Intravenous indometacin in preterm infants with symptomatic patent ductus arteriosus. A population pharmacokinetic study

Abstract

Aims

To characterize the population pharmacokinetics of indometacin in preterm infants with symptomatic patent ductus arteriosus and to investigate the influence of various factors on the response to treatment.

Methods

Data were collected from 35 infants (gestational age 25–34 weeks; postnatal age 1–77 days) in neonatal units in Belfast and Copenhagen. Infants received an initial course of up to three doses of intravenous indometacin (0.1–0.2 mg kg⁻¹) as considered appropriate by the treating physician. For those infants who did not respond to therapy or in whom the ductus reopened, a second course was sometimes given. Population analysis of the 185 plasma concentrations obtained was conducted using NONMEM and pharmacokinetic and demographic differences between responders and nonresponders were compared.

Results

The concentration-time course of indometacin was best described by a one-compartment model. The final population parameter estimates of clearance (CL) and volume of distribution (V) (standardized to the median weight of 1.17 kg) were 0.00711 l h⁻¹ and 0.266 l, respectively. CL increased from birth by approximately 3.38% per day and V by approximately 1.47% per day. Concomitant digoxin therapy resulted in a 30% decrease in V. Interindividual variability in CL and V was 41% and 21%, respectively. Interoccasion variability for CL was 43%. Residual variability corresponded to a standard deviation of 0.148 mg l⁻¹. Closure occurred in 75% of infants with a plasma concentration ≥0.4 mg l⁻¹ 24 h after the last dose.

Conclusions

Dosing regimens for indometacin should take into account the weight and postnatal age of the infant and any concomitant digoxin therapy. The population estimates can be used to determine typical values of CL and V allowing the prediction of individualized doses of indometacin that should increase the probability of achieving a 24 h plasma concentration ≥0.4 mg l⁻¹. Although the pharmacokinetic estimates will be affected by both interindividual and within-individual variation, it is anticipated that this approach will decrease the variability of exposure and optimize treatment outcome.

Introduction

Functional closure of the ductus arteriosus (DA) occurs spontaneously within 2–4 days of birth in most healthy term and preterm infants [1]. However, persistent patency is frequently observed in preterm infants suffering from respiratory distress syndrome (RDS) [2] and the incidence increases with decreasing gestational age [3] and lower birth weight [2]. Patent ductus arteriosus (PDA) is associated with a number of significant complications including bronchopulmonary dysplasia, intraventricular haemorrhage and necrotizing enterocolitis [4].

The use of indometacin, a prostaglandin synthesis inhibitor, to induce pharmacological closure of the PDA was first described in 1976 [5, 6]. Since then its use in the treatment of clinically significant PDA has been widely documented, but the response to treatment has been extremely variable [7–11]. Studies of indometacin in the newborn have also reported considerable interindividual variability in its pharmacokinetics [9, 10, 12–16]. Traditional pharmacokinetic studies are difficult to perform in infants because of the need for multiple blood samples, which creates ethical and technical problems. Population analysis, however, allows the estimation of pharmacokinetic parameters from sparse data and enables the identification of patient characteristics that have a significant influence on drug disposition, and thereby explain some of the variability between individuals [17]. The aims of this study were to characterize the population pharmacokinetics of indometacin in preterm infants with PDA and to examine potential factors affecting response to treatment.

Methods

Patients and data collection

The study was approved by the Research Ethics Committee of Queen's University Belfast and informed parental consent was obtained for each infant before enrolment in the study.

Diagnosis of PDA was based on clinical findings of systolic and diastolic murmur at the left sternal border, hyperactive precordium and bounding pulses. In 90% of cases this was confirmed by echocardiography. Data were collected prospectively from 19 infants in the neonatal intensive care unit at the Royal Maternity Hospital, Belfast. Indometacin (as the sodium trihydrate salt) was administered intravenously at doses of 0.1–0.2 mg kg⁻¹ as considered appropriate by the treating physician. Up to two subsequent doses were administered at 12–24 h intervals depending on response, which was assessed by clinical examination 12 and 24 h after each dose. For those infants who did not respond to therapy or in whom the DA reopened, a second course was sometimes given. Blood samples were obtained by heel prick 1 h after each i.v. administration and 12, 24, 36 and 48 h after the last dose, where possible. Samples were centrifuged and plasma stored at −20 °C until assayed.

