Morphine is a commonly used analgesic for moderate and severe (postoperative) pain in the intensive care unit (ICU).1 Clinically, the analgesic effect of morphine varies because of a number of factors, such as age,2 sex,3 ethnic origin,4 anxiety,5 and genetic polymorphisms,6 but it may also vary with differences in morphine pharmacokinetics caused by differences in health status, hepatic metabolic capacity, and renal function. These effects are enhanced in ICU patients.7
Morphine is mainly metabolized in the liver through glucuronidation by phase II metabolism enzyme UDP-glucuronosyltransferase 2B78 to morphine-3-glucuronide (M3G) and morphine-6-glucuronide (M6G). Both metabolites and unchanged morphine are excreted by the kidneys.9 The main metabolite M3G may have antagonistic or hyperalgesic effects, which could result in reduced morphine efficacy.10 As a consequence, a better understanding of the pharmacokinetics of morphine and its main metabolite, M3G, may help guide dosing regimes in specific patient groups such as ICU patients.
To date, there are no reports on the pharmacokinetics of morphine and M3G in ICU patients, even though morphine has been extensively investigated in terms of efficacy and side effects in other patient groups.9,11–15 In ICU patients, glucuronidation of morphine through UDP-glucuronosyltransferase 2B7 to M3G and/or renal excretion of morphine and M3G may be impaired with major surgery such as cardiac surgery with cardiopulmonary bypass, critical illness related to septic shock, or multiple organ failure. Furthermore, some of these patients may suffer from acute renal failure. Therefore, we developed a population pharmacokinetic model of morphine and its main metabolite, M3G, to study the glucuronidation of morphine to M3G and elimination of M3G and morphine in ICU patients in comparison with healthy volunteers.
This analysis was based on the results of 2 studies. The first was based on the observations in ICU patients participating in a clinical trial in which pain management for procedure-related pain was evaluated (registered at ClinicalTrials.gov; identifier NCT00558090, November 13, 2007). In this study, both ICU patients after cardiac surgery through sternotomy16 and critically ill patients with an expected duration of mechanical ventilation of >48 hours were included. The study was performed in a 30-bed surgical/medical ICU in a teaching hospital, Nieuwegein, The Netherlands, and was approved by the Ethics Committee of the St. Antonius Hospital. Written informed consent was obtained from ICU patients after cardiac surgery on the day before the cardiac surgery. Written informed consent was obtained from critically ill patients by their next of kin on the first day of admission to the ICU. Whenever possible, written informed consent was also obtained from the critically ill patient himself during or after the study.
The second study was a healthy volunteer study, which was performed in the academic hospital, Leiden, The Netherlands.17 Twenty volunteers were recruited to participate in the protocol after approval was obtained from the Human Ethics Committee (Commissie Medisch Ethiek, Leids Universitair Medisch Centrum, Leiden, The Netherlands: protocol No. P00.034) and after written and oral informed consents were obtained.
In the first study, ICU patients after cardiac surgery through sternotomy and critically ill patients with an expected duration of mechanical ventilation of >48 hours were included.16 Additional inclusion criteria were age between 18 and 85 years and weight between 45 and 140 kg. Exclusion criteria were patients with a known morphine or paracetamol allergy, planned admission to the postoperative anesthesia care unit, pregnancy or breast-feeding, serious neurologic deficits (e.g., coma or brain death), a language barrier, and patients who refused informed consent.16
In the second study, 20 healthy volunteers were enrolled earlier as a part of 2 other studies, and detailed information can be found in the references.17 No naltrexone or other blocking agents were administered in any of the patients.
Morphine Dosing Schemes
In the first study, on admission to the ICU, in all ICU patients, a morphine continuous infusion (2 mg/h) was started, with subsequent doses adapted to individual pain levels resulting in morphine doses ranging from 0.5 to 2.5 mg/h.16 In addition, before a painful procedure (turning of the patient and/or chest drain removal), all ICU patients received a bolus dose of morphine (2.5 or 7.5 mg) on the first postoperative day (cardiac surgery patients) or on the first or second day after admission in the ICU (critically ill patients).16 Also for other procedures, additional bolus doses of morphine, which were registered on the case report form, could be administered at the discretion of the attending intensivist.
In the second study performed in healthy volunteers, 20 young men and women (10 of each sex) received IV morphine bolus 0.10 mg/kg dose followed by an infusion of 0.03 mg/kg/h for 1 hour.17
In the first study, in ICU patients, 2 mL arterial blood was drawn for the determination of serum concentrations of morphine and M3G 4 times daily (3:00 AM, 7:00 AM, 3:00 PM, and 9:00 PM), during their ICU stay. In addition, a blood sample was collected at 5 minutes before and 5 minutes after the painful procedure (turning of the patient and/or chest drain removal), which corresponded to 30 and 40 minutes after the bolus dose of morphine of 2.5 or 7.5 mg, respectively, on the first day of ICU admittance.
In the second study, with healthy volunteers,17 blood samples were collected at fixed times (t = 5, 10, 20, 30, 40, 50, 60, 65, 70, 80, 100, 130, 180, 300, and 420 minutes after morphine bolus dose).
Blood samples were rapidly centrifuged, and serum was stored at −20°C until analysis. Morphine and M3G serum concentrations were determined using a high-performance liquid chromatography tandem mass spectrometry method (see Appendix 1). The lower limit of quantification (LLOQ) was 1 μg/L for morphine and 2 μg/L for M3G. For morphine, within-day coefficients of variation were 5.5% at 5 μg/L and 3.5% at 250 μg/L. For M3G within-day coefficients of variation were 10.4% at 10 μg/L and 3.5% at 500 μg/L. The molecular weights of morphine and M3G are 285 and 461 Da, respectively. For the analytic method of the healthy volunteer study,17 the LLOQ values were 2.0 and 30.0 μg/L for morphine and M3G, respectively.
