#2907 “Use a machine learning model that estimates survival curves to identify patients at high risk of rapid progression of chronic kidney disease.”

Abstract Background and Aims The appropriate modification of care for patients with chronic kidney disease provides an opportunity to slow its progression. Aims: To test a machine learning strategy aimed at selecting patients at high risk of rapid progression of chronic kidney disease and to attempt a medical assessment of the validity of the results of the proposed machine learning model. Method We propose a machine learning-based computational system that extends Cox survival function estimation by clustering into similar groups based on readily available raw laboratory results from patients admitted to a hospital ward for treatment. The system requires a laptop to run already trained classifiers, e.g., neural networks and stored survival curves for patient subgroups. Our learning/testing data consisted of 13 laboratory results including: urine specific gravity, serum and urine protein concentrations, and serum creatinine concentration for 520 patients each. Data from patients (320 persons) whose creatinine concentration increased were clustered into two groups by applying the spectral method. Each cluster was then split into four subgroups according to age and sex. Results Different spectra of impairment affecting the rate of endpoint achievement were correctly identified. Also, the system correctly indicated that the group with the highest risk of renal disease progression was the population of older men with an initial mild degree of renal impairment, clustered in cluster two. Conclusion The machine learning-based system that estimates survival curves can be a reliable tool for accurately identifying patients at high risk of progression of chronic kidney disease. The system operates in accordance with current medical knowledge, therefore it can be a tool for quick preliminary analysis for nephrologists, as well as support for doctors with specialties other than nephrology, thus contributing to reducing the burden of routine nephrology consultations.