Data from 16 infants were obtained from the Department of Neonatology and Paediatrics, Rigshospitalet, Copenhagen. An initial i.v. dose of 0.2 mg kg⁻¹ indometacin was administered to all infants. Five received up to two subsequent doses of 0.2 mg kg⁻¹ at 24 h intervals. The remainder received a second dose of 0.2 mg kg⁻¹ after 6 h and, if necessary, a third dose of 0.1 mg kg⁻¹, 24 h after the first dose. Blood samples were collected via heel prick 1 h after each i.v. administration, immediately prior to subsequent administrations and 6, 24 and 48 h after the last dose. For clinical reasons this schedule was not always adhered to. Samples were centrifuged and plasma stored at −20 °C until assayed.

In addition to accurate information on dosing and times of sampling, the following data were collected for each infant: weight (WT), gestational age (GA), postnatal age (PNA), gender (GEND), Apgar score at 1 min (APG1), comedication [digoxin (DIG), diuretics (DIUR), aminoglycoside antibiotic administration (GLY)], and whether the infant was small for GA (SGA, <10th percentile), suffered from RDS (diagnosed clinically and radiologically) or was receiving parenteral nutrition (TPN). The characteristics of the infants studied are presented in Table 1.

Table 1.

Characteristics of infants studied (n = 35)

Characteristic	Number of infants (%)	Median (range)
Weight (kg)		1.17 (0.57–2.19)
Gestational age (weeks)		29 (25–34)
Postnatal age (days)		14 (1–77)
Male	15 (43)
Apgar score at 1 min <6	13 (37)
Concomitant digoxin administration	15 (43)
Concomitant diuretic administration	17 (49)
Concomitant aminoglycoside administration	17 (49)
Small for gestational age	7 (20)
Respiratory distress syndrome	27 (77)
Parenteral nutrition	22 (63)
Plasma samples per patient		5 (2–10)
Indometacin doses per patient		3 (1–6)
Closure of patent ductus arteriosus	17 (49)

Drug analysis

Indometacin plasma concentrations in samples from Belfast were measured by reversed phase high-performance liquid chromatography with ultraviolet detection at 254 nm. Plasma (0.2 ml) was treated with acetonitrile (1 ml) containing mefenamic acid (0.6 mg l⁻¹, internal standard). After centrifugation, 1 ml of supernatant was added to citrate buffer (pH 3, 1 ml) and extracted twice with ethyl acetate (3 ml). The organic phase was separated and evaporated to dryness at 45 °C under nitrogen. The residue was reconstituted with mobile phase (200 µl) and 100 µl injected onto a Beckman Ultrasphere ODS analytical column (5 µm particle size, 4.6 mm ID × 150 mm), through which a mobile phase of acetonitrile : acetic acid (0.166 m, 48 : 52 v/v) was pumped at 1.8 ml min⁻¹. The retention times for indometacin and mefenamic acid were 6.2 min and 10.5 min, respectively. Intraday and interday coefficients of variation (CV) were ≤8.3% and ≤5.3%, respectively, and the limit of quantification was 100 ng ml⁻¹.

Indometacin plasma concentrations in samples from Copenhagen were measured using a published gas chromatography method with electron-capture detection [18]. The limit of quantification was 50 ng ml⁻¹ and the CV was 6%.

Population pharmacokinetic modelling

Population analysis was performed using NONMEM (version V, level 1.1 with double precision) [19] on a personal computer in conjunction with the DIGITAL Visual Fortran compiler (version 5.0.A). The first-order conditional estimation (FOCE) method was used and the concentration-time course of indometacin was investigated using one- and two-compartment models with first-order elimination. The one-compartment model was found to describe adequately the disposition of indometacin, as the earliest samples were taken 1 h after i.v. administration, thereby allowing for initial distribution of the drug. There was no evidence that a more complex structural model was required. The pharmacokinetic parameters estimated from this model (implemented using PREDPP subroutines ADVAN1 TRANS2) were clearance (CL) and volume of distribution (V).

Interindividual variability was estimated using an exponential error model, as pharmacokinetic parameters are frequently log-normally distributed. The covariance between the variability in CL and V was examined using the $OMEGA BLOCK option. The covariance is a measure of statistical association between the two variables and is related to their correlation [19]. As most infants were studied on at least two occasions, interoccasion variability was investigated using an exponential model. An ‘occasion’ was a set of concentrations obtained after any single dose. Residual variability, which incorporates error arising from inaccuracy in the timing of sample collection and dose administration, assay error and model misspecification, was modelled using additive, proportional and combined error structures.

