Population pharmacokinetics of intravenous indomethacin in neonates with symptomatic patent ductus arteriosus The population pharmacokinetics of intravenous indomethacin were investigated with 665 indomethacin serum concentrations from 83 neonates (mean -± SD: gestational age, 28.8 ± 2.5 weeks; postnatal age, 5.7 ± 4.7 days; birth weight, 1.13 ± 0.40 kg) receiving indomethacin for symptomatic patent ductus arteriosus. A one-compartment open model was used for pharmacokinetic analysis. Hypotheses were tested to determine which developmental and demographic data influenced clearance (CL) and volume of distribution (Varea) In the final regression equation CL and Varea were modeled as a function of body weight and postnatal age (PNA) from 0 to 20 days. Final estimates were as follows: CL (ml/hr) = 2.63 weight (kg) + 0.244 PNA (days) and Varea (L) = 0.28 weight (kg) + 0.0041 PNA (days). The coefficients of variation for interindividual variability in CL and Varea were 77% and 28%, respectively. Intraindividual variability was 19%. These mean population parameter estimates should prove useful in designing dosage regimens to achieve desired indomethacin concentrations for neonates from 0 to 20 days of age with symptomatic patent ductus arteriosus. (CLIN PHARMACOL THER 1991;49:550-7.)
Donald B. Wiest, PharmD, Julianne B. Pinson, PharmD, Peter S. Gal, PharmD, Richard C. Brundage, PharmD, S. Schall, MD, J. Laurence Ransom, MD, Richard L. Weaver, MD, Dilip Purohit, MD, and Y. Brown, MTASCP Charleston, S.C., Greensboro, N.C., and Minneapolis, Minn.
Functional closure of the patent ductus arteriosus (PDA) occurs rapidly after birth in full-term healthy newborn infants.' In contrast, more than 80% of neonates with a birth weight less than 1000 gm will have a PDA that remains open for days or weeks, which may be associated with significant morbidity.2 Heymann et al.3 and Friedman et al.4 first described the use of indomethacin to close the PDA in 1976. Since then indomethacin has gained acceptance as effective and appropriate pharmacologic therapy for PDA closure.5 Despite the frequency of indomethacin
From the Departments of Clinical Pharmacy and Pediatrics, Medical University of South Carolina, Charleston; the Department of Pediatrics, Moses H. Cone Memorial Hospital, Greensboro; and the Department of Pharmaceutics, University of Minnesota, Minneapolis. Presented at the Ninety-second Annual Meeting of the American Society for Clinical Pharmacology and Therapeutics, San Antonio, Texas, March 13-15, 1991. Received for publication Sept. 10, 1990; accepted Jan. 21, 1991. Reprint requests: Donald B. Wiest, PharmD, Medical University of South Carolina, CSB-219. 171 Ashley Ave., Charleston, SC 29425. 13/1/28206
550
use and the literature supporting its use, many questions remain unresolved regarding pharmacodynamic and pharmacokinetic aspects of indomethacin in the neonate. The relationship between indomethacin serum concentrations and ductal closure remains controversial. Several studies conducted in the 1980s reached the conclusion that PDA response correlated with indomethacin plasma concentrations, although the level necessary to close the PDA remained inconclusive.6-1° However, other investigations found no direct relationship between indomethacin serum concentrations and efficacy of FDA closure.' 1-16 This same conclusion was reached by the 1983 national collaborative study on PDA, which investigated the largest study population to date.5 At least seven studies have been conducted that have addressed the disposition of indomethacin in neonates.7,9,10,12,17-19 The results of these studies are summarized in Table I. Analysis of these results reveals that there is extreme variability between neonates with regard to the disposition of indomethacin. Consequently, these kinetic parameters present a challenge in determining a dosage regimen that provides
VOLUME 49 NUMBER 5
Indornethacin in neonates
551
Table I. Pharmacokinetic studies of indomethacin in neonates No.
Author
patients
Evans et al.,17 1979 Thalji et al.," 1979 Thalji et al. ,9 1980 Vert et al.,' 2 1980 Yaffee et al.,19 1980
Brash et al.,7 1981 Yeh et a1.,'° 1985
9 7 17 10 10
Birth weight (gm) 1499 ± 122 1301 ± 72 1253 ± 389 1300 ± 200
18
1000
35 38
1213 ± 309 1200 ± 400
GA (wk)
32
± 0.7
NR 31.3 ± 2.4
29.0 ± 1.5 NR NR 29.7 ± 2.1 32 ± 2
Data are mean values -± SD. GA, Gestational age; PNA, postnatal age; t1,2, elimination half-life; *Based on two infants.
