186-OR: Risk Prediction of Early Decline in Renal Function in Diabetic Kidney Disease with Algorithm Including Fractional Excretion of Glycated Amino Acids Free

Diabetic kidney disease occurs in ca. 40% patients with diabetes. Approximately 1 in 5 patients with type 1 diabetes mellitus (T1DM) develop early decline in renal function (EDRF), requiring renal dialysis after 5 - 20 years. Currently, it is uncertain which patients are at high risk of EDRF. With Joslin Kidney Study investigators, we found patients with T1DM with and without microalbuminuria who later develop EDRF (Decliners) have higher fractional excretion (FE) of 6 glycated amino acids - fructosyl-lysine and 5 advanced glycation endproducts (AGEs), compared to patients with stable renal function (Non-decliners) (Perkins et al., Sci Rep 10, 12709, 2020). No individual FE could classify patients. In this study, we applied artificial intelligence machine learning to develop diagnostic algorithms to classify Decliners and Non-decliners, exploring A1C, log(ACR), 14 glycated and oxidized amino acids in plasma, urine, related FEs and other features. Data from patients with T1DM with stable renal function (n = 52) and EDRF (n = 33) were available. Algorithms were trained and tested on 90%/10% data split, repeated 1000 times, using Extreme Gradient Boosting. Optimum classification required measurement of 3 AGEs: N_ꞷ-carboxymethylarginine (CMA) and glyoxal-derived hydroimidazolone (G-H1) in plasma and urine, and N-carboxymethyl-lysine (CML) in plasma; measured in 15 min by liquid chromatography-tandem mass spectrometry. Optimum algorithm features were: A1C, log[ACR], FE_CMA, FE_G-H1 and plasma CML free adduct. The diagnostic performance for risk prediction of future EDRF was (mean ± SD): sensitivity 0.74 ± 0.09, specificity 0.91 ± 0.04 and accuracy 87 ± 4%. The positive likelihood ratio LR+ of 11.0 indicates that this method gives strong, often conclusive evidence of future EDRF in patients with T1DM. With further validation, including in patients with type 2 diabetes mellitus, this method may markedly improve risk prediction of EDRF.