Innovative adaptive dose simulations in NONMEM to allow simulation-based model evaluation of titration-based dosing, applied to a population dose-hemoglobin model for Jesduvroq (daprodustat)
Sebastiaan C. Goulooze (1), Martijn van Noort (1), Paul van den Berg (1), Shuying Yang (2), Misba Beerahee (2), Kelly M. Mahar (3), Teun M. Post (1)
(1) Leiden Experts on Advanced Pharmacokinetics and Pharmacodynamics (LAP&P), The Netherlands, (2) Clinical Pharmacology Modelling and Simulation, GSK, London, UK, (3) Clinical Pharmacology Modeling & Simulation, GSK, Collegeville, PA, USA
Objectives: Jesduvroq (daprodustat) is a hypoxia-inducible factor prolyl hydroxylase inhibitor (HIF-PHI). It is approved for the treatment of anemia in adult patients with chronic kidney disease (CKD) in Japan. In the United States, it has been approved for the treatment of anemia in adult patients who have been dialysis dependent for at least four months. In the five global phase 3 studies, the daprodustat dose is individualized according to frequent monitoring of hemoglobin (Hgb) using an adaptive-dose algorithm to reach target levels of Hgb. To allow a correct interpretation of the simulation-based evaluation of a population dose-Hgb model for daprodustat [1,2], the study-specific adaptive-dose algorithm was incorporated in simulations within NONMEM.
Methods: The development of the dose-Hgb model using NONMEM 7.5.0 is described in a PAGE 2023 companion poster [2]. As a comparison, standard visual predictive check (VPC) and prediction-corrected VPC were attempted without the use of adaptive dosing. The adaptive-dose algorithm was implemented in the simulation code. This algorithm determined the dose change on the two most recent Hgb observations (and the change in Hgb between those observations) and the current dose level at each point in time during the study, based on a total of 11 rules. The code $ABBREVIATED COMRES=5 was used to keep track of the following variables that are required to determine the dose change:
- simulated Hgb observation for the two most recent visits,
- dose level at the most recent visit,
- visit number of the most recent visit, and
- dose level prior to temporary treatment discontinuation (only for patients that had their treatment discontinued before the most recent visit).
The individual starting dose was retained as this depended on prior erythropoiesis-stimulating agent (ESA) and Hgb level at inclusion. Dropout was modeled in part, in the sense that simulated patients were discontinued when they qualified for rescue treatment due to low Hgb levels. Other potential dropout reasons were not included.
Results: The use of standard VPC and prediction-corrected VPC resulted in an inadequate representation of the variability of Hgb levels due to a mismatch between the individual model parameters of simulated patients and their corresponding daprodustat dose levels over time. While the prediction-corrected VPC has been proposed to also be appropriate in the presence of dose-titration [3], it likely suffers in this scenario from its lack of an adaptation of the dose in its simulations for subjects with low Hgb despite being on the highest dose level. When incorporating the adaptive-dose algorithm in the simulations, the VPC showed good agreement between the simulated and observed data. Because the titration-based VPC does not require a correction of the simulated data, the simulated data can also be used as input to evaluate other aspects of the model (such as the predicted % of patients reaching the target Hgb levels).
Conclusions: A complex study-specific adaptive-dose algorithm was implemented into a simulation workflow in NONMEM, which allowed adequate model evaluation of a daprodustat dose-Hgb model using a VPC and avoided misinterpretation of the model performance. Implementation of such an algorithm is required to accurately apply a simulation-based approach for model evaluation (e.g., VPCs). Furthermore, this algorithm included into NONMEM could possibly be used for clinical trial simulations after inclusion of a more complete dropout model.
References:
[1] Kristensen NR et al. CPT Pharmacometrics Syst Pharmacol 2022;11:1592–603.
[2] van Noort et al. campanion poster at PAGE 2023 and previously presented at ACoP 2022. https://www.go-acop.org/default.asp?abstract=443.
[3] Bergstrand M et al. Prediction-Corrected Visual Predictive Checks for Diagnosing Nonlinear Mixed-Effects Models, AAPS J. 2011 Jun; 13(2): 143–151.