Numerous pharmacokinetic models have been published aiming at more accurate and safer dosing of dexmedetomidine. The vast majority of the developed models underpredict the measured plasma concentrations with respect to the target concentration, especially at plasma concentrations higher than those used in the original studies. The aim of this article was to develop a dexmedetomidine pharmacokinetic model in healthy adults emphasizing linear versus nonlinear kinetics.
The data of two previously published clinical trials with stepwise increasing dexmedetomidine target-controlled infusion were pooled to build a pharmacokinetic model using the NONMEM software package (ICON Development Solutions, USA). Data from 48 healthy subjects, included in a stratified manner, were utilized to build the model.
A three-compartment mamillary model with nonlinear elimination from the central compartment was superior to a model assuming linear pharmacokinetics. Covariates included in the final model were age, sex, and total body weight. Cardiac output did not explain between-subject or within-subject variability in dexmedetomidine clearance. The results of a simulation study based on the final model showed that at concentrations up to 2 ng · ml–1, the predicted dexmedetomidine plasma concentrations were similar between the currently available Hannivoort model assuming linear pharmacokinetics and the nonlinear model developed in this study. At higher simulated plasma concentrations, exposure increased nonlinearly with target concentration due to the decreasing dexmedetomidine clearance with increasing plasma concentrations. Simulations also show that currently approved dosing regimens in the intensive care unit may potentially lead to higher-than-expected dexmedetomidine plasma concentrations.
This study developed a nonlinear three-compartment pharmacokinetic model that accurately described dexmedetomidine plasma concentrations. Dexmedetomidine may be safely administered up to target-controlled infusion targets under 2 ng · ml–1 using the Hannivoort model, which assumed linear pharmacokinetics. Consideration should be taken during long-term administration and during an initial loading dose when following the dosing strategies of the current guidelines.
Dexmedetomidine pharmacokinetic models underpredict the measured plasma target-controlled infusion concentrations that are higher than those used in the model validation studies
The elimination clearance of high hepatic extraction ratio drugs like dexmedetomidine is determined by liver blood flow and not enzyme activity
The data of 48 subjects from two published pharmacokinetic studies were pooled to build a three-compartment pharmacokinetic model with nonlinear elimination clearance that successfully predicted plasma dexmedetomidine concentrations over a wide concentration range
Cardiac output did not explain between-subject or within-subject variability in dexmedetomidine elimination clearance
Dexmedetomidine elimination clearance may decrease with increasing plasma concentrations because it alters the liver blood flow–to–cardiac output ratio in a concentration-dependent manner
Dexmedetomidine is widely known for its anxiolytic, sedative, and analgesic effects due to its central α2- adrenergic agonistic properties, which provide a unique arousable sedation profile with minimal ventilatory effects when used in therapeutic ranges.1,2 Nowadays, dexmedetomidine is approved for short-term use in mechanically ventilated patients admitted to the intensive care unit (ICU) and for periprocedural sedation. A considerable amount of effort has been invested to elicit a deeper understanding and characterization of dexmedetomidine’s pharmacology among all ages and clinical conditions. As a result, numerous and heterogeneous pharmacokinetic (and pharmacodynamic) models have been published. These models aim to develop a more accurate and safer dosing regime that permits a faster onset of the desired effects while diminishing the risk for undesirable cardiovascular effects.3,4 Unfortunately, the vast majority of the developed models underpredict the measured plasma concentrations with respect to the target concentration, especially at higher plasma concentrations than those originally used in the validation studies. For example, Hsu et al.5 using the model developed by Dyck et al.,6 Snapir et al.7 using the model developed by Talke et al.,8 and Weerink et al.9 using the model developed by Hannivoort et al.10 underpredicted the measured plasma concentrations at higher targets. All aforementioned models include a two- or three-compartment mammillary model structure with first-order kinetics. Dutta et al.11 have suggested that dexmedetomidine’s