The research, led by Ava Aghakhani, Darya Rajabi, and Marziyeh Khoshkholgh of Tehran University of Medical Sciences, together with senior authors Solaleh Emamgholipour and Hossein Poustchi, drew on baseline data from the PolyIran-Liver cohort, a trial nested within the well-established Golestan Cohort Study in northern Iran. The analysis focused on 570 men aged 50 to 80 years, a demographic in which cardiometabolic disease burden is high and in which the limitations of BMI-based classification are particularly consequential. By cross-classifying participants according to body mass index and the presence or absence of metabolic syndrome, the investigators created six distinct obesity phenotypes: metabolically healthy obese, metabolically unhealthy obese, metabolically healthy overweight, metabolically unhealthy overweight, metabolically healthy normal-weight, and metabolically unhealthy normal-weight.

Against these six groups, the team measured an unusually broad panel of adiposity and organ-damage markers. Visceral fat thickness, a direct ultrasound estimate of the fat layer lining the abdominal cavity, was assessed alongside liver stiffness, a surrogate measure of hepatic fibrosis and fatty liver burden. Carotid intima-media thickness, an established marker of subclinical atherosclerosis, was also recorded. In parallel, the researchers calculated four lipid- and glucose-based indices from routine blood work: the visceral adiposity index, the lipid accumulation product, the body adiposity index, and the triglyceride-glucose index, commonly abbreviated as the TyG index. The central question was straightforward: which of these measures actually distinguish men who are obese but metabolically healthy from those who are lean but metabolically compromised?

The answer, published as an open-access article, is that no single index does everything, but the complementary strengths of the different measures map neatly onto the clinical problem. Visceral fat thickness, liver stiffness, and the body adiposity index all showed significant differences between obesity groups that shared the same metabolic status, meaning they tracked the sheer burden of excess fat rather than its metabolic consequences. In contrast, the triglyceride-glucose index and the visceral adiposity index separated metabolically healthy from metabolically unhealthy men within the same weight category, exactly the discrimination that BMI cannot provide. The lipid accumulation product stood out as the most versatile marker, differing significantly both across obesity categories at fixed metabolic status and across metabolic categories at fixed obesity status.

Regression analyses reinforced these patterns. When the researchers modeled the odds of elevated liver stiffness, the metabolically unhealthy obese, metabolically unhealthy overweight, and even metabolically healthy obese groups all showed significantly increased odds compared with the reference healthy phenotype, suggesting that excess fat mass itself, regardless of metabolic presentation, loads the liver with fibrogenic risk. The odds of an elevated lipid accumulation product were significantly raised in every non-reference group, underscoring the sensitivity of this simple measure, which is derived from waist circumference and fasting triglycerides, to deviations from metabolic health in either direction. Multiple linear regression models for visceral fat thickness, liver stiffness, and carotid intima-media thickness confirmed the group-level findings.

Perhaps the most sobering result was the one that failed to appear. Carotid intima-media thickness, the ultrasound marker of early arterial wall thickening, showed no significant differences across the six obesity phenotypes. In a cohort of older men, in whom decades of cumulative cardiovascular exposure may have homogenized atherosclerotic burden, the adiposity and metabolic indices did not translate into detectable differences in subclinical vascular disease. The authors are careful to note that this null finding should be interpreted within the specific context of their population and does not rule out associations in younger cohorts, in women, or in populations followed longitudinally rather than assessed at a single time point.

The technical logic behind the lipid-based indices helps explain why they outperform BMI. The visceral adiposity index integrates waist circumference, body mass index, triglycerides, and high-density lipoprotein cholesterol into a sex-specific estimate of visceral fat dysfunction, capturing not just how much fat is present but how pathologically it is behaving. The lipid accumulation product combines waist circumference with fasting triglyceride levels, effectively quantifying the lipid overflow that occurs when adipose tissue storage capacity is exceeded. The triglyceride-glucose index, a simple product of fasting glucose and triglycerides, serves as a proxy for insulin resistance, the central metabolic lesion linking obesity to type 2 diabetes and fatty liver disease. Because all three rely on measurements already collected in routine clinical practice, they could be deployed at scale without new equipment or expense.

The study’s imaging findings carry their own implications. Visceral fat thickness measured by ultrasonography emerged as a meaningful discriminator of obesity phenotypes, offering a radiation-free, bedside alternative to computed tomography and magnetic resonance imaging for quantifying the fat compartment most strongly linked to cardiometabolic risk. Liver stiffness measurement, likewise obtainable by ultrasound-based elastography, identified hepatic involvement even in men classified as metabolically healthy obese, a finding consistent with the growing recognition that so-called healthy obesity is frequently a transitional state in which organ damage precedes overt metabolic decompensation. For screening programs in resource-limited settings, the combination of ultrasound and routine blood chemistry may offer a pragmatic risk-stratification pathway that far outstrips the information content of a height and weight measurement.