In the base model both CL and V were standardized to the median WT of 1.17 kg using an allometric model [20, 21] according to the following equations:

$$\text{CL}={\theta }_1\times {(\text{WT}/1.17)}^{0.75}$$

$$\text{V}={\theta }_2\times \text{WT}/1.17$$

where WT is in kg and θ₁ is the CL, and θ₂ is the V, in the standard individual with WT 1.17 kg. θ₁ and θ₂ were estimated by NONMEM.

Individual Bayesian estimates of the pharmacokinetic parameters were obtained from the base model and plotted against the following covariates: GA, PNA, GEND, APG1, DIG, DIUR, GLY, SGA, RDS and TPN. The graphs were then examined for any relationships and the significant covariates were tested for inclusion in the population model by incorporating them individually into the base model. Continuous covariates were modelled in a linear or nonlinear way, for example using the equations:

$$\text{CL}={\theta }_1\times {(\text{WT}/1.17)}^{0.75}\times (1+{\theta }_3\times \text{COV})$$

where COV is a general continuous covariate and θ₁ and θ₃ are regression parameters estimated by NONMEM.

Categorical covariates were examined using a multiplicative model:

$$\text{CL}={\theta }_1\times {(\text{WT}/1.17)}^{0.75}\times {{\theta }_3}^{\text{CAT}}$$

where θ₁ is the population value for an individual with WT 1.17 kg in the absence of the covariate (CAT = 0) and θ₃ is the fractional change when the covariate is present (CAT = 1). Similar models were used to investigate the effect of covariates on V.

Models were compared by examining the difference in objective function (OF) which is a measure of goodness of fit and equals minus twice the log likelihood of the data. Models forming a full and reduced model pair were compared statistically using the likelihood ratio test [19]. The change in OF caused by inclusion of a covariate is asymptotically χ² distributed with n degrees of freedom (n = the difference in number of parameters between the two models). A decrease in OF >6.63 was considered statistically significant (P < 0.01, 1 degree of freedom).

Goodness of fit was also assessed by examining the precision of parameter estimates, the decrease in interindividual and residual variability, and graphs of residuals (RES), weighted residuals (WRES) and measured indometacin concentrations plotted separately against predicted concentrations. The covariate producing the greatest improvement in fit was added to give an intermediate model and the process repeated until all significant covariates were included and no further statistically significant reduction in OF was obtained. Stepwise backward elimination was used to obtain the final regression model. An increase in OF >7.88 (P < 0.005, 1 degree of freedom) was required to retain the covariate in the final model.

Model validation

Use of the population model for predictive purposes requires evaluation of its performance to assess its accuracy and robustness. Owing to the limited number of infants in the study, model validation using the data splitting technique could not be performed. Instead, internal validation using the bootstrap technique [22, 23] was implemented using Wings for NONMEM (WFN version 401) (written by N. Holford, Department of Pharmacology and Clinical Pharmacology, University of Auckland, New Zealand and available at http://wfn.sourceforge.net/index.htm). One thousand bootstrap samples were generated by random sampling with replacement from the original dataset. The final model was fitted to the 1000 bootstrap samples and the mean parameter estimates and their standard deviations were calculated and compared with those obtained from the original dataset.

Response to treatment

Closure of the PDA was assessed by clinical examination 12 and 24 h after each dose and additionally by echocardiography after the final dose in those cases of apparent nonresponse. Pharmacokinetic and demographic differences between responders and nonresponders were compared statistically. Continuous data were compared using two-tailed t-tests (for unequal variances), whereas categorical data were compared using two-sided Fisher's exact tests.

Results

NONMEM analysis was performed using 185 plasma concentrations (median 0.72 mg l⁻¹, range 0.08–2.37 mg l⁻¹) from 35 infants. Figure 1 shows the individual plasma indometacin concentrations (normalized to a dose of 0.2 mg kg⁻¹) against time relative to the final drug administration. An additive error model best described the residual variability. The allometric scaling model was selected as the base model as this best described the data, giving the lowest value of objective function and the best fit of RES vs. time and RES vs. predicted concentrations. The estimation of interoccasion variability in CL resulted in a decrease in OF of 8.40 and was accompanied by a 14.6% decrease in the residual variability. Estimation of interoccasion variability for V was not supported. For a standard individual with weight 1.17 kg, the population estimates from this model for CL and V were 0.0128 l h⁻¹ and 0.291 l, respectively, with an interindividual variability (CV) of 57.1% for CL and 32.2% for V. The interoccasion variability for CL was 46.2% and the residual variability corresponded to a standard deviation of 0.149 mg l⁻¹.