PNA (days)
± 1.5 7-14 10.8 ± 6.0 9.5 ± 2.2 NR 12
NR 8.8 ± 6.8 10.2 ± 2.2
t172
(hr)
11-20 14.5 ± 3.0 20.7 ± 8.0 33.9 ± 11.7
20.7 ± 5.2 15.4 ± 3.4 32
21.6
(LIkg)
CL (mIlkg1hr)
NR NR 0.80* 38* 0.35 ± 0.21 13 ± 9.5 0.35 ± 0.16 7.6 ± 3.0 0.22 ± 0.01t 9.6 ± 1.23 0.29 ± 4.2t 22.8 ± 4.21; 7.0 0.19-0.61 0.3 9.2
apparent volume of distribution; CL, clearance; NR, not reported.
+Reported in milliliters per hour.
indomethacin concentrations potentially associated with PDA closure with minimal risk of toxicity. Because of the large interpatient variability in indomethacin pharmacokinetics in neonates, it may be more useful to determine the typical pharmacokinetic behavior of the drug in this population rather than in an individual patient. NONMEM (nonlinear mixed-effect model) is a computer program in which the population, as opposed to the individual, is the object of study.2° With this approach one is able to estimate the pharmacokinetic parameters of a population by using sparse data collected during routine clinical care rather than data collected through intensive blood sampling, an advantage that is particularly desirable for the neonatal population. In addition, with the program one is able to establish to what degree patient characteristics (e.g., postnatal age 113NA1 or time of PDA closure) influence pharmacokinetics of the drug. This article describes the use of this approach in determining the pharmacokinetics of indomethacin in the neonatal population when used to treat symptomatic PDA.
METHODS Patients and data collection. A total of 83 patients who received indomethacin for treatment of symptomatic PDA were included in the analysis. All patients had been diagnosed to have a clinically significant PDA according to echocardiographic and clinical criteria. Clinical criteria for diagnosis of PDA included presence of systolic murmur, bounding peripheral pulses, hyperactive precordium, widened pulse pressures, or high or fluctuating ventilatory support requirements. Routine clinical data from 65 patients were collected
between 1986 and 1989 at the Moses H. Cone Hospital Neonatal Intensive Care Unit (NICU) and from 13 neonates at the North Carolina Memorial Hospital (NCMH) NICU. In addition, 5 patients from the NICU at the Medical University of South Carolina (MUSC) who received indomethacin between 1986 and 1987 were included in the analysis. The dose of indomethacin that the infants received varied according to the NICU in which they were treated. Patients at Moses H. Cone Hospital received an initial intravenous dose of 0.2 to 0.25 mg/kg. Subsequent doses ranged from 0.1 to 0.4 mg/kg and were adjusted to increase serum indomethacin concentrations at increments of 0.3 to 0.5 mg/L until clinical and echocardiographically confirmed closure of PDA was observed. The number of doses and time interval between doses varied according to clinical response. Serum indomethacin concentrations were measured 2 and 8 hours after the initial dose and 2 to 4 hours before each subsequent dose. Patients treated at NCMH and MUSC received three doses of intravenous indomethacin according to the protocol suggested by Gersony et al.5 All of these patients received an initial dose of 0.2 mg/kg. Two subsequent doses were given at 12-hour intervals that were on based the age of the neonate at the time of administration of the first dose. Patients less than 48 hours of age received 0.1 mg/kg, patients aged 2 to 7 days received 0.2 mg/kg, and patients greater than 7 days of age received 0.25 mg/kg. Serum indomethacin concentrations were assayed from blood that was scavenged from regularly planned hospital laboratory studies in patients treated at NCMH. The number of serum concentration measure-
CAN 1'HARMAC01. THER
552
Wiest et al.
