Modeling hepatitis D virus kinetics during bulevirtide monotherapy: challenges and solutions

Technical Summary: Modeling Hepatitis D Virus Kinetics During Bulevirtide Monotherapy

Problem Statement
The entry inhibitor Bulevirtide (BLV) was recently approved in Europe for chronic hepatitis D virus (HDV) infection, the most severe form of viral hepatitis. While mathematical modeling is essential for refining treatment strategies, a recently proposed "two-equation" model by El Massoudi et al. (2024) was found to have structural limitations when applied to broader clinical cohorts. This model, which accounts for free virus and infected cells but assumes a fixed number of target cells, is theoretically limited to predicting monophasic viral declines. However, clinical data from patients treated with BLV monotherapy often exhibit complex kinetic patterns, including biphasic declines, flat-partial responses, and viral breakthrough (a subsequent increase in viral load after initial decline). Furthermore, the existing model fails to predict the viral rebound observed in clinical studies after treatment cessation. A critical concern is that relying on this structurally limited model, even when parameters appear "precisely estimated" by standard statistical metrics, leads to incorrect predictions regarding the duration of treatment required to reach a theoretical cure (defined as <1 virion in the body).

Methodology
The authors re-evaluated the El Massoudi et al. model using a new cohort of 38 patients with HDV-related advanced compensated cirrhosis treated with BLV monotherapy (2 mg/day) for up to 96 weeks. All patients received concomitant therapy for hepatitis B virus.

• Data Collection: HDV RNA and alanine aminotransferase (ALT) levels were measured at baseline and at regular intervals up to week 96.
• Modeling Approach: The study utilized Non-Linear Mixed Effects (NLME) modeling implemented in Monolix. The authors fitted the El Massoudi model (Eq. 1) to the data, estimating population parameters (fixed effects) and inter-individual variability (random effects).
• Evaluation Metrics: Parameter precision was assessed using Relative Standard Error (RSE), with the standard industry threshold of RSE < 50% used to define "precise" estimates.
• Diagnostic Analysis: Beyond parameter estimates, the authors conducted rigorous diagnostic checks, including visual predictive checks (VPC), observation-vs-prediction plots, and analysis of individual weighted residuals (IWRES) to assess model fit quality.
• Proposed Solution: To address identified structural limitations, the authors introduced an extended model (Eq. 2) that replaces the fixed target cell assumption with a dynamic equation for target cell production and loss.

Key Results

1. Limitations of RSE-Based Evaluation: In the 38-patient cohort, all population parameter estimates for the El Massoudi model had RSE < 50%, suggesting high precision. However, diagnostic plots revealed that the model failed to reproduce non-monophasic kinetic patterns (biphasic, flat-partial, and breakthrough) observed in the majority of patients. This demonstrates that meeting the RSE < 50% threshold does not guarantee a model is structurally adequate to capture biological reality.
2. Failure to Predict Clinical Outcomes:
   • Time to Cure: Because the model forces a monophasic decline, it significantly underestimates the time required to reach the cure boundary for patients exhibiting flat-partial responses or biphasic declines with a slower second phase.
   • Viral Breakthrough: The model cannot explain viral breakthrough kinetics, where viral load increases after an initial decline, as it structurally precludes such behavior.
   • Post-Treatment Dynamics: The El Massoudi model predicts that viral load remains steady after BLV cessation. In contrast, clinical data and the authors' simulations show a distinct viral rebound upon stopping treatment.
3. Success of the Extended Model: By incorporating target cell dynamics (allowing the number of susceptible cells to change over time), the extended model successfully captured all observed kinetic patterns: monophasic decline, biphasic decline, flat-partial response, and viral breakthrough. Furthermore, the extended model correctly predicted the viral rebound post-treatment cessation, with rebound levels returning to near pre-treatment baselines regardless of BLV efficacy.

Significance and Claims
The paper claims that while the El Massoudi model parameters may be statistically "precise" (low RSE), the model is structurally insufficient for reliable clinical prediction in the context of BLV monotherapy. The authors emphasize that relying on such a model leads to misleading clinical interpretations, particularly regarding the estimated duration of therapy needed for a cure.

The primary contribution of this work is the demonstration that target cell dynamics are essential for accurately modeling HDV kinetics under BLV therapy. The proposed extended model resolves the structural limitations of previous approaches, enabling the accurate prediction of complex viral decline patterns and post-treatment rebound. This refinement is presented as a necessary step for improving the reliability of mathematical models used to guide treatment duration and evaluate the potential for a functional cure in chronic HDV infection.