The nonlinear mixed-effects modeling software NONMEM® 7.3 (Globomax LLC, Hanover, MD) was used, with R® version 3.1.1 (R core team, R foundation, available at http://cran.r-project.org/doc/FAQ/R-FAQ.html#Citing-R. Accessed November 1, 2014.) for visualization of the data. Model building was performed in 4 different steps: (1) selection of the structural model (1-, 2-, or 3-compartment model); (2) choice of the error model; (3) covariate; and (4) model evaluation. Discrimination between different models for the structural and statistical model was made by comparison of the objective function (−2 log likelihood). A value of P < 0.05 was considered statistically significant, which corresponds to for instance a decrease of 3.84 points in the objective function for 1 degree of freedom or 9.49 points for 4 degrees of freedom (χ2 distribution). In addition, goodness-of-fit plots (both observed versus individual- and population-predicted concentrations and conditional weighted residuals versus time and population predictions) were evaluated, with specific emphasis on observed versus population-predicted concentrations. Furthermore, visual improvement of the individual plots, the confidence interval (CI) of the parameter estimates, and the correlation matrix were used to evaluate the model. For the interindividual variability, CV% was calculated using the formula SQRT(EXP(ω2) − 1) × 100%, for the residual error, we used the formula SQRT(σ2) × 100%.
The concentrations of morphine and M3G were expressed as micrograms of morphine units per liter, logarithmically transformed and fitted simultaneously (NONMEM 7.3, ADVAN5). Concerning data below the limit of quantification or detection, we used the methods described by Keizer18 to fit the model based on all available data, including data below limit of quantification. Of the 1506 morphine and M3G concentrations in the data set of the ICU patients, 223 (15%) concentrations and 27 (1.8%) were below LLOQ, and 60 (3.9%) and 1 (0.07%) was below the limit of detection, concentrations which were omitted from the analysis.18 In the healthy volunteer study, 5% (n = 36) of the morphine concentrations and 4.5% (n = 36) of the M3G concentrations were below the LLOQ, and 20 (6.4%) and 22 (7.1%) of these were below the limit of detection, which were all omitted from the analysis, as was done in the original analysis.17
For morphine, a 3-compartment model was preferred over a 2-compartment model, because the model was able to describe the data sets more accurately (improved diagnostics and decrease in objective function value of 65 points; P < 0.001). For the metabolite M3G, a 1-compartment model was used (Fig. 1). Because of the large number of observations in steady state and the relative sparse sampling after bolus injection in ICU patients (both patients after cardiac surgery and critically ill patients), minimization difficulties occurred when all parameters of the 4-compartment model (Fig. 1) were independently estimated, with particularly unstable results for volume of distribution of central compartment of morphine (Vparent). As a result, Vparent in the “healthy volunteers” was fixed to 3.67 L, based on a previous study in which a value of 0.052 L/kg was reported in subjects with a mean weight of 70.6 kg.17 As an alternative approach, Vparent in the ICU patients was fixed to 14.2 L, as reported by Mazoit et al.19 in postoperative patients. This alternative approach was rejected because diagnostics deteriorated, even though the estimates for the parameters of the model other than Vparent were stable.
For the 1-compartment M3G metabolite model, 3 parameters need to be estimated: fraction fm of parent going to metabolite, volume, and elimination clearance. When the observations are dose and concentration solely, only 2 parameters can be estimated independently. For this model to be identifiable, we fixed the volume of the metabolite M3G to a literature-reported value of 23 L.20 With volume of the metabolite fixed, the other 2 parameters can be estimated, relative to the value of the fixed parameter while the results are conditional on the metabolite volume of distribution being identical between the studies and constant over time (Appendix 2). For both ICU patients and healthy volunteers, the same literature value was used.
The interindividual value (post hoc value) of the parameters of the ith subject was modeled by:
where θmean is the population mean and ηi is a random variable with mean zero and variance ω2, assuming log-normal distribution in the population. The intraindividual variability, resulting from assay errors, model misspecifications, and other unexplained sources, was best described with a proportional error model. This means for the jth observed log-transformed morphine and M3G concentration of the ith individual, the relation (Yij) is described by Equation 2:
where cpred is the predicted morphine and M3G concentration and εij is a random variable with a mean of 0 and variance of σ2.
Individual post hoc parameters estimates were plotted independently against covariates and the weighted residuals to visualize potential relationships. The following covariates were tested: age, sex, body weight, body mass index, type of cardiac surgery, serum creatinine concentration (value over time during ICU stay), and study group (cardiac surgery patients, critically ill patients and healthy volunteers). For the study group, in addition to exploring differences between the 3 study groups, differences between 2 groups were also evaluated (ICU patients versus healthy volunteers). Continuous covariates (age, body weight, body mass index, and serum creatinine concentration) were tested linear centered
in which θi represents the individual parameter estimate, COV denotes the covariate, and COVmedian denotes the median value of the covariate for the population. Categorical covariates (study group as 3 or 2 groups, sex, type of surgery) were tested fractional by estimation of an additional parameter on a structural parameter for each subgroup.
Covariates were separately incorporated into the model and considered statistically significant if the objective function decreased to ≥7.9 points (P < 0.005). When >1 significant covariate was found, the covariate-adjusted model with the largest decrease in objection function was chosen as a basis to sequentially explore the influence of additional covariates with the use of the same criteria. Beside the objective function, other criteria as presented under data analysis were considered. This procedure was in particular applied to covariates that could be interrelated such as study group, age, and serum creatinine concentration. Finally, after forward inclusion, a backward exclusion procedure was applied to justify the covariate and was considered statistically significant if the objective function decreased to ≥10.8 points (P < 0.001), while also the criteria as discussed under data analysis and internal validation were considered.
