Exploration of pathological prediction of chronic kidney diseases by a novel theory of bi-directional probability

In the clinic, the pathological types of chronic kidney diseases (CKD) are considered references for choosing treatment protocols. From a statistical viewpoint, a non-invasive method to predict pathological types of CKD is a focus of our work. In the current study, following a frequency analysis of the clinical indices of 588 CKD patients in the department of nephrology, a third-grade class-A hospital, a novel theory is proposed: "bi-directional cumulative probability dichotomy". Further, two models for the prediction and differential diagnosis of CKD pathological type are established. The former indicates an occurrence probability of the pathological types, and the latter indicates an occurrence of CKD pathological type according to logistic binary regression. To verify the models, data were collected from 135 patients, and the results showed that the highest accuracy rate on membranous nephropathy (MN-100%), followed by IgA nephropathy (IgAN-83.33%) and mild lesion type (MLN-73.53%), whereas lower prediction accuracy was observed for mesangial proliferative glomerulonephritis (0%) and focal segmental sclerosis type (21.74%). The models of bi-directional probability prediction and differential diagnosis indicate a good prediction value in MN, IgAN and MLN and may be considered alternative methods for the pathological discrimination of CKD patients who are unable to undergo renal biopsy.