Related Experiment Videos
Identification of metabolic risk phenotypes among U.S. adolescents: A machine learning analysis of NHANES 2017-2023
1School of Nursing, National Taipei University of Nursing and Health Sciences, No. 365, Mingde Rd. Beitou District, Taipei, 11219, Taiwan, ROC.
Background:
Adolescent metabolic risk is multidimensional and may not be fully characterized by individual anthropometric or biochemical indicators considered separately; however, evidence regarding data-derived metabolic-risk phenotypes among adolescents remains limited.
Purpose:
To identify metabolic risk phenotypes among U.S. adolescents using unsupervised machine learning and examine demographic and socioeconomic factors associated with phenotype membership.
Methods:
Data from adolescents aged 12-17 years participating in the National Health and Nutrition Examination Survey (NHANES) 2017-March 2020 and 2021-2023 cycles were analyzed. K-means clustering was performed among 848 adolescents with complete data for seven anthropometric and cardiometabolic variables. Logistic regression examined sociodemographic factors associated with elevated metabolic-risk phenotype membership.
Results:
A two-cluster solution yielded the highest silhouette coefficient (0.289), identifying a lower metabolic-risk phenotype (n = 580, 68.4%) and an elevated metabolic-risk phenotype (n = 268, 31.6%). The elevated-risk phenotype was characterized by substantially higher BMI and waist circumference, together with higher triglycerides and LDL-C cholesterol, lower HDLC, and modestly higher glycated hemoglobin. In the adjusted logistic regression analysis (n = 763), older age was associated with greater odds of elevated-risk phenotype membership (aOR = 1.18, 95% CI 1.07-1.29), whereas income-to-poverty ratio was associated with lower odds (aOR = 0.78, 95% CI 0.70-0.87).
Conclusions:
Unsupervised machine learning identified two exploratory metabolic-risk phenotypes characterized by concurrent differences in adiposity, lipid, and glycemic measures. The findings provide an exploratory basis for investigating multidimensional cardiometabolic assessment approaches relevant to pediatric nursing. Independent replication and longitudinal validation are needed to determine their prognostic and clinical utility.