To support safe and effective use of propofol in nonventilated children after major surgery, a model for propofol pharmacokinetics and pharmacodynamics is described.
After craniofacial surgery, 22 of the 44 evaluated infants (aged 3-17 months) in the pediatric intensive care unit received propofol (2-4 mg . kg-1 . h-1) during a median of 12.5 h, based on the COMFORT-Behavior score. COMFORT-Behavior scores and Bispectral Index values were recorded simultaneously. Population pharmacokinetic and pharmacodynamic modeling was performed using NONMEM V (GloboMax LLC, Hanover, MD).
In the two-compartment model, body weight (median, 8.9 kg) was a significant covariate. Typical values were Cl = 0.70 . (BW/8.9)0.61 l/min, Vc = 18.8 l, Q = 0.35 l/min, and Vss = 146 l. In infants who received no sedative, depth of sedation was a function of baseline, postanesthesia effect (Emax model), and circadian night rhythm. In agitated infants, depth of sedation was best described by baseline, postanesthesia effect, and propofol effect (Emax model). The propofol concentration at half maximum effect was 1.76 mg/l (coefficient of variation = 47%) for the COMFORT-Behavior scale and 3.71 mg/l (coefficient of variation = 145%) for the Bispectral Index.
Propofol clearance is two times higher in nonventilated healthy children than reported in the literature for ventilated children and adults. Based on the model, the authors advise a propofol dose of 30 mg/h in a 10-kg infant to achieve values of 12-14 on the COMFORT-Behavior scale and 70-75 on the Bispectral Index during the night. Wide pharmacodynamic variability emphasizes the importance of dose titration.
TO correct craniosynostosis, most infants undergo surgery in the first years of life. Because of edematous eyelids, separation from parents and the need to stay in the intensive care unit for control of vital signs, and the possible development of neurologic sequelae, these children often experience stress postoperatively. Although propofol is widely used for sedation in the adult intensive care unit, its use is subject to debate in sedated children in the pediatric intensive care unit since the report of five deaths in children receiving high doses (> 5 mg · kg−1· h−1) of propofol.1In general, larger doses of propofol are required in children, and it is suggested that this is because of differences in pharmacokinetics,2sensitivity,3or both.
To date, there are no population models in children investigating the pharmacodynamics to study the variability between and within children. As a pharmacodynamic endpoint, a number of clinical sedation scores have been devised for use in children, in which the COMFORT-Behavior (COMFORT-B) scale4,5would be a reliable alternative to the original, most used COMFORT scale.6The Bispectral Index (BIS) may have benefits in comparison with clinical sedation scores because it assesses sedation continuously and may provide an objective, quantitative measure of the level of sedation.7However, to date, the BIS has only been validated in children older than 1 yr.
Our clinical experiences regarding the use of propofol evaluated by the COMFORT-B in young children in the pediatric surgical intensive care unit (PSICU) have recently been published by Prins et al. 8In the current article, propofol pharmacokinetics and pharmacodynamics characterized by use of the COMFORT-B and BIS on the postoperative sleep pattern in nonventilated infants are described using population modeling, to select appropriate doses in infants and to support the safe and effective use of propofol.
Materials and Methods
The study was performed in the PSICU of the Erasmus Medical Center–Sophia Children's Hospital, Rotterdam, The Netherlands. The study protocol was approved by the ethics committee of the Erasmus Medical Center–Sophia Children's Hospital. Written informed consent was obtained from the parents. The studied infants, the design, the sedative and analgesic regimen, and safety parameters are presented in detail in the article of Prins et al. 8and are shortly repeated as relevant to this article.
Patients
Eligibility criteria included major craniofacial surgery, age between 1 month and 2 yr, and postoperative admission to the PSICU. The children were randomly allocated to receive propofol or midazolam if sedative medication was judged necessary on the basis of the COMFORT-B score (score ≥ 17). Infants were excluded when they had respiratory infections, epilepsy, hypertriglyceridemia or family histories of hypercholesterolemia, or history of allergy to propofol, eggs, or soybean oil.
