Evaluation of a logistic regression model for predicting liver necroinflammation in hepatitis B e antigen-negative chronic hepatitis B patients with normal and minimally increased alanine aminotransferase levels

Abstract

Liver necroinflammation is the indicator for treating patients with chronic hepatitis B (CHB) infection. However, there is no suitable non-invasive index for diagnosing liver necroinflammation. This study aimed to create a non-invasive index to predict liver necroinflammation in patients who lack clear-cut clinical inflammation parameters. Patients who were hepatitis B e antigen (HBeAg)-negative and underwent liver histological diagnosis, had a normal or minimally increased alanine aminotransferase (ALT) level were enrolled. Liver necroinflammation was defined as histological active index ≥4. A logistic regression model (LRM) was established based on the parameters independently associated with liver necroinflammation. Of all 550 patients, 36.73% had necroinflammation. In patients with an abnormal ALT level, the rate of necroinflammation was 52.49%. The area under the curve (AUC) of the ALT level for predicting necroinflammation was 0.655 (95% confidence interval [CI], 0.609-0.702), and that of the HBV DNA level ≥2000 IU/mL combined with an abnormal ALT level was 0.618. By using the LRM, the AUC improved to 0.769 (95% CI, 0.723-0.815) with a Youden index of 0.519 and diagnostic accuracy of 75.3%. The cutoff value ≥0.7 in the LRM had a specificity of 97.4% and positive predictive value of 85.0% for predicting necroinflammation. By using the cutoff value <0.15 in the LRM, the presence of necroinflammation could be excluded with a negative predictive value of 90.8%. This study indicated that the LRM can be used to effectively diagnose liver necroinflammation in HBeAg-negative patients with CHB who have normal or minimally elevated ALT levels.

Keywords: ALT level; HBeAg negative; chronic hepatitis B; liver inflammation; logistic regression model.