The authors emphasize that their conclusions are bounded by the study design and population. The cross-sectional analysis captures a single moment in time and cannot establish whether the measured indices predict future cardiovascular events, diabetes, or liver disease progression. The cohort was exclusively male and exclusively older, spanning ages 50 to 80, so the findings cannot be directly generalized to women or to younger individuals in whom the physiology of adipose tissue distribution and metabolic risk may differ. The data derive from a secondary analysis of the PolyIran-Liver trial, originally funded by Tehran University of Medical Sciences, and the current analysis received no dedicated funding. The work was approved by the ethics committee of Tehran University of Medical Sciences and conducted in accordance with the Declaration of Helsinki, with informed consent obtained from all participants.

Even with those caveats, the study lands at a moment when the medical community is actively searching for alternatives to BMI. Recent debates over the definition of clinical obesity have called for measures that distinguish fat distribution and organ dysfunction from simple mass, and the new findings supply a concrete, low-cost toolkit for doing so in middle-aged and older men. The lipid accumulation product and the visceral adiposity index, the authors conclude, can be considered practical and inexpensive indices for metabolic risk stratification, while visceral fat thickness, liver stiffness, and the body adiposity index add anatomical and organ-level resolution that blood tests alone cannot provide. As health systems confront rising rates of obesity-related disease with constrained resources, the message of this research is that the next leap in obesity medicine may come not from new machines but from smarter use of the measurements clinicians already make every day.

Subject of Research: Comparison of visceral fat, liver stiffness, and lipid-based indices for classifying obesity phenotypes in older men

Article Title: Evaluating obesity phenotypes in men aged 50–80 years: comparative insights from visceral fat, liver stiffness, and lipid-based indices

Article References: Aghakhani, A., Rajabi, D., Khoshkholgh, M., Mohamadi, P., Mohammadi, Z., Emamgholipour, S., & Poustchi, H. (2026). Evaluating obesity phenotypes in men aged 50–80 years: comparative insights from visceral fat, liver stiffness, and lipid-based indices. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02548-9

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02548-9

Keywords: obesity phenotypes, metabolic syndrome, visceral fat thickness, liver stiffness, visceral adiposity index, lipid accumulation product, TyG index, body adiposity index, carotid intima-media thickness, BMI limitations, cardiometabolic risk, Golestan Cohort Study

The core of this research centers on metabolomics — the comprehensive study of metabolites, which are the small molecules involved in metabolism within cells, tissues, or organisms. Traditionally, metabolomic profiling requires invasive procedures and significant processing time, limiting its applicability in regular health assessments. However, the utilization of retinal tissue from the RNFL presents a non-invasive, accessible, and highly informative biomarker source. The retinal nerve fibre layer, composed of unmyelinated axons of retinal ganglion cells, reflects systemic physiological states in a unique way, encompassing both neurological and vascular components critical to understanding overall health and disease progression.

Artificial intelligence becomes indispensable in this context due to the intricate and vast data generated by metabolomic analysis. High-throughput mass spectrometry and other advanced biochemical profiling instruments produce rich datasets with thousands of measured metabolites. Extracting meaningful patterns that correlate with disease risk and mortality from this data demands robust computational methods. AI algorithms, particularly deep learning architectures, excel at detecting subtle, multidimensional relationships within the dataset that human analysts might overlook. The research team deployed these AI models to integrate metabolomic signals from the RNFL and correlate them with longitudinal health outcomes, including incidences of cardiometabolic conditions and mortality statistics.

One particularly compelling aspect of the study is the establishment of a predictive metabolomic signature from retinal tissue, which showed remarkable accuracy in stratifying individuals by their risk of fatal and non-fatal cardiometabolic events. Cardiometabolic diseases — encompassing conditions such as coronary artery disease, stroke, diabetes, and related metabolic disorders — remain leading causes of morbidity and mortality worldwide. Current risk assessments rely heavily on clinical metrics and blood markers, which, while informative, might miss subtle signals discernible deep within tissue-specific metabolomic landscapes. The RNFL’s metabolic profile offers an unprecedented window into systemic disease dynamics at an early, potentially reversible stage.