Figure 1.

Individual plasma indometacin concentrations (normalized to a dose of 0.2 mg kg⁻¹) vs. time relative to the final drug administration. Data from the same patient are linked

The addition of several covariates to the base model reduced the OF by >6.63. The factors affecting CL were PNA, DIUR and DIG, and those affecting V were PNA and DIG. The greatest improvement in fit was caused by the addition of PNA on CL (ΔOF = 18.22), hence this covariate was added to give an intermediate model. The subsequent steps in the model-building process are shown in Table 2. The influence of PNA on both CL and V was best described by an exponential relationship. Concomitant therapy with digoxin was associated with a reduction of V. The addition of further covariates failed to reach significance. Results from the backward elimination procedure are shown in Table 3. The exclusion of each covariate from the full model increased the OF by >10, and therefore, all factors were retained in the final model.

Table 2.

Summary of stepwise model building

Model	PK parameters	OF	ΔOF	P-value
Base	CL ∼ WT	−347.08	–	–
	V ∼ WT
Step 1	CL ∼ WT + PNA	−365.30	−18.22	<0.001
	V ∼ WT
Step 2	CL ∼ WT + PNA	−372.30	−7.00	<0.01
	V ∼ WT + PNA
Full	CL ∼ WT + PNA	−385.23	−12.93	<0.001
	V ∼ WT + PNA + DIG

CL, clearance (l h⁻¹); V, volume of distribution (l); WT, weight (kg); PNA, postnatal age (days); DIG, concomitant digoxin therapy; OF, objective function; ΔOF, change in objective function. A ΔOF >6.63 is statistically significant (P < 0.01, 1 degree of freedom).

Table 3.

Results of backward elimination process

Covariate deleted	ΔOF	P-value
PNA in CL model	15.08	<0.001
PNA in V model	10.92	<0.001
DIG in V model	12.93	<0.001

CL, clearance; V, volume of distribution; PNA, postnatal age; DIG, concomitant digoxin therapy; ΔOF, change in objective function from full model (CL = θ₁ × (WT/1.17)^0.75 × e^{θ₃ × PNA}; V = θ₂ × WT/1.17 × e^{θ₄ × PNA} × θ₅^DIG). A ΔOF > 7.88 is statistically significant (P < 0.005, 1 degree of freedom).

The structural and variance population parameter estimates of the final model and their corresponding CV are shown in Table 4. For a standard individual with a WT of 1.17 kg the estimates of CL and V at birth were 0.00711 l h⁻¹ and 0.266 l, respectively, resulting in an elimination half-life (t_1/2) of 25.9 h. V was reduced to 0.186 l and t_1/2 to 18.1 h if the infant received digoxin. CL and V increased from birth by approximately 3.38% and 1.47% per day, respectively. Interindividual variability (CV) in CL and V were reduced from 57.1% to 40.9% and 32.2% to 20.7%, respectively, by the inclusion of the covariates PNA and DIG, explaining some of the variability in CL and V between individuals. The correlation coefficient for the population parameter variability for CL and V was 0.83. Interoccasion variability for CL was 43.2%. The residual variability (SD) was 0.148 mg l⁻¹, which corresponds to CVs of 82% and 9% at plasma indometacin concentrations of 0.18 and 1.68 mg l⁻¹, respectively.

Table 4.

Population parameter estimates of the final model from the original dataset and the 1000 bootstrap samples

	Original dataset		1000 bootstrap samples
Parameter	Estimate	CV%	Mean estimate	CV%	Percent difference
θ1	0.00711	19.0	0.00720	18.9	1.3
θ2	0.266	8.2	0.264	8.4	0.8
θ3	0.0338	23.8	0.0344	24.3	1.7
θ4	0.0147	27.5	0.0150	27.4	2.0
θ5	0.699	8.4	0.702	8.8	0.4
ωCL (CV%)	40.9	46.1	39.6	19.7	3.3
ωV (CV%)	20.7	40.2	20.2	19.1	2.5
πCL (CV%)	43.2	66.3	42.2	29.8	2.4
σ (mg l−1)	0.148	23.6	0.142	10.7	4.2