MAY 1991
Table II. Patient demographic data Variable
Mean ± SD
Range
1.13 ± 0.66
0.20-4.80
Gender Male Female Patients
Indomethacin serum concentration measurements (mg/L) No. levels per patient No. doses per patient Birth weight (kg) Gestational age (wk) Apgar score (1 min) Apgar score (5 mm) Postnatal age (days) Serum creatinine (mg/di) Incidence of PDA closure (%) Incidence of concomitant furosemide administration (%)
45 38 83
665
8.0 ± 4.4 3.6 ± 1.8 1.125 ± 0.399 28.8 ± 2.5 4.9 ± 2.5 6.8 ± 2.1 5.7 ± 4.2 1.3 ± 0.5
1-23 1-10
0.517-2.560 25-36 0-9 0-10 1-20
0.3-3.4
60 (81) 38 (46)
PDA, Patent ductus arteriosus.
ments and time interval between measurements varied in this group of patients. Serum concentrations for patients at MUSC were obtained under study protocol whereby the first blood sample was collected just before the third dose. Subsequent samples were obtained 1/2, 1, 2, 6, 24, and 72 hours after the third dose. Demographic data for each patient, number of doses, and concentrations measured in the population are summarized in Table II. Indexes of renal function (serum creatinine, blood urea nitrogen, and urine output) were recorded for each patient before, during, and after indomethacin treatment. In addition, the time of PDA closure as evidenced through echocardiogram or resolution of clinical symptoms (systolic murmur, bounding peripheral pulses, hyperactive precordium, widened pulse pressures, or high or fluctuating ventilatory support requirements) was determined in each patient. Serum concentrations of indomethacin were assayed by HPLC.2t The coefficient of variation for this assay was less than 5% for indomethacin concentrations in the range of 0.1 to 5.0 mg/L. Pharmacokinetic model. A one-compartment open model with first-order elimination and zero-order input was used to describe the concentration-time course of indomethacin. The PREDPP program subroutines ADVAN1 and TRANS2 were selected to define this model 22 The pharmacokinetic model was written in a recursive form for intravenous bolus administration as described by Beal et al.22 No steady-state situations for indomethacin were assumed. Statistical model. The interindividual variabilities in clearance (CL) and volume of distribution (Varea)
were modeled with proportional error according to the following equations: CL, = TVCL(1 + Varea,j
TIJCL)
= TVV.,a(1 +
my)
Va,i are the true (unknown) pharmacokinetic parameters for the jth individual as predicted by the regression model, TVCL and TVVa,a are the population means of CL and Vaa, lliCL and are identically distributed random variables with zero mean and variances w2cL and (02V, respectively, that distinguish the jth individual's CL and Varea from those predicted by the regression model. The intraindividual variability, or residual error, was modeled with additive error: in which CLJ and
=
+
in which Cii is the ith observed concentration for the jth individual and Cmij is the ith observed concentration predicted by the model for the jth individual. This value includes the contribution of the and terms; eij is the residual intrapatient variability term, representing independent identically distributed statistical errors with mean zero and variance cr2. Data analysis. All data were analyzed on a VAX 11/785 computer (Digital Equipment Corp., Maynard, Mass.) with the NONMEM program version 2, level 1.4, with double precision. Estimates were sought for (1) population means of CL and Va,a, (2) variance (032) of and TIN/area, (3) variance (o-2) of the residual random error (e), (4) standard errors of the parameter estimates, and (5) value of the objective function, which is equal to minus twice the log likelihood of the data lnL).
(2
VOLUME 49 NUMBERS
Indomethacin in neonates
553
Table III. Development of full regression model: hypotheses tested Hypotheses tested Clearance Did gender affect CL? Did GA 5_30 wk affect CL? Did PNA affect CL? Did 1-min Apgar scores -55 affect Did 5-min Apgar scores :58 affect Did PDA closure affect CL? Did serum creatinine affect CL? Did concomitant furosemide affect Volume of distribution Did gender affect wk affect Va? Did GA Did PNA affect Did 1-min Apgar scores 55 affect Did 5-min Apgar scores 58 affect Did PDA closure affect Varea? Did serum creatinine affect Did concomitant furosemide affect
Difference in 2-1nL versus base model
p Value
Conclusion addressed
0.042 2.953 50.54 0.180 4.672 27.905 21.312
NS NS
No No Yes No No Yes Yes Yes
CL? CL?
CL?
Varea? Varea?
V,?
Varea?
