Population pharmacokinetics of enterally administered cisapride in young infants with gastro-oesophageal reflux disease

Abstract

Aims

To investigate the pharmacokinetics of enterally administered cisapride suspension in young infants being treated for gastro-oesophageal reflux disease.

Methods

Plasma cisapride concentrations in 49 subjects (weight: 825–5010 g; n=108 samples, median two per patient; concentration: 14.8–170 ng ml⁻¹) were fitted to a one-compartment model with first-order absorption and elimination in the NONMEM program using a logarithmic transformation of the observed and predicted concentrations. Fitting was achieved using the first order conditional estimation (FOCE) method with interaction between the interpatient and intrapatient variabilities. The interpatient variance of clearance (CL/F) and volume of distribution (V /F) and their covariance were estimated using an exponential error model. Intrapatient (residual) variance was estimated using an additive model.

Results

The clearance of cisapride was shown to be linearly related to current body weight, slope: 0.538. The typical population values of CL/F, V /F and Ka (absorption rate constant) were 0.538 l h⁻¹ kg⁻¹, 21.9 l, and 2.58 h⁻¹, respectively. The population coefficients of variation (CV%) for CL/F and V /F were 34.4% and 84.3%, respectively. The squared coefficient of correlation between random effects for CL/F and V /F was 0.45. The intrapatient variance was 0.15. V /F and K a were not influenced significantly by any patient characteristic.

Conclusions

Cisapride pharmacokinetics in infants with reflux disease were satisfactorily described by a one-compartment model. Current weight should be taken into account when calculating maintenance cisapride doses in these infants.

Introduction

Gastro-oesophageal reflux disease is a common problem in young infants with the potential for considerable morbidity, and even mortality [1–3]. It is associated not only with the classic symptoms of regurgitation, failure to thrive, and oesophagitis, but also with chronic bronchopulmonary diseases such as reactive airways disease, pneumonia and atelectasis. In the preterm infant, reflux disease has been implicated in aspiration pneumonia and bronchopulmonary dysplasia, and in the aetiology of apnoea [4–7].

Cisapride is a prokinetic agent which acts by stimulating gastrointestinal motor activity through an indirect mechanism involving the release of acetylcholine mediated by postganglionic nerve endings in the myenteric plexus of the gut [8]. It promotes normal motility patterns throughout the gut by accelerating gastric emptying and improving gastroduodenal motor function, thus it is beneficial in the treatment of disorders associated with decreased gastrointestinal motility, such as reflux disease, functional dyspepsia and gastroparesis. Cisapride lacks the antidopaminergic and central depressant effects of some other prokinetic drugs such as metoclopramide and domperidone which have limited their use in infants and older children [4, 5, 8]. Despite the increased use of cisapride [6–9] there has been little published pharmacokinetic data in adults, and none whatsoever in paediatric patients. As liver metabolism is often functionally immature in preterm infants [10, 11], its kinetics in this group may be of special significance. Appropriate pharmacokinetic data is valuable for setting efficacious doses while minimizing potential adverse effects such as problems with urinary frequency [12, 13] and possible cardiac rhythm disturbances recently described in adults [14].

Pharmacokinetic studies in neonates often are difficult to conduct and analyse. There may be very limited amounts of data per patient (e.g. 1–2 samples) because of strict limitations on the frequency and volume of blood sampled, and characteristics such as weight and physiological function may change rapidly and unpredictably over the study period. Population methods offer an attractive solution for studying pharmacokinetic response in these circumstances [15]. The aim of this study therefore was to employ a parametric population approach (NONMEM) to model the basic pharmacokinetic responses following enteral administration of a cisapride suspension to young infants, and to quantitatively determine the influence of any patient-related characteristics on its disposition.

Methods

Patients, blood sampling, and data collection

This study was approved by the ethics committees of the University of Queensland and the Royal Women’s Hospital, Brisbane. Verbal information on the reason for the study and the procedures involved was provided to the parent(s) and written parental permission was obtained for each patient before entry to the study. This study did not require any extra blood collection since surplus plasma was available following routine clinical biochemistry and haematology testing.

Data were obtained from 52 infants in a special care nursery who were being treated for reflux disease, and from whom a total of 119 plasma samples were obtained. Characteristics of the study patients are contained in Table 1. A commercially available cisapride suspension of 1 mg ml⁻¹ (Prepulsid® Janssen-Cilag, Sydney, NSW, Australia) was administered enterally at doses of 0.11–0.45 mg kg⁻¹, 15 min before feeds, every 6 h (44 infants) or 8 h (8 infants).

Table 1.

Mean (range) values for characteristics of infants enrolled in the study.

Cisapride assay

Plasma concentrations of cisapride were measured in 100 μl of plasma by a h.p.l.c. method previously developed by us [16] which used a base-stable column and fluorescence detection. The correlation coefficient for calibrations was >0.99, absolute recovery was >82%, within-day and between-day coefficients of variation, (CV%) were <10%, and inaccuracy was <6%. The lower limit of quantification was 5 ng ml⁻¹.