Patient characteristics of the group in which no sedation was necessary (nonagitated group) and the group in which sedation was needed (agitated group) are presented in table 1. Infants who received midazolam could be used for pharmacokinetic and pharmacodynamic analysis before midazolam administration if more than two COMFORT-B observations were available for the description of the postoperative sleep pattern in the agitated group. These infants are represented in table 1as the agitated, no sedative group. All patients had normal hepatic and renal functions.
Anesthesia
Standardized anesthesia was induced with thiopental (5 mg/kg) or sevoflurane and fentanyl (2.5 μg/kg), and the infants were paralyzed with vecuronium (0.1 mg/kg). Thereafter, the infants underwent intubation and mechanical ventilation. Anesthesia was maintained with isoflurane, oxygen, and air, and fentanyl was given as needed. A median total dose of 17.9 (10.0–32.9) μg/kg fentanyl was administered during surgery. Approximately 2 h before extubation, a loading dose of acetaminophen (40 mg/kg) was administered rectally. After the operation, the patients were admitted to the PSICU for a minimum of 24 h, depending on the clinical condition.
Sedative and Analgesic Regimen
Pharmacodynamic data collection was started at arrival at the PSICU. The COMFORT-B score, which has been validated in pediatric intensive care, was used as a pharmacodynamic endpoint.4,5The COMFORT-B scale assesses six behavioral items: alertness, calmness, muscle tone, body movement, facial tension, and crying (nonventilated children) or respiratory response (ventilated children). All items range from 1 (no distress) to 5 (severe distress), resulting in a total score varying from 6 to 30. The interobserver reliability represented by linearly weighted κ was greater than 0.65 for all nurses and the principal investigator. In addition, the BIS was recorded continuously and noted at 15-min intervals (BIS® A 2000 version 3.12; Aspect Medical Systems, Natick, MA; with pediatric BIS® sensors). The BIS ranges from 100 (awake) to 0 (isoelectric electroencephalogram). Propofol, 6% (Department of Clinical Pharmacy, St. Antonius Hospital, Nieuwegein, The Netherlands9,10), was given by a central venous line into a running saline infusion by a B. Braun Medical infusion pump (Melsungen, Germany) to a summed rate of 3 ml/h. For propofol, the doses were increased or decreased as needed up to a maximum of 4 mg · kg−1· h−1. When patients were inadequately sedated with 4 mg · kg−1· h−1propofol, midazolam was added. One patient received an additional dose of 0.1 mg/kg followed by 0.05 mg/kg midazolam, one patient received two doses of 0.1 mg/kg, and two patients received a single bolus of midazolam. These responses were excluded from the analysis. To determine whether restlessness was induced by pain, the trained nurses also obtained the visual analog scale score. Patients received the standard four daily doses of 120–240 mg acetaminophen rectally.11In more than 99% of the observations, the visual analog scale score was less than 4.
Blood Sampling
Arterial samples (250 μl) were taken before the start of the propofol infusion; at approximately 30 or 45, 60 or 90, and 120 min after the start of the propofol infusion; three times in steady state; just before and 1 h after dose adjustment; just before stopping; and 15 or 30, 45 or 60, 120, and 150 min after the end of the infusion.
Analytical Methods
Propofol concentrations were measured in whole blood using high-performance liquid chromatography with fluorescence detection as described in a previous study from our laboratory.3,12Blood samples were collected in oxalate tubes and stored at 4°C until analysis (within 1 week). The limit of quantification was 0.035 mg/l, and the between-day coefficients of variation were less than or equal to 6.0%.
