Population Pharmacokinetic Analysis of Digoxin in Chinese Neonates and Infants

Transcription

1 J Pharmacol Sci 125, (2014) Journal of Pharmacological Sciences The Japanese Pharmacological Society Full Paper Population Pharmacokinetic Analysis of Digoxin in Chinese Neonates and Infants Yan Gong 1, Yewei Chen 1, Qin Li 1, and Zhiping Li 1, * 1 Children s Hospital of Fudan University, Department of Pharmacy, No. 399, Wanyuan Road, Minhang District, Shanghai , China Received November 14, 2013; Accepted March 9, 2014 Abstract. To obtain more information regarding the influence of various covariates on the disposition of digoxin in Chinese neonates and infants, routine clinical pharmacokinetic data were retrospectively collected from 131 hospitalized patients. A nonlinear mixed effects modeling (NONMEM) method was applied to the data. A one-compartment/first-order absorption model was employed to estimate the influence of total body weight (allometric power model), postnatal age, serum creatinine, gender, presence of heart congestive failure, and concomitant medications on apparent total clearance and apparent drug distribution of digoxin. Pharmacokinetic parameter estimates for CL/F and V/F were L h 1 kg 1 and 15.7 L/kg, respectively. Total body weight and postnatal age were identified as the important factors affecting total clearance of digoxin; total body weight was the covariate identified to influence the apparent distribution volume. Both internal (bootstrap method, visual predictive checks, and normalized prediction distributed error) and external validation supported the stable and predictive performance of the final model. We concluded that the model can be used to choose an appropriate dose regimen in Chinese neonates and infants. Keywords: digoxin, neonate, nonlinear mixed effects modeling, allometric scaling, pharmacokinetics Introduction Digoxin is a cardiac glycoside, which has been widely prescribed for the treatment of congestive heart failure, atrial fibrillation, and arrhythmias of supraventricular origin. With its narrow therapeutic range [ ng/ml in adults (1)], large pharmacokinetic variability and side effect profile, routine therapeutic drug monitoring is essential for digoxin in clinical use. Historically, digoxin dosages commonly used in neonates and infants have been empirically calculated on a mg/kg or mg/m 2 basis, extrapolated from adult dosing. Neonates and young infants are not little adults ; their postnatal anatomical, physiological, and psychological function continue to develop during the first year of life (2). Various pharmacokinetic models for digoxin have been *Corresponding author. Zhipinglifudan@163.com Published online in J-STAGE doi: /jphs.13233FP developed in Chinese patients (3, 4), but only in adults. We therefore retrospectively collected data and used them to develop a population model for Chinese neonates and infants. Previous population pharmacokinetics studies in pediatric patients (non Chinese) demonstrated that the clearance of digoxin was influenced by the following demographic variables: total body weight, postnatal age, serum creatinine, presence of congestive heart failure, gender, and other concomitant medications (5 14). All the candidate variables were investigated in this study. Materials and Methods Data resources According to the Ethical Guidelines for Clinical Research, the study was approved by the Research Ethics Committee of Children s Hospital of Fudan University (Ethical code: [2013]96). The registration number in the Chinese Clinical Trial Registry is: ChiCTR-RCH