pharmacokinetic profile exhibited nonlinear kinetics. During a recent dexmedetomidine trial conducted at our hospital, we noticed that there was bias in the predictions differing from the target-controlled infusion model for plasma concentrations above the 3.0 ng · ml–1 target-controlled infusion targets,9 indicating a potential nonlinear increase in exposure with increasing dexmedetomidine target concentrations. The aim of the article was to investigate whether dexmedetomidine clearance is nonlinear, to establish whether between-subject or within-subject differences in cardiac output explain this apparent nonlinearity, to determine the clinical importance of this phenomenon when following the current recommended dose regimen guidelines, and to establish the potential impact on the use of the currently available (linear) Hannivoort dexmedetomidine model.10
Materials and Methods
Study Design and Data Collection
The data of two previously published clinical trials,9,10 conducted in healthy subjects at the Department of Anesthesiology at the University Medical Center Groningen (Groningen, The Netherlands) were pooled to build a pharmacokinetic model. Both trials were conducted in accordance with the Declaration of Helsinki and in compliance with good clinical practice and the applicable regulatory requirements. Both trials explored the effects of increasing target concentrations of dexmedetomidine stepwise using target-controlled infusion. Ethical approval was obtained from an independent medical ethics review committee (Medisch Ethische Toetsings Commissie) by the Association for the Evaluation of Ethics in Biomedical Research (Stichting Beoordeling Ethiek Biomedisch Onderzoek, Assen, The Netherlands). Both studies were registered in the ClinicalTrials.gov database (NCT03143972 and NCT01879865). Written informed consent was obtained before performing a standard medical screening, which included a thorough medical history and a physical examination. Exclusion criteria included a body mass index greater than 30 kg · m–2 or less than 18 kg · m–2, an age less than 18 yr or above 70 yr, pregnancy, or currently breastfeeding. Alcohol consumption, smoking, and use of concomitant (illegal) drugs were not allowed during study participation. Inclusion was stratified according to age and sex in both studies. Subjects were in fasting conditions during each study period.
Trial by Hannivoort et al.
For the first study, 18 subjects were included. For a thorough description of the study demographics, data collection, and analytic methods, please refer to Hannivoort et al.10 In short, during the trial, an initial 20-s 6-µg · kg–1 · h–1 infusion was started. Ten minutes after the start of this infusion, a dexmedetomidine target-controlled infusion was started with stepwise increasing targets of 1, 2, 3, 4, 6, and 8 ng · ml–1. Each target was maintained for 30 min using the model developed by Dyck et al.6 Arterial blood samples were obtained at baseline, 2 min after the initial 20-s infusion, before each increase in target concentration (at 10 min and every 30 min thereafter), during the target- controlled infusion administration period, and at 0, 2, 5, 10, 20, 60, 120, and 300 min during the postinfusion period. To avoid hypertension, the infusion rate was limited to 6 µg · kg–1 · h–1 for the first three infusion steps and was increased to 10 µg · kg–1 · h–1 for the two highest target-controlled infusion targets of 6 and 8 ng · ml–1. The upper and lower limits of quantification for the analytical method were 0.02 and 20 ng · ml–1.
Trial by Weerink et al.
In the second study, 30 subjects were included. For a thorough description of the study demographics, data collection, and analytic methods, please refer to Weerink et al.9 Only data from the first period (dexmedetomidine alone and not in combination with remifentanil) were used for the current analysis. Target-controlled infusion was started with stepwise increasing targets of 1, 3, 4, 5, and 8 ng · ml–1 every 40 to 50 min using the previously developed dexmedetomidine model by Hannivoort et al.10 Arterial blood samples were drawn at baseline and at pseudo-steady state before changing the target concentration. Once the infusion was stopped, dexmedetomidine samples were drawn at 2, 5, 10, 20, 60, 120, 300, and 420 min. Dexmedetomidine infusion was limited to 6 µg · kg–1 · h–1 for the first three infusion steps and was increased to 10 µg · kg–1 · h–1 for the two highest targets of 5 and 8 ng · ml–1. The protocol was, however, amended to lower the highest dexmedetomidine target from 8 to 6 ng · ml–1 due to safety concerns. The upper and lower limits of quantification for the analytical method were 0.05 and 20 ng · ml–1.
