Introduction

The global prevalence of obesity and its associated health concerns has been increasing dramatically over the last 50 years across human populations1. This increase in obesity prevalence may have its origins in emerging mismatches between past and present environments, particularly in current versus past energy availability and expenditure. Originally, Homo sapiens communities’ sustenance depended exclusively from various combinations of ancestral hunting, gathering and, eventually, farming. Additionally, in ancestral populations parity was higher and individuals were exposed to higher risks of infection, predation or attacks by wildlife. Thus, energy availability was lower and energetic demands higher than those imposed by current industrial contexts. Within pre-industrial contexts, then, alleles associated with biological mechanisms that optimize energy efficiency and increase energy storage are hypothesized to have been advantageous and, thus, be positively selected2. Two main alternatives hypotheses have been proposed to explain these predispositions. The “Thrifty gene” hypothesis suggests that genes promoting efficient energy storage and fat accumulation were favored by natural selection because they improved survival during periods of food scarcity3. In contrast, the “Drifty gene” hypothesis proposes that, once predation pressures diminished in human evolutionary history, genetic drift allowed variation in appetite-regulating genes to accumulate, leading to increased susceptibility to obesity in modern environments3,4. Yet, in industrialized settings the same genetic predispositions and the biological mechanisms associated with them can increase the risk of developing phenotypes (observable physical characteristics) associated with obesity. The consequences of these mismatches are illustrated by the increase in obesity incidence in individuals who migrate from traditional agro-pastoral communities to urban, heavily industrialized contexts5.

Importantly, obesity is linked to increased risks of a broad range of health problems including diabetes, heart disease, mental health issues and negative reproductive outcomes6,7. Consequently, a number of anthropometric methods have been developed to detect and monitor obesity over the years. Traditional anthropometric indices to assess obesity include Body Mass Index (BMI), Waist-to-Hip Ratio (WHR), and Waist-to-Height Ratio (WHtR). While BMI reflects overall body mass relative to height, WHR and WHtR are widely used indicators of central fat distribution and abdominal obesity. These methods, however, have been criticized for not reflecting aspects of body composition that are critically important to accurately identify health risk profiles in diverse populations. In particular, evidence suggests that these indices may overestimate or underestimate health risks for some ancestry groups8,9,10, raising questions about their usefulness in regions with high genetic admixture11. Similar concerns have been discussed extensively in the literature since the introduction of BMI in 1974 by Ancel Keys12, and more recently in the context of body shape research13,14,15,16. These works emphasize that anthropometric indices, including body shape descriptors, have inherent methodological limitations and should be interpreted with caution11. These inaccuracies cannot only lead to ill-informed health policies but also, as a result, heighten health inequalities affecting historically marginalized groups with particular ancestries11. Human body morphologies vary occupying a continuous and multivariate space. This complexity can be driven by complex interactions between developmental, genetic, and environmental factors. Thus, understanding obesity and its health risks requires methodological approaches that capture meaningful variation in body size and shape. Nevertheless, current clinical assessments often rely on discrete thresholds and categorical classifications, such as BMI cutoffs, to estimate obesity-related risks, potentially overlooking the nuanced and continuous nature of body form variation. Regarding this, recently, Rubino et al. (2025) recommended that BMI should be used only as a surrogate measure of health risk at a population level, for epidemiological studies, or for screening purposes, rather than as an individual measure of health6.

Fig. 1
Fig. 1
Full size image

Evolution of methods to study overweight and obesity, and the role of genetic ancestry.

The concerns described above have led to the development of novel ways to study and assess obesity in different populations across time and space. New diagnostics tools include new technologies including 3D scanning, video-based models and image processing, which are illustrated in Fig. 1. Traditional and widespread ways of detecting overweight and obesity were based on anthropometric measurements combined on several indices that, as a norm, provide an easy and accessible way to approach in a statistical way the study of the phenotype of interest. Yet, the obesity phenotype is an intrinsically complex one, particularly in terms of the variety of patterns in which fat can be distributed in the human body. Proper assessment of such patterns requires capturing a variety of 3D geometric profiles. Gold-standard imaging techniques such as computed tomography (CT) and magnetic resonance imaging (MRI) have been extensively used to assess obesity phenotypes17,18,19. These modalities allow precise quantification of adipose tissue distribution, differentiation between visceral and subcutaneous fat compartments, and evaluation of ectopic fat deposition, all of which are strongly associated with metabolic risk19. Studies employing CT and MRI have demonstrated that central fat accumulation, rather than total body fat alone, is a stronger predictor of cardiometabolic disease20. While their high cost and limited accessibility in large, population-based studies constrain their widespread application, they remain critical benchmarks against which newer, more scalable approaches, such as 3D body surface scanning, can be validated. To that effect it is necessary to utilize 3D devices and statistical models that can accommodate this type of data. Attaining a proper, comprehensive and accurate evaluation of the obesity phenotype, is critical to achieving greater refinement in its study, needed to complement precision strategies at other population structure levels (genomic, epigenomic, etc). It is important to note, however, that principal component analysis PCA–based body shape phenotyping, while capturing variation beyond traditional indices, also faces constraints regarding resolution, interpretability, and clinical applicability13,15,16. As highlighted by the 2024 Lancet Obesity Commission, the current challenge in obesity research lies in refining its conceptual and clinical definition, rather than relying solely on anthropometric or morphometric phenotyping21.