2 2 Y Gong et al ( In the model-developing group, routine clinical pharmacokinetic data (125 observations) were retrospectively collected from 107 patients in Children s Hospital of Fudan University ( ). The patients were all given the drug in the form of an elixir (0.005%, 30 ml/ bottle; Beijing Double-Crane high SCI. & TECH. Natural Pharmaceutical Co., Ltd.) twice a day. The doses were calculated on a ml/kg basis. We included the hospitalized patients (age < 1 year) who were administered digoxin orally for more than 6 days and had their blood drawn at least 6 h after the drug administration. Patients who suffered serious hepatic or renal dysfunction were excluded. All the patients were hospitalized and treated under the supervision of medical and nursing staff. Digoxin concentration measurement was determined by the micro-particle enzyme immunoassay (MEIA), using the AxSYM Digoxin III assay method (Abbott Laboratories, Abbott Park, IL, USA). The lowest measurable concentration using this method was 0.3 ng/ml. Both the intra- and inter-assay coefficients of variation were less than 10%. The data collected were dose regimen of digoxin, date and time of blood sampling, steady drug concentrations, postnatal age (PNA), total body weight (TBW), gender, serum creatinine (SCR), presence of congestive heart failure (CHF), and concomitant medications (diuretics and angiotensin converting enzyme inhibitor). We reconfirmed the information before model developing. Population pharmacokinetics analysis was performed on NONMEM software (version VII, level 1.0; Icon Development Solutions, Ellicott City, MD, USA). The internal validation method of visual predicted check and normalized prediction distributed error were carried out on the add-on package of R (2.10.1). Model development A one-compartment and first-order absorption model was employed to build the digoxin population pharmacokinetic model. The ADVAN2 TRANS2 subroutine and the first-order conditional estimation (FOCE) method with h e interaction were used throughout the process of model building. In the process of model development, candidate covariates were added to the structure model to identify its significance (P < 0.05, 1 degree of freedom). Continuous covariates (TBW, PNA, SCR) were implemented using an allometric model, using the following equation (15): P i = P p *(Cov/Cov Median ) k, where P i represents the individual parameter estimate of the ith subject, P p represents the population parameter estimates, Cov is the covariate, and k is the exponent. Size (TBW) can be incorporated into the model using the allometric size approach, by modifying the above equation as follows (15): P i = P std *(W i /70) PWR, where P std represents the individual parameter estimate with a body weight of 70 kg, W i represents the body weight of the ith subject, and PWR is an exponent with 0.75 for clearance, and 1 for distribution of volume. After the stage of forward selection, covariates need to be re-evaluated by improving the test level (P < 0.01, freedom degree 1) in the stage of backward elimination. In order to avoid the effect of fluctuation on the stability of the model, the value of Ka was fixed at the stage of stacking the covariates. The residual error was evaluated with additive, proportional, and combined residual error models, respectively. Model evaluation Internal model evaluation: A non-parametric bootstrap method (Wings for NONMEM), a visual predictive check (VPC), and a normalized prediction distribution error (NPDE) analysis were carried out after the model building. The bootstrap method involves repeated random re-sampling with replacement from the original dataset for 1000 times. The values of estimated parameters from the bootstrap procedure were compared with those estimated from the original dataset. No significant differences between the data were required for the evaluation. The VPC used Monte Carlo simulation to generate concentration-time profiles of 1000 patients. Then the observed concentration-time data were graphically superimposed on the median values and the 5th and 95th percentiles of the simulated concentration time profiles. The model was deemed precise if the observed concentration data were approximately distributed within the 5th to 95th prediction interval. NPDE analysis yields information on how accurate the model predicts the median value of observations and their variability. An NPDE add-on package for the open source statistical package R, designed to compute NPDE, could make statistical tests for the 1000 simulated datasets and generate visual evaluation scatter plots. The diagrams showed no trends or bias in the simulation datasets, which indicated a good predictive performance of the final model. External model evaluation: An additional 24 patients (information given in Table 1) were enrolled for the external evaluation of the final model. Estimates of the model were fixed during the process and drug concentrations were predicted by the calculation of the final model. The predictive performance of the model was evaluated

3 PPK Analysis of Digoxin in Chinese Children 3 Table 1. Summary of patient data Characteristic Original group Validation group Number of patients (male/female) 107 (66/41) 24 (16/8) Number of observations (male/female) 125 (77/49) 24 (16/8) Postnatal age (d) ± 98.8* ± Total body weight (kg) 5.5 ± ± 0.4 Serum creatinine (mmol/l) 26.7 ± ± 3.5 CHF (+) Spironolactone (+) Enalapril (+) 39 2 Digoxin dose (mg kg 1 d 1 ) 7.5 ± ± 2.1 Drug concentrations (ng/ml) 1.1 ± ± 0.2 *Mean value ± standard deviation. Table 2. Significant fixed effect factors Model OFV DOFV P < 0.05 Incorporate one covariate individually: TBW-CL a Yes b Yes PNA-CL Yes SCR-CL Yes TBW-V Yes PNA-V Yes Incorporate two covariates: TBW-PNA-CL a Yes b No TBW-SCR-CL a No b No TBW-CL TBW-V a Yes b Yes TBW-CL PNA-V a Yes b No TBW-CL SCR-V a No b No OFV, the objective function value (the minimum value of model in each successful run); DOFV, the difference between OFVs of the two models; a, the OFVs of the models incorporating body weight using allometric size model approach; b, the OFVs of the models incorporating body weight using allometric approach. by mean prediction error (MPE) and mean absolute prediction error (MAE), which were calculated by the following formula: 1 n MPE = ( C C ), (n = 1, 2,, i) n i= 1 n i= 1 predi obsi 1 n MAE = ( C C ), (n = 1, 2,, i), predi obsi, where C predi represents the predicted concentration of the ith subject and C obsi represents the observed concentration for the ith subject. Results Model development The preliminary pharmacokinetic analysis indicated that the proportional model was suitable for the description of residual error in the study. The OFV of the one compartment structure model was The parameter estimates of CL/F, V/F, and Ka were 1.77 L/h, 29.2 L, and h 1, respectively. When adding the candidate covariates separately to the structure model, the model of CL/F was markedly affected by TBW, PNA, and SCR; the model of V/F was markedly affected by TBW and