Study Drug Administration and Data Collection
Dexmedetomidine administration was conducted using RUGLOOP II software (Demed, Belgium) using a syringe pump with an Orchestra module DPS (Orchestra Base A, Fresenius Kabi, Germany); the software also recorded all vital parameters, pump infusion rates, and eventual case report form annotations during the execution of the clinical trials. Continuous cardiac output was measured through a FloTrac sensor connected to a Vigileo monitor (both from Edwards Lifesciences, USA) in the trial by Hannivoort et al.10 and through a hemodynamic monitor (EV1000 monitor with FloTrac sensor; Edwards Lifesciences, USA) in the trial by Weerink et al.9
Pharmacokinetic Model Development
Data set creation, goodness of fit plots, and simulations were performed using R statistical software (R Foundation for Statistical Computing, 2018; version 3.5.0). Nonlinear mixed effects modeling was conducted using the NONMEM (version 7.3.0) software package (ICON Development Solutions, USA). In the first trial by Hannivoort et al.,10 there were concentrations below the limit of quantification limit. Values below the limit of quantification were handled using the M3 method according to Ahn et al.12
Model Building and Parameter Estimation
The set of differential equations were solved using a second-order approximation method (LAPLACE option) using the ADVAN13 subroutine. The a priori model assumed weight-based allometric scaling on all pharmacokinetic parameters using a fixed exponent of 1 on volume of distribution terms and an exponent of 0.75 on clearance terms.13 Interindividual variability was incorporated using a log-normal distribution. Residual variability was modeled using a proportional and additive error model. Candidate models were compared using the Akaike information criterion, and the model with the lowest Akaike criterion was taken forward. Covariates were screened by exploring plots of random effects versus covariates and were included in the model if inclusion resulted in a decrease in the Akaike criterion. The covariates that were tested for inclusion in the model were age (yr), total body weight (kg), height (cm), and sex. The derived parameters body surface area, body mass index, and lean body mass were also tested.
Testing Cardiac Output as a Covariate on Dexmedetomidine Clearance
The influence of cardiac output on clearance was tested according to equation 1. This equation allows simultaneous estimation of the effect of a covariate on between- and within- subject variability on a pharmacokinetic parameter.14
In equation 1, FCO represents the effect of cardiac output (CO) on dexmedetomidine clearance. In the model, FCO was a proportional factor on dexmedetomidine clearance. The parameters θBSV and θWSV are estimated to quantify the influence on dexmedetomidine clearance of between-subject variability in baseline cardiac output (CObaseline) compared to the median in the population (6.6 l · min–1) and within-subject longitudinal changes from baseline cardiac output. Cardiac output was included in the data set as a time-varying covariate with backward constant interpolation between observations.
In-sample predictive performance variability was evaluated according to Varvel et al.15 using the median performance error and the median absolute performance error as described in equation 2.
where CP indicates the dexmedetomidine plasma concentration, and PE indicates the performance error.
Model precision and parameter identifiability along with the variance of the parameters random effects were used to numerically assess the models. Goodness of fit plots and visual predictive check plots were used to visually assess the model performance. The visual predictive checks were constructed according to Bergstrand et al.16 and were based on 1,000 simulations. To graphically represent the difference in performance between the linear Hannivoort model and our final nonlinear model, individual level visual predictive checks were conducted for six subjects from our data set that had the highest median performance error values according to the linear Hannivoort model.
Visual predictive check simulations were based on a sample of 1,000 virtual subjects with characteristics randomly sampled from the multivariate distribution of observed patient characteristics from our study. Plasma target-controlled infusion simulations were based on the Hannivoort model that assumes first-order pharmacokinetics, taking into account the maximal recommended infusion rate of 6 µg · kg–1 · h–1 . Effect-site target-controlled infusion simulations were based on the ke0 from the dexmedetomidine pharmacokinetic–pharmacodynamic model for the Modified Observer’s Assessment of Alertness and Sedation (MOAA/S) score17 (ke0 = 4.85 · 10–2 min–1) and were based on the PKPD Tools for Excel package.18 The ICU simulations consisted of a loading dose of 1 µg · kg–1 over 10 min followed by a constant rate of 0.7 μg · kg–1 · h–1 for 1 h and, alternatively, followed by the maximum recommended constant rate infusion of 1.4 µg · kg–1 · h–1 over 24 h according to the European Medicinal Agency Summary of Medicinal Product Characteristics19 for sedation in the ICU.