Genomic DNA was isolated from EDTA blood (MasterAmp; Epicenter Technologies, Madison, WI). Cytochrome P450 (CYP) 2B6 mutations 516G>T, 785A>G, and 1459C>T were analyzed (alleles *4, *5, *6, *7, and *9). Polymerase chain reaction–restriction fragment length polymorphism analyses were performed as described previously13with the exception of using Bst NI as restriction enzyme instead of Sty I. Analysis for the 1459C>T polymorphism was performed using primers 5′-CTGTTGCAGTGGACATTTG-3′: and 5′- ATCTCACTCCTGCACTCAC-3′ in a polymerase chain reaction with an initial step of 7 min at 94°C, followed by 30 cycles of (1 min at 94°C, 1 min at 57°C, 1 min at 72°C) and concluded with a final extension step of 6 min at 72°C. The polymerase chain reaction product was digested with Bgl II. CYP2C9*2 and *3 and CYP2C19*2 and *3 analyses were performed on the LightCycler® (Roche Diagnostics, Mannheim, Germany), using the CYP2C9 and CYP2C19 kits (Roche Diagnostics), respectively.
Data Analysis
The Non-Linear Mixed Effect Modeling (NONMEM) program (version V; GloboMax LLC, Hanover, MD)14was used for population analysis. S-plus (version 6.2; Insightful software, Seattle, WA) was used to visualize the data. NONMEM estimates the mean pharmacokinetic and pharmacodynamic parameters of the population and the interindividual variability and the residual error, minimizing the objective function (−2 log likelihood). The NONMEM option of the first-order conditional estimation (method 1) with η–ϵ interaction was used. Model development was performed in four steps: (1) choice of the structural pharmacokinetic or pharmacodynamic model, (2) choice of the residual model, (3) covariate analysis, and (4) internal validation of the model. Discrimination between different models was made by comparison of the objective function. A value of P < 0.005, representing a decrease of 7.8 in the objective function, was considered statistically significant. In addition, the diagnostic plots (observed vs. individually predicted, observed vs. population predicted, time vs. weighted residuals, and population predictions vs. weighted residuals) for examining bias and precision, the confidence interval of the parameter estimates, the correlation matrix, and visual improvement of the individual plots were used to evaluate the model.
Covariate Analysis
Covariates were plotted independently against the individual post hoc parameter estimates and the weighted residuals to identify their influence. Tested covariates were body weight, age, body surface area, body mass index (if height was known), and sex. The pharmacokinetic parameters were also tested for correlation with heart frequency, blood pressure, triglycerides, and the CYP isoforms (2B6*4, *5, *6, *7, *9, 2C9*2, 2C9*3, 2C19*2, 2C19*3). In addition, the influence of the total dose of fentanyl administration during surgery on the pharmacodynamics was assessed.
Potential covariates were separately incorporated, and a significant covariate that most reduces the objective function was left in the model. Additional covariates had to reduce this objective function further to be retained in the model. The choice of the model was further evaluated as discussed above.
Validation
A bootstrap resampling method was used to assess the stability of the parameter estimates and the robustness of the final model.15A bootstrap involves repeated random sampling to produce another data set of the same size but with a different combination of individuals. The mean parameter values and coefficients of variation (CVs) of the bootstrap replicates were compared with the estimates of the original data set.
Pharmacokinetic Model
The parameters of a two-compartment model were fitted to the log-transformed data and parameterized in terms of volume of steady state (Vss), volume of the central compartment (Vc), clearance (Cl), and intercompartmental clearance (Q) using subroutine ADVAN 5. The central volume was related to the volume of distribution at steady state as
The individual value of the parameters of the i th subject was modeled by
where θmeanis the population mean and ηiis assumed to be a Gaussian random variable with zero mean and variance ω2. The residual error was described with a proportional error model. This means for the j th observed log-transformed concentration of the i th individual the relation (Yij):
where cpredis predicted transformed propofol concentration and ϵijis a random variable with mean zero and variance ς2.
Simulation
Pharmacodynamic Model
Depth of sedation was characterized with postoperative natural sleep pattern (PNSP) and propofol effect (PEF).
where Sijis the j th observed sedation level in the i th subject.
The postoperative natural sleep pattern (PNSP) was described as a function of three equations allowing the depth of sedation to increase and decrease during the first postoperative night in the absence of a sedative.
in which BSL represents the level of sedation at arrival at the PSICU, PAEFF represents the postanesthesia effect, and CNR represents the circadian night rhythm.