Latin America, with its rich history of biological and cultural admixture, is an example of a region likely to be affected by the misclassification of obesity phenotypes. The sub-continent underwent extensive admixture between Native Americans and people arriving from other continents, particularly Europe and Africa22,23,24,25. Most genetic studies have examined this process using a “tri-hybrid” model that includes variation in overall Native American, European and Sub-Saharan African ancestry across regions and amongst individuals23,24,25, with small and geographically-restricted East Asian ancestry also reported26. Recent research using novel haplotype-based methods, showed that Native American ancestry components in Latin Americans correspond geographically to the present-day genetic structure of Native groups, and that sources of non-Native ancestry, and admixture timings, match documented migratory flows25. As a whole, these and other studies help to document a fine-grained landscape of genetic and non-genetic factors that arose after five centuries of extensive admixture and migratory movements in the region.

The Latin American fine-grained admixture dynamics described above represents a fruitful scenario to detect genetic markers associated to phenotypes via admixture mapping approaches. Indeed, the CANDELA initiative has investigated samples from Latin American admixed populations to help verify the presence of previously reported genetic markers, including Single Nucleotide Polymorphisms (SNPs), and to report new markers associated to several phenotypes such as facial shape27, eyebrow shape28, skin and/or hair pigmentation, dental size and shape29, pain sensitivity30, cancer risk31,32, ear morphology27, and geometric-morphometric obesity-related traits33,34,35,36. These examples, along with other research published elsewhere on admixed groups5,22,37,38,39 demonstrate that complex traits do present ancestry-driven patterns that are likely to result in diverse metabolic profiles.

Our goal is to take advantage of well-known Latin American admixture dynamics to assess whether traditional indices are appropriate to accurately identify obesity and health risk across populations. Our results should help inform public health strategies in Latin America that are specific to each country’s population structure. Here, we test the hypothesis that obesity classifications based on traditional anthropometric indices present discrepancies across genetic ancestry groups, which renders universal diagnostic criteria for obesity-associated health risks inappropriate. To this aim, we investigate the links between traditional anthropometric indices associated with obesity, genetic ancestry and health risks. Finally, we propose the development of new methodological approaches that go beyond traditional indices and incorporate a deeper understanding of body shape variation, acknowledging that such phenotyping is complementary to, rather than a substitute for, comprehensive clinical stratification frameworks21.

Methods

Data integration and preprocessing

We integrated five (5) databases containing anthropometric indices, Ancestry-Informative Markers (AIMs), and/or 3D body shape to assess the influence of genetic ancestry on body shape and obesity risk from a total of seven (7) countries. Specifically, we analyzed data from the CANDELA consortium (Colombia, Mexico, Chile, Brazil, Peru), as well as from three projects in Argentina: Raíces, Patagonia3DLab, and ECHA (Emoción, Cognición y Hábitos Alimentarios), and the SER project (Society, Environment, and Reproduction Research) based in Guatemala. All the databases contain the anthropometric indices on which our analyses are based: Body Mass Index (BMI), Waist-Hip Ratio (WHR), and Waist-to-Height Ratio (WHtR)35,40,41,42,43,44. All studies but SER included Genetic Ancestry-Informative Markers. That said, SER participants’ indigenous ancestry is well established, allowing us to classify them safely within the America AIM category45. Inclusion criteria for participants were adults older than 18 years, availability of complete anthropometric and genetic ancestry data, and absence of conditions affecting body composition such as pregnancy or severe musculoskeletal disorders. Sample sizes per site are provided in Table 1.

Database descriptions

CANDELA Consortium40: This large-scale study, which included 7,235 participants collected in 2012, was designed to investigate genetic and environmental influences on physical and facial traits. This database includes individuals of both sexes from five (5) cities in five (5) Latin American countries: Porto Alegre in Brazil; Arica in Chile, Medellín in Colombia; Mexico City in Mexico; and Lima in Peru. Further details regarding the CANDELA consortium (http://www.ucl.ac.uk/silva/candela) study can be found in Ruiz-Linares et al. (2014). The CANDELA consortium project obtained its own ethical clearance from the University College London (UK, protocol number 3352/001), and was approved by all participating institutions: Universidad Nacional Autónoma de México (México), Universidad de Antioquia (Colombia), Universidad Peruana Cayetano Heredia (Perú), Universidad de Tarapacá (Chile), Universidade Federal do Rio Grande do Sul (Brazil) and University College London (UK). Informed consent was obtained from all participants.

Raíces41, Patagonia3DLab35, and ECHA42: These projects investigates anthropometric and morphological diversity in Argentine populations43. Data collection took place in: 2018 for Raíces, 2016 for Patagonia3DLab, and 2022 for ECHA. They all include anthropometric indices and Raíces also includes 3D body models and AIM data in 96 individuals. The Raíces project was approved by the Research Ethics Committee of the Northern Programmatic Area of the Ministry of Health of the province of Chubut (Regional Hospital of Puerto Madryn, Chubut, Argentina) under protocol number 19/17 (approved on September 4, 2017). The Patagonia 3D Lab project, prior to Raíces, was approved by the same Ethics Committee under protocol number 010/16 (approved on June 9, 2016).

