TyG-Centered Endocrine–Metabolic Architecture in Patients with Thyroid Dysfunction: A Systems Phenotyping Study

Background: Metabolic dysfunction is a complex process arising from coordinated interactions among insulin resistance, dyslipidemia, hepatic dysfunction, obesity, low-grade inflammation, and endocrine alterations. The triglyceride–glucose (TyG) index is a simple, reproducible surrogate of insulin resistance, but most studies have evaluated it through isolated association-based analyses. We characterized the endocrine–metabolic architecture associated with TyG in patients with thyroid dysfunction using a systems-phenotyping approach. Methods: In this retrospective, single-center, cross-sectional study of 387 adults evaluated for thyroid dysfunction, unsupervised phenotypes were derived by principal component analysis (PCA), K-means, Gaussian mixture models (GMM), and hierarchical (Ward) clustering using fully measured primary metabolic variables (fasting glucose, triglycerides, HDL- and LDL-cholesterol, body mass index, alanine, and aspartate aminotransferase). TyG was then examined as a descriptor of the identified architecture. Model-based latent profile analysis (BIC, entropy) and a censoring-aware, rank-based Gaussian graphical model (GGM) of partial correlations were additionally estimated. Results: Three overlapping, clinically interpretable phenotypic partitions of an underlying metabolic continuum were identified: a metabolically preserved phenotype (MP), an intermediate hepatic-dominant phenotype (IHD), and a severe insulin-resistance-dominant phenotype (SIRD), characterized by obesity, atherogenic dyslipidemia, and hyperuricemia. TyG values increased monotonically across the three partitions (medians 8.24, 8.74, and 8.87 in MP, IHD, and SIRD, respectively; p < 0.001), providing an accessible single-number summary of the metabolic gradient. Cluster separation was modest (silhouette = 0.17), consistent with a continuum, but the three-profile solution was supported by latent profile analysis (ΔBIC = −96.7; relative entropy = 0.75) and was reproduced across frameworks (adjusted Rand index of 0.75 with GMM and 0.46 with Ward). In the censoring-aware GGM, metabolic and thyroid variables formed two near-orthogonal blocks (maximum absolute metabolic–thyroid partial correlation: 0.16; median: 0.05). Conclusions: The systems phenotyping approach identified three clinically interpretable phenotypic partitions (MP, IHD, SIRD) along a metabolic continuum in patients evaluated for thyroid dysfunction. TyG values increased monotonically across partitions (medians 8.24, 8.74, 8.87) and provided a pragmatic single-number summary of the principal metabolic axis. The IHD phenotype indicates that hepatic dysregulation can emerge as a distinct early stage rather than solely a late consequence of insulin resistance. Metabolic and thyroid axes were near-orthogonal, positioning thyroid autoimmunity as a parallel modulatory layer. These findings support TyG as an accessible descriptor for multidimensional endocrine–metabolic risk stratification.