Analysis Data Set
The final analysis data set included 48 subjects, 18 subjects from the trial by Hannivoort et al.,10 and 30 were from the trial by Weerink et al.9 From the trial by Hannivoort et al.,10 two sessions, separated by a washout, were included. From a total of 762 observations, 29 plasma concentration were below the lower limit of quantification in the trial by Hannivoort et al.10 There were no samples below the lower limit of quantification in the trial by Weerink et al.9 Dexmedetomidine concentrations ranged from 0.02 to 14.4 ng · ml–1. A stratified inclusion resulted in a balanced population: 24 subjects were male, and 24 subjects were female. The mean age was 42.2 yr (range, 18 to 70 yr). The mean total body weight was 74.8 kg (SD, 13.3 kg; and range, 49 to 110 kg), and height was 176.2 cm (SD, 10.7 cm; and range, 155 to 203 cm). In total, 17 subjects were between 18 and 34 yr old, 15 were between 35 and 50 yr old, and 16 were between 51 and 70 yr old . Cardiac output measurements were available at a resolution of 1 measurement every 10 s, resulting in a median of 728 (range, 194 to 1,375) cardiac output measurements per subject included in the data set. Figure S1 of Supplemental Digital Content (https://links.lww.com/ALN/C743) shows the measured cardiac output against the measured dexmedetomidine plasma concentrations for all subjects included in the analysis.
Structural Model Development
Linear versus Nonlinear Elimination Kinetics
A three-compartment mamillary model with linear pharmacokinetics from the central compartment was used as the starting point for the model development. This first model showed considerable bias in the highest predicted concentrations when compared to the measured observations (Supplemental Digital Content fig. S2, https://links.lww.com/ALN/C743). Although the linear pharmacokinetic model adequately described the data from the trial by Hannivoort et al.10 (the data set on which this model was developed, upper panel in Supplemental Digital Content fig. S2, https://links.lww.com/ALN/C743), it did not adequately describe the observations from the trial by Weerink et al.9 (lower panel in Supplemental Digital Content fig. S2, https://links.lww.com/ALN/C743). Subsequent modifications to this starting model are summarized in the model development log as presented in table 1.
Nonlinear dexmedetomidine clearance was tested in the model using equation 3:
where the clearance at a particular plasma concentration ranges between CL0 and CL0 · (1 – IMAX) depending on the plasma concentration of dexmedetomidine (Cp). C50 determines the concentration at which 50% of inhibition takes place, and the exponent (γ) determines the steepness of the relationship. IMAX represents the theoretical maximum inhibitory effect at which dexmedetomidine clearance decreases depending on the plasma concentration. In the tested models, IMAX was fixed to 1 (i.e., dexmedetomidine clearance decrease to 0 at very high concentrations) to avoid numerical difficulties with the parameter estimation. Inclusion of equation 3 in the three-compartment model resulted in a significant decrease in the Akaike criterion of 169 points. Figure 1 depicts the post hoc clearance for all subjects in the data set as a function of the dexmedetomidine plasma concentration. Supplemental Digital Content figure S3 (https://links.lww.com/ALN/C743) shows that this nonlinear kinetic model adequately described both data sets. Interoccasion variability was tested on all pharmacokinetic parameters but did not result in a decrease in the model Akaike criterion. The additive error term in the model for the trial by Weerink et al.9 was negligible and was therefore removed.
Covariates were tested after accepting the model with nonlinear elimination and once interindividual variability was incorporated. In our final model, as shown in equations 4 to 8 and table 2, the volumes of distribution (V1, V2, and V3) scaled linearly with FSIZE, the intercompartment clearance (Q1 and Q2) scaled with size according to an exponent of 0.75, and CL0 scaled with FSIZE according to an exponent of 0.75 and decreased with increasing age as function of FAGE.
The estimates of the final model can be found in table 2.
