JMIR Publications

ABSTRACT

Background:

Early detection of hepatocellular carcinoma (HCC) in patients with chronic hepatitis B (HBV) is challenging. There is no globally accepted, highly sensitive, specific, simple, feasible, and cost-effective clinical prediction model for early HBV-related HCC screening and diagnosis.

Objective:

This study seeks to enhance prediction by combining the ESPL1 biomarker with clinical features to create and validate an HBV-related HCC risk model.

Methods:

Patients with chronic HBV were split into a training set and an external testing set based on treatment timing. A Lasso logistic regression was applied to the training set to develop a model for early HBV-related HCC diagnosis, which was then validated using the external testing set. The model's predictive performance and clinical utility were assessed through discrimination, calibration, and decision curve analysis. A web-based calculator was created for clinical application. Using X-tile, two optimal HBV-related HCC risk cutoffs were identified, categorizing patients in the external testing set into low, medium, and high-risk groups, and a cumulative incidence curve for HBV-related HCC was created.

Results:

The study included 621 chronic HBV patients, with 373 in the training set and 248 in the test set. Multivariate logistic regression on the training set identified age, ESPL1, and log (AFP) as independent predictors of HBV-related HCC, with ORs and 95% CIs of 1.08 (1.05–1.12), 1.01 (1.00–1.01), and 2.55 (1.95–3.33). In the training set, the Hosmer-Lemeshow test yielded P = 0.596, with a model C-index of 0.922 (95% CI: 0.890–0.954) for HCC prediction and bootstrap resampling for internal validation showed a C-index of 0.923 (95% CI: 0.890–0.950). In the testing set, the model had a C-index of 0.958 (95% CI: 0.929–0.988) and a calibrated C-index of 0.958 (95% CI: 0.926–0.986). Calibration curves confirmed good calibration in both sets. The X-tile tool determined optimal HBV-related HCC risk cutoffs at 4% and 24%, classifying patients into low, medium, and high-risk groups. Their 5-year cumulative HCC incidences were 5.1%, 21.1%, and 31.3%, showing a significant difference (P < 0.001).

Conclusions:

The ESPL1-based HBV-related HCC risk model is a valuable tool for accurately and easily predicting HCC risk in chronic HBV patients. Physicians can use this user-friendly calculator for personalized risk assessments, enabling tailored treatment plans and follow-up schedules.