Log-likelihood profiling was used to calculate nonparametric CIs for the parameters.21 Furthermore, the normalized prediction distribution error (NPDE) method was used.22,23 The NPDE method was implemented using the NPDE add-on software package, which was run in R version 3.1.1 (NPDE package version 2.0). In this study, the entire dataset was simulated 2000 times in NONMEM, based on a model in which the θ, ω, and σ values obtained from the final model were fixed to their final value and subsequently each observed concentration was compared with the reference distribution of the simulated data points. The results of NPDE method are visualized in different graphs: (1) a histogram showing the distribution of the NPDEs, which are expected to follow a normal distribution; (2) a scatterplot NPDE versus time; and (3) a scatterplot NPDE versus predicted concentrations. The software performed standard statistical tests on the normal distribution. The Wilcoxon signed rank test indicates whether the mean of the NPDE is significantly different from 0, whereas the Fisher exact test for variance determines whether the variance is significantly different from 1.
With the developed pharmacokinetic model, simulations were performed to establish which serum concentrations of morphine and M3G would be achieved in healthy volunteers, in ICU patients with a normal renal function (creatinine serum concentration, 80 μmol/L [0.90 mg/dL]), and in ICU patients with renal failure (creatinine serum concentration, 250 μmol/L [2.83 mg/dL]), after receiving a continuous infusion of 2 mg/h for 24 hours.
The analysis was based on 1506 morphine and 1506 M3G serum concentrations obtained from 135 ICU patients, i.e., 117 cardiac surgery patients and 18 critically ill patients.16 From the 20 healthy volunteers, 311 serum concentrations of morphine and 311 serum concentrations of M3G were available.17 Patient characteristics and healthy volunteer characteristics are summarized in Table 1.
The pharmacokinetic model used is depicted in Figure 1. The morphine data were best described with a 3-compartment model, parameterized in terms of volume of the central compartment (Vparent), 2 peripheral compartments (Vperiph1 and Vperiph2), intercompartmental clearances (Q1 and Q2), glucuronidation clearance of morphine to M3G (CLm, M3G), and non-M3G clearance (CLnon-M3G). The volume of distribution (Vparent) for the ICU patients was 17.1 L (CI, 10.6–24.2). M3G pharmacokinetics were best described by a 1-compartment model, in which the metabolite model is analog to a 1-compartment pharmacokinetic absorption model, with fm (bioavailability) as absorption fraction. In this study, fm represents the fraction of the parent (morphine) transformed to M3G, with Vm, M3G as a volume of distribution of M3G, CLm, M3G as a glucuronidation clearance of morphine, and CLe, M3G as an elimination clearance of M3G. For the residual or intraindividual variability, an additive error model was used with log-transformed data (which is comparable with a proportional model with nontransformed data), with different errors for the study groups: “healthy volunteers” and “ICU patients.”
In the covariate analysis, serum creatinine concentration on the elimination clearance of M3G (CLe, M3G) was a significant (P < 0.001) covariate, resulting in a decrease in objective function value by 297 points (P < 0.001). The serum creatinine concentration was implemented using an exponential function with a scaling factor (k2), which was −0.88 (CI, −0.772 to −0.995; Table 2).
Implementation of the creatinine concentration on CLe, M3G also resulted in improved diagnostics, especially for the postoperative cardiothoracic patients and ICU patients (diagnostic plot “population-predicted versus observed concentrations”). However, after implementation of this covariate, there was some misspecification in the goodness-of-fit plots, mainly in the healthy volunteer group, which could be explained by the fact that there was no intraindividual and interindividual variability in the healthy volunteer group, because all individuals had a normal renal function (creatinine clearance of 80 μmol/L).
The covariate study group (healthy volunteers versus ICU patients) on CLe, M3G (elimination clearance of M3G) was a second significant covariate and improved the model compared with the simple model. Adding this covariate as a study group, multiplication factor on CL2 resulted in a decrease of objective function value by 80 points (P < 0.001), and diagnostic plots of the model largely improved in comparison with the simple model. Further differentiation in 3 study groups (healthy volunteers, cardiac surgery patients, and critically ill patients) did not improve the results. Then, the creatinine as a covariate was incorporated in the study group “ICU patients,” thereby further improving the model (decrease in objective function value by 150 points, P < 0.001). The goodness-of-fit plots improved significantly, compared with the previous model, especially for the ICU patients (diagnostic plot “population-predicted versus observed concentrations”). Overall, the estimated CLe, M3G was 4.13 times higher in healthy volunteers compared with ICU patients with a median creatinine concentration of 80 μmol/L (Table 2), which equals an estimated reduction of 76% in ICU patients. All these results are reported on the condition that the M3G volume of distribution is the same in ICU patients and healthy volunteers.
In this study, total systemic clearance of morphine (CLparent) was estimated as 1.11 L/min (CLnon-M3G of 0.54 L/min; CI, 0.479–0.606; (assuming the creatinine concentration of 80 mmol/L) plus CLm, M3G of 0.57 L/min; CI, 0.534–0.615; Table 2).
In all patients, serum creatinine concentration was an additional covariate for CLnon-M3G (which is CLparent − CLm, M3G), which resulted in a decrease in objective function value of 106 points (P < 0.001). The influence of serum creatinine concentration on CLnon-M3G was described using an exponential function with a scaling factor (k1) of −1.47 (CI, −1.14 to −1.82; Table 2). Implementation of creatinine concentration as covariate on CLnon-M3G also resulted in improved diagnostics, especially the diagnostic plot “population-predicted versus observed concentrations” for the critically ill patients. No other covariates could be identified.
