Lifestyle-metabolic subtypes of insulin resistance: a cross-center latent class and interpretable machine learning study

Background Insulin resistance (IR) is a highly heterogeneous metabolic state jointly influenced by lifestyle behaviors and metabolic factors. However, population-based approaches integrating multidimensional phenotypic information for identifying IR-related phenotypes and conducting individualized risk stratification remain limited. Objective To identify lifestyle-metabolic phenotypes associated with IR using latent class analysis (LCA) and to develop an interpretable multiclass phenotype classification model for identifying and characterizing LCA-derived subtypes to inform precision stratified interventions. Method Participants were recruited from two independent databases: the Chengdu Health Examination Database of Wenzhou District People’s Hospital and NHANES. LCA was conducted using lifestyle and metabolic characteristics to identify IR-related lifestyle-metabolic subtypes. Seven feature selection methods and eight multiclass machine learning algorithms were applied to classify LCA-derived subtypes. The optimal model was externally validated in NHANES, and SHAP analysis was used to elucidate global and individual-level classification patterns and feature contributions. Result LCA identified five lifestyle-metabolic subtypes with significantly different IR prevalence across classes (29.89%–79.61%, p < 0.001). Compared with the reference class (Class 2), Class 4 and Class 5 were associated with higher odds of IR, with ORs of 7.724 (95% CI: 6.771–8.827) and 7.935 (95% CI: 7.100–8.877), respectively. Among the multiclass machine learning models, random forest (RF) demonstrated the best performance and showed strong capability in identifying LCA-derived subtypes. SHAP analysis revealed that carbohydrate and fat intake and BMI were key contributors to subtype classification. Dietary characteristics contributed substantially to the identification of Class 1, Class 3, and Class 5, whereas metabolic characteristics showed greater contributions to the classification of Class 4 and Class 5. Conclusion This study identified five IR-related subtypes characterized by distinct lifestyle-metabolic profiles across two independent databases, highlighting the heterogeneity of IR phenotypes. An interpretable machine learning approach effectively classified LCA-derived subtypes and elucidated key contributors underlying subtype classification. These findings provide insights into lifestyle-metabolic patterns associated with IR and support the development of precision-oriented intervention strategies.

Authors

Institutions

Publication Details

Journal
Frontiers in Nutrition
Published
2026-11-18
DOI
https://doi.org/10.3389/fnut.2026.1972783
Primary Topic
Diabetes, Cardiovascular Risks, and Lipoproteins
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Lifestyle-metabolic subtypes of insulin resistance: a cross-center latent class and interpretable machine learning study

Zhao Xia He, Ru Meng Yao, Ning Ma, Hui Zhen Cai et al.
2 links
Frontiers in Nutrition
Diabetes, Cardiovascular Risks, and Lipoproteins
article

Lifestyle-metabolic subtypes of insulin resistance: a cross-center latent class and interpretable machine learning study

Zhao Xia He, Ru Meng Yao, Ning Ma, Hui Zhen Cai, Hong Xia He, Yi Qian Wang, Kang Chen
article en

Abstract

Background Insulin resistance (IR) is a highly heterogeneous metabolic state jointly influenced by lifestyle behaviors and metabolic factors. However, population-based approaches integrating multidimensional phenotypic information for identifying IR-related phenotypes and conducting individualized risk stratification remain limited. Objective To identify lifestyle-metabolic phenotypes associated with IR using latent class analysis (LCA) and to develop an interpretable multiclass phenotype classification model for identifying and characterizing LCA-derived subtypes to inform precision stratified interventions. Method Participants were recruited from two independent databases: the Chengdu Health Examination Database of Wenzhou District People’s Hospital and NHANES. LCA was conducted using lifestyle and metabolic characteristics to identify IR-related lifestyle-metabolic subtypes. Seven feature selection methods and eight multiclass machine learning algorithms were applied to classify LCA-derived subtypes. The optimal model was externally validated in NHANES, and SHAP analysis was used to elucidate global and individual-level classification patterns and feature contributions. Result LCA identified five lifestyle-metabolic subtypes with significantly different IR prevalence across classes (29.89%–79.61%, p < 0.001). Compared with the reference class (Class 2), Class 4 and Class 5 were associated with higher odds of IR, with ORs of 7.724 (95% CI: 6.771–8.827) and 7.935 (95% CI: 7.100–8.877), respectively. Among the multiclass machine learning models, random forest (RF) demonstrated the best performance and showed strong capability in identifying LCA-derived subtypes. SHAP analysis revealed that carbohydrate and fat intake and BMI were key contributors to subtype classification. Dietary characteristics contributed substantially to the identification of Class 1, Class 3, and Class 5, whereas metabolic characteristics showed greater contributions to the classification of Class 4 and Class 5. Conclusion This study identified five IR-related subtypes characterized by distinct lifestyle-metabolic profiles across two independent databases, highlighting the heterogeneity of IR phenotypes. An interpretable machine learning approach effectively classified LCA-derived subtypes and elucidated key contributors underlying subtype classification. These findings provide insights into lifestyle-metabolic patterns associated with IR and support the development of precision-oriented intervention strategies.

Frontiers in NutritionVol. 13
Ningxia Medical University (CN)
Good health and well-being
Diabetes, Cardiovascular Risks, and Lipoproteins
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon