Development of an adaptive dosing approach for doxorubicin in paediatric cancer patients
C. Siebel (1), G. Würthwein (1), C. Lanvers-Kaminsky (1), G. Hempel (2), J. Boos (1)
(1) Department of Paediatric Haematology and Oncology, University Children’s Hospital of Muenster, Muenster, Germany; (2) Department of Pharmaceutical and Medical Chemistry - Clinical Pharmacy, University of Muenster, Muenster, Germany
Objectives: Anthracyclines, such as doxorubicin, are known for causing potentially irreversible cardiotoxicity. Paediatric cancer patients are at the highest risk to suffer from long-term cardiac side effects due to the generally long life expectancy of childhood cancer survivors. The reduction of variability in systemic therapy intensity (drug exposure and peak concentrations) holds promise to better balance tumour efficacy and the risk of cardiac toxicity. Dosing algorithms that reflect individual differences in pharmacokinetics are therefore needed. A priori dose adaptations that take into account relevant covariates might offer a possibility to reduce variability in particular during the first administration of the drug. A further attempt to reduce variability could be provided by adaptive drug administration based on a single or few drug concentration measurements and subsequent Bayesian estimation of individual pharmacokinetic parameters.
Methods: The impact of a priori dose adaptations based on a dosing formula derived from a published population pharmacokinetic model for doxorubicin in paediatric cancer patients was investigated [1]. The dosing formula takes into account individual body surface area and age which have been identified as predictive covariates during model building [2]. Hypothetical, dose-adjusted AUC was calculated using the proposed dosing formula and the maximum a posteriori clearance estimates from a paediatric patient population (94 children from the EPOC-MS-001-Doxo trial).The model-expected AUC of an 18-year-old boy with median demographics served as target for dose calculation. Dose-adjusted AUC values and the actual observed AUC values were normalised to the target AUC and compared with respect to bias (median prediction error), precision (median absolute prediction error) and the probability of target attainment. To assess the predictive power of the population pharmacokinetic model in a reduced sampling situation truncated datasets with 1 - 3 observations were generated based on the full dataset from the EPOC-MS-001-Doxo trial. Bayesian clearance estimates were computed in NONMEM version 7.3 (POSTHOC option with MAXEVAL = 0) [3] using the truncated datasets and the full dataset and compared. Bias, precision and the percentage of clearance values within 10 % and 20 % error range were calculated. Optimal sampling time points for 1 - 3 sample designs were identified based on Ds-optimality criteria using the optimal design software PopED [4]. Identification of optimal sampling times was performed based on data of the EPOC patient cohort as the EPOC population was considered to represent typical paediatric cancer patients.
Results: Using data from the EPOC population consideration of body surface area and age for dose calculation suggests to achieve a predefined target AUC without relevant bias (-2.5 %, 95 % CI -8 – 3 %). However, only a small decrease in precision between observed (21 %, 95 % CI 18 – 23 %) and dose-adjusted AUC values (17 %, 95 % CI 13 – 19 %) could be observed (p < 0.01, Wilcoxon signed rank test). The percentage of AUC attaining the range of 80 – 125 % around the target AUC was 58.5 % for observed and 69.1 % for dose-adjusted AUC values (p > 0.05, McNemar’s chi-squared test). Bayesian forecasting of individual clearance seems to be sufficiently accurate and precise with median absolute prediction errors not higher than 13.1 % (95 % CI 9.5 – 17.5 %) for estimation based on a single observation. Exemplarily, for a treatment regimen with a 1 h infusion and a dose of 30 mg/m² (corresponding to the AIEOP-BFM ALL 2017 protocol) optimal sampling times for estimation of clearance were 8 h (1 sample design), 4 h + 16 h (2 sample design) and 1 h + 5 h + 23 h (3 sample design) after the start of infusion.
Conclusions: A standardised dosing concept that individualises doxorubicin doses based on body surface area and age seems to be beneficial given the current situation of diverse dosing strategies in particular for very young children. However, a relatively small reduction in variability of drug exposure can be expected with a priori dose adaptations. Bayesian forecasting suggests to allow accurate and precise prediction of individual clearance. Adaptive doxorubicin dosing based on the most informative sampling strategy might therefore provide a possibility to better control variability.
References:
[1] Völler S, Hempel G, Wurthwein G, et al. Towards a Model-Based Dose Recommendation for Doxorubicin in Children. Clinical Pharmacokinetics 2016.
[2] Völler S, Boos J, Krischke M, et al. Age-Dependent Pharmacokinetics of Doxorubicin in Children with Cancer. Clinical Pharmacokinetics 2015;54:1139–49.
[3] S Beal, LB Sheiner, A Boeckmann, RJ Bauer. NONMEM Users’s Guides (1989-2009). Ellicott City, MD, USA: Icon Development Solutions, 2009.
[4] Nyberg J, Ueckert S, Strömberg EA, Hennig S, Karlsson MO, Hooker AC. PopED: an extended, parallelized, nonlinear mixed effects models optimal design tool. Comput Methods Programs Biomed 2012;108(2):789–805.