Finally, we analyzed the study groups “ICU patients” and “healthy volunteers” separately. In these models, the parameters were comparable, thereby strengthening the findings for our final model (data not shown).
Table 2 lists all parameter estimates with their CIs obtained of the final model. Final diagnostics plots for morphine and M3G are shown in Figures 2 to 4. No bias was observed in the plots “conditional weighted residuals versus time” and “conditional weighted residuals versus population-predicted concentrations” (data not shown).
Log-likelihood profiling confirmed the high precision of the parameter estimates (Table 2). Figure 5 shows the results of the NPDE for morphine and M3G. The histogram follows a normal distribution expected by the solid line. Although there was a slight overestimation of the interindividual variability, no trend was observed in the NPDE versus time or versus predicted concentrations. The values of the mean and variance are given below each graph, with * and *** indicating a significant difference from 0 and 1 at the P < 0.05 and P < 0.001, respectively, level as determined by the Wilcoxon signed rank test and Fisher exact test for variance.
The simulations based on the final pharmacokinetic model presented in Figure 6, A and B show morphine and M3G serum concentrations in a typical healthy volunteer, a typical ICU patient with normal renal function (creatinine concentration, 80 μmol/L [0.90 mg/dL]), and a typical ICU patient with an impaired renal function (creatinine concentration, 250 μmol/L [2.83 mg/dL]) on a 24-hour morphine infusion of 2 mg/h. Figure 6A shows that similar morphine concentrations can be expected in ICU patients compared with healthy volunteers, except in case of impaired renal function. Figure 6B shows that in ICU patients, increased M3G concentrations can be expected in comparison with healthy volunteers. When renal function is impaired in ICU patients, even higher M3G concentrations are anticipated. Figure 6C shows that a 33% reduction in the maintenance dose for ICU patients with renal failure (creatinine concentration, 250 μmol/L) would result in morphine plasma concentrations similar to healthy volunteers and ICU patients with normal renal function (creatinine concentration, 80 μmol/L). In addition, Figure 6D shows that despite equal morphine concentrations on this dose reduction for ICU patients with substantial renal dysfunction, higher M3G concentrations can be expected.
To quantify the glucuronidation and elimination clearance of morphine in ICU patients in comparison with healthy volunteers, we developed a pharmacokinetic model of morphine and M3G. We found that ICU patients had a significantly decreased elimination clearance of M3G (estimated 76%) compared with healthy volunteers. Furthermore, serum creatinine concentration was a covariate for elimination clearance of M3G in ICU patients and for the unchanged morphine clearance in all patients. Therefore, substantial accumulation of M3G can be expected in ICU patients with a 24-hour morphine infusion. This effect is even more pronounced in patients with impaired renal function (Fig. 6). Under the assumption that the M3G volume of distribution is the same in ICU patients and healthy volunteers, our data show a very large reduction (estimated 76%) in elimination clearance of M3G (CLe, M3G) in ICU patients compared with healthy volunteers. Alternate explanations, equally supported by the data, are that the fraction of morphine metabolized to M3G is increased in critically ill patients or that both increased metabolism to M3G and decreased clearance are present. In the absence of a study in which the metabolite is administered to these populations, our data cannot distinguish between these scenarios.
Under these assumptions, we identified a very large reduction (estimated 76%) in elimination clearance of M3G (CLe, M3G) in ICU patients compared with healthy volunteers. Although the estimated elimination clearance of M3G in healthy volunteers of 0.175 L/min (10.5 L/h) is similar to those reported in the literature,24 this estimated reduction of 76% in this parameter in ICU patients was independent of renal function. Thus, in ICU patients with normal renal function, substantially higher M3G levels may be expected on 24-hour continuous infusion, with even higher M3G levels in patients with renal failure (Fig. 6B).
The clinical relevance of high M3G levels in ICU patients is unclear. Although controversial, M3G may antagonize the analgesic effect and play a role in the development of tolerance and hyperalgesia.9 Data supporting this possibility arise from animal studies in which M3G administration antagonizes morphine and M6G analgesia.25–27 However, these findings could not be reproduced in 2 healthy volunteer studies,20,28 although the short study period of 2 hours may have played a role. Moreover, in a recent study by Swartjes et al.29 in mice, morphine produced hyperalgesia without a significant contribution from M3G. In contrast, in postoperative patients receiving morphine as an IV infusion, M3G had an antinociceptive effect.19 The authors noted that this effect was moderate, and because of the very long transfer half-time from injection site to effect compartment, a significant effect of M3G did not occur before the 9th to 18th hour after initiation of analgesic treatment. In our study, cardiac surgery patients had a mean infusion duration of 21 hours, whereas critically ill patients received IV morphine infusions for a mean duration of 123 hours. In humans, symptoms of altered pain behavior, such as hyperalgesia or allodynia, have been reported in cancer patients chronically treated with high-dose morphine, and in several of these reports, very high levels of morphine and M3G and accumulation of M3G relative to morphine or M6G were observed.9 Thus, the very large decrease in elimination clearance of M3G in ICU patients, resulting in substantially increased M3G concentrations, may be clinically relevant, in particular when prolonged morphine infusions are used.