Pharmacokinetic analysis

Population pharmacokinetic modelling was performed in double precision using NONMEM IV, level 2.0 [17]. A preliminary graphical analysis of cisapride plasma concentration vs time after dosing showed that the one-compartment open model with first-order absorption would be adequate to describe the data. Estimates were sought for the typical population values of CL/F, V/F and Ka. Elimination half-life (t_1/2) was calculate as, t_1/2=ln (2). V /F÷(CL/F), the influence of a lag time from the absorption site (gastro-intestinal tract) was also assessed. The model was implemented using the PREDPP library model subroutine ADVAN2 and the transformation routine TRANS2. The first-order conditional estimation (FOCE) method with interaction between interindividual and intraindividual random-effects was used. Goodness-of-fit was assessed by plotting the unweighted residuals (observed minus model-predicted concentration), and by the analysis of residuals weighted by the standard deviations produced during a NONMEM run; weighted residuals have unit variance and have correlations among observations obtained within individuals removed.

Variance model

The interpatient variability was modelled as follows:

where CL/F_j and V /F_j are individual model-predicted values of clearance and volume of ditribution, respectively (corrected for oral bioavailability, F ) in the jth subject; η_CL/Fj and η_{V /Fj} are random effects. The latter were treated as single, normally distributed parameters with zero mean and symmetric covariance matrix:

where ω₁₁, ω₁₂ are interindividual variances for CL/F and V /F, respectively, and ω₁₂ (=ω₂₁) is a covariance between random effects for CL/F and V /F, respectively. The covariance (expressed as the squared correlation coefficient) was calculated by:

Residual intraindividual variability was modelled as follows:

where ln (C_ij) and ln (C_pred,_ij) are the natural logarithms of the ith value in the jth individual for observed and model-predicted concentrations, respectively. ε_ij is a normally distributed parameter with zero mean and variance σ².

The following patient characteristics were tested as potential covariates: current body weight (WT); birth weight (BWT); gestational age (GA); postnatal age (PNA); serum albumin concentration (ALB), theophylline coadministration (THE).

The effect of a covariate was considered to have improved the fit if there was decrease in the objective function value (Obj) compared with the base pharmacokinetic model (containing no covariate). Hypothesis testing for the influence of a covariate could be tested formally in a full-reduced model set by comparison of the reduction in the Obj with the critical value of χ²_{1, 0.01} >6.7 [17].

Results

Characteristics of the 52 infants enrolled in the study are shown in Table 1. Cisapride concentrations in several samples were less than 5 ng ml⁻¹ and were excluded from the analysis. Furthermore, preliminary data screening revealed the presence of apparent outliers in five samples at low cisapride concentrations (7.1–14.1 ng ml⁻¹), the inclusion of these points resulted in failure of the covariance step. Therefore, the population model was developed using the data from 49 patients (weight: 825–5010 g; number of samples: 108, median 2 per patient; concentration: 14.8–170 ng ml⁻¹). After several analyses with various intrapatient error models it was found that the logarithmic transformation of observed and the population model-predicted concentrations would guarantee homoscedasticity of the residual variance, and the additive error model was applied.

Figure 1a and b show, respectively, semilogarithmic plots of cisapride plasma concentration vs postdose sampling time, and the observed and model-predicted plasma cisapride concentrations (both normalized to a 0.2 mg kg⁻¹ dose) vs postdose blood sampling time. Table 2 contains values of the fixed-effects parameters, derived parameters (e.g. elimination half-life, absorption half-life), and the variance (random-effects) parameters of the final model. The estimated parameters of the structural and variance models were estimated with a CV% of less than 50% (calculated by expressing the asymptotic standard error as a percentage of the estimated parameter value). Estimation of the covariance (squared correlation coefficient, 0.45) was shown to be an essential part of the random-effects model; omission of this term (i.e. assumption of a diagonal variance matrix) resulted in failure of the program to produce the (covariance) matrix containing the estimated standard errors of the population parameters. The available data were not sufficient to support an interpatient variance model for Ka, most likely because there usually was only a maximum of 1 sample per subject drawn during the absorption phase. Exclusion of the interpatient variance parameter for Ka, i.e. ETA(3), had no effect on the Obj value.

Figure 1.

a) Semi-logarithmic plot of plasma cisapride concentration vs postdose sampling time. Connected data points are from the same patient; b) Semi-logarithmic plot of pairs of (○) observed (•) population model-predicted concentration (based on WT) vs postdose sampling time. Observed and population model-predicted concentrations have been normalized to a dose of 0.2 mg kg⁻¹.

Table 2.

Summary of model building.

