Caixia Mei, Yang Li, Hailin Ma, Xueling Li, Sheng Jiang
In this single-center, short-term retrospective study, PCA-derived metabolic phenotype dimensions, particularly the blood pressure axis (PC1), were associated with differential eGFR trajectories and albuminuria status. The attenuation of the PC1-eGFR association after UACR adjustment generates the hypothesis that blood pressure-associated renal function decline may involve albuminuric pathways. Temporal validation confirmed PCA loading structure stability, though predictive model generalization was limited. These findings are hypothesis-generating given the observational design, limited UACR availability, short followup, and absence of independent external validation. Prospective, multi-center studies with comprehensive phenotyping are required before clinical translation can be considered.
BACKGROUND: Type 2 diabetes (T2D) exhibits substantial phenotypic heterogeneity, yet the relationship between multidimensional metabolic phenotypes and longitudinal renal function decline remains poorly characterized, particularly regarding albuminuria status.
OBJECTIVE: To identify metabolic phenotype dimensions using principal component analysis (PCA) in a Chinese T2D cohort and evaluate their associations with estimated glomerular filtration rate (eGFR) trajectories, rapid eGFR decline, cardiometabolic comorbidities, and KDIGO 2024 risk categories incorporating urinary albumin-to-creatinine ratio (UACR).
METHODS: In this retrospective cohort study of 1,427 patients from Dazhou Tongchuan District People's Hospital (January 2024-July 2026), baseline metabolic variables were subjected to PCA with Varimax rotation (N = 621 with ≥6 variables). Multiple imputation by chained equations (MICE, 20 imputations) addressed missing data. UACR was calculated from simultaneously measured urine creatinine and microalbumin (374 observations). Longitudinal eGFR trajectories were analyzed using linear mixed models (LMM; N = 162, 415 observations), with UACR-adjusted sensitivity analysis (N = 103). Rapid eGFR decline (annual slope < -3 mL/min/1.73 m2/year) was predicted using logistic regression with bootstrap internal validation (1,000 resamples). Temporal validation was performed by comparing PCA structures derived from pre- versus post-intervention cohorts. The 3-component solution was compared against a 4-component alternative using predictive performance and loading stability. Subgroup analyses assessed the consistency of the months × PC1 interaction across age, sex, baseline eGFR, glycemic control, and sodium-glucose cotransporter-2 inhibitor (SGLT2i) use strata.
RESULTS: Three principal components explained 53.1% of total variance: PC1 (Blood Pressure Axis, 20.2%), PC2 (Metabolic Syndrome Axis, HDL-/TG+, 17.3%), and PC3 (Renal Function Axis, eGFR+/UA-, 15.6%). Parallel analysis suggested 4 components; the 4-component model improved predictive discrimination (AUC: 0.633 vs. 0.699; LRT P = 0.001), but PC4 was dominated by HbA1c and LDL-C without distinct physiological interpretation, and the 3component solution was retained on grounds of parsimony and interpretability. Temporal validation demonstrated excellent stability of PCA loading structures (Pre vs. Post: PC1 r = 0.961, PC2 r = 0.928, PC3 r = -0.976). Hopkins statistic (0.994) confirmed continuous data structure; MICE intraclass correlation coefficients ranged from 0.617 to 0.719. Compared with excluded patients, included patients had higher HbA1c (9.1% vs. 8.3%, P < 0.001), SBP (132 vs. 126 mmHg, P < 0.001), and longer diabetes duration (62 vs. 41 months, P < 0.001). In covariate-adjusted LMM, only PC1 was associated with differential eGFR trajectories (months × PC1: β = -0.135 per month, P = 0.028; annualized: -1.62 mL/min/1.73 m2/year per 1-SD PC1; approximately -3.55 mL/min/1.73 m2/year between extreme tertiles). This association was substantially attenuated after UACR adjustment (P = 0.792). Subgroup analyses showed a consistent direction of the months × PC1 interaction across all examined strata. PC1 was associated with macroalbuminuria (OR 1.76, P = 0.007) and PC3 was inversely associated with albuminuria (OR 0.65, P = 0.014). For rapid eGFR decline prediction (65/119, 54.6%), addition of PCA dimensions yielded a modest, non-significant improvement (AUC: 0.562 → 0.636; bootstrap-corrected: 0.674, 95% CI 0.588-0.762; LRT P = 0.315). In the UACR-available subcohort (N = 35, 18 events; exploratory only), AUC increased from 0.837 to 0.925; however, given the extremely small sample and quasi-complete separation, this value is almost certainly overestimated and should not be cited as evidence of predictive utility.
CONCLUSIONS: In this single-center, short-term retrospective study, PCA-derived metabolic phenotype dimensions, particularly the blood pressure axis (PC1), were associated with differential eGFR trajectories and albuminuria status. The attenuation of the PC1-eGFR association after UACR adjustment generates the hypothesis that blood pressure-associated renal function decline may involve albuminuric pathways. Temporal validation confirmed PCA loading structure stability, though predictive model generalization was limited. These findings are hypothesis-generating given the observational design, limited UACR availability, short followup, and absence of independent external validation. Prospective, multi-center studies with comprehensive phenotyping are required before clinical translation can be considered.