Multidimensional health phenotypes in adolescents identified through unsupervised machine learning: associations with cardiometabolic risk, physical fitness, psychosocial and academic variables
Academic Journal of Health Sciences · 16.10.2026 · 41 (5)
DOI: 10.65735/ajhs.2026.41.5.3 ↗Kurz gefasst
Fachartikel in der Zeitschrift Academic Journal of Health Sciences, online erschienen am 16.10.2026. Verfasst von Rodrigo Yañez-Sepulveda und 7 weiteren Autor:innen.
Abstract (Originalsprache)
Introduction
Adolescent health is inherently multidimensional, emerging from interactions between physiological, behavioral, and psychosocial factors. Conventional surveillance strategies frequently rely on isolated indicators, limiting the identification of integrated risk profiles.
Objective
This study aimed to identify integrative health phenotypes in adolescents using unsupervised machine learning and to examine their association with cardiometabolic risk, physical fitness, psychosocial, and academic variables.
Methods
A cross-sectional sample of 1,143 adolescents aged 12–17 years was analyzed. Sixteen standardized variables spanning anthropometric measures, blood pressure, physical fitness components, lifestyle behaviors, academic performance, and psychological indicators were included. K-means clustering was applied following standardized preprocessing procedures, including imputation, outlier screening, and Z-score normalization. The optimal number of clusters was determined using the elbow method, silhouette coefficient, and internal validity indices.
Results
A three-cluster solution was retained (silhouette: 0.173 [male sex] and 0.115 [female sex]), identifying: (1) a 'fitness' phenotype characterized by favorable body composition, higher cardiorespiratory fitness and muscular fitness, and better academic performance; (2) an 'intermediate' phenotype with near-average profiles across domains; and (3) a 'cardiometabolic risk' phenotype defined by central adiposity, elevated blood pressure, reduced functional capacity, and less favorable psychosocial indicators. Between-cluster differences were statistically significant across most of physiological and psychosocial variables (p Conclusion: Unsupervised learning can identify coherent, multidimensional health phenotypes in adolescents using non-clinical, school-based data. Data-driven phenotyping may support scalable, integrative frameworks for adolescent health monitoring within big data–enabled public health systems.
Zitieren
APA
Yañez-Sepulveda, R., López, P. J. T., Miño, C., Ridgell, L., Quiroz-Cardenas, F., Cardena, D. Q., Montenegro-Espinosa, J. A., & Gil, J. F. L. (2026). Multidimensional health phenotypes in adolescents identified through unsupervised machine learning: associations with cardiometabolic risk, physical fitness, psychosocial and academic variables. Academic Journal of Health Sciences, 41(5). https://doi.org/10.65735/ajhs.2026.41.5.3
Vancouver
Yañez-Sepulveda R, López PJT, Miño C, Ridgell L, Quiroz-Cardenas F, Cardena DQ, et al. Multidimensional health phenotypes in adolescents identified through unsupervised machine learning: associations with cardiometabolic risk, physical fitness, psychosocial and academic variables. Academic Journal of Health Sciences. 2026;41(5). doi:10.65735/ajhs.2026.41.5.3
BibTeX
@article{YaezSepulveda2026,
author = {Yañez-Sepulveda, Rodrigo and López, Pedro Juan Tárraga and Miño, Camila and Ridgell, Lucille and Quiroz-Cardenas, Fiorella and Cardena, Doris Quiros and Montenegro-Espinosa, Jose Adrian and Gil, Jose Francisco lopez},
title = {Multidimensional health phenotypes in adolescents identified through unsupervised machine learning: associations with cardiometabolic risk, physical fitness, psychosocial and academic variables},
journal = {Academic Journal of Health Sciences},
year = {2026},
volume = {41},
number = {5},
doi = {10.65735/ajhs.2026.41.5.3}
}
EUVIMED