Validation of Clinical Response Rates Extrapolated With Dose-Response Estimation Methods in Chronic Hepatitis Delta

OBJECTIVES: Extrapolation is important in clinical research to predict long-term outcomes from short-term data when follow-up is limited. This research aims to demonstrate that dose-response extrapolation methods are more viable for predicting long-term data with a small sample size and to validate the accuracy of these predictions by comparing them with actual data.

METHODS: Individual patient-level data for bulevirtide (BLV) 2mg and 10mg treatments from the MYR301 trial was divided into training (baseline to week 48) and testing (week 48 to week 72) data sets. Several dose-response estimation methods were employed to predict the proportion of responders (complete, virologic, and ALT normalization response) at weeks 96, 108, and 144 for both BLV doses. These methods included standard (time-series analysis and expected maximum effect model [EMAX]) and non-standard (nonlinear least squares estimation [NLSE] and NLSE with Box-Cox transformation using log-logistic and Weibull distributions) techniques. The best-fit model was selected based on the Akaike Information Criterion (AIC) and visual inspection. Extrapolations from all models were compared with actual data at weeks 96, 108, and 144, using mean square error (MSE) to assess accuracy.

RESULTS: The NLSE with Box-Cox transformation using log-logistic distribution was selected as the best-fit model based on the minimum AIC and visual inspection. Comparisons of predictions from the best-fitted model with actual data at weeks 96, 108, and 144 showed the lowest MSE for all endpoints, indicating that the selected model provided the most accurate extrapolated values.

CONCLUSIONS: Non-linear regression models can be effective alternatives to conventional linear and EMAX models for estimating long-term responder rates, particularly in clinical trials with limited participants. These findings support the broader application of advanced nonlinear regression models to estimate long-term response rates, providing robust evidence for HTA submissions in demonstrating long-term treatment cost-benefit, thereby enhancing decision-making and planning for long-term patient management.