Figure 2 presents the goodness of fit plots of the final model, which showed no clear systematic trend or bias, confirming the adequacy of the model to describe the concentration-time profiles of dexmedetomidine. The final (log-)likelihood profiles can be found in Supplemental Digital Content figure S4 (https://links.lww.com/ALN/C743). Supplemental Digital Content figure S5 (https://links.lww.com/ALN/C743) depicts the prediction- and variance-corrected visual predictive check for the final model. The parameter uncertainty was considered acceptable and ranged between 4 and 32.7% for the fixed effects and 30.5 to 40.8% for the random effects. The final model had a median absolute performance error of 21.7% and a median performance error of –2.3%. Figure 3 presents the individual-level visual predictive check for the linear pharmacokinetic model (blue lower panels) and the nonlinear pharmacokinetic model (gray upper panels) for six subjects from our data set. In this graph, it is clear that the highest concentrations are not contained in the 95% prediction interval for the linear pharmacokinetic model.
We found that females had a higher estimated FSIZE by a factor of 1.18 compared to males (equation 4; change in Akaike criterion = –11). In addition, age was a significant covariate on CL0 (change in Akaike criterion = –7), and CL0 decreased with age as shown in equation 5. On average, for every 10 yr of age, the clearance decreased by 5.1%. There were no significant correlations between C50 and the tested covariates.
We found that cardiac output did not explain between-subject or within-subject variability in dexmedetomidine clearance. Implementation of equation 1 in the model did not significantly decrease the objective function value (change in Akaike criterion = +1.2; P value for likelihood ratio test = 0.25 for 2 degrees of freedom). The estimated parameters for θBSV and θWSV were 0.0211 and 0.0037 and indicate that the estimated effect of cardiac output on clearance is negligible. According to equation 1, these estimates denote (1) that an individual with a cardiac output that is 50% of the population median (6.6 l · min–1) has a dexmedetomidine clearance that is 6.7% lower compared to the typical clearance and (2) that a relative change from baseline cardiac output of 50% is associated with a 1.2% decrease in dexmedetomidine clearance. Prediction-variance–corrected visual predictive checks stratified by covariate for total body weight (Supplemental Digital Content fig. S6, https://links.lww.com/ALN/C743), sex (Supplemental Digital Content fig. S7, https://links.lww.com/ALN/C743), and age (Supplemental Digital Content fig. S8, https://links.lww.com/ALN/C743) can be found in the supplemental materials.
Simulation of Target-controlled Infusion Using the Hannivoort Model
To better illustrate the differences between the models and the possible implications in the clinical setting, we simulated target-controlled infusion administrations according to the linear pharmacokinetic model by Hannivoort et al.10 with increasing target concentrations. The aim of the simulation was to identify the concentration range in which predictions from our final nonlinear model would deviate from the expected concentrations according to a target-controlled infusion system based on the linear pharmacokinetic model by Hannivoort et al.10 Figure 4 shows the simulated plasma concentrations for the target- controlled infusion system based on the linear pharmacokinetic model by Hannivoort et al.10 and the predicted plasma concentrations and the 95% prediction interval for our final nonlinear model. The predicted concentrations from the final nonlinear model and the target-controlled infusion model are in good agreement until 2 ng · ml–1. Beyond this target concentration, a discrepancy is seen between the predicted concentrations according to the target-controlled infusion model and the predictions according to the nonlinear model with 62.1% of the simulated individuals being above the target at a target of 3 ng · ml–1 to 80.4% at 5 ng · ml–1. The inaccuracy of the linear target-controlled infusion model, expressed as the ratio between the predicted concentration according to our final nonlinear model and the target concentration, was 0.89 (95% prediction interval = 0.7 to 1.2) at 1 ng · ml–1, 0.97 (0.7 to 1.6) at 2 ng · ml–1, 1.1 (0.7 to 2.0) at 3 ng · ml–1, 1.2 (0.7 to 2.4) at 4 ng · ml–1, and 1.4 (0.8 to 2.9) at 5 ng · ml–1.
Another scenario that we explored was effect-site target- controlled infusion. Figure 5 shows effect-site target- controlled infusion for increasing effect-site targets and the resulting predicted plasma concentrations for the linear pharmacokinetic model by Hannivoort et al.10 (blue lines and shaded area) and our final nonlinear model (gray lines and shaded area). The median predicted CMAX was 1.28-fold higher (95% prediction interval: 0.4 to 2.8) at 0.1 ng · ml–1, 1.28-fold higher (0.4 to 2.5) at 0.2 ng · ml–1, 1.27-fold higher (0.4 to 2.2) at 0.5 ng · ml–1, 1.29-fold higher (0.4 to 2.2) at 1.0 ng · ml–1, 1.26-fold higher (0.4 to 2.0) at 2.0 ng · ml–1, and 1.22-fold higher (0.4 to 2.0) at 3.0 ng · ml–1.
