The article is devoted to the review of existing studies on the use of artificial intelligence and machine learning tools in the field of dialysis. The search databases were Elibrary, PubMed, Scopus. The selection of metainformation was performed by the fields "TITLE" and "ABSTRACT" for the queries: artificial intelligence AND (dialys* OR hemodialys*); machine learning AND (dialys* OR hemodialys*). Queries were built under the condition of the absence of restrictions on the word forms of key concepts. The publication period was not limited. Since there were many full-text articles found (669), manual selection was carried out independently by three experts: two nephrologists and 1 specialist in the field of AI. The principles of article selection by experts: the presence of AI algorithms, solving a specific problem in dialysis, the algorithm was developed specifically for dialysis patients, and is not an accompanying one when solving another problem. An analysis of the sources conducted showed that one of the most common problems in dialysis solved by AI tools is the problem of predicting the adequacy of the dialysis treatment program. As a rule, in these problems, the adequacy of dialysis is understood as the adequacy of the removal of nitrogen metabolism products, but an indirect sign of the lack of adequacy can also be the presence of complications after the procedures, the risk of which is what artificial intelligence algorithms are trained to identify. AI algorithms are also used to select therapy for the treatment of renal anemia and for the restoration of phosphorus-calcium metabolism, since here it is necessary to consider too many different factors, including those tracked in dynamics. Quite many studies using AI are used to predict the survival of patients receiving dialysis. Also, artificial intelligence algorithms are used to solve the problem of directly predicting the period when a patient with chronic kidney disease should start renal replacement therapy. A critical analysis of the work revealed a number of problems: small sample size, insufficient test control both during the analysis of results and at the preliminary stage, and the irreproducibility of many studies.
Лакман et al. (Fri,) studied this question.