Population pharmacokinetics of intravenous sufentanil in critically ill patients supported with extracorporeal membrane oxygenation therapy

This was a prospective, cohort study conducted at the cardiac intensive care unit in Severance Cardiovascular Hospital, a university-affiliated tertiary care hospital in Seoul, Republic of Korea, between January 2016 and June 2017. The study was approved by the Institutional Review Board (IRB No.: 4-2014-0919) of Severance Hospital and was registered at Clinicaltrials.gov (NCT02581280). Written informed consent was obtained from the patients or the legal surrogates of unconscious patients. This study complied with the Strengthening the Reporting of Observational studies in Epidemiology (STROBE).

Twenty patients aged 19 years or older, who received sufentanil-based analgesia and sedation during VA-ECMO, were enrolled in this study. The exclusion criteria were younger than 19 years, known allergy to sufentanil, and taking any medication that could cause potential drug-drug interactions or alter sufentanil concentrations.

ECMO patients received sufentanil for maintenance of analgesia and sedation supplemented as needed with midazolam. Sufentanil dosing in our centre was based on patients’ body weight, with initial infusion doses of 12.5 (< 60 kg) or 17.5 μg h⁻¹ (≥60 kg). An initial bolus of 3 (< 60 kg) or 5 mg (≥ 60 kg) midazolam was given, with an initial infusion dose of 4.5 mg h⁻¹. Management of pain should be guided by routine pain assessment of Nonverbal Pain Scale. Nurses assessed the depth of sedation, indicating the prevalent Richmond Agitation Sedation Scale (RASS) in their work shift. Analgesia and sedation protocol was to keep patients deeply sedated during the first few days of ECMO followed by intermediate or light sedation before ECMO discontinuation when possible. The infusion rates of sufentanil and midazolam were modified to achieve a target RASS score, and each dose adjustment was recorded.

Data on demographics, organ function, ECMO, vital signs, and drug dosing were collected from the electronic medical records.

The ECMO circuit included a centrifugal blood pump with a pump controller (Capiox® SP-101, Terumo Inc., Tokyo, Japan), an air-oxygen mixer (Sechrist® Ind., Anaheim, CA, USA), and conduit tubing (Capiox® EBS Circuit with X coating, Terumo Inc.). The days on ECMO, ECMO flow rate, and ECMO pump speed were recorded.

The plasma concentrations of sufentanil were analysed using a validated HPLC system (Agilent Technologies, CA, USA) coupled with a 4000 Qtrap liquid chromatograph-mass spectrometer (ASICX, Concord, Ontario, Canada). The plasma samples were denatured with acetonitrile containing 0.5 μg mL⁻¹ prazosin as an internal standard. The mixture was vortexed and centrifuged at 150,000×g for 10 min at 4 °C. HPLC was performed on a Kinetex C₁₈ analytical column (4.6 × 50 mm; particle size 2.6 μm; Phenomenex, Torrance, CA, USA) with a mobile phase consisting of 0.1% formic acid in acetonitrile at a flow rate of 0.055 mL min⁻¹. The lower limit of quantification for sufentanil was 0.02 μg L⁻¹. The assay was validated between 0.02 and 10 μg L⁻¹ with inter- and intra-assay coefficients of variation of < 15%.

The population PK model was developed using a first-order conditional estimation method with an interaction (FOCE+I) algorithm in the nonlinear mixed effects modelling software NONMEM® version 7.4 (ICON Development, Ellicott City, MD, USA). Pirana® ver. 2.9.2 and Xpose® ver. 4.0 (http://​xpose.​sourceforge.​net) in R® ver. 3.2.4 (http://​www.​r-project.​org) were used to visualise and evaluate the models. One-, two-, and three-compartment models were evaluated as the structural PK models. Inter-individual variability (IIV) for the PK parameters was modelled assuming a log-normal distribution: θ_i = θ_Pop × EXP(η_i), where θ_i is the individual value of the parameter θ in the ith individual, θ_Pop is the population value of this parameter, and η_i is a random variable with mean zero and variance ω_η² [18]. Proportional models for residual variability was used: c_ij = cp_ij × (1 + ε_ij) in which c_ij is the jth observed concentration of the ith individual, cp_ij is the corresponding predicted concentration, and ε_ij is a random variable with mean zero and variance σ².

