A diagnostic model of autoimmune hepatitis in unknown liver injury based on noninvasive clinical data.

All the diagnostic criteria of autoimmune hepatitis (AIH) include histopathology. However, some patients may delay getting this examination due to concerns about the risks of liver biopsy. Therefore, we aimed to develop a predictive model of AIH diagnostic that does not require a liver biopsy. We collected demographic, blood, and liver histological data of unknown liver injury patients. First, we conducted a retrospective cohort study in two independent adult cohorts. In the training cohort (n = 127), we used logistic regression to develop a nomogram according to the Akaike information criterion. Second, we validated the model in a separate cohort (n = 125) using the receiver operating characteristic curve, decision curve analysis, and calibration plot to externally evaluate the performance of this model. We calculated the optimal cutoff value of diagnosis using Youden's index and presented the sensitivity, specificity, and accuracy to evaluate the model in the validation cohort compared with the 2008 International Autoimmune Hepatitis Group simplified scoring system. In the training cohort, we developed a model to predict the risk of AIH using four risk factors-The percentage of gamma globulin, fibrinogen, age, and AIH-related autoantibodies. In the validation cohort, the areas under the curve for the validation cohort were 0.796. The calibration plot suggested that the model had an acceptable accuracy (p > 0.05). The decision curve analysis suggested that the model had great clinical utility if the value of probability was 0.45. Based on the cutoff value, the model had a sensitivity of 68.75%, a specificity of 76.62%, and an accuracy of 73.60% in the validation cohort. While we diagnosed the validated population by using the 2008 diagnostic criteria, the sensitivity of prediction results was 77.77%, the specificity was 89.61% and the accuracy was 83.20%. Our new model can predict AIH without a liver biopsy. It is an objective, simple and reliable method that can effectively be applied in the clinic.