Critical Care Guidelines Recommended Dosing Simulation
To further illustrate the implications of the nonlinear pharmacokinetics of dexmedetomidine we simulated the current European Medicines Agency (Amsterdam, The Netherlands)–approved dosing regimen for sedation in the ICU,19 consisting of a weight-based constant rate infusion (µg · kg–1 · h–1). The upper panel of figure 6 shows that the predicted CMAX according to our final nonlinear model reaches 3.2 ng · ml–1 (95% prediction interval, 2.2 to 4.8) after the initial loading dose, whereas the predicted CMAX according to the linear pharmacokinetic model by Hannivoort et al.10 is 1.5 ng · ml–1 (95% prediction interval, 1.1 to 2.0). The lower panel in figure 6 depicts the maximum recommended continuous administration of dexmedetomidine and shows a good agreement between the median predicted plasma concentrations for both models. However, figure 6 suggests that with time, the variability in the population increases, with the 95% prediction interval increasing from 0.95 to 2.02 ng · ml–1 1 h after the dose to 1.80 to 20.84 ng · ml–1 toward the end of the 24-h infusion period. At the end of the 24-h infusion, 62.8% of the simulated subjects had plasma concentrations above the median concentration predicted with the linear model, 24.3% twice as high and 7.3% five times higher than the concentrations predicted by the linear model.
In this pooled analysis, we have shown that dexmedetomidine administered to healthy subjects exhibits nonlinear clearance, which may lead to higher than expected plasma concentrations with increasing drug exposure. Dyck et al.6 were the first to suggest nonlinearity in dexmedetomidine pharmacokinetics. In line with Dyck et al.,6 Dutta et al.11 hypothesized that a dose-related reduction in cardiac output resulted in a decrease in hepatic blood flow and consequently a reduced dexmedetomidine clearance. Interestingly, Dutta et al.11 concluded that based on their data from 10 healthy volunteers, there was no statistically significant difference in the fit of a cardiac output independent and dependent pharmacokinetic model. In 2012, Iirola et al.20 reported a positive association between cardiac output and dexmedetomidine clearance in eight patients admitted to the ICU with a continuous infusion of dexmedetomidine. Unfortunately, Iirola et al.20 did not compare a model with cardiac output against a nonlinear pharmacokinetic model without cardiac output, like the model we are proposing. Such a comparison would have given an indication of whether cardiac output is a better surrogate than the dexmedetomidine plasma concentration for describing the nonlinear dexmedetomidine pharmacokinetic model. At the same time, it is noteworthy that the subjects in the study by Iirola et al.20 received concomitant medication known/suspected to decrease cardiac output (e.g., propofol), which might have confounded the association between dexmedetomidine plasma concentration and cardiac output.
Our analysis based on continuous cardiac output measurements in 48 healthy volunteers has shown that cardiac output is not a significant predictor for between-subject variability in dexmedetomidine clearance. We also showed that changes from baseline cardiac output are not correlated with within-subject longitudinal changes in dexmedetomidine clearance.
Bloor et al.21 studied the hemodynamic changes, including cardiac output, induced after IV dexmedetomidine bolus doses of 0.25, 0.5, 1.0, and 2.0 µg · kg–1 administered over 2 min. These authors found that cardiac output (and systolic blood pressure) in the lowest dose group was not different from placebo. In the trial by Hannivoort et al.10 and the trial by Weerink et al.,9 the dexmedetomidine infusion rate was limited to 6 µg · kg–1 · h–1 for target concentrations less than 5 ng · ml–1 and to 10 µg · kg–1 · h–1 for higher targets. These rate limitations imply that the bolus doses given by the target-controlled infusion system immediately after changing the target-controlled infusion target were limited to a rate of 0.20 µg · kg–1 and 0.33 µg · kg–1 per 2 min, respectively. In that respect, the absence of a significant dexmedetomidine-induced decrease in cardiac output in our data (as shown in Supplemental Digital Content fig. S1, https://links.lww.com/ALN/C743) is not surprising and completely in line with the seminal work by Bloor et al.21 The 0.5, 1.0, and 2.0 µg · kg–1 bolus doses studied by Bloor et al.21 were infused at higher infusion rates than what is recommended nowadays on the drug label (1 µg · kg–1 over 10 min, i.e., 6 µg · kg–1 · h–1), and the finding of decreased cardiac output in these dose groups is therefore less representative of the current use of dexmedetomidine.