Critical value = 6.6 (x2, p = 0.01; df = 1). CL, Clearance; NS, not significant; GA, gestational age;
V,
apparent volume
Minimizing the objective function provided by each NONMEM run is equivalent to maximizing the likelihood of the data. Hypothesis testing can be performed by monitoring changes in the objective function when one or more parameters in the model are first estimated iteratively and then restricted to a fixed value. This difference is approximately x2 distributed with degrees of freedom equal to the number of parameters with fixed values in the restrictive model. Development of the final regression model. The first stage in the model-building phase was initiated with a minimum number of parameters that were suspected to influence indomethacin disposition. Alternate statistical models were tested during this phase to determine which model would afford the best fit of the data. Careful examination of scatterplots of weighted residual (WRES) versus time and WRES versus predicted concentration guided the selection of the most appropriate statistical model. Additional parameters were incorporated into the initial regression model in a stepwise fashion to develop the full regression model. Any fixed effect that reduced the objective function by more than 6.6 (x2, p
Donald B. Wiest, PharmD, Julianne B. Pinson, PharmD, Peter S. Gal, PharmD, Richard C. Brundage, PharmD, S. Schall, MD, J. Laurence Ransom, MD, Richard L. Weaver, MD, Dilip Purohit, MD, and Y. Brown, MTASCP Charleston, S.C., Greensboro, N.C., and Minneapolis, Minn.
Functional closure of the patent ductus arteriosus (PDA) occurs rapidly after birth in full-term healthy newborn infants.' In contrast, more than 80% of neonates with a birth weight less than 1000 gm will have a PDA that remains open for days or weeks, which may be associated with significant morbidity.2 Heymann et al.3 and Friedman et al.4 first described the use of indomethacin to close the PDA in 1976. Since then indomethacin has gained acceptance as effective and appropriate pharmacologic therapy for PDA closure.5 Despite the frequency of indomethacin
From the Departments of Clinical Pharmacy and Pediatrics, Medical University of South Carolina, Charleston; the Department of Pediatrics, Moses H. Cone Memorial Hospital, Greensboro; and the Department of Pharmaceutics, University of Minnesota, Minneapolis. Presented at the Ninety-second Annual Meeting of the American Society for Clinical Pharmacology and Therapeutics, San Antonio, Texas, March 13-15, 1991. Received for publication Sept. 10, 1990; accepted Jan. 21, 1991. Reprint requests: Donald B. Wiest, PharmD, Medical University of South Carolina, CSB-219. 171 Ashley Ave., Charleston, SC 29425. 13/1/28206
550
use and the literature supporting its use, many questions remain unresolved regarding pharmacodynamic and pharmacokinetic aspects of indomethacin in the neonate. The relationship between indomethacin serum concentrations and ductal closure remains controversial. Several studies conducted in the 1980s reached the conclusion that PDA response correlated with indomethacin plasma concentrations, although the level necessary to close the PDA remained inconclusive.6-1° However, other investigations found no direct relationship between indomethacin serum concentrations and efficacy of FDA closure.' 1-16 This same conclusion was reached by the 1983 national collaborative study on PDA, which investigated the largest study population to date.5 At least seven studies have been conducted that have addressed the disposition of indomethacin in neonates.7,9,10,12,17-19 The results of these studies are summarized in Table I. Analysis of these results reveals that there is extreme variability between neonates with regard to the disposition of indomethacin. Consequently, these kinetic parameters present a challenge in determining a dosage regimen that provides
VOLUME 49 NUMBER 5
Indornethacin in neonates
551
Table I. Pharmacokinetic studies of indomethacin in neonates No.
Author
patients
Evans et al.,17 1979 Thalji et al.," 1979 Thalji et al. ,9 1980 Vert et al.,' 2 1980 Yaffee et al.,19 1980
Brash et al.,7 1981 Yeh et a1.,'° 1985
9 7 17 10 10
Birth weight (gm) 1499 ± 122 1301 ± 72 1253 ± 389 1300 ± 200
18
1000
35 38
1213 ± 309 1200 ± 400
GA (wk)
32
± 0.7
NR 31.3 ± 2.4
29.0 ± 1.5 NR NR 29.7 ± 2.1 32 ± 2
Data are mean values -± SD. GA, Gestational age; PNA, postnatal age; t1,2, elimination half-life; *Based on two infants.
PNA (days)
± 1.5 7-14 10.8 ± 6.0 9.5 ± 2.2 NR 12
NR 8.8 ± 6.8 10.2 ± 2.2
t172
(hr)
11-20 14.5 ± 3.0 20.7 ± 8.0 33.9 ± 11.7
20.7 ± 5.2 15.4 ± 3.4 32
21.6
(LIkg)
CL (mIlkg1hr)
NR NR 0.80* 38* 0.35 ± 0.21 13 ± 9.5 0.35 ± 0.16 7.6 ± 3.0 0.22 ± 0.01t 9.6 ± 1.23 0.29 ± 4.2t 22.8 ± 4.21; 7.0 0.19-0.61 0.3 9.2
apparent volume of distribution; CL, clearance; NR, not reported.