SER project44: SER is a longitudinal project that began in 2000 to examine the interactions between society, environment, and reproduction in a Maya-Kachikel population from Guatemala. For the analyses in this article we included anthropometric data including BMI, WHR and WHtR and 3D data collected in 2023. SER’s data collection and use was approved by the Research Ethics Board (REB) at Simon Fraser University (SFU) (Application Numbers: 2012s0668 and 2016s0576).

The total sample size of these five datasets provides a final sample of 7,757 individuals (of both sexes, except for the Guatemala sample, which contains only females) from seven Latin American countries: Colombia, Mexico, Chile, Brazil, Peru, Argentina, and Guatemala.

Table 1 Summary of integrated datasets by country, detailing available data types and ancestry estimation methods, and mean values for Weight, WC (Waist Circumference), and HC (Hip Circumference).
Full size table

Genetic ancestry: definition and estimation

Genetic ancestry measures the genetic variation in DNA to assess the geographical origins of individuals’ ancestors46. Genetic ancestry is a multidimensional continuum that can, for example, reflect the proportion of an individual’s ancestry originating from Africa, the Americas, Asia or Europe at a continental level, and can also be assessed using finer scales (e.g. Chacón-Duque et al. 201825)46,47,48,49. In contrast to race and ethnicity, genetic ancestry is directly linked to genetic variation, which in turn may be related to particular biological processes47,49. As such, genetic ancestry captures a portion of the biological variation between and within groups.

Despite the continuous and multifactorial nature of genetic ancestry, we rely on broad categorical groupings in this study because they offer a practical and reproducible way to summarize complex genetic information. These groupings, such as individuals with predominantly Amerindian, European, or African ancestry, serve as useful proxies to identify general trends. However, we acknowledge that these categories are simplifications of a much richer underlying variation. Future approaches that incorporate ancestry as a continuous variable or explore ancestry at more local levels using specific genomic regions may provide even more precise insights into how genetic background shapes human morphology and metabolic health.

The concept of ancestry can be defined as a description of a person’s origin from his or her genetic lineage. However, its use in scientific research and public discourse has generated debates about its meaning and implications. The confusion between genetic ancestry, race and ethnicity has led to misinterpretations and problematic use of these categories in fields such as medicine and anthropology. Race and ethnicity are social constructs influenced by historical, political, and cultural factors, whereas genetic ancestry is based on DNA variability and allows estimating the geographic origin of a person’s ancestors47. This approach has limitations, as genetic diversity is a continuum without fixed boundaries, and categorizing populations based on genetics alone can lead to misinterpretations. While genetic ancestry can provide relevant information in biomedical studies, it should not be used as a proxy for race or ethnicity, as these categories reflect social experiences that impact health and life chances. Ignoring this dimension may reinforce biological determinism and generate biased conclusions49. Therefore, it is crucial to employ these concepts accurately, recognizing both their differences and their interactions.

To accurately characterize population structure, it is essential to define ancestry and the genetic markers used for its estimation. Continuing in the line of genetic ancestry, Ancestry-Informative Markers (AIMs) are genetic variants specifically selected for their ability to differentiate ancestral populations through allele frequency differences across geographical groups. In the datasets analyzed here, AIMs correspond to Single Nucleotide Polymorphisms (SNPs), as used in the CANDELA study and related initiatives40,50,51. In admixed populations, such as those in Latin America, AIMs allow for the estimation of ancestry proportions, reducing potential bias due to population structure, a well-known confounding factor in association studies. Based on these SNP markers, individuals can be characterized by their proportion of major continental ancestries (commonly African, European, and Amerindian), enabling the exploration of how genetic ancestry relates to phenotypic variation.

In this sense, Latin America’s admixed populations exemplify the complex interplay of genetic ancestry, environment, and culture in shaping obesity phenotypes. By combining AIM-based genomics with 3D morphometrics, we move beyond the limitations of traditional anthropometry toward precision public health. The approach used here highlights several aspects of interest regarding the biological and non-biological causes of obesity. It should be noted that the exploration of new approaches to study obesity, overcoming the limitations of traditional anthropometric indices, has clinical relevance since it aims to address diagnostic gaps existing in real-world settings.

Anthropometric indices and risk classification across ancestry groups

Individuals at high-risk of metabolic and cardiovascular diseases were identified using categorical thresholds based solely on anthropometric indices using WHO criteria based on obesity risks52. Although ethnicity-specific BMI and WHR cut-off values have been proposed53, we applied WHO thresholds to ensure consistency and comparability of classifications across the multi-country, genetically diverse sample analyzed in this study. For BMI, individuals were categorized as “Overweight” (BMI 25.0–29.9), “Obesity class I” (30.0–34.9), “Obesity class II” (35.0–39.9), and “Obesity class III” (\(\ge\) 40.0), reflecting increasing severity of obesity-related health risks. The categories “Underweight” (BMI <18.5) and “Normal weight” (BMI 18.5–24.9) were not considered obesity high-risk. WHR risk thresholds were set at >0.85 for women and >0.90 for men, indicating increased risk for cardiometabolic diseases54,55. For WHtR, a cut-off of >0.50 was used, which has been widely associated with elevated risk for central adiposity and metabolic syndrome54. These classifications are regularly used to identify at-risk individuals across populations, and to evaluate health disparities within and between genetically admixed groups. This analysis was carried out both at the country level and within each country, stratifying the data by AIM groups.

