Development and validation of a diagnostic model for prevalent liver-related events in patients with non-alcoholic fatty liver disease

The burden of liver-related events (LREs) among hospitalized patients with non-alcoholic fatty liver disease (NAFLD) is substantial. We developed a diagnostic model based on routine clinical parameters to estimate the probability of prevalent LREs status ascertained during the index hospitalization stay and evaluated its discriminative performance against conventional non-invasive scoring systems. This retrospective study enrolled hospitalized patients with nonalcoholic fatty liver disease (NAFLD) treated at a tertiary A-level hospital in Fujian between July 2008 and March 2025. The patients were randomly assigned in a 6:4 ratio to a training set and a validation set. Candidate factors associated with prevalent LREs were identified using the least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression. A nomogram and web-based algorithm incorporating the four identified factors were constructed. Internal validation was performed using bootstrapping and leave-one-out cross-validation, and external validation was performed on the MIMIC-IV database. A total of 1,013 patients were included, with 608 in the training set and 405 in the validation set. The overall prevalence of liver-related events (LREs), the primary outcome, was 15.7% (159/1,013), with comparable rates between the derivation (16.3%; 99/608) and validation (14.8%; 60/405) cohorts ( p = 0.529). Four independently associated factors were identified: age, diabetes mellitus, prothrombin time, and platelet count. The nomogram demonstrated superior discriminative capacity for identifying patients with LREs (training AUC, 0.873; internal validation AUC, 0.916), outperforming the APRI, FIB-4, and ALBI. External validation confirmed the acceptable discriminative performance (AUC = 0.749). Net reclassification index (NRI) and integrated discrimination improvement (IDI) analyses indicated improved reclassification of predicted probabilities for in-hospital LREs (all p < 0.05). We developed a diagnostic model for NAFLD-related LREs using routinely available baseline characteristics. The nomogram exhibited good discriminative performance for identifying patients with concurrent LREs and demonstrated overall acceptable transportability across cohorts, performing better than conventional scoring systems.