+Reported in milliliters per hour.
indomethacin concentrations potentially associated with PDA closure with minimal risk of toxicity. Because of the large interpatient variability in indomethacin pharmacokinetics in neonates, it may be more useful to determine the typical pharmacokinetic behavior of the drug in this population rather than in an individual patient. NONMEM (nonlinear mixed-effect model) is a computer program in which the population, as opposed to the individual, is the object of study.2° With this approach one is able to estimate the pharmacokinetic parameters of a population by using sparse data collected during routine clinical care rather than data collected through intensive blood sampling, an advantage that is particularly desirable for the neonatal population. In addition, with the program one is able to establish to what degree patient characteristics (e.g., postnatal age 113NA1 or time of PDA closure) influence pharmacokinetics of the drug. This article describes the use of this approach in determining the pharmacokinetics of indomethacin in the neonatal population when used to treat symptomatic PDA.
METHODS Patients and data collection. A total of 83 patients who received indomethacin for treatment of symptomatic PDA were included in the analysis. All patients had been diagnosed to have a clinically significant PDA according to echocardiographic and clinical criteria. Clinical criteria for diagnosis of PDA included presence of systolic murmur, bounding peripheral pulses, hyperactive precordium, widened pulse pressures, or high or fluctuating ventilatory support requirements. Routine clinical data from 65 patients were collected
between 1986 and 1989 at the Moses H. Cone Hospital Neonatal Intensive Care Unit (NICU) and from 13 neonates at the North Carolina Memorial Hospital (NCMH) NICU. In addition, 5 patients from the NICU at the Medical University of South Carolina (MUSC) who received indomethacin between 1986 and 1987 were included in the analysis. The dose of indomethacin that the infants received varied according to the NICU in which they were treated. Patients at Moses H. Cone Hospital received an initial intravenous dose of 0.2 to 0.25 mg/kg. Subsequent doses ranged from 0.1 to 0.4 mg/kg and were adjusted to increase serum indomethacin concentrations at increments of 0.3 to 0.5 mg/L until clinical and echocardiographically confirmed closure of PDA was observed. The number of doses and time interval between doses varied according to clinical response. Serum indomethacin concentrations were measured 2 and 8 hours after the initial dose and 2 to 4 hours before each subsequent dose. Patients treated at NCMH and MUSC received three doses of intravenous indomethacin according to the protocol suggested by Gersony et al.5 All of these patients received an initial dose of 0.2 mg/kg. Two subsequent doses were given at 12-hour intervals that were on based the age of the neonate at the time of administration of the first dose. Patients less than 48 hours of age received 0.1 mg/kg, patients aged 2 to 7 days received 0.2 mg/kg, and patients greater than 7 days of age received 0.25 mg/kg. Serum indomethacin concentrations were assayed from blood that was scavenged from regularly planned hospital laboratory studies in patients treated at NCMH. The number of serum concentration measure-
CAN 1'HARMAC01. THER
552
Wiest et al.
MAY 1991
Table II. Patient demographic data Variable
Mean ± SD
Range
1.13 ± 0.66
0.20-4.80
Gender Male Female Patients
Indomethacin serum concentration measurements (mg/L) No. levels per patient No. doses per patient Birth weight (kg) Gestational age (wk) Apgar score (1 min) Apgar score (5 mm) Postnatal age (days) Serum creatinine (mg/di) Incidence of PDA closure (%) Incidence of concomitant furosemide administration (%)
45 38 83
665
8.0 ± 4.4 3.6 ± 1.8 1.125 ± 0.399 28.8 ± 2.5 4.9 ± 2.5 6.8 ± 2.1 5.7 ± 4.2 1.3 ± 0.5
1-23 1-10
0.517-2.560 25-36 0-9 0-10 1-20
0.3-3.4
60 (81) 38 (46)
PDA, Patent ductus arteriosus.