To assess whether the distributions of anthropometric indices varied across genetic ancestry groups, we applied the non-parametric Kruskal–Wallis test separately for male and female participants. This test is suitable for comparing the distribution of ranks across three or more independent groups when the assumption of normality is not met. It is used to assess whether the groups originate from the same distribution and is appropriate when the grouping variable is categorical, as is the case for ancestry categories56. Post hoc pairwise comparisons were conducted using Dunn’s test, with Bonferroni correction to account for multiple testing57. This analysis was applied to BMI, WHR, and WHtR. Additionally, we recoded each anthropometric index into binary risk categories (low vs. high disease risk) and performed chi-square tests of independence on contingency tables stratified by sex to test whether the application of current diagnostic thresholds leads to unequal classification of risk across ancestry groups. For WHR and WHtR, we maintained our standard binary cut-offs, whereas for BMI we grouped “Underweight” and “Normal weight” into a “low-risk” category, and combined “Overweight” and all obesity classes (I–III) into a “high-risk” group.

Additionally, the Body Shape Index (ABSI) and the Hip Mass Index (HI), two more recent ones that offer a complementary perspective on disease risk, were calculated. The ABSI, developed to be independent of weight and height, has been associated with all-cause and cardiovascular disease mortality in various population-based studies58,59. On the other hand, the HI, which assesses hip mass, has also shown associations with adverse health outcomes60,61.

Descriptive analysis by country and by ancestry-informative marker

Subsequently, individuals were reclassified based on their genetic ancestry using AIMs. We used an arbitrary threshold of >70% of per-individual AIMs belonging to each parental population to maximize the chances of detecting ancestry-dependent anthropometric variation62. As a result of this reclassification, a subset of 2,970 individuals (1,878 females and 1,092 males) was identified as having predominantly Native American, European, or African ancestry (see Fig. 2). These three groups represent distinct genetic profiles derived from the commonly used tri-hybrid model (America, Europe, Africa). It is important to clarify here that we are not assuming that our sample can be dissected into three continental groups, which do not represent biological entities per se. Rather, we just simply implement a rotation of the AIMs’ statistical space in order to maximize the anthropometric variation potentially due to differences in genetic ancestry.

Fig. 2
Fig. 2
Full size image

Sample distribution across age groups, countries, and ancestry-informative markers (AIM) categories, stratified by sex. (a) Age quartile distribution of the sample, showing differences in the number of individuals per age group and sex. (b) Country-wise distribution of the dataset after AIM filtering, highlighting the relative representation of each country. (c) Distribution of individuals according to their predominant AIM category (Europe, America, or Africa), considering only those with at least 70% ancestry membership.

To explore disparities in risk classification, we constructed Venn diagrams to visualize the overlap of high-risk participants identified by different anthropometric indices (BMI, WHR and WHtR) across countries and AIM categories (Europe, America & Africa) (see Fig. 3). These visualizations provided insight into the extent of agreement or divergence among classification methods. Additionally, statistical comparisons were performed using the Kruskal-Wallis test to determine whether significant differences existed in the distribution of anthropometric indices across groups. These steps were crucial for assessing the degree of consistency or disparity in risk categorization across traditional anthropometric measures.

We conducted a Principal Component Analysis (PCA) to examine the underlying structure of individuals’ anthropometric data, and to explore the position of individuals according to their anthropometric indices’ threshold and or their genetic ancestry. Our objective was to identify the Eigenvectors (Principal Components) that best explain the variance in the data. We then visualized the PCA results to explore clustering patterns and assess the influence of ancestry on body shape variation. We applied all analyses separately for male and female sub-samples to account for sex-based differences in body shape and disease risk classification. This comprehensive analytical approach enabled us to robustly evaluate inter-country and inter-ancestry variation in body shape, revealing patterns beyond geopolitical borders. These PCA analyses were conducted using the PCA module from the Scikit-learn library (version 1.6.1)63 in Python 3.11.13.

Fig. 3
Fig. 3
Full size image

Sample of Venn diagrams used to visualize the overlap of individuals classified as high-risk according to BMI, WHR, and WHtR thresholds.

3D human shape variation

To analyze body shape variation, we obtained 3D models from two sources: direct 3D scans and video-based reconstructions. High-resolution 3D scans were acquired using the Structure Sensor scanner for a subset of individuals from the Raíces and Patagonia3DLab databases43,64. Additionally, 3D body reconstructions were generated from video recordings using the body2vec_mesh model (https://github.com/aletrujim/body2vec_mesh), which integrates multiple Deep Learning approaches: background removal with BremNet36, pose estimation with OpenPose65, and 3D human body generation based on PIFuHD66. The ECHA and SER databases included only the video-based reconstructions, without direct 3D scans.

To ensure comparability across datasets, all 3D models were aligned, rotated, and scaled using Procrustes analysis within each database. Procrustes analysis is a geometric method used to standardize 3D shapes by removing differences in position, rotation, and scale while preserving the intrinsic shape variation67. This technique aligns all models to a common reference framework, ensuring that differences observed across individuals are due to actual morphological variation rather than differences in orientation or size68. By applying Procrustes superimposition to our dataset, we were able to compare body shape variation across individuals and populations in a standardized way, facilitating subsequent statistical analyses of shape differences. The patterns of variation in body shape were then explored through PCA, which allowed the identification of the main axes of morphological differences between individuals. These analyses were performed using the SciPy library (version 1.15.3)69 and Open3D library (version 0.19.0)70 in Python 3.11.13.