For estimation of the interindividual variability of the baseline, log-normal distributions were assumed. This means for the i th individual:
where BSLmeanis the population mean and ηiis a Gaussian random variable with zero mean and variance ω2.
Postanesthesia effect (PAEFF) was assumed to wash out in time postoperatively by an Emaxmodel, resulting in a decrease of the depth of sedation to a maximum estimated score (Smax) for the COMFORT-B and 100 (awake) for the BIS.
where PAEmaxis the maximal effect from BSL to the maximal score Smax, TPSis the time (minutes) postsurgery, T50,PSis the time (minutes) postsurgery at half maximum postanesthesia effect, and γ is the steepness of the time-versus -response relation. Interindividual variability of T50,PSand γ were assumed to be log-normally distributed.
Circadian night rhythm (CNR) was modeled by
where O denotes the onset of the natural night dip in minutes from 12:00 h. The end of the circadian night dip (wake-up time) was assumed at 7:00 h, because at this time point, the light is turned on, nursing care is optimized, and parents arrive at the PSICU. A (COMFORT-B or BIS units) is the amplitude of the night dip, and 2π/Fr (minutes) is the frequency of the oscillations.
Propofol effect (PEF) was related to the pharmacokinetic model–predicted individual propofol concentration (Cij) by a simple Emaxmodel:
where Emax,iis the maximum possible propofol effect (equal to Smax− 6 on the COMFORT-B scale and 100 on the BIS scale) in the i th subject, assuming that the response will reach the maximum effect at doses sufficiently higher than 4 mg · kg−1· h−1propofol. EC50is the propofol concentration (mg/l) at half maximum effect, in which the interindividual variability was assumed to be log-normally distributed. The residual error in the COMFORT-B score and BIS was best characterized by a proportional and an additive error model, respectively.
where Yijrepresents the observed effect in the i th subject at the j th time point.
Results
A median of 11 blood samples per infant were collected from 22 evaluable propofol patients. The pharmacokinetics of propofol were best described with a two-compartment model. In some of the patients, the central line had not been primed, for which we added a lag time (ALAG) for a subpopulation to the model to describe the delay of delivery. Body weight (median, 8.9 kg) incorporated as a power function was found to be a significant covariate for elimination clearance, thereby reducing the interindividual variability (CV%) in clearance from 27% to 20%. A slope-intercept model or a weight-proportional model resulted in the same decrease in objective function. The addition of other covariates (arterial blood pressure, heart frequency, triglycerides, CYP isoforms [2B6 *5, *6, *7, 2C9*2, 2C9*3, 2C19*2], age, body mass index, body surface area, and sex) to the model did not improve the quality of fit. The pharmacokinetic parameter values and precision of the basic model, the body weight power-adjusted model, and the values obtained from the bootstrapping are shown in table 2. The fits of 250 bootstrap replicates of the data set demonstrated the stability of the model. Individual fits of the model for a median situation and the most biased situation of the final model (body weight power model) to the observed data are shown in figure 1.
Simulations
The simulations using the pharmacokinetic model previously developed by Knibbe et al. ,3Rigby-Jones et al. ,16and Schüttler and Ihmsen17overestimated the observed propofol concentrations in our patients (fig. 2), indicating that the pharmacokinetics in our study population of awake children are distinctly different.
