Peripheral blood routine parameters-based MLP-ANN model discrimination of ulcerative colitis from Crohn's disease and prediction of disease activity

Abstract

AbstractObjective Peripheral blood routine parameters (PBRPs) are promising markers to reveal inflammatory response and immune function of patients with inflammatory bowel disease (IBD). Therefore, this study aimed to discriminate ulcerative colitis (UC) and Crohn's disease (CD), and predict the disease activity using a PBRPs-based multilayer perceptron artificial neural network (MLP-ANN) model. Methods An MLP-ANN model was established using 18 PBRPs from 146 CD patients, 88 UC patients and 505 healthy controls. The performance for UC and CD discrimination and prediction were evaluated using the area under the receiver operating characteristic curve (AUC). Results The lymphocyte to monocyte ratio (LMR) is the most useful candidate marker in 18 PBRPs for screening IBD patients [AUC = 0.815, 95% confidence interval (CI): 0.780–0.851, sensitivity 79.4%, specificity 73.5%]. The MLP-ANN model based on five optimal PBRPs exhibited well performance for UC and CD prediction (AUC = 0.971, 95% CI: 0.928–1, sensitivity 97.8%, specificity 98.6%, accuracy 97.5%). Besides, the MLP-ANN model exhibited superior performance on correctly predicting active and remissive UC patients (AUC = 0.979, 95% CI: 0.943–1.00, sensitivity 100%, specificity 85.3%, accuracy 95.5%) based on four optimal PBRPs, as well as active and remissive CD patients (AUC = 0.832, 95% CI: 0.737–0.927, sensitivity 78.0%, specificity 78.3%, accuracy 80.8%) based on five optimal PBRPs. Conclusion The PBRPs-based MLP-ANN model provides a simple, rapid and reliable tool for discriminating UC and CD and predicting the activity of both UC and CD patients.