In our study, serum creatinine concentration was a covariate for elimination clearance of M3G (CLe, M3G) in all ICU patients, independent of the estimated decrease of 76% in this parameter in ICU patients in comparison with healthy volunteers. Also, for unchanged morphine clearance (CLnon-M3G), serum creatinine concentration was a significant covariate in all patients, possibly because the covariate “study group” could not be identified as a significant covariate. Although the influence of serum creatinine concentration on these clearance parameters was best described using a negative exponential function, the literature suggests that clearance of morphine glucuronides deteriorates as a result of renal failure in many diverse patient populations.30–34 In these studies, clearance of M3G and M6G were significantly correlated with creatinine clearance.30,31 In addition, Mazoit et al.19 found, in a pharmacokinetic/pharmacodynamics study, that M3G clearance was markedly decreased in postoperative patients receiving IV and IM morphine after several types of surgery. These results suggest that significant accumulation of morphine and M3G may occur in ICU patients, with or without renal failure. Considering the potential antinociceptive and hyperalgesic activity of M3G, its accumulation in critically ill patients may be clinically relevant. Our simulations showed that a 33% reduction in the maintenance morphine dose for ICU patients with increased creatinine levels would result in equal morphine serum concentrations compared with healthy volunteers and ICU patients with normal renal function (Fig. 6C). However, despite this dose reduction of morphine, higher M3G concentrations can be expected in ICU patients with renal dysfunction (Fig. 6D). Thus, we propose that the 33% morphine dose reduction derived in ICU patients with renal failure should be tested in a prospective study before clinical implementation. A similar stepwise approach for morphine dosing in young infants recently proved successful in optimizing morphine dosing in postoperative patients younger than 1 year.35,36
There are some limitations of our study. In this analysis, we were not able to differentiate between the 2 subgroups of ICU patients in our study population, i.e., cardiac surgery patients versus critically ill patients. We realize that these 2 subgroups have many physiologic differences that may result in different pharmacokinetic parameters. One explanation for the lack of difference between these 2 groups may be the imbalance in patient numbers (117 vs 18 patients). We also hypothesize that pharmacokinetic differences between these 2 groups are mainly located in volumes of distribution, which are expected to be larger in critically ill patients. Obviously, data were insufficient to properly estimate the central volume of distribution directly after bolus injection in the ICU patients. Because we fixed the central volume (Vparent) of the healthy volunteers to the value reported for this data set17 and estimated a multiplication factor for this value for ICU patients, we cannot make any statements about the CIs of Vparent in healthy volunteers and ICU patients. Thus, future research in ICU patients should focus on obtaining more samples after bolus injection to study the influence of the ICU subgroups on central volume of distribution. Doing so would also allow the resulting model to simulate morphine and M3G concentrations after bolus injection, which was not performed in this study (Fig. 6).
For the metabolite model to be identifiable (see section Methods and Appendix 2), the volume of distribution of M3G (Vm, M3G,) for both healthy volunteers and patients had to be fixed to a value reported in literature in healthy volunteers (23 L).21 With 1 of the 3 parameters of the metabolite model fixed to a literature value, the other 2 parameters can be estimated relative to this value. However, the volume of distribution of M3G may be different in ICU patients compared with healthy volunteers, even without specific evidence to support this hypothesis. Applying the different volumes of distribution to the 2 groups would have affected the ratio of the M3G clearance for the 2 groups “healthy volunteers” and “ICU patients” directly, resulting in different parameter estimates (Appendix 2). The only way to evaluate possible differences in volume of distribution between healthy volunteers and ICU patients is to administer M3G to healthy volunteers and ICU patients, which may be ethically and practically unfeasible. An alternative parameterization is possible where no assumptions about the fraction of dose metabolized into M3G (f) are made, thereby solving the clearance and volume of distribution of the metabolite as apparent values, Vm, M3G/f and CLe, M3G/f. However, this parameterization would have assumed a constant f, which is in contrast to the results of this pharmacokinetic analysis, where we found that the fraction of morphine metabolized to M3G increases with decreasing renal function. For this reason, we preferred fixing the Vm, M3G to a literature-reported value over alternative approaches. Thus, our conclusions on the ratio of the metabolite clearance should be interpreted with the assumption that the volumes of distribution of the metabolite were the same between healthy volunteers and ICU patients.