Pharmacodynamics
The total data set included a median of 15 (3–25) COMFORT-B scores and 73 (3–101) BIS observations per infant from 21 propofol patients, 9 natural sleep patients who received no sedative, and 13 natural sleep patients until midazolam administration. Table 3summarizes the estimated pharmacodynamic parameters for the full model (postoperative natural sleep pattern and propofol effect) for the COMFORT-B and the BIS. All infants arrived comfortable and lightly sedated at the PSICU (BSL), starting with a COMFORT-B score of 10.4 (CV 16%) and a BIS value of 79 (CV 7%). In the agitated infants during the postoperative night, the narcotic effect washed out earlier, indicated by a smaller T50,PS(518 vs. 1,580 min for the COMFORT-B and 1,044 vs. 2,052 min for the BIS). The steepness value of the washout effect (γ) for the BIS was 8, whereas the steepness for the COMFORT-B was not found to be significantly different from 1. During the night, the infants were “deeper” asleep, which was implemented in the model using the dip of a circadian rhythm. The start of the dip was estimated at 20:00 h (equal to 480 min from 12:00 h) on the COMFORT-B with an amplitude of 3.5 units and 17:30 h (equal to 330 min) on the BIS with an amplitude of 14.5. For the agitated infants receiving propofol during the night, a night dip could not be estimated. Propofol was started at a median time of 19:00 h, which is equal to 5.5 h after surgery. The induced BIS depression as a function of the propofol concentration showed considerable intersubject variability (CV 145%). The bootstrap validation (100 times) confirmed the precision of the parameters. Figure 3Ashows a median fit of a nonagitated infant who received no sedative, with a reduction in response during the night. Figures 3B and Cshow a median and a worse fit of the sleep pattern of an agitated infant and the influence of propofol. Figure 4illustrates the simulated relation among time, propofol infusion rate, propofol concentration, and predicted population response in terms of depth of sedation using COMFORT-B and BIS. The difference between a 10-kg infant and a 5-kg infant is shown at the infusion rate of 18 mg/h. The difference in postoperative natural sleep pattern between infants who did or did not become agitated is shown at the propofol infusion rate of 0 mg/h.
There was not enough evidence to support sex, age, body weight, and total dose of fentanyl during surgery as covariates on the pharmacodynamic parameters.
Discussion
To support safe and effective use of propofol during the first night after major surgery in nonventilated infants younger than 1.5 yr, a population model for the influence of propofol pharmacokinetics and pharmacodynamics on the depth of sedation was described, assessed using COMFORT-B and BIS.
Clearance in postsurgical healthy nonventilated infants was found to be two times higher than reported in the literature for ventilated children and adults.3,16,17Based on the pharmacokinetic model, propofol doses must be doubled in this pediatric group to obtain similar blood concentrations. We believe that the higher estimate of the Cl (0.70 l/min) in an infant with a body weight of 8.9 kg (2.64 l/min standardized to an adult of 70 kg) in our study compared with 0.27 l/min (0.030 l · kg−1· min−1) reported in the literature can be partly explained by the effect of the surgery and the condition of the patients. Rigby-Jones et al. 16found that patients aged 1 week to 12 yr undergoing cardiac surgery had reduced values for metabolic clearance (−26%). Cardiac patients in general show a reduced cardiac output, which may effect the propofol elimination because the clearance of propofol (a high-extraction drug) is dependent on liver blood flow. In addition, mechanical ventilation may be of influence on the clearance of propofol. In patients with trauma and those in the surgical intensive care unit, increasing the positive end-expiratory pressure during mechanical ventilation has been shown to decrease total hepatic blood flow.18Murat et al. 2reported a large clearance of 0.44 l/min (0.049 l · kg−1· min−1) in spontaneously breathing children aged 1–3 yr with minor burns after a single dose of 4 mg/kg. Healthy ventilated children undergoing anesthesia did show a lower estimate of the clearance.17,19The model developed by Schüttler and Ihmsen17for healthy ventilated children undergoing anesthesia from 2 yr of age showed less overprediction of the blood concentration than the model developed by Knibbe et al. 3and Rigby-Jones et al. 16for ventilated children after cardiac surgery. They also found a smaller value of the central volume compared with our model (5–12 vs. 19 l), which may be a consequence of the relation of the central volume to the volume of distribution of steady state. Body weight partially explained the interpatient variability in Cl. The influence of a slope-intercept model, a proportional model, or a power model with a power scaling parameter of 0.61 on the clearance was comparable in the range of 4.8–12.5 kg. We choose the power model because an allometric three-fourths power model has been used with success for interspecies scaling.20As with other studies, age was not found to be a significant covariate.16,21In addition, the genetic expression of the investigated CYP isoforms did not explain the interindividual differences in the clearance. 2B6 would be predominantly involved and, at a lower rate, 2C9 and 2C19 in the minor metabolic hydroxylation pathway.22The homogeneous patient characteristics and the relatively small number of patients may account for the unexplained interpatient variability.