Results

Significant differences were found across all three anthropometric indices (BMI, WHR, WHtR) among genetic ancestry groups, indicating systematic variation in body composition (see Fig. 4). Differences were most pronounced for WHR and WHtR, which showed higher Kruskal–Wallis statistics than BMI, suggesting that these fat distribution measures are more sensitive to ancestry-related variation (see Table 2). Among females, Native American ancestry was associated with significantly higher BMI compared to African and European groups, which did not differ significantly from each other. WHR and WHtR differed significantly across all ancestry groups, with Native American women exhibiting the highest values, followed by Europeans and Africans. In males, Native Americans also had the highest values for all indices, followed by Europeans and Africans, with significant differences observed between African and European men across all measures, including BMI.

Fig. 4
Fig. 4
Full size image

Distribution of anthropometric indices by ancestry group, stratified by sex. Boxplots represent the distribution of values for each anthropometric index across ancestry groups (African, American, and European). Dashed red lines indicate the current cutoff points recommended by the World Health Organization52 for identifying increased health risk.

Table 2 Summary of Kruskal–Wallis and Pearson Chi-squared test results for differences across genetic ancestry groups in each anthropometric index, stratified by sex. Both tests show consistent statistically significant differences across ancestry groups.
Full size table
Table 3 Post hoc Dunn test results for pairwise comparisons of ancestry groups (Africa, America, Europe) across anthropometric indices, stratified by sex. The table reports Z-scores, unadjusted p-values (P.unadj), and Bonferroni-adjusted p-values (P.adj). All indices show statistically significant differences between most ancestry pairs, with the most pronounced differences observed in comparisons involving the American group.
Full size table
Fig. 5
Fig. 5
Full size image

Proportional risk distribution across indices and sex.

Post hoc Dunn tests revealed significant differences between all three ancestry groups, with the differences between African and European males being the least pronounced (see Table 3). In general, WHR emerged as the most discriminative index between ancestry groups, showing consistently strong differences. These results highlight the importance of accounting for ancestry when interpreting anthropometric measures and assessing cardiometabolic risk. Overall, the distribution of BMI, WHR, and WHtR varies significantly by ancestry, indicating that applying uniform diagnostic thresholds across populations may lead to misclassification-overestimating risk in some groups and underestimating in others. Differences in risk classification are shaped not only by health variation, but also by how risk is measured and where thresholds are drawn. This points to a potential structural bias, as current thresholds do not reflect population-specific body composition patterns. Notably, individuals with Native American ancestry were consistently classified at higher risk across all indices and sexes, underscoring the need for ancestry-adjusted criteria to improve equity and accuracy in risk assessment).

Additional analyses including waist circumference (WC) showed significant differences across ancestry groups in both sexes (p<0.001, Kruskal–Wallis). However, effect sizes were consistently greater for WHR and WHtR than for WC and BMI.

Chi-square tests conducted on contingency tables of risk categories by ancestry group and sex revealed statistically significant differences across all anthropometric indices (see Table 2). These results indicate that the current diagnostic thresholds produce unequal classifications of cardiometabolic risk across ancestry groups. Figure 5 illustrates marked disparities in the proportion of individuals classified as “at risk” with consistently higher prevalence among those with predominantly Native American ancestry across nearly all indices and both sexes. While the magnitude of these differences varies by index, the overall pattern persists, suggesting that the issue cannot be solely attributed to the technical properties of a given index. Rather, it reflects a potential structural diagnostic bias arising from the use of cut-off points that have not been calibrated to diverse body phenotypes. If existing criteria could be reliably applicable across populations, the proportion of individuals classified as high risk would reflect true underlying health disparities rather than artifacts of measurement. While it is acknowledged that disease burden is not necessarily distributed homogeneously across populations, the consistent and pronounced over and under representation of certain ancestry groups in high-risk categories suggests that current thresholds may not adequately account for the variation in ancestry-specific body composition patterns among populations. Thus, the observed disparities likely reflect a combination of true epidemiological differences and structural biases in diagnostic criteria.

Considering the analysis by country and information marker of ancestry (using a >70% ancestry membership threshold), our results suggest that the health risk profile varied more in our sample amongst individuals with predominant European ancestry. This sub-sample also had a higher concentration of high-risk individuals in terms of BMI than other groups. We also observed that the American ancestry predominant sub-sample displayed a high prevalence of risk across all anthropometric indices, with a substantial overlap in high-risk categories for WHR\(\cap\)WHtR, with 26.7% (female) and 38.0% (male) of individuals in this group simultaneously exceeding the risk thresholds for both indices. This prevalence was notably higher than that observed in the European and African predominant ancestry groups, highlighting a distinct pattern of central adiposity-related risk within this population.

The Kruskal-Wallis test revealed significant differences in anthropometric indices (BMI, WHR, WHtR) across AIM categories in both female and male populations (see Table 4). Our results show that in females body shape variation is influenced by genetic ancestry when considering highly informative AIM classifications (H = 9.66, p = 0.008). Consistent with our hypothesis we also observed differences among male AIM groups (H = 8.88, p =0.012).

Table 4 Summary of Kruskal-Wallis test results for anthropometric indices across ancestry-informative marker (AIM) groups and national populations, separately for females and males.
Full size table
Fig. 6
Fig. 6
Full size image

Principal Component Analysis (PCA) of female colored by combined risk categories derived from BMI, WHR, and WHtR. Each point represents an individual, with colors indicating combinations of obesity classifications and associated disease risk levels according to WHO criteria. Black squares represent the subset of individuals with available 3D models, whose extreme body shapes are shown along the PCA axes.