A scatterplot of observed concentrations vs the final population model-predicted concentrations is shown in Figure 2a. The weighted residuals were mostly less than ±2 units and were distributed symmetrically about zero (Figure 2b). The only covariate which influenced cisapride pharmacokinetics was WT via the expression, CL/F (l h⁻¹)=0.538. WT(g)/1000 (model 2, Table 2). Implementation of a full slope-intercept intercept relationship (model 13, Table 2) indicated that while the addition of an intercept term (θ₁) was statistically significant (P<0.01), there was no greater reduction in the Obj than obtained using model 2. Furthermore, the intercept value was extremely small (4.16×10⁻⁸) and was estimated with very high uncertainty (CV, 205%). The posterior Bayesian estimates of CL/F (obtained by the POSTHOC step in NONMEM) vs WT are shown in Figure 3. Summary population parameter values are shown in Table 3.

Figure 2.

a) Plot of observed concentration vs population model-predicted concentration (based on WT). The line of identity (perfect agreement, slope=1) is indicated; b) Plot of weighted residual value vs concentration predicted by the final model. The zero ordinate of perfect agreement is indicated (solid line).

Figure 3.

Plot of NONMEM-generated posterior Bayesian estimates of CL/F from the final population model, vs WT. Data points representing estimates within individual patients are joined.

Table 3.

Population pharmacokinetic parameters of cisapride.

Discussion

This is the first time that the pharmacokinetics of cisapride have been reported in paediatric patients. Such information is desirable when attempting to develop appropriate dosage regimens and treatment strategies in preterm infants susceptible to reflux disorders. The range of the observed plasma concentrations was similar to that previously reported in adults [9, 18]. A multicompartment kinetic model was employed in one of these studies following intravenous administration [9], however, in the current study the simpler one-compartment model satisfactorily described cisapride kinetics, both in terms of the higher precision associated with parameter estimation and the fact that rounding errors forced premature termination of some runs using a two-compartment model. As expected in a population analysis with a relatively small number of subjects, there was greater precision associated with estimation of the fixed-effects parameters than the random-effects [17].

The interindividual variability in CL/F and V /F was estimated using an exponential error model which has the dual advantage of avoiding negative parameter values, and which is suitable for pharmacokinetic data which frequently have a log-normal distribution. Cisapride appeared to be rapidly absorbed as shown by the typical value of the first order rate constant of 2.68 h⁻¹ (half-life; 16.1 min). Delayed gastric emptying is a well-known feature of reflux disease [3], but the addition of a lag-time did not improve the fit. However, we point out that the very limited data points in the absorptive phase probably contributed to the relatively high degree of uncertainty (CV 34.9%) in the estimation of Ka, compared with CL/F (CV 7.6%).

In adults, the average CL/F value of 0.33 l h⁻¹ kg⁻¹ (calculated from plasma data reported previously by one of us [18]), lies within the 95% prediction interval for CL/F (0.27–1.06 l h⁻¹ kg⁻¹) in the present study, assuming a log-normal distribution. The typical population value for V /F was not influenced significantly by any factor, including body weight. Initially, this result was somewhat surprising since body weight varied several-fold from <1500 g (20% of infants) to >3000 g (30% of infants). The interpatient variability in V /F (84.3%) was more than twice that associated with CL/F (34.5%). One can anticipate that having more plasma samples per patient within a dosing interval will facilitate a body weight effect on V /F separate from body weight on CL/F; this would reduce the unexplained interindividual variability in V /F. A linear regression of V /F on weight using a Least Squares Trimmed algorithm [19] gave a slope of 1.9 l kg⁻¹ which can be compared with a mean value of 2.4 l kg⁻¹ reported in healthy adults [18]. It is possible the relatively lower V /F of cisapride may reflect a lower proportion of the body weight as fat in young infants, compared with adults [11], since the distribution of relatively lipophilic drugs such as cisapride (oil:water partition coefficient ≈1.0×10⁴) [20] may be influenced markedly by the amount of fatty tissues [21].

Because CL/F increased proportionately with WT, the elimination half-life was inversely related to WT. A half-life of 7 h would be expected for an infant at the population average weight of 2.5 kg, whereas a 750 g infant and a 3 kg infant would have expected half-lives of 37.5 h and 9.4 h, respectively. This can be compared with a range of 7.7 h to 10.1 h in healthy adult subjects, and 7.8 h to 15.8 h in elderly subjects, after single and repeated oral doses [18]. Clearly, the dependency of the posterior Bayesian estimates of CL/Fwith WT is important for setting maintenance cisapride doses. Furthermore, they have important ramifications for setting dosage intervals in infants of different weight, and for the revision of dosing intervals as body weight changes within individuals.

In conclusion, this study reports for the first time the pharmacokinetics of cisapride in young infants. While cisapride is being used increasingly to treat reflux disease [5, 6, 22], there is still very little information on the important relationship between cisapride plasma concentrations and any objective pharmacological response(s) in this distressing condition. Accordingly, the results of this study have paved the way for a much larger population study which will integrate pharmacokinetic and pharmacodynamic data, the latter being obtained using intraoesophageal pH monitoring [3].