Machine learning identifies waist‑height ratio (WHtR) as the strongest determinant of diabetes and prediabetes in children and adolescents - A comprehensive national nutrition survey

Aug 5, 2026

Rising rates of prediabetes and type 2 diabetes (T2D) among adolescents in India require precise identification of risk factors that enable earlier prevention. This study applies machine learning to determine the most informative anthropometric and demographic characteristics for diabetes and prediabetes in ages 10–19 using India’s Comprehensive National Nutrition Survey (2016–2018). Nine supervised algorithms evaluate predictive performance via multiple accuracy metrics.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Abstract

Background

Objective

Methods

Results

Conclusions

Introduction

International Journal of Diabetes in Developing Countries [https://doi.org/10.1007/s13410-025-01531-9](https://doi.org/10.1007/s13410-025-01531-9) Machine learning identifies waist‑height ratio (WHtR) as the strongest determinant of diabetes and prediabetes in children and adolescents: A comprehensive national nutrition survey Kirti Chauhan5 · Thozhukat Sathyapalan1 · Usman Malabu4 · Shashank Rameshchandra Joshi2 · Shri Kant Singh3 · Harshal Deshmukh4 Received: 9 October 2024 / Accepted: 1 July 2025 © The Author(s) 2025 Abstract Background Given the increasing incidence of prediabetes and type 2 diabetes (T2D) in the adolescent population in India, it is essential to identify the risk factors associated with these conditions. Understanding the risk factors associated with prediabetes and T2D can lead to timely interventions to prevent and potentially avert long-term health complications. Objective This study aims to use machine learning algorithms to identify the best anthropometric and demographic characteristics associated with prediabetes and diabetes in Indian children and adolescents ages 10–19. Methods The study utilizes the Comprehensive National Nutrition Survey conducted in 2016–2018 in India. The study sample includes children and adolescents aged 10–19 years. We used nine supervised machine learning algorithms to classify, assess, and identify the best model for ascertaining the risk of diabetes among adolescents in India. Various indices were used to evaluate the classification algorithms, such as the ‘accuracy score’, ‘F1 score’, ‘recall score’, ‘precision score’, and ‘area under the curve’ (i.e. , AUC) . Results were obtained based on the model with higher precision and accuracy in predicting the risk of diabetes among study subjects. Cutoff points for prediabetes were between 5.7 and 6.4 mmol/l and diabetes greater than 6.4 mmol/l. Results The study comprised 12,318 children and adolescents (6333 males and 5985 females) . The prevalence of diabetes and prediabetes in the study population was 11%(n = 1888), while the prevalence of diabetes alone was 0.6%(n = 233) . WHtR was the most crucial feature in predicting prediabetes/diabetes, with an optimum cutoff of 0.62, a sensitivity of 0.93, and an AUC of 0.79. Conclusions The findings derived from our machine learning analysis underscore the significance ofWHtR as a cost-effective and valuable tool for diabetes and prediabetes screening among adolescents in India. Keywords Waist-height ratio · Diabetes · Prediabetes · Machine learning · Random forest * Harshal Deshmukh [harshal.deshmukh@jcu.edu.au](harshal.deshmukh@jcu.edu.au) Kirti Chauhan [kirti070197@gmail.com](kirti070197@gmail.com) Thozhukat Sathyapalan[thozhukat.sathyapalan@hyms.ac.uk](thozhukat.sathyapalan@hyms.ac.uk)[ ](thozhukat.sathyapalan@hyms.ac.uk)Usman Malabu [usman.malabu@jcu.edu.au](usman.malabu@jcu.edu.au)[ ](usman.malabu@jcu.edu.au)Shashank Rameshchandra Joshi [shashank.sr@gmail.com](shashank.sr@gmail.com) Shri Kant Singh [sksingh31962@gmail.com](sksingh31962@gmail.com) 1 Department of Diabetes, Endocrinology and Metabolism, Allam Diabetes Centre, Hull University Teaching Hospitals, Hull HU3 2PA, UK 2 Department of Endocrinology, DM, FACE, MD, FRCP, FACP, Joshi Clinic, Mumbai, Maharashtra, India 3 Department of Survey Research and Data Analytics, International Institute for Population Sciences, Mumbai, Maharashtra, India 4 James Cook University, Australia, Townsville, QLD, Australia 5 Translational Research and Biostatistics, All India Institute of Ayurveda, New Delhi, India Introduction Diabetes is a leading cause of mortality and morbidity, affecting millions worldwide, including individuals as young as children and adolescents. The Centers for Disease Control and Prevention (CDC) estimated that the number of individuals under the age of 20 years, the prevalence of type 2 diabetes (T2D), has increased by 95% from 2001 to 2017 alone in the USA [1] . At

FAQs

What does the study aim to identify for Indian children and adolescents?

It aims to use machine learning to determine the best anthropometric and demographic characteristics associated with prediabetes and diabetes.

What dataset and age group were used in the study?

The analysis uses the Comprehensive National Nutrition Survey conducted in India from 2016 to 2018, including participants aged 10–19 years.

Which feature was most important for predicting diabetes and prediabetes?

Waist-height ratio (WHtR) was the most crucial feature, with an optimum cutoff of 0.62, sensitivity of 0.93, and AUC of 0.79.

Aug 5, 2026

Upload by Clementine