Unlike our finding with AIMs, the distribution of anthropometric indices did not vary amongst countries for either sex. The Kruskal-Wallis test for females (H = 7.32, p = 0.292) and males (H = 4.97, p = 0.420) suggested that body shape variation does not significantly differ between national populations. These findings contrast with the AIM-based analysis, emphasizing that genetic ancestry rather than geopolitical borders may play a more substantial role in anthropometric variation.

The ABSI and HI anthropometric indices were calculated for all participants. For the ABSI, the mean was 0.077599 (\(\pm 0.005284\)), consistent with the average values (\(\sim 0.078\)) reported in Krakauer’s original studies (2012). The HI had a mean of 0.180953 (\(\pm 0.007604\)), an acceptable value given its range is similar to that of the ABSI. As expected, strong positive correlations were observed between traditional central adiposity metrics. For example, BMI showed a very high correlation with WC (\(r = 0.82\)) and HC (\(r = 0.83\)). Similarly, WHtR showed an extremely high correlation with WC (\(r = 0.90\)) and HC (\(r = 0.74\)), confirming its usefulness as a robust indicator of central adiposity. As expected, the correlation matrix also validates the statistical independence of ABSI and HI with respect to height and BMI. The correlation of ABSI with height (\(r = -0.08\)) and BMI (\(r = -0.01\)) was close to zero. Similarly, the correlation of HI with height (\(r = -0.01\)) and BMI (\(r = -0.18\)) was also very low. These findings suggest that ABSI and HI are unique measures that are not redundant with more common metrics, justifying their inclusion as complementary indicators in risk analysis.

Our PCA analysis of the BMI, WHR and WHtR anthropometric indices shows that the first two principal components explain most of the variance in both female and male groups (see Fig. 6). In the female group, the first principal component (PC1) explains approximately 80.77% of the variance, while the second principal component (PC2) explains 17.85%. For the male group, PC1 explains 83.10% of the variance and PC2 explains 12.46%. In terms of the eigenvectors, in both groups PC1 is dominated by similar contributions from the three anthropometric indices, with slightly different values: in female, BMI (0.5571), WHR (0.5362) and WHtR (0.6341), while in male, BMI (0.5675), WHR (0.5576) and WHtR (0.6058). This suggests that this component captures a combination of the three measures as a general pattern of body variation. On the other hand, PC2 presents negative values for BMI and WHtR, and positive values for WHR in both groups. In females, the values are BMI (-0.6662), WHR (0.7444) and WHtR (-0.0441), while in male they are BMI (-0.6662), WHR (0.7434) and WHtR (-0.0602). This indicates that the second principal component mainly differentiates between BMI and WHR, with a smaller impact of WHtR. Overall, these results suggest that the relationships between BMI, WHR and WHtR are similar in both sexes, albeit with small differences in the proportion of variance explained and in the weights of the variables in the principal components.

Figure 7 presents the distribution of the first principal component (PC1) from the principal component analysis (PCA) among female participants, stratified (colored) by combined risk categories derived from BMI, WHR, and WHtR indicators. A clear upward trend in PC1 values is observed across increasing risk levels, with the lowest scores found in women classified as underweight or normal weight with low disease risk, and the highest scores in those classified as “Obesity class III” with high disease risk. This pattern suggests that PC1 captures morphological variations related to increased body mass and central fat accumulation, highlighting a shape-based gradient associated with cardiometabolic risk.

The effect of Generalized Procrustes Analysis (GPA) on 3D body model alignment is illustrated in Fig. 8. The left panel depicts the unaligned configurations, where each point cloud corresponds to a distinct model. Significant variability in position, scale, and orientation is observed, hindering direct shape comparison. In contrast, the right panel presents the models after Procrustes alignment to a common reference (highlighted in red). This process eliminates differences attributable to translation, rotation, and scaling, facilitating a precise assessment of shape variation. The convergence of the point clouds around the reference confirms the efficacy of the alignment procedure.

Fig. 7
Fig. 7
Full size image

Distribution of Principal Component 1 (PC1) for female participants across combined risk categories defined by BMI, WHR, and WHtR. Each boxplot represents the spread of PC1 scores within a specific category. A progressive increase in PC1 values is observed with increasing levels of obesity and disease risk, suggesting a morphological gradient in body shape associated with cardiometabolic risk profiles.

Fig. 8
Fig. 8
Full size image

Three-dimensional point distribution of body models before (left) and after (right) Generalized Procrustes Analysis. (a) Each color represents an individual model, showing variation in size, orientation, and position. (b) After alignment, models are superimposed onto a common reference (red), enabling clearer comparison of shape by eliminating non-shape differences.

Discussion

Our findings suggest that different genetic ancestry backgrounds are associated with distinct body types or morphotypes (body shape profiles). For instance, some ancestry sub-samples tend to show greater accumulation of fat in the abdominal area, while others show a more generalized distribution. These differences are not always captured by traditional indicators, which can lead to under- or overestimation of obesity and, consequently, derive into a noisy framework of obesity-related health risks on specific, under-represented populations. Therefore, understanding how body shape varies with ancestry is essential for improving how we define and assess obesity and its associated risks, particularly in admixed populations such as those in Latin America71. Indeed, our study demonstrates that in Latin American populations even broad genetic ancestry categories can significantly influence body shape and obesity-related risk stratification.

