: a review. JAMA Intern. Med. 184, 818–829 (2024).

   Article 
   CAS 
   PubMed 
   PubMed Central 
   
   Google Scholar 

8. Jensen, M. D. et al. 2013 AHA/ACC/TOS guideline for the management of overweight and obesity in adults: a report of the American College of Cardiology/American Heart Association task force on practice guidelines and the obesity society. Circulation 129, S102–S138 (2014).

   Article 
   PubMed 
   
   Google Scholar 

9. Weng, J. et al. Standards of care for type 2 diabetes in China. Diabetes Metab. Res. Rev. 32, 442–458 (2016).

   Article 
   PubMed 
   PubMed Central 
   
   Google Scholar 

10. Itani, L. & El Ghoch, M. Waist-to-height ratio cut-off points for central obesity in individuals with overweight across different ethnic groups in NHANES 2011–2018. Nutrients 16, 3838 (2024).

    Article 
    PubMed 
    PubMed Central 
    
    Google Scholar 

11. Chen, Y. Q., Hu, C. & Wang, Y. Attributable risk function in the proportional hazards model for censored time-to-event. Biostatistics 7, 515–529 (2006).

    Article 
    PubMed 
    
    Google Scholar 

12. Shen, Y. & Hu, G. Prevalence of clinical obesity in US adults, 2017–2020. Obesity 33, 1968–1976 (2025).

    Article 
    PubMed 
    
    Google Scholar 

13. Ayuzo Del Valle, N. C. et al. Beyond BMI: a comprehensive approach to pediatric obesity assessment. Front. Pediatr. 13, 1597309 (2025).

    Article 
    PubMed 
    PubMed Central 
    
    Google Scholar 

14. Yao, Z. et al. Prevalence of clinical obesity versus BMI-defined obesity among US adults: a cohort study. Lancet Diabetes Endocrinol. 13, 647–649 (2025).

    Article 
    PubMed 
    
    Google Scholar 

15. Carrillo-Larco, R. M., Guzman-Vilca, W. C., Tarazona-Meza, C., Xu, X. & Bernabe-Ortiz, A. Prevalence of pre-clinical and clinical obesity in adults: pooled analysis of 56 population-based national health surveys. PLOS Glob. Public Health 5, e0004838 (2025).

    Article 
    PubMed 
    PubMed Central 
    
    Google Scholar 

16. Fourman, L. T. et al. Implications of a new obesity definition among the All of Us cohort. JAMA Netw. Open 8, e2537619 (2025).

    Article 
    PubMed 
    PubMed Central 
    
    Google Scholar 

17. World Obesity Federation. World Obesity Atlas 2025 https://www.worldobesity.org/resources/resource-library/world-obesity-atlas-2025 (2025).

18. Tucker, M. E. Critics misunderstand Lancetobesity definition, author says. Medscape https://www.medscape.com/viewarticle/critics-misunderstand-lancet-obesity-definition-author-says-2025a10003me (2025).

19. Alexander, L. C. Obesity Medicine Association raises concerns over Lancet Commission’s new recommendations on obesity diagnosis. Obesity Medicine Association https://obesitymedicine.org/blog/obesity-medicine-association-raises-concerns-over-lancet-commissions-new-recommendations-on-obesity-diagnosis/ (2025).

20. European Association for the Study of Obesity (EASO). EASO response to the Lancet Commission report on obesity diagnosis and management. https://easo.org/easo-response-to-the-lancet-commission-report-on-obesity-diagnosis-and-management/ (2025).

21. Viera, A. J. Predisease: when does it make sense?. Epidemiol Rev. 33, 122–134 (2011).

    Article 
    PubMed 
    
    Google Scholar 

22. Geng, T. et al. Relationship between control of cardiovascular risk factors and chronic kidney disease progression, cardiovascular disease events, and mortality in Chinese adults. J. Am. Coll. Cardiol. 84, 1313–1324 (2024).

    Article 
    CAS 
    PubMed 
    
    Google Scholar 

23. Sudlow, C. et al. UK Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 12, e1001779 (2015).

    Article 
    PubMed 
    PubMed Central 
    
    Google Scholar 

24. The All of Us Research Program Investigators. The “All of Us” research program. N. Engl. J. Med. 381, 668–676 (2019).

    Article 
    
    Google Scholar 

25. Ramos, R. et al. Validity for use in research on vascular diseases of the SIDIAP (Information System for the Development of Research in Primary Care): the EMMA study. Rev. Esp. Cardiol. (Engl. Ed.) 65, 29–37 (2012).

    Article 
    PubMed 
    
    Google Scholar 

26. Browning, L. M., Hsieh, S. D. & Ashwell, M. A systematic review of waist-to-height ratio as a screening tool for the prediction of cardiovascular disease and diabetes: 0.5 could be a suitable global boundary value. Nutr. Res. Rev. 23, 247–269 (2010).

    Article 
    PubMed 
    
    Google Scholar 

27. Lily. Lily0906/clinical_obesity_analysis_code: initial public code release. Zenodo https://doi.org/10.5281/zenodo.20256910 (2026).