TVCL = θ₁ × (WT/1.17)^0.75 × e^{θ₃ × PNA}; TVV = θ₂ × WT/1.17 × e^{θ₄ × PNA} × θ₅^DIG. TVCL, Typical population value for clearance; TVV, typical population value for volume of distribution; WT, weight (kg); PNA, postnatal age (days); DIG, 1 if the infant received digoxin and 0 otherwise; ω_CL, interindividual variability in CL; ω_V, interindividual variability in V; π_CL, interoccasion variability in CL; σ, residual variability; CV%, percentage coefficient of variation; Percent difference, the percentage difference between the mean estimate from the bootstrap samples and the estimate from the original dataset.

When the analysis was repeated with CL and V standardized to a weight of 70 kg instead of the median weight of 1.17 kg, there was no effect on the parameter estimates. At birth, CL and V were 0.153 l h⁻¹ 70 kg⁻¹ and 15.9 l 70 kg⁻¹, respectively. If these estimates are used to predict CL and V for a weight of 1.17 kg, the results equal those obtained in the final model (0.00711 l h⁻¹ and 0.266 l, respectively). All other fixed and random effects parameter estimates were identical in the two models.

Since different assay methods were used at the two sites, the possibility of improving model fit by describing residual variability using two random error terms was investigated using the following equation:

$$C_{\text{ij}}=C_{\text{pred,ij}}+{\varepsilon }_{1,\text{ij}}\times \text{SITE}+{\varepsilon }_{2,\text{ij}}\times (1-\text{SITE})$$

where C_ij is the measured and C_pred,ij the model predicted indometacin concentration of the jth individual at the ith sampling time and ɛ_1,ij and ɛ_2,ij are random variables with zero means and variances of and arising from Belfast (SITE = 1) and Copenhagen (SITE = 0) data, respectively. This resulted in a decrease in OF of only 3.26, and thus the original model was used.

The scatterplot in Figure 2a shows the predicted concentrations obtained from the final model are in reasonable agreement with the measured concentrations and vary randomly around the line of identity. The scatterplot of WRES vs. predicted concentrations (Figure 2b) showed a random distribution with most WRES lying within 2 units of the null ordinate. No outliers or trends were apparent, indicating a good fit of the model to the data.

Figure 2.

(a) Measured indometacin concentrations vs. population predicted indometacin concentrations. (b) Weighted residuals vs. population predicted concentrations. The lines of perfect agreement are indicated in bold. □, samples taken from infants receiving digoxin; ▪, samples taken from infants not receiving digoxin

The parameter estimates from the original dataset were within 5% of the bootstrapped means (Table 4), demonstrating that the model is robust.

The median individual Bayesian estimates for CL and V in our population, obtained from the final model, are shown in Table 5. These estimates are determined using the measured concentrations, in contrast to the population parameter estimates which are derived from covariate information. Table 5 also shows the median predicted population t_1/2 and WT normalized values for CL and V.

Table 5.

Individual Bayesian estimates obtained from the final population model

Parameter	Median (P5, P95)
Clearance (ml h−1)	10.0 (5.0, 38.5)
Volume of distribution (l) (no digoxin)	0.274 (0.175, 0.511)
Volume of distribution (l) (with digoxin)	0.251 (0.169, 0.440)
Half-life (h)	17.1 (7.6, 34.3)
Clearance (ml h−1 kg−1)*	9.7 (3.5, 41.6)
Volume of distribution (l kg−1)* (no digoxin)	0.257 (0.204, 0.551)
Volume of distribution (l kg−1)* (with digoxin)	0.213 (0.154, 0.284)

P₅, 5th percentile; P₉₅, 95th percentile.

^*Calculated using the weight of each individual.

Closure of the PDA was achieved in 17 of the 35 infants (49% success rate). Comparison between those infants in whom treatment resulted in successful closure of the PDA and those in whom treatment failed, revealed a significantly higher plasma concentration 24 h after the last dose [0.620 (0.282 SD) vs. 0.366 (0.364 SD) mg l⁻¹, P = 0.027, Figure 3], a significantly lower PNA (11 vs. 25 days, P = 0.008) and a significantly lower percentage receiving concomitant treatment with diuretics (24 vs. 72%, P = 0.007).

Figure 3.

