AI Blind Spots in South Asian Cardiometabolic Risk: A Scoping Review of Food and Social Determinant Measurement

Cardiometabolic disease, including cardiovascular disease, type 2 diabetes, dyslipidemia, and metabolic syndrome, has historically and disproportionately affected South Asian patient populations globally1. Across both South Asia and the South Asian diaspora, individuals experience an earlier onset, more aggressive progression, and an overall higher mortality rate from cardiometabolic conditions compared to most other ethnic groups2,3. Importantly, these disparities still persist even at lower BMI thresholds and amongst individuals who have “normal” body weight, reflecting how there is a unique attribution to metabolic patterns including atherogenic dyslipidemia, insulin resistance, and visceral adiposity4. Although these metabolic profiles have been documented, risk assessment tools and frameworks regarding cardiometabolic disease continue to be anchored in Eurocentric phenotypes and fail to capture the nuances of South Asian risk factors5. Nutritional exposures add a layer of complexity. South Asian dietary habits and patterns, whether based on traditionality, can be shaped by contemporary food environments, or migration-adapted ways, with underlying metabolic vulnerabilities that can magnify cardiometabolic risk6. Nutritional components such as refined carbohydrates, energy-dense oils, ultra-processed food (UPF), culturally culinary practices, and changing food accessibility in both South Asia and Western diasporic contexts can contribute to elevated metabolic burden7. Although this is understood, these exposures are rarely measured in a consistent and culturally competent way. As a result, the nutritional determinants of cardiometabolic disease in this population remain poorly understood and consistently undermeasured8. The rapid rise of use in artificial intelligence (AI) and machine learning (ML) in cardiometabolic research has further amplified the consequences of these measurement gaps. AI are increasingly informing and providing patient risk stratification, dietary assessment, cardiometabolic prediction and clinical decision support. However, AI systems are only able to learn from the variables encoded in the training data that are provided to them. Due to the lack of datasets from South Asian specific metabolic indicators, the omission of culturally relevant dietary constructs, and the omission of social and structural determinants, the resulting models embed blind spots that produce systemic misclassification and inequitable performance9. Taken together, these consistent trends reveal a critical gap: although cardiometabolic disease in South Asians is shaped by a dynamic interaction between metabolic traits, nutrition-related exposures, and social/structural determinants, the measurements of these determinants remain fragmented and poorly aligned within equitable development of AI/ML systems. This scoping review addresses this gap by systematically examining how studies evaluating cardiometabolic risk in South Asian populations measure social, structural, and commercial determinants of nutrition, as well as how these measures intersect with AI and ML-based prediction. Through the synthesis of data, this review will critically identify blind spots, inconsistencies, and fairness to outline stronger and more equitable pathways for AI systems in South Asian health.