Noninvasive diagnosis of significant liver fibrosis in patients with chronic hepatitis B using nomogram and machine learning models.

This study aims to construct and validate noninvasive diagnosis models for evaluating significant liver fibrosis in patients with chronic hepatitis B (CHB). A cohort of 259 CHB patients were selected as research subjects. Through random grouping, 182 cases were included in the training set and 77 cases in the validation set. The nomogram was developed based on univariate analysis and multivariate regression analysis. Various machine learning models were employed to construct prediction models for significant liver fibrosis. The area under the ROC curve (AUC), sensitivity, specificity, NPV, PPV, and F1 score were used to evaluate the diagnostic performance. The new nomogram had excellent diagnostic efficiency (AUC 0.806, 95% CI: 0.740-0.872). Compared with other traditional noninvasive diagnostic models, the nomogram demonstrated higher AUC values and better prediction performance. Among six machine learning models, the random forest (RF) model achieved the highest AUC (AUC 0.819, 95% CI: 0.720-0.906). Finally, the importance of all variables in the RF model was ordered to illustrate the contribution of different variables, providing the clinical factors associated with the risk of significant liver fibrosis. This new nomogram may more reliably than other traditional models and the RF model demonstrated superior accuracy among six machine learning models.