Another issue may be the correlation between covariates within the study groups. ICU patients were older and had higher average serum creatinine concentrations. By systematically testing of the influence of all these covariates in different functions based on the predefined statistical criteria, we were able to identify which covariates were the most predictive, which ultimately generated the model presented in this study. In this respect, we emphasize that we used creatinine concentration as a covariate, whereas glomerular filtration rate in these patients would be preferable. However, equations used to calculate glomerular filtration rate based on the creatinine concentrations in critically ill patients are frequently inaccurate.37
Finally, we could not differentiate among unchanged morphine clearance (CLnon-M3G), which theoretically consists of unchanged morphine clearance, M6G clearance, and potential clearance through other pathways. In the ideal study design, concentrations of M6G and/or excretion of morphine and metabolites in urine should be included in the analysis to distinguish between these routes. Including this information may also lead to different covariates for subparameters of CLnon-M3G, because in our analysis, serum creatinine could be identified for this parameter only. It seems reasonable that serum creatinine will account for the part of this parameter that is responsible for excretion of unchanged morphine, but not for glucuronidation to M6G for instance. However, the results of our final model seem valid, because the percentage of the IV morphine dose that was converted through CLm, M3G to M3G and through CLnon-M3G was 44% and 56%, respectively, in a typical healthy volunteer, which is in line with previous reports for healthy volunteers. In those reports, it was reported that 44% to 55% of the morphine dose was converted to M3G, regardless of the morphine dose.9
In conclusion, in ICU patients, elimination clearance of M3G appeared significantly decreased compared with healthy volunteers, under the assumption that the M3G volume of distribution is the same in ICU patients and healthy volunteers. As a result, increased M3G concentrations, which are even more pronounced with increased serum creatinine concentrations, may be anticipated in ICU patients. Model-based simulations show that in ICU patients with renal failure, compared with healthy volunteers and ICU patients with normal renal function, a 33% reduction in the maintenance dose would result in equal morphine serum concentrations, even though M3G concentrations remain increased. Future pharmacodynamic investigations are needed to clarify target concentrations in this population. E
Determination of Morphine and Morphine-3-Glucuronide in Serum by Liquid Chromatography Tandem Mass Spectrometry
For the construction of the calibration line, human pool serum is spiked with adequate volumes of morphine and morphine-3-glucuronide (M3G; Cerilliant, TX) in methanol/water to give a concentration of 0 to 2500 μg/L. Fifty microliters of the standards was stored in Eppendorf vials (VWR, Amsterdam, The Netherlands) in a −20°C freezer until analysis. Samples of 200 μL, standards, Blanco, and quality controls were transferred in an Eppendorf 1.5-mL vial. The protein was precipitated with 700 μL acetonitrile, which contained an adequate amount of 2H3-Morphin (2H3-M; Cerilliant) and 2H3-morphine-3-glucuronide (2H3-M3G; Cerilliant) as internal standard and 100 μL 1 mM zinc sulfate. The vials were vortexed for 2 minutes and centrifuged for 5 minutes at 13,000 rpm. The supernatant (200 μL) was transferred in a glass tube and dried under a gentle stream of nitrogen at 50°C. The residues were reconstituted in 100 μL of 0.1% (v/v) formic acid in water. Twenty microliters was injected by an Ultimate 3000 autosampler (Dionex, Amsterdam, The Netherlands) and pumped by a HPG680 pump (Dionex) on a 3 μm, 120Å, 50 × 2.1-mm YMC-pack ODS-AQ column (YMC Inacom, Overberg, The Netherlands) with an ODS precolumn (Phenomenex, Utrecht, The Netherlands) at 30°C. This HPLC part of the equipment was controlled by Chromeleon (Dionex). The eluent was monitored by a Quattro micro API tandem mass spectrometer (Waters, Etten-Leur, The Netherlands). Peak areas of reaction ions from morphine and M3G and the internal standards 2H3-M and 2H3-M3G were obtained in the multiple reaction mode and integrated by data software Masslynx 4.1 (Waters). m/z was 165.0 (285.9 > 165.0) for morphine and 286.0 (461.9 > 286.0) for M3G. For the internal standards, m/z was 165.0 (288.9 > 165.0) for 2H3-M and 289.0 (464.9 > 289.0) for 2H3-M3G. All the samples were calculated by the internal standard method with weighing factor 1/(Y2). The mobile phase consisted of 0.1% formic acid in water with 3% acetonitrile (Lichrosolv; Merck BV, Amsterdam, The Netherlands) as a modifier. At a flow rate of 0.5 mL/min, the retention times of morphine, M3G, 2H3-M, and 2H3-M3G were 6.70, 7.18, 6.65, and 5.10 minutes, respectively. Total analysis time was 10 minutes. All analytes were analyzed within 1 run.
An Overview of Possible Metabolite Modeling Parameterizations
Let an IV administered drug with linear 1-compartment metabolite pharmacokinetics be considered. The following equations can be used:
where fm is the fraction of parent drug metabolized to a specific metabolite; CLm is the metabolite formation clearance; CLt,p is the total parent drug clearance; kel,m is the elimination rate constant of the metabolite; CLe is the metabolite elimination clearance; Vm is the metabolite volume of distribution; and AUCm is the area under curve of the metabolite time-concentration data.
Regardless of the assumptions made for metabolite modeling, the parameters CLt,p, AUCm, and kel,m can always be calculated and used as the basis for computing the pharmacokinetic parameters of the metabolite.
Scenario 1: No Assumption About Any of the Metabolite Model Parameters
If no assumptions are made about any of the metabolite pharmacokinetic parameters, then fm is not identifiable. Thus, only apparent values can be calculated for the metabolite distribution volume and elimination clearance:
Consequence: No bias is produced with this method; however, it is important to acknowledge that the clearance and volume are estimated dependent on fm, which is unknown.
Scenario 2: Assumptions are Made About the Metabolite Pharmacokinetics to Make the Model Identifiable
Another approach to identify the metabolite pharmacokinetics is to replace some parameter values with their assumed values based on the literature. These assumed parameters, here denoted by boldface, are used to make the rest of the parameters identifiable.
Scenario 2.1: Assumed Fraction of Drug Metabolized by a Specific Pathway
If fm is replaced by an assumed value of fm, then
Consequence: The parameter estimates will be conditional on the value of the assumed parameter. Therefore, if the assumed value of fm used in the analysis was larger than the true value of fm, this would lead to a larger estimated value of metabolite elimination clearance and volume of distribution (positive bias).
Scenario 2.2: Assumed Metabolite Formation Clearance
If CLm is replaced by an assumed value of CLm, then
Consequence: The parameter estimates will be conditional on the value of the assumed parameter. Therefore, if the assumed value of CLm used in the analysis was larger than the true value of CLm, this would lead to a larger estimated value of metabolite elimination clearance and volume of distribution (positive bias).
Scenario 2.3: Assumed Metabolite Distribution Volume
If Vm is replaced by an assumed value of Vm, then
Consequence: The parameter estimates will be conditional on the value of the assumed parameter. Therefore, if the assumed value of Vm used in the analysis was larger than the true value of Vm, this would lead to a larger estimated value of metabolite elimination clearance and fm (positive bias).