Plasma indometacin concentrations 24 h after the last dose vs. individual patients. □, infants in whom treatment resulted in successful closure of the PDA; ▪, infants in whom treatment failed

Discussion

Since 1976, when the use of indometacin for the treatment of PDA was first described, several traditional pharmacokinetic studies [9, 10, 12, 14–16] and one population analysis [13] have investigated its pharmacokinetics in preterm infants. Considerable variability in the pharmacokinetic parameters has been reported and patient characteristics such as PNA, GA and birth weight have been identified as having a significant influence on indometacin disposition [9, 10, 12–16].

Yaffe et al.[15] reported significantly greater values for the mean total body CL and mean V in infants of birth weight >1 kg compared with those in lighter babies. In the current study, the influence of WT on both CL and V was best described by an allometric scaling model [21, 24]. This was used as the base model which, having accounted for the effect of size (as predicted by WT), then allowed investigation of the effects of maturation (GA and PNA) on the pharmacokinetics of indometacin.

PNA was the most influential covariate examined, as it produced the greatest reduction in OF when it was incorporated into the model and resulted in a marked decrease in interindividual variability in both CL and V. The model predicts that CL increases from birth at a rate of approximately 3.4% per day. In adults, indometacin is metabolized by O-demethylation and N-deacylation, and both the parent drug and its metabolites are conjugated with glucuronic acid [25]. Only a small proportion of the dose is excreted unchanged in the urine. Preterm infants have an impaired metabolic activity due to immature hepatic microsomal enzyme systems and decreased capacity for conjugation with glucuronic acid [26]. Increase in indometacin CL with advancing PNA is probably due to the maturation of hepatic enzyme systems with a small contribution from increased renal function.

Others have noted a significant correlation between CL and PNA [9, 10, 12, 13, 15]. Individual estimates for CL in our population are in good agreement with these studies, with the exception of the population analysis [13], which reported much lower values for CL, ranging from 2.6 to 7.5 ml h⁻¹ kg⁻¹ for a 1-kg infant over the first 20 days of life. This is possibly due to the wider range of PNA (1–77 days) in our study. The median estimated CL in our study is approximately 4–10 times lower than that reported in adults [27]. Comparison of the mean parameter estimate of clearance, standardized to a weight of 70 kg (0.153 l h⁻¹ 70 kg⁻¹), with that for adults (5.62 l h⁻¹ 70 kg⁻¹) indicates that with indometacin preterm infants do not respond kinetically as ‘small adults’ [27].

The final model predicts that V increases from birth at a rate of approximately 1.5% per day. As this is slower than the increase in CL, it results in a decrease in the t_1/2 of indometacin with advancing PNA. Several authors have documented a significant inverse correlation between t_1/2 and PNA, and have also reported a significant positive correlation between CL and PNA. However, no mention was made of any relationship between V and PNA [9, 10, 12]. Although indometacin is highly bound to plasma albumin, it is also extensively bound to tissues [28]. In preterm infants, total body water comprises up to 90% of body weight, whereas fat accounts for only about 1%[26]. This contrasts with term neonates in whom total body water is 75% and body fat is about 15% of body weight. At birth, perfusion of muscle and various tissues is often erratic, but improves with maturation [26]. The increase in indometacin V with advancing PNA is probably due to increased tissue binding of the drug resulting from maturational changes in body composition, and improved delivery of drug to the tissues.

Individual estimates for V in our population are in good agreement with those from previous studies [9, 10, 12, 13]. Each of our infants with a PNA ≥30 days (n = 5) had an estimated V >0.34 l kg⁻¹, which is above the lower limit of the reported range in adults (0.34–1.57 l kg⁻¹) [27]. This is not surprising as tissue perfusion improves dramatically in the first few weeks of life, and the most substantial changes in body composition occur in the first 6 months of life [26].

In the final model, V was decreased by 30% in those infants receiving concomitant digoxin therapy. When this finding was incorporated, interindividual variability in V fell from 26.0% to 20.7%. An interaction between indometacin and digoxin in preterm infants has been reported previously [29], and resulted in a significant increase in the serum concentrations and t_1/2 of digoxin, which corresponded with a decrease in urine output. As far as we are aware, there have been no previous reports of an effect of digoxin on the pharmacokinetics of indometacin. It is possible that when indometacin is added to digoxin therapy, the latter is bound preferentially to tissues, resulting in a decreased V of indometacin. Alternatively, a decrease in the volume of extracellular fluid resulting from digoxin administration may produce a decreased V for indometacin.