ments and time interval between measurements varied in this group of patients. Serum concentrations for patients at MUSC were obtained under study protocol whereby the first blood sample was collected just before the third dose. Subsequent samples were obtained 1/2, 1, 2, 6, 24, and 72 hours after the third dose. Demographic data for each patient, number of doses, and concentrations measured in the population are summarized in Table II. Indexes of renal function (serum creatinine, blood urea nitrogen, and urine output) were recorded for each patient before, during, and after indomethacin treatment. In addition, the time of PDA closure as evidenced through echocardiogram or resolution of clinical symptoms (systolic murmur, bounding peripheral pulses, hyperactive precordium, widened pulse pressures, or high or fluctuating ventilatory support requirements) was determined in each patient. Serum concentrations of indomethacin were assayed by HPLC.2t The coefficient of variation for this assay was less than 5% for indomethacin concentrations in the range of 0.1 to 5.0 mg/L. Pharmacokinetic model. A one-compartment open model with first-order elimination and zero-order input was used to describe the concentration-time course of indomethacin. The PREDPP program subroutines ADVAN1 and TRANS2 were selected to define this model 22 The pharmacokinetic model was written in a recursive form for intravenous bolus administration as described by Beal et al.22 No steady-state situations for indomethacin were assumed. Statistical model. The interindividual variabilities in clearance (CL) and volume of distribution (Varea)
were modeled with proportional error according to the following equations: CL, = TVCL(1 + Varea,j
TIJCL)
= TVV.,a(1 +
my)
Va,i are the true (unknown) pharmacokinetic parameters for the jth individual as predicted by the regression model, TVCL and TVVa,a are the population means of CL and Vaa, lliCL and are identically distributed random variables with zero mean and variances w2cL and (02V, respectively, that distinguish the jth individual's CL and Varea from those predicted by the regression model. The intraindividual variability, or residual error, was modeled with additive error: in which CLJ and
=
+
in which Cii is the ith observed concentration for the jth individual and Cmij is the ith observed concentration predicted by the model for the jth individual. This value includes the contribution of the and terms; eij is the residual intrapatient variability term, representing independent identically distributed statistical errors with mean zero and variance cr2. Data analysis. All data were analyzed on a VAX 11/785 computer (Digital Equipment Corp., Maynard, Mass.) with the NONMEM program version 2, level 1.4, with double precision. Estimates were sought for (1) population means of CL and Va,a, (2) variance (032) of and TIN/area, (3) variance (o-2) of the residual random error (e), (4) standard errors of the parameter estimates, and (5) value of the objective function, which is equal to minus twice the log likelihood of the data lnL).
(2
VOLUME 49 NUMBERS
Indomethacin in neonates
553
Table III. Development of full regression model: hypotheses tested Hypotheses tested Clearance Did gender affect CL? Did GA 5_30 wk affect CL? Did PNA affect CL? Did 1-min Apgar scores -55 affect Did 5-min Apgar scores :58 affect Did PDA closure affect CL? Did serum creatinine affect CL? Did concomitant furosemide affect Volume of distribution Did gender affect wk affect Va? Did GA Did PNA affect Did 1-min Apgar scores 55 affect Did 5-min Apgar scores 58 affect Did PDA closure affect Varea? Did serum creatinine affect Did concomitant furosemide affect
Difference in 2-1nL versus base model
p Value
Conclusion addressed
0.042 2.953 50.54 0.180 4.672 27.905 21.312
NS NS
No No Yes No No Yes Yes Yes
CL? CL?
CL?
Varea? Varea?
V,?
Varea?
Critical value = 6.6 (x2, p = 0.01; df = 1). CL, Clearance; NS, not significant; GA, gestational age;
V,
apparent volume
Minimizing the objective function provided by each NONMEM run is equivalent to maximizing the likelihood of the data. Hypothesis testing can be performed by monitoring changes in the objective function when one or more parameters in the model are first estimated iteratively and then restricted to a fixed value. This difference is approximately x2 distributed with degrees of freedom equal to the number of parameters with fixed values in the restrictive model. Development of the final regression model. The first stage in the model-building phase was initiated with a minimum number of parameters that were suspected to influence indomethacin disposition. Alternate statistical models were tested during this phase to determine which model would afford the best fit of the data. Careful examination of scatterplots of weighted residual (WRES) versus time and WRES versus predicted concentration guided the selection of the most appropriate statistical model. Additional parameters were incorporated into the initial regression model in a stepwise fashion to develop the full regression model. Any fixed effect that reduced the objective function by more than 6.6 (x2, p