Moreover, the AI-driven approach circumvents several challenges traditionally associated with biomarker discovery. By automating feature extraction and selection processes, the system reduces biases inherent in manual analysis and enhances reproducibility across different populations and scanning platforms. The model’s adaptability means it can continuously improve with added data, reflecting new patient cohorts or emerging health trends. This dynamic learning capability is crucial for tailoring personalized health strategies and could revolutionize how clinicians approach preventive care for high-risk individuals.

The implications of this research extend beyond mortality prediction. Given the retina’s embryological origin as an extension of the central nervous system, metabolomic analysis of the RNFL could potentially illuminate mechanisms underlying neurodegenerative diseases and other systemic disorders with metabolic underpinnings. Early detection and intervention in these conditions depend on sensitive biomarkers capable of tracking disease evolution at a granular molecular level, a role this retinal metabolomics-AI fusion is uniquely positioned to fulfill.

This paradigm shift also underscores the emerging importance of integrating cross-disciplinary expertise — combining ophthalmology, biochemistry, computational science, and clinical epidemiology — to harness AI’s full potential in medicine. The study’s success is a testament to how advanced imaging and metabolomic profiling platforms, coupled with cutting-edge computational algorithms, can unveil biological insights that were previously unattainable. It charts a roadmap for future investigations aiming to expand AI-driven metabolomics to other accessible tissues or biofluids.

Beyond the science, the prospect of a rapid, non-invasive, and highly accurate diagnostic tool has profound public health implications. Cardiometabolic diseases place an enormous burden on healthcare systems through chronic morbidity and acute life-threatening events. Early identification of at-risk individuals, enabled by this retinal metabolomics analytics, could facilitate timely lifestyle or pharmacological interventions to mitigate disease progression, ultimately lowering population-level mortality rates.

While the study demonstrates tremendous promise, implementation in clinical practice will require further validation through large-scale, multicenter trials and longitudinal studies. Ensuring consistency across diverse demographic groups and linking retinal metabolomic profiles with genetic, environmental, and lifestyle factors remains an essential next step. Additionally, ethical considerations concerning data privacy, AI transparency, and accessibility must be addressed to realize equitable deployment of these advanced diagnostic tools.

Future research directions hinted by these findings include expanding the metabolite database specific to retinal tissues and enhancing AI models to dissect complex interactions between metabolic pathways. Researchers anticipate that refinement in AI explainability methods will also play a critical role in gaining clinicians’ trust and facilitating regulatory approval processes. Ultimately, this confluence of AI and retinal metabolomics represents a paradigm shift towards personalized, predictive, and preventive healthcare.

This study exemplifies how leveraging the synergy between artificial intelligence and metabolomic profiling can unravel subtle biological signatures crucial for disease prediction. The retinal nerve fibre layer emerges as a valuable bio-indicator, capable of reflecting systemic health statuses through molecular fingerprints decipherable by AI. As researchers continue to integrate diverse data layers, from genomics to imaging, the future of precision medicine appears increasingly intertwined with such interdisciplinary innovations.

It is anticipated that the adoption of AI-driven metabolomic approaches alongside traditional clinical assessments will soon become standard practice for evaluating cardiometabolic health. Such a future promises not just improved patient outcomes but also a profound transformation in our understanding of disease biology at the molecular level. This new capability may usher in an era where a simple retinal scan can provide comprehensive insights into an individual’s mortality risk and guide targeted interventions well before clinical symptoms manifest.

By harnessing the retina’s unique accessibility combined with cutting-edge AI analytics, the research team has positioned retinal metabolomics at the frontier of clinical diagnostics. This pioneering work not only informs strategies to combat cardiometabolic disease but also establishes a methodological blueprint for leveraging metabolomic signatures from other tissues to tackle a broad spectrum of diseases.

In conclusion, the marriage of artificial intelligence with metabolomics of the retinal nerve fibre layer heralds a new age in biomedicine, where non-invasive molecular diagnostics can robustly inform health risk profiling. With continued innovation and validation, this technology holds the potential to significantly reduce global mortality burdens by enabling earlier, more precise interventions against cardiometabolic and potentially other systemic diseases.

Subject of Research: Artificial intelligence-driven metabolomics analysis of the retinal nerve fibre layer to predict mortality and cardiometabolic disease risks.

Article Title: Artificial intelligence-driven metabolomics of retinal nerve fibre layer to profile risks of mortality and cardiometabolic diseases.

Article References: Yang, S., Xin, Z., Li, H. et al. Artificial intelligence-driven metabolomics of retinal nerve fibre layer to profile risks of mortality and cardiometabolic diseases. Nat Commun 16, 11039 (2025). https://doi.org/10.1038/s41467-025-66979-z

Image Credits: AI Generated

DOI: https://doi.org/10.1038/s41467-025-66979-z