Am J Nephrol. 2026 Oct 20:1. doi: 10.1159/ajn/ablag032. Online ahead of print.
ABSTRACT
Background Diabetic kidney disease (DKD) is the leading cause of CKD globally. Despite marked heterogeneity in progression, risk stratification remains anchored to single-timepoint measures that fail to capture individual patient trajectories. We aimed to identify clinically meaningful DKD subgroups from real-world longitudinal data and characterise their molecular signatures using integrative multi-omics. Methods In this retrospective cohort study, unsupervised clustering of longitudinal creatinine and haematocrit trajectories was performed in the Salford Kidney Study (SKS; n=335 eligible for clustering, n=303 DKD) and validated in two independent cohorts: Scottish Health Research Register (SHARE; n=824) and CERCA-Diabetes (Spain; n=187). Prognostic performance was assessed by C-statistic, time-dependent area under the curve (AUC), IDI (integrated discrimination improvement), and decision curve analysis relative to the Kidney Failure Risk Equation (KFRE) and Kidney Disease Improving Global Outcomes (KDIGO) risk categorisation. Multi-omic characterisation (proteomics, metabolomics, DNA methylation) was performed in an 87-patient SKS subset using Data Integration Analysis for Biomarker Discovery using Latent Components (DIABLO). Results Two groups with markedly different progression rates were identified: high-risk (SKS-HR; n=173) and low-risk (SKS-LR; n=162). SKS-HR showed eGFR decline nearly four times faster and a higher rate of the composite kidney failure endpoint (HR 4.05 [95% CI 2.77-5.92] adjusted for age, sex, and eGFR; 2.95 [1.90-4.56] after further adjustment for haemoglobin, proteinuria, HbA1c, and cardiovascular burden; both p<0.0001). The signal was specific to renal progression (fully adjusted HR 5.98 [3.32-10.79], p<0.0001) and absent for cardiovascular or renal death (HR 1.00 [0.58-1.71], p=0.984). The subgroup-outcome association was independently replicated in SHARE (p=3.83×10⁻²⁷) and CERCA-Diabetes (p=1.72×10⁻⁴). The subgrouping added prognostic information complementary to the KFRE, raising the C-statistic from 0.689 to 0.721 when added to static KFRE and 0.711 to 0.732 when added to longitudinal KFRE. KDIGO risk categories showed limited discriminatory capacity in this predominantly G3b-G4 cohort. Hypothesis-generating multi-omic profiling showed FDR-significant enrichment for lipid transport pathways and for an independent DKD progression proteomic signature; all multi-omic findings require independent validation. Conclusions Longitudinal creatinine and haematocrit trajectories, available from routine clinical records, identify DKD patients at high risk of kidney failure with prognostic information beyond current risk tools. The high-risk subgroup warrants prospective validation and evaluation of whether subgroup assignment can improve risk stratification beyond existing clinical tools.
PMID:42900647 | DOI:10.1159/ajn/ablag032