The likelihood ratio test was used to evaluate statistical significance between nested models where a decrease in the objective function value (OFV), a statistical equivalent to the − 2 log likelihood of the model, of at least 3.84 was considered statistically significant for an added parameter (χ² distribution, degrees of freedom (df) = 1, p < 0.05). In addition, bias of the goodness-of-fit plots (observed versus population predicted concentrations, observed versus individual predicted concentrations, conditional weighted residuals (CWRES) versus population predicted concentrations, and CWRES versus time after dosing), visual improvement of individual plots, confidence intervals of parameter estimates, and shrinkage were assessed. The aim of this study was to examine the potential effect of various covariates of the model structural parameters. The following covariates were investigated: sex, age, weight, lean body weight, body mass index, tympanic body temperature, total plasma protein, partial pressure of carbon dioxide, plasma pH, estimated glomerular filtration rate, serum creatinine, total bilirubin, alanine transaminase, aspartate transaminase, use of continuous renal replacement therapy, ECMO pump speed, and ECMO flow rate. The estimated parameters were plotted against each covariate to identify its influence. Continuous covariates (Cov) were incorporated into the structural model with centering on their median values within the population and tested using power (1), linear (2), and exponential (3) equations:where θ_TV is the typical value of the parameter and θ_Cov quantifies the covariate effect. The covariate model building was carried out in a stepwise process. In the forward selection, a P value of < 0.05 was used (a decrease in OFV of at least 3.84, df = 1), while in the backward elimination, a P value of < 0.01 was applied (a decrease in the OFV of at least 6.64, df = 1).

To evaluate stability in the final model, a non-stratified bootstrap analysis was performed using the PsN Toolkit [19]. A bootstrap with 5000 runs was performed on the final model to evaluate the internal validity of the parameter estimates and their corresponding 95% confidence intervals (CIs). The model performance was evaluated by means of prediction-corrected visual predictive checks (pc-VPCs). One thousand datasets were simulated from the final model, and the median and 90% CI of the simulated data were plotted along with the observed concentrations. To illustrate the effect of body temperature and total plasma protein level on predicted sufentanil concentrations, Monte Carlo simulations were performed using PK parameters from the final model. We assumed that sufentanil was continuously infused for 120 h with a rate of 12.5 or 17.5 μg h⁻¹, which were the doses most frequently used in our ECMO patients. The median parameter values for the patient population were obtained with five different levels of body temperature (33, 35, 36.7, 38 and 39 °C) and four different total plasma protein levels (2, 4, 6, and 8 g dL⁻¹) by simulating 1000 individuals in each case.

Twenty ECMO patients were included, with a median age of 55 years, median weight of 69.4 kg, and median APACHE II score of 29 at the initiation of ECMO support. All the patients received mechanical ventilation and started ECMO during the first 12 h after the onset of myocardial infarction (MI). The median duration of VA-ECMO and sufentanil infusion was 138 and 110 h, respectively. Nine patients concurrently received continuous venovenous hemodiafiltration (CVVHDF) during ECMO (Table 1). Desired sedation level was reached or nearly reached (observed RASS = target RASS ± 1) in all patients. A population PK analysis was conducted with 106 plasma samples from the 20 patients. Concentration records that were below the lower limit of quantification were excluded from the analysis.

A two-compartment model parameterised in terms of systemic clearance (CL), central volume of distribution (V1), peripheral volume of distribution (V2), and intercompartmental clearance (Q) was preferred to a one- and three-compartment model. Residual variability was described with a proportional residual error model. IIVs were included for CL and V2, since they significantly improved model performance. The structural model had an OFV of − 343.1. In the univariate covariate analysis, body temperature and total bilirubin were identified as significant covariates of CL, resulting in a drop in OFV of − 9.1 and − 6.0 points, respectively. In addition, total protein and lean body weight were significant covariate candidates of V2 with ∆ OFVs of − 9.3 and − 3.4 points, respectively. After the forward selection and backward elimination, total bilirubin for CL and lean body weight for V2 were removed.

Thus, the final PK model is described as follows:

The final parameter estimates and IIVs along with their bootstrap CIs are provided in Table 2. All parameters had acceptable relative standard error values. The population mean estimates were contained within the 95% CIs of the bootstrap results.

Figure 1 a–d show the goodness-of-fit plots. Both the population predictions and individual predictions were distributed relatively symmetrically around the line of unity when plotted against the observations. In addition, the CWRES versus time and versus population predictions did not show any trends and were approximately uniformly distributed around zero in the final model. The pc-VPCs revealed that the 5th to 95th percentiles of the predicted data overlaid most of the observed data, indicating good precision of the PK model (Fig. 2). The final covariate model was then used for simulations. Figure 3 shows the Monte Carlo simulated sufentanil concentrations during 120-h infusion using two different dosing regimens, stratified by the body temperature and total plasma protein level. The target concentrations were set to 0.3–0.6 μg L⁻¹. Overall, the concentrations of sufentanil are increased in patients with low body temperature and low total plasma protein levels.

Figure 3 a and c reveal the effect of different body temperatures on sufentanil concentrations. With a dose of 17.5 μg h⁻¹, the concentrations of sufentanil from 24 to 120 h were within the target concentrations in patients with a body temperature of 35–38 °C, whereas they were above the target concentrations in patients developing hypothermia (33 °C) and under the target concentrations in patients with high fever (39 °C).

Figure 3b and d show the effect of different total plasma protein levels on sufentanil concentrations. With a dose of 17.5 μg h⁻¹, sufentanil concentrations were within target concentrations in most patients, whereas a dose of 12.5 μg h⁻¹ was low for patients with total plasma protein levels of 4–8 g dL⁻¹.