#2756 Metabolomic prediction of end stage renal disease in large national biobanks

Abstract

Abstract Background and Aims Identifying at-risk individuals likely to progress to end-stage renal disease as early as possible is important to effectively concentrate preventative interventions. Existing predictive tools, like the kidney failure risk score, are effective for individuals who already have moderate to severe kidney disease. Here we examine whether comprehensive metabolomics can improve performance in individuals with mildly impaired kidney function. Method Using an NMR metabolomics platform we measured >200 biomarkers including creatinine, albumin, lipids, fatty acids, amino acids, glycolysis metabolites and inflammation markers in blood samples from all 500,000 participants of the UK Biobank, who have >10 years of linked prospective follow-up data. We used machine learning to develop a score to predict kidney disease progression based on biomarkers, age and sex. We compared its performance to eGFR in a held out set of individuals in UK Biobank, and replicated in 200,000 participants of the Estonian Biobank. Results Among individuals with prevalent diabetes, hypertension or eGFR < 90 the biomarker score performs as well as or better than eGFR for all individuals. In particular it outperforms among individuals with only moderate (eGFR 60-75) or mild (eGFR 75-90) impairment of kidney function with an AUC increase from 0.63 to 0.72 and 0.48 to 0.70, respectively (Figure). This improvement is driven by contributions from numerous biomarkers, including lipids and inflammation markers. Furthermore, because these biomarkers capture diverse metabolic risk, the biomarker score also strongly predicts major adverse cardiovascular outcomes. Conclusion This large dataset allows detailed investigation of the value of metabolomics in identifying future risk of severe kidney disease early. In particular these biomarkers can highlight individuals who do not already have very low eGFR.