Data-Driven Multidimensional Clinical Phenotypes and Longitudinal Changes in Type 2 Diabetes Mellitus: A Retrospective Cohort Study
- Design
- Retrospective cohort
- Drugs
- Class unspecified
- Assessment
- Not yet reviewed by a person · added Sep 27, 2026
This study was added automatically and has not been reviewed. Until it is, the collection publishes no population match, weight-loss explanation, study tier, findings, population or funding details for it: a check of automatically extracted details found too many errors to show them as fact. The abstract below is the source as retrieved.
01Source
The source, as retrieved
Abstract
Background: Type 2 diabetes mellitus (T2DM) is clinically heterogeneous, and routinely available renal, inflammatory, and hepatic measures may add information beyond glycemic assessment. Methods: Among 811 screened adults, 809 had a standardized baseline assessment and six-month follow-up; two were excluded because a required baseline clustering variable was missing, leaving 807 complete cases for phenotype derivation. Twelve-month data were available for 320 participants. Nine baseline variables were standardized, with UACR and CRP analyzed after ln(x + 1) transformation. Candidate two- to six-cluster solutions were assessed using silhouette, Calinski-Harabasz, and Davies-Bouldin indices, 200 bootstrap resamples, and a consistent-transformation sensitivity analysis. Longitudinal outcomes were evaluated using generalized estimating equations including time, phenotype, and their interaction. Results: The final stored four-cluster solution comprised a comparatively favorable-profile phenotype (n = 294), a cardiorenal-fibrotic phenotype (n = 239), an inflammatory-metabolic phenotype (n = 141), and a severe hyperglycemic-obesity phenotype (n = 133). Internal separation was modest (silhouette 0.118; Calinski-Harabasz 87.6; Davies-Bouldin 2.118), bootstrap agreement was moderate (median adjusted Rand index 0.445), and the consistent ln(x + 1) sensitivity solution showed substantial agreement with the stored assignment (adjusted Rand index 0.776). Significant phenotype-by-time interactions were observed for HbA1c, BMI, TyG, triglycerides, UACR, eGFR, FIB-4, and CRP after false-discovery-rate correction, but not for LDL cholesterol. Baseline characteristics did not differ detectably between participants with and without twelve-month follow-up after correction. Marked treatment changes occurred by T2, particularly increased use of GLP-1 receptor agonists and SGLT2 inhibitors. Conclusions: The four phenotypes provide an exploratory multidimensional description of this cohort. Longitudinal differences should not be attributed to phenotype biology alone because regression to the mean and treatment intensification are plausible contributors; external validation remains necessary.