Our model, most notably equation 3, implies that dexmedetomidine clearance decreases with increasing dexmedetomidine plasma concentrations. Saturation of metabolic clearance at higher concentrations is not expected to have an influence on dexmedetomidine clearance because dexmedetomidine is considered a high extraction ratio drug. According to the well stirred liver model, liver blood flow and not enzyme activity is the main determinant of clearance for high extraction ratio drugs. This reasoning is supported by experimental work by Kohli et al.22 and Wang et al.,23 who observed no influence of CYP2A6 metabolizer status on dexmedetomidine pharmacokinetics.
We hypothesize that dexmedetomidine alters the liver blood flow–to–cardiac output ratio in a concentration- dependent manner (in line with the other hemodynamic effects of dexmedetomidine21,24,25 ). Such a phenomenon could account for the apparent lack of correlation between cardiac output and dexmedetomidine clearance and the concentration-dependent nature of the decrease in clearance. Interestingly, a similar phenomenon was observed for propofol, another high extraction ratio drug. Peeters et al.26 demonstrated that in critically ill patients, there was no relationship between propofol clearance and cardiac output, whereas hepatic blood flow (measured by sorbitol) was positively correlated with propofol clearance. Although the analysis by Peeters et al.26 might have been confounded by the presence of extrinsic factors, such as comedication, their work supports our hypothesis that, in some cases, the liver blood flow–to–cardiac output ratio is not constant, leading to no (or poor) correlation between cardiac output and drug clearance.
Our final nonlinear three-compartment pharmacokinetic model successfully predicted dexmedetomidine concentrations over a wide concentration range. We found that weight, age, and sex are significant covariates for the pharmacokinetics of dexmedetomidine. In addition to weight-based allometric scaling, we identified the importance of incorporating age into the model to enhance accuracy of allometric scaling for the estimation of the clearance. Clearance was also inversely proportional to the age in several previous models.3 Despite the lack of a clear consensus as to whether age correlates with the clearance or volumes of distribution, three studies20,27,28 have reported positive results of age as covariate over a wide age range. With reference to sex, this is the first data set that demonstrates variation between females and males. Differences have also been reported in cardiac output and liver perfusion between females and males in addition to the body fat composition,29 which may account for the pharmacokinetic differences between the sexes. Supplemental Digital Content figure S9 (https://links.lww.com/ALN/C743) depicts the result of a simulation in subjects with different covariates, providing evidence that male subjects with increased age and a low total body weight might be subjects at risk of nonlinear pharmacokinetics.
Our final model assumes that clearance is completely inhibited at very high dexmedetomidine concentrations. This assumption was challenged by an attempt to estimate an IMAX parameter in the model, describing the maximum proportional decrease in dexmedetomidine clearance at very high concentrations. However, we encountered numerical difficulties when trying to estimate IMAX, and this approach was abandoned. According to our final model, the mean maximal decrease in dexmedetomidine clearance across individuals was 55.2% (range, 23.3 to 82.7%). To estimate an IMAX parameter from experimental data, higher dexmedetomidine concentrations than the ones reported in our study are likely necessary.
In our study, the plasma concentrations were from arterial blood samples taken during (45% of total number of samples) or after (55%) drug infusion. Incomplete mixing of first-pass and recirculated drug concentrations, most notably in the presence of reduced cardiac output, affects measured plasma concentrations from samples taken during drug infusion.30 As a consequence, the apparent nonlinearity that we described could have originated from the amplifying effect of incomplete mixing rather than from a concentration-dependent decrease in clearance. To test this hypothesis, we excluded all observations that were collected during or shortly after (less than 5 min) drug infusion and reestimated the linear three-compartmental model and the nonlinear model on the remaining 419 (55% of total number of observations) measured dexmedetomidine concentrations. The results showed strong statistical evidence in favor of the nonlinear model (Δ objective function value = –63; P value likelihood-ratio test < 0.001), thereby demonstrating that potential incomplete mixing of arterial samples during drug infusion is not confounding our conclusions.