4 4 Y Gong et al Table 3. Summary of the NONMEM analyses Equation OFV Basic model CL/F (L/h) = q 1 V/F (L) = q 2 Ka = q 3 q 1 = 1.77 q 2 = 29.2 Ka = q 3 = (fix) Model CL/F (L/h) = q 1*(TBW/70) 0.75 V/F (L) = q 2 q 1 = 10.3 q 2 = 26.8 Model CL/F (L/h) = q 1*(TBW/70) 0.75 *(PNA/125) q4 V/F (L) = q 2 q 1 = 11.2 q 2 = 37.8 q 4 = Model CL/F (L/h) = q 1*(TBW/70) 0.75 *(PNA/125) q4 *(SCR/18) q5 V/F (L) = q 2 q 1 = 10.9 q 2 = 39.2 q 4 = q 5 = Model CL/F (L/h) = q 1*(TBW/70) 0.75 *(PNA/125) q4 V/F (L) = q 2*(TBW/70) q 1 = 10.4 q 2 = 1100 q 4 = Model CL/F (L/h) = q 1*(TBW/70) 0.75 *(PNA/125) q4 *(SCR/18) q5 V/F (L) = q 2*(TBW/70) q 1 = 10.4 q 2 = 1100 q 4 = q 5 = *Ka = was fixed throughout the stage of model developing. PNA (Table 2). TBW was chosen as the first significant covariate adding to the clearance. We have tried different approaches in incorporating body weight to the model and found that the allometric size model is more suitable for this study (Table 2). After forward selection and backward elimination of all the covariates (Table 3), the final model presented as follows: CL/F(L*h 1 *70 kg 1 ) = 10.4*(TBW/70) 0.75 *(PNA/125) ; V/F(L/70 kg) = 1100*(TBW/70). Diagnostic plots for the structure model and the final model are shown in Fig. 1 and Fig. 2, respectively. The final model provided a better fit for the prediction of steady drug concentrations rather than the structure model. Model evaluation Internal evaluation: The estimated covariates of the final model and 1000 bootstrap replicates for digoxin are summarized in Table 4. In this study, 929 re-samplings of the original dataset were successfully performed, which indicates a qualified stability for the final model. The VPC of digoxin is given in Fig. 3. During this procedure, most model-predicted concentrations were within the 95% CI ( ng/ml, calculated by R) of the model-predicted concentrations. Only 17 modelpredicted concentrations were outside of the limits. This result supports the precise performance of the final model. The NPDE scatter plots (Fig. 4) show low trends or bias in the simulation datasets, which meet the requirements of the evaluation. External evaluation: A plot of predicted concentrations vs. observed concentrations is shown in Fig. 5. The difference between the predicted concentrations and observation data ranged from 20.7% to 32%. MPE and MAE were ± ng/ml and ± ng/ml, respectively. The results of the external model evaluation indicate a good prediction and low bias of the final model.