Closure of the PDA occurred in 49% of infants. Response rates in previous studies vary from 44%[8] to 91%[11]. This is attributed to biological factors such as birth weight [30], GA [31] or PNA [32], which may affect the inherent ability of the DA to constrict permanently, and pharmacokinetic variability that may result in subtherapeutic plasma concentrations [10, 33].

Our results showed the mean plasma concentration 24 h after the last dose was significantly higher in infants who responded to treatment. Several studies have reported a correlation between plasma concentration and response to treatment. Brash et al.[10] found that response of the DA was related to indometacin plasma concentration 24 h after the last dose, with failure being associated with concentrations <0.25 mg l⁻¹. Yeh et al.[9] observed a significant correlation between plasma concentration 12 h after a dose and response of the DA at 24 h, with a 50% or greater chance of closure with concentrations ≥0.6 mg l⁻¹. Two other studies have noted an increased response with greater exposure to the drug [12, 16]. In contrast, several studies in which indometacin plasma concentrations were measured shortly after drug administration failed to demonstrate any relationship with subsequent closure of the DA [7, 8]. However, indometacin concentrations were unlikely to correspond to the time that closure occurred.

Mean PNA was significantly lower in the closure group. It has been proposed that failure of indometacin treatment in older infants may be due to histological changes in the DA rendering it less sensitive to prostaglandin inhibition [32]. This is a possible explanation, but none of the infants in the failure group with a PNA >21 days had a plasma concentration >0.26 mg l⁻¹ 24 h postdose. PNA was found to have a significant influence on the pharmacokinetics of indometacin and failure in older infants may therefore be due to inadequate dosing resulting from the increased CL and decreased t_1/2 in these infants. This is supported by a study which showed no significant difference in response rates in older children when dosing was individualized [33].

The percentage of infants receiving concomitant furosemide therapy was significantly lower in the closure group. Yeh et al.[34] suggested that concomitant use of furosemide may prevent the renal side-effects of indometacin. However, Romagnoli et al.[35] found this was not the case and reported no negative effects on indometacin efficacy. In contrast, two studies documented an association between administration of furosemide and development of PDA in preterm infants, probably resulting from an increased renal synthesis of prostaglandin E₂[36, 37]. It has also been suggested that furosemide may compete for plasma protein binding sites and increase glomerular filtration of unbound indometacin [25]. This is a possible explanation for the results of our study. Concomitant diuretic therapy was not present in our final model, as its effect did not reach significance after the inclusion of PNA and DIG. However, in the initial modelling stage, diuretic therapy was associated with an increase in CL and a significant decrease in OF. Therefore, it is possible that inadequate plasma concentrations of indometacin may contribute to the higher failure rate observed in those infants receiving furosemide.

Our results suggest that maintaining indometacin plasma concentration (ideally for 24 h) above a threshold concentration (0.4 mg l⁻¹) is important to obtain closure of the PDA. The final population model indicates that WT, PNA and concomitant digoxin therapy should be taken into account when designing dosing regimens for indometacin. In the UK, dosing recommendations are the same as those used in the US National Collaborative Trial [7]. After administration of a standard first dose of 0.2 mg kg⁻¹, doses are adjusted over the first 8 days of life and thereafter remain constant. For an infant of 1.17 kg, our final population model predicts a t_1/2 of 22.3 h at a PNA of 8 days, and of 16.1 h at a PNA of 25 days (11.2 h if receiving digoxin). To maintain the indometacin plasma concentration above a threshold concentration, the older infant, or an infant receiving digoxin, must receive either a larger dose or more frequent administration than the younger child.

In conclusion, the internal validation using the bootstrapping technique demonstrates the predictive accuracy and robustness of the population model obtained in this study. Therefore, the model can be used with confidence to determine typical values for indometacin CL and V for preterm infants with PDA, depending on their WT, PNA and whether they are also receiving digoxin treatment. These estimates will enable prediction of individualized doses of indometacin, which should increase the probability of achieving a 24 h plasma concentration ≥0.4 mg l⁻¹. Although the success of treatment is not guaranteed, as pharmacokinetics will be affected by both interindividual and within-individual variation, it is anticipated that this approach will decrease the variability in exposure and optimize treatment outcome. It is intended to subject the model to external validation, which is the most stringent method for testing a model. Simulation studies will then be used to generate dosing regimens which will be used for a more detailed investigation of the pharmacokinetic/pharmacodynamic relationships for indometacin.