Back

A Deep Learning-Based Predictive Algorithm for Metabolic Syndrome Detection in the U.S. Population

Correa Segade, C.; Solozabal, R.; Hammouri, Z. A. A.; Gomez-Peralta, F.; Rossman, H.; Vidal, J. C.; Klonoff, D. C.; Segal, E.; Matabuena, M.

2026-06-02 endocrinology
10.64898/2026.05.24.26354007 medRxiv
Show abstract

Objective To develop clinically operational, population-representative risk-score models for detecting metabolic syndrome (MetS) in U.S. adults by incorporating the NHANES survey design. Research Design and Methods We analyzed 36,812 U.S. adults from NHANES 1988--2018. Seven models of increasing clinical complexity were trained and evaluated, ranging from basic demographics to full biochemical panels. We used a new deep-learning methodology for survey data with a predictive uncertainty quantification model. Results A model combining anthropometrics, vital signs and a basic lipid panel achieved an AUC of 0.923 at an estimated cost of 0.40 eur per individual. Adding diabetes-specific biomarkers, including fasting plasma glucose (FPG) and glycated hemoglobin (HbA1c), yielded only marginal improvements. Conclusions This low-cost population-representative screening tool for MetS may help identify at-risk individuals and support data-driven public health interventions.

Matching journals

The top 9 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.