Keywords: chronic kidney disease; progression; machine learning; artificial intelligence, artificial neural network

Aim:

The progression of chronic kidney disease towards its final stages represents a major health problem, which is why the design of new prognostic models is a priority. The aim is estimate the progression of chronic kidney disease based on artificial intelligence algorithms.

Method:

A retrospective, longitudinal, analytical cohort study was conducted in 1,117 patients with chronic kidney disease from January 1, 2022 to December 31, 2022. For the design of the prognostic models, a training cohort of 467 patients was used, while the validation cohort consisted of 650 patients. Five artificial intelligence algorithms were designed: decision tree, support vector machine, Naïve bayes, random forest and artificial neural network.

Results:

Proteinuria, cardiovascular disease, albumin, age, dyslipidemia, creatinine, malnutrition, hemoglobin, diabetes mellitus and uric acid were identified as prognostic variables for progression. The five algorithms showed good indicators of statically efficiency. The artificial neural network was the best prognostic model with overall correct classification percentages of 97.8 % in the training sample and 94.7 % in the validation sample. The discriminatory capacity was very good, 0,982 ROC area.

Conclusions:

The designed artificial intelligence prognostic models contributed to the risk stratification of chronic kidney disease progression. The variables included are easy to determine and interpret, so these models are useful for implementation in daily clinical practice.