Clear differences in anthropometric profiles were observed across individuals with different genetic ancestry backgrounds. Notably, waist-based indices (WHR and WHtR) identified a higher proportion of at-risk individuals among those with predominantly Amerindian ancestry, while BMI tended to classify more individuals with predominantly European ancestry as at risk. This supports the hypothesis that universal diagnostic thresholds may inadequately capture obesity-related health risks in admixed populations. These findings align with previous research performed on the CANDELA dataset regarding ancestry-linked anthropometric variation34 and extend them by integrating 3D morphometrics to reveal geometric patterns (e.g., visceral adiposity distribution). These traits are clearly geometric-morphometric aspects of the phenotype that are not captured by traditional indices.

Beyond anthropometric indices, it is important to acknowledge that discrepancies in obesity-related risk classification may become even more pronounced when integrating other clinical and metabolic dimensions, such as biochemical markers (e.g., lipid profile, glucose, insulin), physiological parameters (e.g., blood pressure, resting heart rate), family history of metabolic disease, and lifestyle factors72,73. Incorporating these variables alongside ancestry-informed anthropometric and morphometric measures would provide a more comprehensive and personalized assessment of cardiometabolic risk, particularly in genetically heterogeneous populations like those of Latin America [REV-10]74,75.

We previously explored the contribution of genomic ancestry and socioeconomic status to obesity in a sample of admixed Latin Americans from the CANDELA dataset and reported lack of consistency among-indexes when ascertaining obesity34. In other words, the proportion of obesity was heavily dependent on the index and the population. We suggested that genomic ancestry has a significant influence on anthropometric measurements, especially on central adiposity, and that better approaches to overweight and obesity phenotypes are needed in order to obtain more precise reference values.

In Navarro et al. (2020) we attempted to improve upon classical anthropometric measures of obesity and explored 3D image-based computational approaches to capture the distribution of abdominal adipose tissue as an aspect of shape35. We reported shape indicators to be good predictors of the behavior of classical measurements, and evaluated the accuracy of 3D features to describe body shape, overweight and obesity related traits. More recently, we developed more accurate 3D models and representations of body shape, which enabled more precise, systematic, and fast measuring capabilities43,64. The shape indicators reported in said papers proved to be accurate predictors of classical indices, adding geometric characteristics that reflect more properly the shape of the bodies under study. This dataset included raw point clouds and parameterized 3D body surfaces, enabling both anthropometric and geometric morphometric analyses. The study demonstrated that 3D-derived shape variables could capture meaningful variation in body morphology and were reliable predictors of traditional anthropometric indices. Importantly, the dataset facilitates the investigation of population-specific body shape patterns, relevant for studying obesity and metabolic health and provides a framework for developing more precise and ancestry-informed phenotypic assessments.

Our team has also propose an automatic 3D body shape-based descriptor and classifier aimed at improving the assessment of obesity beyond traditional anthropometric methods33. This computational approach leverages three-dimensional imaging to enable more precise characterization of body shape, particularly in identifying abdominal regions most strongly associated with obesity. By analyzing clustering patterns, the method also allows for the detection of potential risk thresholds, thus supporting large-scale epidemiological studies.

Ancestry-driven variation in obesity phenotypes

From an evolutionary perspective, the disparities detected here may reflect ancestral adaptations to distinct energetic environments. The thrifty genotype hypothesis76 posits that alleles promoting efficient fat storage were advantageous in pre-industrial societies with intermittent food scarcity. Such alleles could disproportionately influence central adiposity in Amerindian-descended groups, whose evolutionary history involved extreme climatic and nutritional pressures (e.g., Andean high-altitude adaptations)77. Conversely, European ancestry components may be associated with peripheral fat deposition, as seen in our BMI-dominant risk cluster-a pattern potentially linked to colder climate adaptations78. However, broad tri-hybrid models need to be further refined to implement more fine-grained ancestry panels that are necessary to detect more local adaptive past processes or drift-derived patterns of variation (e.g. See25).

Pitfalls of one-size-fits-all diagnostics

Our Venn analyses revealed that only 62% of high-risk individuals were concurrently identified by BMI, WHR, and WHtR, with ancestry group-specific disagreement rates (Amerindian: 28%; European: 19%). This inconsistency mirrors critiques of BMI’s validity in South Asian and African populations79,80 and underscores two critical gaps.

These discrepancies highlight two key limitations in current obesity assessment strategies. On the clinical side, an over-reliance on BMI may lead to the underdiagnosis of individuals at metabolic risk, particularly among Amerindian-dominant populations who may present a “healthy” BMI yet show elevated WHtR values indicative of central adiposity. On the methodological side, traditional anthropometric indices are unable to distinguish between subcutaneous fat, which may be metabolically neutral or protective, and visceral fat, which is strongly associated with adverse health outcomes. This distinction, however, becomes apparent through the application of 3D shape analyses, such as principal component analysis (PCA), which allow for a more precise characterization of body fat distribution. These two gaps are clearly exposed in our analyses of Guatemalan Maya-Kaqchikel cohort (SER project), where 34% of women classified as “healthy” by BMI fell into high-risk WHtR categories, echoing disparities observed in Indigenous populations globally81. These findings align with Rubino et al. (2025), and support the use of complementary measures alongside BMI to better identify at-risk individuals across diverse ancestries6.

