Woo Jin Jung (1)*, Seon Jong Park (1)*, Sun ae Yu (2)*, Thi Quyen Tran (1), Thi Lien Ngo (1), Sung Yoon Yang (1), Chung Hee Lee (1), Jung-Woo Chae (1), Hwi-yeol Yun (1) *Both of authors were contributed equally for this work as first-presenter
(1) College of Pharmacy, Chungnam National University, Korea, (2) Ministry of Food and Drug Safety, Korea
Objectives: As an anti-epileptic drug, phenobarbital is widely used upon infant/neonates. Already there are many population pharmacokinetic (popPK) models of phenobarbital in pediatrics had been published, yet their predictivity hasn’t tested externally. In this study, every model was assessed on external dataset which it hasn’t been fitted with. Simulations using literature-based parameter and prior-data-based parameter were done. Over the cases, model parameter estimation and simulation results will be compared so that the number prior data for reasonable Bayesian forecasting can be assessed.
Methods: By literature searching, models of phenobarbital for neonates were gathered. For external validation of those models, independent dataset of 28 neonate patients had gathered. As a measurement of the model’s external prediction performance, PE% and IPE%(individual prediction error) plus, NDPE(normalized prediction distribution error) method were used. Utilizing software NONMEM® 7.4.1, models’ parameters were estimated. Default setting of significant numbers in the model file was corrected considering the fitted dataset’s circumstances. Allometric factors and scaling factors also applied into some of the parameters to make selected models adapt to the dataset. On plotting outputs, R 3.6.2, Rstudio 1.2.5033 were used. To compare the errors between the models, plotting was done in box plots. As an error model, if it is not stated in the paper, combined error model was chosen for initial approach.
Results: Total seven models (Grasela[3], Marsot[4], Moffett[5], Shellhaas[6], Voller[7], Vucicevic[8], Yukawa[9] et al.) were selected. Each model’s prediction performance was plotted. Bayesian forecasting gotten better as the number of the prior data had increased. With one point of prior data, most of the model showed increased PE% compared to the original model simulation. At least ‘two or three prior data points’ is found out to be needed for preserved predictive performance.
Conclusions: Structural model, co-administrated drugs and allometric factors were thought to be important on predictive performance since most of the neonate patients varies widely with demographics. Though all the model couldn’t show full prediction performance since the external dataset have limited covariates that can be utilized, many of the models shown reasonable prediction. On comparing between prior data conditions, decreased prediction performance gotten improved when introducing more prior data to them.
References:
[1] K. A. Barraclough et al., “Evaluation of limited sampling methods for estimation of tacrolimus exposure in adult kidney transplant recipients,” Br. J. Clin. Pharmacol., vol. 71, no. 2, pp. 207–223, 2011, doi: 10.1111/j.1365-2125.2010.03815.x.
[2] J. Geweke and C. Whiteman, “Chapter 1 Bayesian Forecasting,” Handb. Econ. Forecast., vol. 1, no. 05, pp. 3–80, 2006, doi: 10.1016/S1574-0706(05)01001-3.
[3] T. H. Grasela and S. M. Donn, “Neonatal population pharmacokinetics of phenobarbital derived from routine clinical data,” Dev. Pharmacol. Ther., vol. 8, no. 6, pp. 374–383, 1985, doi: 10.1159/000457062.
[4] A. Marsot, V. Brevaut-Malaty, R. Vialet, A. Boulamery, B. Bruguerolle, and N. Simon, “Pharmacokinetics and absolute bioavailability of phenobarbital in neonates and young infants, a population pharmacokinetic modelling approach,” Fundam. Clin. Pharmacol., vol. 28, no. 4, pp. 465–471, 2014, doi: 10.1111/fcp.12042.
[5] B. S. Moffett et al., “Phenobarbital population pharmacokinetics across the pediatric age spectrum,” Epilepsia, vol. 59, no. 7, pp. 1327–1333, 2018, doi: 10.1111/epi.14447.
[6] R. A. Shellhaas, C. M. Ng, C. H. Dillon, J. D. E. Barks, and V. Bhatt-Mehta, “Population pharmacokinetics of phenobarbital in infants with neonatal encephalopathy treated with therapeutic hypothermia,” Pediatr. Crit. Care Med., vol. 14, no. 2, pp. 194–202, 2013, doi: 10.1097/PCC.0b013e31825bbbc2.
[7] S. Völler et al., “Model-based clinical dose optimization for phenobarbital in neonates: An illustration of the importance of data sharing and external validation,” Eur. J. Pharm. Sci., vol. 109, no. May, pp. S90–S97, 2017, doi: 10.1016/j.ejps.2017.05.026.
[8] K. Vučićević et al., “Nonlinear mixed effects modelling approach in investigating phenobarbital pharmacokinetic interactions in epileptic patients,” Eur. J. Clin. Pharmacol., vol. 71, no. 2, pp. 183–190, 2015, doi: 10.1007/s00228-014-1778-7.
[9] M. Yukawa et al., “Population pharmacokinetics of phenobarbital by mixed effect modelling using routine clinical pharmacokinetic data in Japanese neonates and infants: An update,” J. Clin. Pharm. Ther., vol. 36, no. 6, pp. 704–710, 2011, doi: 10.1111/j.1365-2710.2010.01220.x.
Reference: PAGE () Abstr 9413 [www.page-meeting.org/?abstract=9413]
Poster: Methodology - Model Evaluation