Dynamic prediction of liver cirrhosis risk in chronic hepatitis B patients using longitudinal clinical data

Objectives: In longitudinal studies, serum biomarkers are often measured longitudinally which is valuable to predict the risk of disease progression. Previous risk prediction models for liver cirrhosis restrict data to baseline or baseline and a single follow-up time point, which failed to incorporate the time-dependent marker information. The aim of this study is to develop risk model in patients with chronic hepatitis B for dynamic prediction of cirrhosis by incorporating longitudinal clinical data. Methods: Data from the hospital-based retrospective cohort at the Third Affiliated Hospital of Sun Yat-sen University, from 2004 to 2016, were analyzed. Using the multilevel logistic regression model, the time-dependent marker information and individual characteristics were taken as input, and the risk of at different time as the output. Results: At the end of follow-up, 8.8% of patients progressed to cirrhosis, the average estimate values of hepatitis B virus DNA and alanine aminotransferase demonstrated a downward trend, the aspartate aminotransferase/alanine aminotransferase ratio showed a flat trend overall. The important predictors were as follows: age, oral antiviral treatment, hepatitis B virus DNA. This risk prediction model had an area under the receiver operator characteristic curve of 0.835 (95% confidence interval: 0.772–0.899) and 0.809 (95% confidence interval: 0.708–0.910) in the derivation and validation sets, respectively. Conclusion: Longitudinal prediction model can be used for dynamic prediction of disease progression and identify changing high-risk patients.