摘要

Background: lmmunoglobulin A nephropathy (IgAN), a dominant glomerulonephritis in China, has presented challenges in its early non-invasive diagnosis and accordingly has drawn considerable attention regarding the need to develop effective easy-to-conduct methods. Methods: In this retrospective study, a support vector machine-based classifier was trained to obtain a minimum subset with the highest discerning power between IgAN and non-IgAN cases in China based on 36 biochemical indicators connected with a feature-selection procedure. Results: Our analyses indicated 19 biochemical indicators with differential distributions between IgAN and non-IgAN cases, indicating their potential as classifiers. Further examination for the discerning power of all k-feature combinations indicated a 5-feature combination, ALB + CK + Cr + HDL + CA125 + TB, which gave the best accuracy, 79.71%, in classifying all training data into the 2 subtypes of nephropathy. Moreover, two combinations, ALB + CK + AFP + AST and TP + Glu + DB + CH, were gender-specific, giving the best classification accuracies of 81.90% and 80.22% for male and female patients, respectively. These 3 classifiers achieved classification accuracies of 75.36%, 72.00% and 84.09% in the entire, the male and the female independently validated datasets, respectively. Conclusions: Blood biochemical indicators could distinguish between IgAN and non-IgAN cases with a bioinformatic algorithm, providing a promising method to diagnose the subtypes of nephropathy.

全文