Discrimination of urine infrared spectral biomarkers for early-stage chronic kidney disease patients using attenuated total reflectance fourier transform infrared spectrometry

Chronic Kidney Disease (CKD) is a highly prevalent non-communicable disorder lacking a gold standard method for diagnosis. Early-stage CKD remains undiagnosed and untreated leading to the disease progression. The study aimed to discriminate urine samples from CKD patients and healthy groups using an ATR-FTIR spectrometer. Forty-five healthy and a hundred CKD urine samples were included. Five replicates of three microliters of urine were dropped onto a crystal and air-dried as a film on a portable ATR-FTIR spectrometer. Spectra were analyzed using multivariate analysis combined with machine learning. The PCA scores plot showed the discrimination of the CKD group, both early and late stage, from healthy in the CH and fingerprint regions. For a screening approach, machine learning were introduced to the healthy and early-stage CKD samples. Prediction models were generated using six machine learning models. The neural networks (NN) model demonstrating the best performance, achieving 85 % sensitivity, 73 % specificity and 77 % accuracy on test dataset. Thus, ATR-FTIR data combined with machine learning shows potential as a medical tool with high performance for screening minimal urine volumes of CKD patients.