Work

AI and Conventional Statistical Approaches to Evaluate Indices for Diabetes Screening. DICA 2022-23

Year: 2025

Type: article

Abstract: Background Early detection of diabetes mellitus (DM) is important for prevention of cardiovascular risk. Anthropometric indices such as body-mass index (BMI), waist-to-height ratio (WHtR), A Body Shape Index (ABSI), waist circumference (WC) and body fat percentage could be used to classify DM patients. We aim to assess the ability of these indices to predict diabetes using Receiver-Operating-Characteristic (ROC) curves and artificial intelligence models like CART decision trees. Material and methods Cross-sectional, population-based, representative survey (DICA Study). 2020 participants aged ≥ 20 were selected through two-stage stratified sampling (health area, sex, age, urban/rural residence). DM defined as fasting glucose ≥ 126 mg/dL, HbA1c ≥ 6.5 %, or reported diagnosis/treatment. Area under the ROC curve (AUC) and Youden index cut-off was calculated; an additional sex-specific CART was trained and provided sensitivity (Se), specificity (Sp), and positive and negative predictive values (PPV, NPV). Results Diabetes prevalence was 13.6% overall. - ROC curves. Men: highest AUC for ABSI 0.729, WHtR 0.728 and WC 0.710; BMI 0.625. Women: WHtR 0.752 and WC 0.732; ABSI 0.689; BMI 0.696. Optimal cut-offs: men: ABSI 0.81 (Se 0.74; Sp 0.62) and WHtR 0.57 (0.76; 0.62); women: WHtR 0.54 (0.79; 0.62) and WC 85.5 cm (0.79; 0.59). - CART trees. The formula used on the tree with the best results was: Diabetes ∼ WHtR + Fat percentage + Body mass. Men: Se 0.75, Sp 0.61, PPV 0.27, NPV 0.93. Women: Se 0.44, Sp 0.79, PPV 0.22, NPV 0.92. The multivariate tree model did not improve discrimination over individual ROC curves. Conclusions WHtR was overall the best anthropometric predictor of diabetes, especially in women. BMI and ABSI showed lower utility, while CART trees provided no additional predictive ability. The simplicity and solid performance of WHtR supports its use as a primary-care screening tool. Funding Ref. PI20/00068, Carlos III Health Institute. Key messages • Waist-to-height ratio (WHtR) outperforms BMI, WC & ABSI for spotting undiagnosed diabetes in Spanish adults, delivering a quick, no-cost screen for primary care. • CART AI offers no extra gain; simple WHtR alone matches complex models, proving low-tech antropometric indices can rival machine learning for diabetes risk detection.

Source: European Journal of Public Health

DOI: 10.1093/eurpub/ckaf161.1016

Authors: M Moreno, J M Huerta, N A Cabrera, S M Colorado, M Ballesta, M D Chirlaque

Institutions: Instituto Murciano de Investigación Biosanitaria, Centro de Investigación Biomédica en Red de Epidemiología y Salud Pública, Universidad de Murcia

Language: English

Topic: Artificial Intelligence in Healthcare

Subfield: Health Information Management

Field: Health Professions

Domain: Health Sciences

SDG: Good health and well-being

Open Access status: gold

Estimated APC paid: $3,181