The large pharmacodynamic interindividual variability and residual error in BIS and COMFORT-B emphasize the complexity of depth of sedation in infants. Young children can vary in depth of sedation in the absence of sedatives as a result of day–night rhythm, the presence of parents and medical staff, hunger, light, and noise.5,23Especially at lighter sedation levels, noise has a greater effect on the BIS.24To account for natural variation, data of infants not receiving a sedative and until sedative administration were used to describe a postanesthesia effect (PAEFF) and a night dip (CNR). For adults, a similar PAEFF has been described after coronary artery bypass grafting by assuming a virtual drug that washes out over time.25Because stress and severe discomfort entail risks, a complete natural sleep pattern of agitated infants could not be described. The administration of the sedative may cover the night dip, which could not been estimated in the agitated children. The EC50of propofol for the reduction of the BIS was different from that of the COMFORT-B, indicating that the two measurements are not interchangeable measures of the propofol effect in a spontaneously breathing child. Courtman et al. 26and Crain et al. 27also suggest that the BIS and COMFORT are only moderately correlated: A child can be comfortable but fully awake. The use of the BIS has the advantage that it assesses sedation continuously and may allow more objective assessment of sedation. It gives additive information and can be useful for patients who are difficult to assess clinically. The use of the maximal estimated score of 20 on the COMFORT-B scale and a smaller number of observations make it difficult to determine which sedation scale is more sensitive in this population based on the EC50, but in lightly sedated children, the COMFORT-B seems more advantageous. The COMFORT-B has never been used before as a pharmacodynamic instrument in a pharmacokinetic–pharmacodynamic propofol analysis, but the effect of propofol on BIS in adults has been described. Interestingly, the sensitivity of infants to propofol, defined as EC50, seems comparable to that in adults. Defining the Emaxas the maximum effect seen on the BIS, Bouillon et al. 28estimated an EC50of 3.07 mg/l (CV 12.1%), and Doufas et al. 29estimated a value of 2.4 mg/l (CV 30%). By fixing the Emaxto 100, Calvo et al. 30estimated the EC50on 3.91 mg/l (41%), which may suggest that infants only require higher doses because of differences in pharmacokinetics rather than pharmacodynamics. In general, the sensitivity to propofol between infants is quite variable. Unfortunately, no explanation could be found based on patient characteristics as age, body weight and sex. In this narrow age group, the potential stressful environment resulting from inability to see, separation from parents, and unknown voices may play a major role.
Based on the population pharmacodynamic model, we advise a propofol infusion rate of 30 mg/h for a 10-kg nonventilated infant to achieve a COMFORT-B score between 12 and 14, 6 h after surgery during the night, which corresponds to BIS values of 70–75 (fig. 4). The considerable variability emphasizes the importance of drug titration to a maximum of 4 mg · kg−1· h−1. Further pharmacodynamic studies in larger groups of children are needed to explain the variability in response and help clinicians to improve individualization. For drugs such as propofol, this is especially important because of the troublesome reports in the literature regarding the safety of the use of propofol in children beyond procedures.
The authors thank the staff hospital pharmacists of the Division of Pharmacy of the Erasmus Medical Center, Rotterdam, The Netherlands, for their help and cooperation, in particular Arnold G. Vulto, Pharm.D., Ph.D. (Professor of Hospital Pharmacy and Practical Pharmacotherapy), Lidwien M. Hanff, Pharm.D., and Ron A. A. Mathôt, Pharm.D., Ph.D., and the medical and nursing colleagues of the Pediatric Surgical Intensive Care Unit of the Erasmus Medical Center–Sophia Children's Hospital, Rotterdam, The Netherlands.