Background/Objectives: Autoimmune hepatitis (AIH) is a chronic immune-mediated inflammatory liver disease that, if not diagnosed and treated promptly, leads to cirrhosis and liver failure. Data on AIH in Central Asia, including Kazakhstan, remain limited. The aim of this study was to characterize the clinical profile of AIH in a Kazakhstani patient cohort, determine the timeliness of diagnosis, and develop an interpretable machine learning model for detecting liver fibrosis based on routine clinical and laboratory parameters. Methods: A retrospective observational study of adult patients with a diagnosis of AIH between 2015 and 2025 was conducted. Demographic, laboratory, instrumental, and histological data of patients with AIH were extracted from medical records. All statistical analyses were performed using SPSS 22.0. Results: The study included 240 patients with a mean age of 49.3 ± 14.3 years; 87.1% of patients were women. The Random Forest model showed the best results: ROC-AUC of 0.803 ± 0.057, PR-AUC of 0.868 ± 0.044, Brier of 0.180 ± 0.017, sensitivity of 0.816, and specificity of 0.641. SHAP analysis confirmed that platelet count, age, INR, disease duration, and bilirubin and albumin levels made the greatest contribution to the prognosis. Conclusions: This retrospective observational study of AIH in Kazakhstan identified a patient population characterized by late diagnosis and advanced disease stages at presentation, a high frequency of overlapping autoimmune liver diseases, and a significant burden of metabolic and extrahepatic autoimmune comorbidities. The results demonstrate that an interpretable machine learning model based on routine biomarkers can effectively detect fibrosis and provide clinically interpretable risk factors.
Ashimova et al. (Wed,) studied this question.