Scenario 2.4: Assumed Metabolite Elimination Clearance
If CLe is replaced by an assumed value of CLe, then
Consequence: The parameter estimates will be conditional on the value of the assumed parameter. Therefore, if the assumed value of CLe used in the analysis was larger than the true value of CLe, this would lead to a larger estimated value of fm and metabolite volume of distribution (positive bias).
Name: Sabine J. G. M. Ahlers, PharmD, PhD.
Contribution: This author helped conduct the study, collect the data, analyze the data, and prepare the manuscript.
Attestation: Sabine J. G. M. Ahlers approved the final manuscript, attests to the integrity of the original data and the analysis reported in this manuscript, and is the archival author.
Name: Pyry A. J. Välitalo, PhD.
Contribution: This author helped coanalyze the data, prepare the manuscript, and revise the manuscript.
Attestation: Pyry A. J. Välitalo approved the final manuscript.
Name: Mariska Y. M. Peeters, PharmD, PhD.
Contribution: This author helped coanalyze the data, prepare the manuscript, and revise the manuscript.
Attestation: Mariska Y. M. Peeters approved the final manuscript.
Name: Laura van Gulik, MD, PhD.
Contribution: This author helped collect the data and revise the manuscript.
Attestation: Laura van Gulik approved the final manuscript.
Name: Eric P. A. van Dongen, MD, PhD.
Contribution: This author helped conduct the study, design the study, and revise the manuscript.
Attestation: Eric P. A. van Dongen approved the final manuscript.
Name: Albert Dahan, MD, PhD.
Contribution: This author helped coanalyze the data and revise the manuscript.
Attestation: Albert Dahan approved the final manuscript.
Name: Dick Tibboel, MD, PhD.
Contribution: This author helped conduct the study, design the study, and revise the manuscript.
Attestation: Dick Tibboel approved the final manuscript.
Name: Catherijne A. J. Knibbe, PharmD, PhD.
Contribution: This author helped design the study, analyze the data, and revise the manuscript.
Attestation: Catherijne A. J. Knibbe approved the final manuscript and attests to the integrity of the original data and the analysis reported in this manuscript.
This manuscript was handled by: Avery Tung, MD.
The authors thank the staff and nurses of the Department of Anaesthesiology, Intensive Care and Pain Management and the staff of the Department of Clinical Pharmacy of the St. Antonius Hospital for their contribution to this study. René Mooren is acknowledged for his contribution of the analytic measurements. Maurice Wang and Elke Krekels are acknowledged for facilitating the NPDE analysis. Jeroen Diepstraten and Margreke Brill are acknowledged for their support in data analysis.
1. Barr J, Fraser GL, Puntillo K, Ely EW, Gélinas C, Dasta JF, Davidson JE, Devlin JW, Kress JP, Joffe AM, Coursin DB, Herr DL, Tung A, Robinson BR, Fontaine DK, Ramsay MA, Riker RR, Sessler CN, Pun B, Skrobik Y, Jaeschke RAmerican College of Critical Care Medicine. American College of Critical Care Medicine. . Clinical practice guidelines for the management of pain, agitation, and delirium in adult patients in the intensive care unit. Crit Care Med. 2013;41:263–306
2. Macintyre PE, Jarvis DA. Age is the best predictor of postoperative morphine requirements. Pain. 1996;64:357–64
3. Aubrun F, Salvi N, Coriat P, Riou B. Sex- and age-related differences in morphine requirements for postoperative pain relief. Anesthesiology. 2005;103:156–60
4. Cepeda MS, Farrar JT, Roa JH, Boston R, Meng QC, Ruiz F, Carr DB, Strom BL. Ethnicity influences morphine pharmacokinetics and pharmacodynamics. Clin Pharmacol Ther. 2001;70:351–61
5. Ozalp G, Sarioglu R, Tuncel G, Aslan K, Kadiogullari N. Preoperative emotional states in patients with breast cancer and postoperative pain. Acta Anaesthesiol Scand. 2003;47:26–9
6. Allegri M, De Gregori M, Niebel T, Minella C, Tinelli C, Govoni S, Regazzi M, Braschi A. Pharmacogenetics and postoperative pain: a new approach to improve acute pain management. Minerva Anestesiol. 2010;76:937–44
7. Sessler CN, Grap MJ, Ramsay MA. Evaluating and monitoring analgesia and sedation in the intensive care unit. Crit Care. 2008;12(suppl 3):S2
8. Lötsch J. Opioid metabolites. J Pain Symptom Manage. 2005;29:S10–24
9. Andersen G, Christrup L, Sjøgren P. Relationships among morphine metabolism, pain and side effects during long-term treatment: an update. J Pain Symptom Manage. 2003;25:74–91
10. Martini C, Olofsen E, Yassen A, Aarts L, Dahan A. Pharmacokinetic-pharmacodynamic modeling in acute and chronic pain: an overview of the recent literature. Expert Rev Clin Pharmacol. 2011;4:719–28
11. Roberts G. A review of the efficacy and safety of opioid analgesics post-craniotomy. Nurs Crit Care. 2004;9:277–83
12. Papaleontiou M, Henderson CR Jr, Turner BJ, Moore AA, Olkhovskaya Y, Amanfo L, Reid MC. Outcomes associated with opioid use in the treatment of chronic noncancer pain in older adults: a systematic review and meta-analysis. J Am Geriatr Soc. 2010;58:1353–69
