Performance of ChatGPT4.0 in predicting advanced liver fibrosis in patients with metabolic dysfunction-associated steatotic liver diseases

Accurate prediction of advanced liver fibrosis is essential for managing patients with metabolic dysfunction-associated steatotic liver disease (MASLD). Conventional diagnostic models like the FIB-4, APRI and NFS scores have limitations in sensitivity and specificity. We evaluated the performance of ChatGPT4.0 in predicting advanced fibrosis compared to traditional scoring systems. A total of 341 biopsy-confirmed MASLD patients were retrospectively analyzed. ChatGPT4.0’s predictions were compared to liver biopsy results. Input variables included ethnicity, MASLD diagnosis, age, sex, body mass index, diabetes status, hypertension status, albumin, SGOT, SGPT, and platelet levels. Performance metrics were calculated and compared to conventional scores for detecting advanced fibrosis. Among the patients, 26.1% had advanced fibrosis. ChatGPT4.0 achieved an accuracy of 76.25%, AUROC of 0.763, sensitivity of 65.17%, and specificity of 80.16%. It outperformed NFS, and performed comparably to FIB-4 and APRI. It showed strong predictive values with a PPV of 53.7% and NPV of 86.7%. The weighted F1 score (0.4439) demonstrated balanced performance across fibrosis stages, with superior specificity (comparison with APRI of 68.54%) aiding in reduction of false positives. In summary, ChatGPT4.0 demonstrates strong potential as a clinical decision-support tool for predicting advanced liver fibrosis in MASLD. Its performance is comparable to conventional models.