Download references

Author notes

1. These authors contributed equally: An Pan, Shengping Yang, Rafael Ramos, Xianli Li, Jinchi Xie

2. These authors jointly supervised this work: An Pan, Shouling Wu, Eric Ravussin, José Manuel Fernández-Real, Francesco Rubino

Authors and Affiliations

1. Department of Epidemiology and Biostatistics, Center for Obesity and Diabetes Research, School of Public Health, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China

   An Pan 
   (潘安), Xianli Li 
   (李娴丽), Jinchi Xie 
   (谢金池) & Tingting Geng 
   (耿婷婷)

2. Pennington Biomedical Research Center, Louisiana State University System, Baton Rouge, LA, USA

   Shengping Yang & Eric Ravussin

3. Institut Universitari d’Investigació en Atenció Primària Jordi Gol (IDIAP Jordi Gol), Barcelona, Spain

   Rafael Ramos & Jordi Blanch

4. Network for Research on Chronicity, Primary Care, and Prevention and Health Promotion (RICAPPS), Girona, Spain

   Rafael Ramos

5. Biomedical Research Institute Girona (IdIBGi), Girona, Spain

   Rafael Ramos & José Manuel Fernández-Real

6. Department of Medical Sciences, University of Girona, Girona, Spain

   Rafael Ramos & José Manuel Fernández-Real

7. Division of Endocrinology, Metabolism, Diabetes and Cardiology, Department of Medicine, University of Colorado Anschutz Medical Campus, Aurora, CO, USA

   Robert H. Eckel

8. Department of Diabetes, Università Cattolica del Sacro Cuore, Rome, Italy

   Geltrude Mingrone

9. Department of Medical and Surgery Sciences, Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Rome, Italy

   Geltrude Mingrone

10. School of Cardiovascular and Metabolic Medicine and Sciences, King’s College London, London, UK

    Geltrude Mingrone & Francesco Rubino

11. North China University of Science and Technology, Tangshan, China

    Jingyuan Jiang 
    (江靖源)

12. Department of Cardiology, Kailuan General Hospital, Tangshan, China

    Hongmin Liu 
    (刘红敏), Shuohua Chen 
    (陈朔华) & Shouling Wu 
    (吴寿岭)

13. Centro de Investigación Biomédica en Red Fisiopatología de la Obesidad y Nutrición (CIBEROBN), Instituto de Salud Carlos III, Madrid, Spain

    José Manuel Fernández-Real

14. Bariatric and Metabolic Surgery, King’s College Hospital, London, UK

    Francesco Rubino

Contributions

A.P., S.Y., R.R., X.L. and J.X. contributed equally to this work and share first authorship. A.P. led the statistical analysis plan, coordinated multi-cohort harmonization, oversaw analysis of the UKB study and contributed to writing of the manuscript, especially the Methods and results. S.Y. performed cohort-specific data extraction and quality control for the AoU study, managed literature synthesis and contributed to drafting the results and Supplementary Information. R.R. led cohort-specific data extraction and quality control for the SIDIAP database, contributed to literature synthesis and assisted in drafting the results and Supplementary Information. X.L. led the analysis of the UKB study, performed cohort-specific data extraction and contributed to drafting the results and Supplementary Information. J.X. led the analysis of the Kailuan Study, and contributed to writing the Methods and results. Co-senior authors F.R., J.M.F.-R., E.R. and S.W. jointly supervised the work and share senior authorship. F.R. conceived the study, led coordination across study sites, contributed to the interpretation of the results and provided major input in the draft of the manuscript and its revisions. J.M.F.-R. contributed to conceiving the study, oversaw analysis of the SIDIAP data, contributed to interpretation of the overall results, elaborated the first preliminary draft of the manuscript and supervised revisions. E.R. supervised methodological analysis of the AoU study and supervised revisions. S.W. oversaw analysis of the Kailuan Study, contributed to interpretation of the overall results and supervised revisions. R.H.E. and G.M. contributed to study design and discussion and interpretation of data and provided critical input for preparation of the manuscript and its revisions. H.L., T.G., J.J., S.C. and J.B. contributed to cohort-specific data extraction, harmonization and technical validation of EHR-derived variables. All authors reviewed and approved the final manuscript and take full responsibility for the integrity of the data and analysis.

Corresponding authors

Correspondence to
An Pan 
(潘安) or Francesco Rubino.

Competing interests

A.P., S.Y., R.R., X.L., J.X., H.L., J.J., S.C., T.G., S.W., J.M.F.-R. and J.B. declare no competing interests. G.M. reports consulting fees from Novo Nordisk, Boehringer Ingelheim, Eli Lilly, Medtronic, Fractyl and Recor. She is also a scientific advisor for Keyron, MetaDeq, GHP Scientific and Jemyll. R.H.E. reports receiving consulting fees from Novo Nordisk, the Healthy Aging Company and WW International. E.R. reports consulting fees from Energesis Pharmaceuticals, Eli Lilly, Merck, Amway, Novo Nordisk, Altimmune, AbbVie, Amgen, Orbimed and Boehringer Ingelheim. F.R. reports research grants from Ethicon (Johnson & Johnson), Novo Nordisk and Medtronic, consulting fees from Kailera and speaking honoraria from Medtronic, Ethicon, Novo Nordisk, Eli Lilly, Amgen, Regeneron and Astra Zeneca. F.R. is also a member of the data safety monitoring board for GI Metabolic Solutions and the president of the Metabolic Health Institute, a non-profit organization.

Peer review information

Nature thanks Louis J. Aronne, who co-reviewed with Aditi Rao; and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions