[Constructing the Bayesian network models to explore the factors related to glomerular and tubular injury]

Abstract

Objective: To construct Bayesian network (BN) models to explore the factors related to glomerular injury (GI) and tubular injury (TI). Methods: A cross-sectional study was carried out. From April to November 2019, Shanxi Provincial People's Hospital performed an opportunistic screening for chronic kidney disease in 10 counties of Shanxi Province. The general data and laboratory results of blood and urine samples were collected. Chi-square test and logistic regression were used to explore the related factors of GI and TI, which were included in the construction of BN models with max-min hill-climbing (MMHC) algorithm. Results: A total of 12 269 participants were included, there were 5 198 males and 7 071 females, with a median age of 58 (40-91) years. The prevalence of GI and TI was 12.7% (1 561/12 269) and 11.6% (1 425/12 269), respectively. The BN model consisted of 8 nodes and 10 edges for GI, and 11 nodes and 17 edges for TI, respectively. BN models showed that age and glycated hemoglobin were direct related factors for GI, while gender and fasting blood glucose were indirect related factors for GI. Age, gender, fasting blood glucose and glycosylated hemoglobin were direct related factors for TI. Additionally, the area under the receiver operating characteristic curve (AUC) was 0.761 (95%CI: 0.746-0.777) and 0.753 (95%CI: 0.736-0.769) for GI and TI BN models, respectively. Conclusions: BN models allow for identifying the complex network relationships among the factors related to GI and TI. Meanwhile, Bayesian risk reasoning can provide reference value for the clinical prevention of GI and TI.