Скрининг хронической болезни почек у детей с помощью алгоритмов машинного обучения.

Abstract

Late diagnosis of chronic kidney disease (CKD) in children is common. Among the main reasons: lack of awareness among parents and medical personnel, nonspecific symptoms, and difficulties in carrying out diagnostic procedures in children. This leads to serious health consequences for children, including progression of the disease and the need for long-term dialysis therapy or kidney transplantation. P u r p o s e : to identify signs and symptoms in children that have a non-linear impact on CKD using a decision tree (DT) algorithm. M a t e r i a l s a n d m e t h o d s of the study: data were obtained from a single-center prospective cohort study (2011–2022) involving 128 children with CKD stages 1–4 and 30 children in the control group aged 0 to 18 years. An analysis of the anamnesis, hereditary factors, the early period of the child’s development, the results of clinical, paraclinical and genetic examination was carried out. The model was built using a machine learning (ML) algorithm using the decision tree (DR) method. R e s u l t s . The decision tree model identified three variables that jointly influence CKD: protein loss, red blood cells in the urine, and the T598T polymorphic marker of the IL4 gene. The model predicts CKD on the training set with an accuracy of 98.9% [97.3; 100.0]%, sensitivity 97.8% [95.1; 100.0]%, specificity 100.0% [100.0; 100.0]%, ROC-AUC = 100.0% [99.9; 100.0]%.; describes 95.7% [89.1; 100.0]% variance. The resulting regression model is of excellent quality (>90%), because ROC-AUC is 0.98 on the test sample. During the study, the value of the cut-off point (cut-off) of the VLP was determined, which is equal to 0.5. C o n c l u s i o n s . Biomarkers have been identified that will help primary care physicians identify CKD in children at early stages of development. These variables can be easily examined in outpatient and primary care settings. This information may help raise awareness of the diagnosis. Healthcare providers can form groups of patients for more detailed examination, which will reduce the likelihood of wasted time and improve early detection of diseases.