5 PPK Analysis of Digoxin in Chinese Children 5 Fig. 1. Diagnostic scatter plots of the structure model. a: observed concentrations (DV) vs. individual predicted concentrations (IPRED). b: observed concentrations (DV) vs. population predicted concentrations (PRED). c: conditional weighted residuals (CWRES) vs. time. d: conditional weighed residuals (CWRES) vs. population predicted concentrations (PRED). Fig. 2. Diagnostic scatter plots of the final model. a: observed concentrations (DV) vs. individual predicted concentrations (IPRED). b: observed concentrations (DV) vs. population predicted concentrations (PRED). c: conditional weighted residuals (CWRES) vs. time. d: conditional weighed residuals (CWRES) vs. population predicted concentrations (PRED). Table 4. Parameters Parameter estimates of the final model and bootstrap results Final model Bootstrap n = 1,000 Estimate RSE (%) Median 95% CI Clearance: CL/F (L/h) = q 1*(TBW/70) 0.75 *(PNA/125) q3 Volume of distribution: V/F (L) = q 2*(TBW/70) CL/F (L h 1 70 kg 1 ) V/F (L 70 kg 1 ) q Inter-individual variability (%) CL/F V/F Residual error model (%) Proportional RSE, relative standard error; CI: confidence interval. Discussion To our knowledge, this is the first population pharmacokinetic study of digoxin concentrations in the neonates and infants in China. In this study, the pharmacokinetic parameter estimates of digoxin for neonates and infants are L h 1 kg 1 (10.4 L h 1 70 kg 1 ) for CL/F and 15.7 L/kg (1100 L/70 g) for V/F. These estimates are considered to be reasonable according to the range summarized by Wells et al. (16) in 1992 ( L h 1 kg 1 for CL/F, L/kg for V/F). Compared to the clearance for the Japanese (0.298 L h 1 kg 1 ),

6 6 Y Gong et al Spanish (0.237 L h 1 kg 1 ), and Thai (0.322 L h 1 kg 1 ) pediatric patients within the same age range, the clearance in our study is much less. It may reflect a lower elimination in Chinese neonates and infants. Fig. 3. Visual predictive check obtained from 1000 simulations of the database. Dotted lines indicate the nonparametric 95% confidence interval of the simulations, whereas the points indicate the observed concentrations. Fig. 5. Plot of predicted concentrations vs. observed concentrations in validation group. Fig. 4. Goodness of fit evaluated by normalized prediction distributed error (NPDE). a: the plots of theoretical quantiles vs. sample quantiles; b: the plots of frequency vs. NPDE; c: the plots of NPDE vs. time (hour); d: the plots of NPDE vs. predicted concentrations.