13. Patanwala AE, Keim SM, Erstad BL. Intravenous opioids for severe acute pain in the emergency department. Ann Pharmacother. 2010;44:1800–9
14. Pergolizzi J, Böger RH, Budd K, Dahan A, Erdine S, Hans G, Kress HG, Langford R, Likar R, Raffa RB, Sacerdote P. Opioids and the management of chronic severe pain in the elderly: consensus statement of an International Expert Panel with focus on the six clinically most often used World Health Organization Step III opioids (buprenorphine, fentanyl, hydromorphone, methadone, morphine, oxycodone). Pain Pract. 2008;8:287–313
15. Walder B, Tramèr MR. Analgesia and sedation in critically ill patients. Swiss Med Wkly. 2004;134:333–46
16. Ahlers SJ, van Gulik L, van Dongen EP, Bruins P, van de Garde EM, van Boven WJ, Tibboel D, Knibbe CA. Efficacy of an intravenous bolus of morphine 2.5 versus morphine 7.5 mg for procedural pain relief in postoperative cardiothoracic patients in the intensive care unit: a randomised double-blind controlled trial. Anaesth Intensive Care. 2012;40:417–26
17. Sarton E, Olofsen E, Romberg R, den Hartigh J, Kest B, Nieuwenhuijs D, Burm A, Teppema L, Dahan A. Sex differences in morphine analgesia: an experimental study in healthy volunteers. Anesthesiology. 2000;93:1245–54
18. Keizer RJ Pharmacometrics in Early Clinical Drug Development [Dissertation]. 2010 Netherlands Utrecht University
19. Mazoit JX, Butscher K, Samii K. Morphine in postoperative patients: pharmacokinetics and pharmacodynamics of metabolites. Anesth Analg. 2007;105:70–8
20. Penson RT, Joel SP, Clark S, Gloyne A, Slevin ML. Limited phase I study of morphine-3-glucuronide. J Pharm Sci. 2001;90:1810–6
21. Lindbom L, Pihlgren P, Jonsson EN, Jonsson N. PsN-Toolkit—a collection of computer intensive statistical methods for non-linear mixed effect modeling using NONMEM. Comput Methods Programs Biomed. 2005;79:241–57
22. Brendel K, Comets E, Laffont C, Laveille C, Mentré F. Metrics for external model evaluation with an application to the population pharmacokinetics of gliclazide. Pharm Res. 2006;23:2036–49
23. Comets E, Brendel K, Mentré F. Computing normalised prediction distribution errors to evaluate nonlinear mixed-effect models: the npde add-on package for R. Comput Methods Programs Biomed. 2008;90:154–66
24. Hasselström J, Eriksson S, Persson A, Rane A, Svensson JO, Säwe J. The metabolism and bioavailability of morphine in patients with severe liver cirrhosis. Br J Clin Pharmacol. 1990;29:289–97
25. Faura CC, Olaso MJ, Garcia Cabanes C, Horga JF. Lack of morphine-6-glucuronide antinociception after morphine treatment. Is morphine-3-glucuronide involved? Pain. 1996;65:25–30
26. Gong QL, Hedner J, Björkman R, Hedner T. Morphine-3-glucuronide may functionally antagonize morphine-6-glucuronide induced antinociception and ventilatory depression in the rat. Pain. 1992;48:249–55
27. Smith MT, Watt JA, Cramond T. Morphine-3-glucuronide—a potent antagonist of morphine analgesia. Life Sci. 1990;47:579–85
28. Penson RT, Joel SP, Bakhshi K, Clark SJ, Langford RM, Slevin ML. Randomized placebo-controlled trial of the activity of the morphine glucuronides. Clin Pharmacol Ther. 2000;68:667–76
29. Swartjes M, Mooren RA, Waxman AR, Arout C, van de Wetering K, den Hartigh J, Beijnen JH, Kest B, Dahan A. Morphine induces hyperalgesia without involvement of μ-opioid receptor or morphine-3-glucuronide. Mol Med. 2012;18:1320–6
30. Pauli-Magnus C, Hofmann U, Mikus G, Kuhlmann U, Mettang T. Pharmacokinetics of morphine and its glucuronides following intravenous administration of morphine in patients undergoing continuous ambulatory peritoneal dialysis. Nephrol Dial Transplant. 1999;14:903–9
31. Peterson GM, Randall CT, Paterson J. Plasma levels of morphine and morphine glucuronides in the treatment of cancer pain: relationship to renal function and route of administration. Eur J Clin Pharmacol. 1990;38:121–4
32. Säwe J, Odar-Cederlöf I. Kinetics of morphine in patients with renal failure. Eur J Clin Pharmacol. 1987;32:377–82
33. Wolff J, Bigler D, Christensen CB, Rasmussen SN, Andersen HB, Tønnesen KH. Influence of renal function on the elimination of morphine and morphine glucuronides. Eur J Clin Pharmacol. 1988;34:353–7
34. Milne RW, Nation RL, Somogyi AA, Bochner F, Griggs WM. The influence of renal function on the renal clearance of morphine and its glucuronide metabolites in intensive-care patients. Br J Clin Pharmacol. 1992;34:53–9
35. Knibbe CA, Krekels EH, van den Anker JN, DeJongh J, Santen GW, van Dijk M, Simons SH, van Lingen RA, Jacqz-Aigrain EM, Danhof M, Tibboel D. Morphine glucuronidation in preterm neonates, infants and children younger than 3 years. Clin Pharmacokinet. 2009;48:371–85
36. Krekels EH, Tibboel D, de Wildt SN, Ceelie I, Dahan A, van Dijk M, Danhof M, Knibbe CA. Evidence-based morphine dosing for postoperative neonates and infants. Clin Pharmacokinet. 2014;53:553–63
37. Baptista JP, Neves M, Rodrigues L, Teixeira L, Pinho J, Pimentel J. Accuracy of the estimation of glomerular filtration rate within a population of critically ill patients. J Nephrol. 2014;27:403–10