3D morphometrics: precision and pragmatism

Our Procrustes-based shape analysis represent a formalization of the three-dimensional shape of the human body in our sample, and identified three ancestry-correlated morphotypes. Individuals with predominant Amerindian ancestry tended to exhibit a “centralized” body shape, characterized by higher PC2 scores and a distribution pattern resembling apple-shaped obesity, which is often linked to higher visceral fat accumulation and cardiometabolic risk. In contrast, those with predominant European ancestry displayed a more “diffuse” body shape, with lower PC1 and PC2 scores, corresponding to a gynoid fat distribution typically concentrated around the hips and thighs-an arrangement often considered metabolically less harmful. Admixed individuals, as expected, presented a spectrum of “intermediate” morphotypes, reflecting heterogeneous body shape configurations and risk profiles that do not fit neatly into existing diagnostic categories.

While 3D scanning offers unparalleled resolution, it still needs a scalability strategy in order to achieve a clinical concrete using. Here we propose a two-tiered approach. We propose a two-tiered approach to bridge this gap. In research and clinical settings, high-resolution 3D scanners combined with fine-grained ancestry informative marker (AIM) panels can be employed to refine ancestry-specific thresholds for obesity-related risk assessment. For broader public health applications, more accessible and scalable technologies, such as AI-assisted video morphometry tools like body2vec_mesh43, could enable large-scale screening and monitoring of body shape and composition, particularly in under-resourced settings. This dual strategy balances precision with accessibility, advancing the integration of shape-based metrics into both precision medicine and population-level interventions.

Navigating the ancestry-race-ethnicity triad

Genetic ancestry, as quantified here via AIMs, is a biological construct distinct from the sociopolitical dimensions of race/ethnicity46. However, their entanglement in Latin America-where European ancestry often correlates with higher socioeconomic status23 demands cautious interpretation. For instance, the elevated obesity risk in our European-dominant group may reflect gene-environment interactions (e.g., urbanized diets) rather than innate biological differences. Future studies should integrate socio-demographic covariates (e.g., income, urbanization) to disentangle genetic and environmental effects, avoiding deterministic narratives82.

Future directions and policy implications

To move toward more equitable and effective strategies for obesity assessment, we highlight three key areas of action. First, ancestry-aware guidelines should be developed by leveraging data from admixed populations to define region-specific obesity thresholds. Mexico’s implementation of WHtR-based standards serves as a valuable precedent in this regard83. Second, the democratization of technology is essential to expand access to advanced phenotyping methods. This includes validating low-cost tools, such as smartphone-based 3D scanning, for use in resource-limited settings36,84,85,86. Finally, promoting global health equity requires addressing the stark underrepresentation of Latin American populations in genetic and obesity research. Currently, less than 5% of genomics studies related to obesity include data from this region87, limiting the generalizability and applicability of findings in global health contexts. To achieve these goals, it is crucial to increase the collection of high-quality data on obesity, cardiometabolic diseases, and fat distribution, ideally including 3D body composition measures and body shape from all ancestry groups in any given populations. Such data are essential for identifying accurate links between morphotypes and health risks and, in so doing, create novel, ancestry-informed predictors of risk and establishing diagnostic thresholds that reflect the biological diversity of global populations more accurately. Without this foundation, current models risk perpetuating structural biases and diagnostic inaccuracies, particularly in regions with high genetic admixture and limited representation in biomedical research. Moreover, our approach also reinforces the importance of policy engagement, necessary to the implementation of novel methodological approaches aimed to guarantee more sophisticated diagnostic procedures and a broader, more egalitarian access of patients from different socio-economic contexts to it. However, the underlying biological background and fine-grained information must be leveraged responsibly-acknowledging the societal dimensions of health disparities while refining tools to detect them.

Conclusion

Our findings demonstrate that genetic ancestry is a significant determinant of anthropometric variation and cardiometabolic risk classification in Latin American populations. Across both sexes, BMI, WHR, and WHtR varied systematically by ancestry group, with Native American ancestry consistently associated with higher values for all indices and a greater proportion of individuals classified as high risk. Among these measures, WHR emerged as the most discriminative index, followed closely by WHtR, highlighting their value in capturing ancestry-specific differences in body fat distribution. Analyses based on AIMs revealed clear associations between genetic ancestry and anthropometric indices, whereas country-level comparisons showed no significant variation, underscoring the primacy of genetic ancestry over geopolitical boundaries in shaping body composition. Our analysis also confirms that the ABSI and HI indices are statistically independent of BMI and height. This independence validates their use as complementary tools for a more accurate assessment of health risk in diverse populations. Additionally, PCA confirmed a consistent morphological gradient across risk categories, with PC1 capturing combined variation in BMI, WHR, and WHtR, and aligning strongly with cardiometabolic risk stratification. Finally, 3D body shape alignment using Generalized Procrustes Analysis proved effective for standardizing models and isolating shape variation, reinforcing the utility of geometric morphometric approaches for refining obesity risk assessment. Collectively, these results highlight both the limitations of applying uniform diagnostic thresholds across admixed populations and the need for ancestry-informed, shape-based criteria to improve the accuracy and equity of cardiometabolic risk evaluation in diverse populations.