7 PPK Analysis of Digoxin in Chinese Children 7 During the selection of covariates, serum creatinine showed a statistically significant effect on CL/F when it was tested individually. However, with the incorporation of other covariates, total body weight and postnatal age, were identified as important factors affecting total clearance of digoxin (CL/F). Similarly, total body weight was the covariate identified to influence the apparent distribution volume (V/F). The presence of congestive heart failure and potential drug interactions did not significantly affect pharmacokinetic parameters for digoxin in Chinese neonates and infants (age < 1 year), and development is the leading factor affecting the disposition of digoxin. These findings are different from the pediatric population pharmacokinetic studies that have been conducted on digoxin in Japan, Spain, and Thailand (5 14), It can be determined that body weight was incorporated into the model using different methods in these 4 investigations. Body weight was linearly added to the Thai patients model, exponentially included in the Japanese model, and allometrically incorporated into our model. That can be the reason of our results. Growth and development are two major factors that affect the disposition of digoxin in young children. However, these two covariates share a degree of correlation and are not mutually exclusive. Allometric scaling is a mechanistic approach that has a strong theoretical and empirical basis for this. West et al. (18, 19) explained that the 3/4 power law was derived from a general model that described how essential materials are transported through space-filled fractal networks of branching tubes. Fixing the power exponent of body weight allowed investigators to delineate secondary covariate effects from the effect of size. In addition, the allometric scaling also allowed the direct comparison of pediatric estimates with adults when a weight standard of 70 kg was used (20). Pediatricians in our hospital have been cautioned about the drug interaction between digoxin and calciumchannel blockers (CCB). Several studies (21 23) have reported that the two-drug combination can lead to the increase of steady-state plasma digoxin concentrations and the decrease of apparent clearance. As a result, the two drugs were not co-administered in clinical practice in our hospital. It has reported (24) that erythromycin had no effect on the transport of digoxin, Azithromycin also had little effect on the transport of digoxin or any of its metabolites. The variable of CCB and macrolide antibiotics have not been incorporated into the covariates selection of the study. On the other hand, the drug concentration was measured using the AxSYM Digoxin III assay method. Although this method demonstrated little digoxin-like immune-reactive factors (DLIS) interference, the influence of DLIS cannot be absolutely guaranteed. DLIS can lead to the overestimation of the serum digoxin concentrations, with a resulting underestimate of apparent clearance. A more precise and accurate approach (e.g., LC/MS/MS) should be established to improve the measurement of drug concentrations in further investigations. Acknowledgment Prof. David Knoppert (Canada) assisted us with the use of English grammar. We gratefully acknowledge his kindness. Conflicts of Interest The authors indicated no potential conflicts of interest. References 1 Cauffield JS, Gums JG, Grauer K. The serum digoxin concentration: ten questions to ask. Am Fam Physician. 1997;56: Wood AJ, Kearns GL, Abdel-Rahman SM, Blowey DL, Leeder JS, Kauffman RE. Developmental pharmacology drug disposition, action, and therapy in infants and children. N Engl J Med. 2003;349: Chen R, Zou S, Wang M, Jiang Y, Xue H, Qian CY, Xia ZL. Population pharmacokinetics of digoxin in elderly patients. Eur J Drug Metab Ph. 2012: Zhou X, Gao Y, Guan Z, Li Z, Li J. Population pharmacokinetic model of digoxin in older Chinese patients and its application in clinical practice. Acta Pharmacol Sin. 2010;31: Yukawa E, Honda T, Ohdo S, Higuchi S, Aoyama T. Populationbased investigation of relative clearance of digoxin in Japanese patients by multiple trough screen analysis: an update. J Clin Pharmacol. 1997;37: Suematsu F, Minemoto M, Yukawa E, Higuchi S. Population analysis for the optimization of digoxin treatment in Japanese paediatric patients. J Clin Pharm Ther. 1999;24: Suematsu F, Yukawa E, Yukawa M, Minemoto M, Ohdo S, Higuchi S, et al. Population-based investigation of relative clearance of digoxin in Japanese neonates and infants by multipletrough screen analysis. Eur J Clin Pharmacol. 2001;57: Yukawa E, Suematu F, Yukawa M, Minemoto M, Ohdo S, Higuchi S, et al. Population pharmacokinetics of digoxin in Japanese patients. Clin Pharmacokinet. 2001;40: Yukawa E, Akiyama K, Suematsu F, Yukawa M, Minemoto M. Population pharmacokinetic investigation of digoxin in Japanese neonates. J Clin Pharm Ther. 2007;32: Yukawa M, Yukawa E, Suematsu F, Takiguchi T, Ikeda H, Aki H, et al. Population pharmacokinetic investigation of digoxin in Japanese infants and young children. J Clin Pharmacol. 2011;51: Petcharattana S. Population pharmacokinetics of digoxin in Thai pediatric patients. J Med Assoc Thai. 2009;92: Martin-Suarez A, Falcao AC, Outeda M, Hernández FJ, González MC, Quero M, et al. Population pharmacokinetics of digoxin in pediatric patients. Ther Drug Monit. 2002;24: Nagaraja NV, Derendorf H. Population pharmacokinetics of digoxin in Egyptian pediatric patients: impact of one data point

8 8 Y Gong et al utilization. Am J Ther. 2001;9: Desoky ESE, Madabushi R, Amry SEDA, Bhattaram VA, Derendorf H. Application of two-point assay of digoxin serum concentration in studying population pharmacokinetics in egyptian pediatric patients with heart failure: does it make sense? Am J Ther. 2005;12: Li Z, Chen Y, Li Q, Cao D, Shi W, Cao Y, et al. Population pharmacokinetics of piperacillin/tazobactam in neonates and young infants. Eur J Clin Pharmacol. 2013: Wells TG, Young RA, Kearns GL. Age-related differences in digoxin toxicity and its treatment. Drug Safety. 1992;7: West GB, Brown JH, Enquist BJ. A general model for the origin of allometric scaling laws in biology. Science. 1997;276: West GB, Brown JH, Enquist BJ. The fourth dimension of life: fractal geometry and allometric scaling of organisms. Science. 1999;284: Anderson BJ, Allegaert K, Holford NH. Population clinical pharmacology of children: modelling covariate effects. Eur J Pediatr. 2006;165: Bauer LA, Horn JR, Pettit H. Mixed-effect modeling for detection and evaluation of drug interactions: digoxin-quinidine and digoxin-verapamil combinations. Ther Drug Monit. 1996;18: Rameis H, Magometschnigg D, Ganzinger U. The diltiazemdigoxin interaction. Clin Pharmacol Ther. 1984;36: Belz GG, Aust PE, Munkes R. Digoxin plasma concentrations and nifedipine. Lancet. 1981;317: Hughes J, Crowe A. Inhibition of P-glycoprotein mediated efflux of digoxin and its metabolites by macrolide antibiotics. J Pharmacol Sci. 2010;113: