Systematic evaluation of plasma and urine metabolites to predict adverse kidney-related outcomes in chronic kidney disease: The GCKD study

Rationale & Objective

Accurate risk prediction of adverse kidney-related outcomes in individuals with chronic kidney disease (CKD) is essential to guide personalized clinical interventions. Metabolites measured in plasma and urine may offer additional prognostic information beyond established clinical predictors.

Study Design

Prospective German CKD cohort study.

Setting & Participants

5,217 individuals with predominantly CKD stage G3 at baseline, followed for a median of 6.5 years (IQR 6.5-6.5).

Exposure(s) or • Predictor(s)

Baseline metabolite levels measured via untargeted mass spectrometry (Metabolon Inc.) in plasma (N=5,144; 1,096 metabolites) and urine (N=5,088; 1,129 metabolites).

Outcome(s)

(i) Kidney failure (KF), defined as initiation of maintenance dialysis, kidney transplantation, or death due to untreated KF; (ii) composite kidney endpoint (CKE), defined as KF, a ≥40% decline in estimated glomerular filtration rate (eGFR), or an eGFR <15 mL/min/1.73 m².

Analytical Approach

Time-to-event data were analyzed using subdistribution hazard models to predict outcomes. A component-wise boosting algorithm was employed to select metabolites in various scenarios, including models based on plasma, urine, or combined data. Predictive performance was assessed at multiple time points using time-dependent area under the curve (AUC) and Brier scores, and compared to benchmark models.

Results

Several individual metabolites were predictive of KF or CKE and provided marginal improvements beyond established prognostic factors (age, sex, eGFR, and urinary albumin-to-creatinine ratio [UACR]). For example, plasma pseudouridine increased the AUC at year 6 by 0.012 (95% CI: 0.005–0.018) when added to the clinical model.

Multi-metabolite models developed under different scenarios included a median of 53 metabolites (range: 14-86). Some metabolites, such as plasma N2,N5-diacetylornithine and urine 1-palmitoyl-2-oleoyl-GPC (16:0/18:1), were selected more often than others. Predictive performance of all models declined over time. For KF, models achieved AUCs ≥0.89 at year 2 and ≥0.85 at year 6; for CKE, AUCs were ≥0.85 and ≥0.77, respectively. Compared to benchmark models with only clinical variables, metabolite-based models performed better but not to a clinically meaningful extent. AUC values were comparable to those reported for other published KF prediction models.

Limitations

Use of single baseline measurements and semi-quantitative nature of the metabolite measurements.

Conclusions

While certain metabolites improved prediction of adverse kidney-related outcomes, the observed gains were not clinically meaningful. Nonetheless, these findings may provide insights into metabolic pathways and processes related to the progression of CKD, thereby complementing current knowledge. Further research is warranted to refine predictive models and explore the biological relevance of identified metabolites.