Recently, a research team led by Dr. Jeong-Hoon Lee from Seoul National University College of Medicine made significant strides in predicting the risk of liver-related outcomes (LROs) after functional cure of chronic hepatitis B (HBV). The team developed a novel machine learning model called "PLAN-B-CURE," which greatly enhances the accuracy of predicting liver-related risks, enabling more personalized monitoring strategies for patients. The study was published online in the authoritative Journal of Hepatology.
The development and application of the PLAN-B-CURE model mark an important step forward in managing chronic HBV. The model not only improves prediction accuracy but also provides a scientific foundation for long-term patient management. It is expected to become an essential tool for managing patients after functional HBV cure in the future.
Even after hepatitis B surface antigen (HBsAg) clearance, patients remain at risk for hepatocellular carcinoma (HCC) and liver function decompensation. This study aimed to develop and validate a machine learning model to predict the risk of LROs, including HCC, liver decompensation, and liver-related death, following HBsAg clearance.
The research team included 4,787 patients who achieved HBsAg clearance between 2000 and 2022 from six centers in Korea and a regional database in Chinese Hong Kong. These patients were divided into a training group (n=944), an internal validation group (n=1,102), and an external validation group (n=2,741). The team developed three machine learning-based prediction models and compared them across the groups. Ultimately, the PLAN-B-CURE model, built using a gradient boosting algorithm and seven variables (age, gender, diabetes, alcohol consumption, cirrhosis, serum albumin, and platelet count), showed the best predictive performance in the training group.
The PLAN-B-CURE model significantly outperformed previous HCC prediction models in terms of prediction accuracy. In the training group, the model’s C-index was 0.82 (compared to 0.63-0.70 for traditional models), the area under the receiver operating characteristic (ROC) curve was 0.86 (vs. 0.62-0.72 for traditional models), and the area under the precision-recall curve (PRC) was 0.53 (vs. 0.13-0.29 for traditional models), all of which were statistically significant (P<0.01). Moreover, the model demonstrated excellent calibration and was successfully validated in both the internal and external validation groups.
Based on the PLAN-B-CURE model’s risk predictions, patients were categorized into low, medium, and high-risk groups. The study found that patients in the low-risk group had a significantly lower 5-year incidence of liver-related outcomes compared to other groups, indicating the model’s effectiveness in distinguishing patients of varying risk levels.
Dr. Jeong-Hoon Lee stated that the PLAN-B-CURE model will provide a more precise tool for monitoring chronic hepatitis B patients after functional cure. By assessing individual liver-related risks, doctors can create more targeted follow-up plans, improving patients’ quality of life and prognosis. As the model continues to be promoted and applied, it is expected to benefit many more HBV patients in the future.