To identify the prediction discrepancies between our final nonlinear pharmacokinetic model and the linear pharmacokinetic model by Hannivoort et al.10 that is currently available in target-controlled infusion systems, we simulated different scenarios. For plasma target-controlled infusion, we found that the model predictions for both models for plasma targets less than 2 ng · ml–1 were very consistent. Conversely, when aiming for higher targets, caution is advised, and small increments in targets should be used due to nonlinear clearance. Effect-site target-controlled infusion simulations highlighted the discrepancies between both models when utilizing specific effect compartment concentrations. Effect-site target-controlled infusion differs from the previous plasma concentration target-controlled infusion because the aimed concentration is to be found at the (theoretical) compartment concentration where the desired effect is expected, in this case sedative effect in the central nervous system. The effect-site target-controlled infusion has the advantage of a quicker appearance of the desired effect because the system administers an initial bolus to achieve the target concentration at the effect site. Unfortunately, this initial bolus can lead to adverse effects as the plasma concentrations can abruptly peak. Even though there was a difference between the linear and nonlinear model plasma concentrations, the CMAX did not increase in the target range from 0.1 to 3.0 ng · ml–1 with a maximum ratio between 1.22 and 1.28. As a final example of the implications of the model, we simulated the current guideline-recommended doses for sedation for ventilated patients in the ICU.19 Figure 6 clearly demonstrates that the predicted concentration would have doubled in the nonlinear model resembling the real plasma concentrations compared to the previous linear model during the initial 1 µg · kg–1 loading dose administration. In addition, figure 6 also identifies that long-term infusion (24 h) can cause higher plasma concentrations in a minority of subjects with a low individual C50 value. Subjects in the extremes of age were not included in the trial. Future studies should examine very elderly as well as pediatric subjects and multiple comorbid patients. An enhanced understanding of the etiologic mechanisms of nonlinearity are needed to further understand the pharmacology of dexmedetomidine, and if proven favorable for patient outcome, commercially available target-controlled infusion pumps should include a more advanced algorithm to include the solution of nonlinear kinetics. Today, target-controlled infusion steering algorithms implemented in clinically available pumps support only linear kinetics.31
In this study, we developed a nonlinear three-compartment pharmacokinetic model that predicted dexmedetomidine at various concentration ranges. As opposed to previous studies, our model may permit a more accurate dosing at high plasma concentrations (greater than 2 ng · ml–1). We identified weight, age, and sex as significant covariates in our pharmacokinetic model. Administration of up to 2 ng · ml–1 may be safe to utilize in the previously developed dexmedetomidine pharmacokinetic model with linear or first-order pharmacokinetics. Caution is advised in the ICU for patients at risk during long-term administration and during an initial loading dose when following the dosing strategies of the current dosing guidelines.
The authors thank Rob Spanjersberg, R.N. (University Medical Center Groningen, Groningen, The Netherlands), for coordinating and facilitating this research project, as well as all research personnel that collected the data of the two clinical studies. The authors also thank all of the volunteers who contributed to the two previous clinical trials and David A. Newhall, M.D. (North Bristol National Health Service Trust, Bristol, United Kingdom), for his help with thoroughly reviewing the article.
Supported by the Department of Anesthesiology, University of Groningen and University Medical Center Groningen, Groningen, The Netherlands.
Dr. Struys’s research group/department received (over the last 3 yr) research grants and consultancy fees from the Medicines Company (Parsippany, New Jersey), Masimo (Irvine, California), Fresenius (Bad Homburg, Germany), Dräger (Lübeck, Germany), Paion (Aachen, Germany), and Medtronic (Dublin, Ireland). Dr. Struys also receives royalties on intellectual property from Demed Medical (Temse, Belgium) and Ghent University (Ghent, Belgium). He is an editorial board member and director for the British Journal of Anesthesia and associate editor for Anesthesiology. He was not involved in the editorial process of this publication. The other authors declare no competing interests.