Prospective clinical validation of the Eleveld propofol pharmacokinetic-pharmacodynamic model in general anaesthesia.

BACKGROUND
Target-controlled infusion (TCI) systems incorporating pharmacokinetic (PK) or PK-pharmacodynamic (PK-PD) models can be used to facilitate drug administration. Existing models were developed using data from select populations, the use of which is, strictly speaking, limited to these populations. Recently a propofol PK-PD model was developed for a broad population range. The aim of the study was to prospectively validate this model in children, adults, older subjects, and obese adults undergoing general anaesthesia.


METHODS
The 25 subjects included in each of four groups were stratified by age and weight. Subjects received propofol through TCI with the Eleveld model, titrated to a bispectral index (BIS) of 40-60. Arterial blood samples were collected at 5, 10, 20, 30, 40, and 60 min after the start of propofol infusion, and every 30 min thereafter, to a maximum of 10 samples. BIS was recorded continuously. Predictive performance was assessed using the Varvel criteria.


RESULTS
For PK, the Eleveld model showed a bias < ±20% in children, adults, and obese adults, but a greater bias (-27%) in older subjects. Precision was <30% in all groups. For PD, the bias and wobble were <5 BIS units and the precision was close to 10 BIS units in all groups. Anaesthetists were able to achieve intraoperative BIS values of 40-60 using effect-site target concentrations about 85-140% of the age-adjusted Ce50.


CONCLUSIONS
The Eleveld propofol PK-PD model showed predictive precision <30% for arterial plasma concentrations and BIS predictions with a low (population) bias when used in TCI in clinical anaesthesia practice.

Target-controlled infusion (TCI) systems are used to facilitate anaesthetic drug administration, particularly for short-acting drugs used in anaesthesia. TCI systems calculate the drug dose and infusion rate necessary to achieve and maintain a targeted drug concentration in the plasma or an effectcompartment based on a pharmacokinetic (PK) or PKpharmacodynamic (PK-PD) model. Widespread clinical experience has been gained and TCI is considered a mature technology. 1 Current TCI systems incorporate PK or PK-PD models developed for specific patient groups, such as adults, children, older patients, and obese adults. Extrapolation of a model developed for one population to a patient from a different population is discouraged because of the uncertainty of whether model predictions and TCI drug administration will be appropriate in the other population. This may result in under-or overdosing, leading to superficial anaesthesia or delayed recovery. Some degree of extrapolation is unavoidable in routine clinical practice because the diversity across patients and clinical conditions exceeds that which was present in the study subjects from which the PK and PK-PD models currently available in TCI systems were developed.
Recently, Eleveld and colleagues 2 developed a propofol PK-PD model that can be incorporated in a TCI system used to administer propofol for general anaesthesia and sedation. This second generation model 3 was developed using PK-PD data from different studies from a broad, diverse, population, including data from neonates, children, adults, and older subjects, and including both volunteers and patients with conditions such as obesity, alcoholism, liver cirrhosis, and cancer, along with varying approaches to concomitant drug administration, such as opioids. In general, greater diversity in the data used to develop a model increases the chance of a match between the model and a given clinical situation, 4 thus reducing the risks associated with extrapolation. The PK predictive performance of the Eleveld model has been evaluated in adults 5 and the underweight. 6 It is not currently known how accurately the Eleveld PK-PD model predicts plasma propofol concentrations and bispectral index (BIS) values when used by a TCI system for propofol administration to a group of patients with diverse characteristics. The aim of the current study was to validate prospectively the Eleveld propofol PK-PD model in a broad range of patients, from children to older patients, including obese adults, by assessing its predictive performance for arterial plasma concentrations and BIS in routine practice using TCI. Our hypothesis was that its predictive precision is not significantly worse than specialised models and that it is clinically applicable for all subgroups. The secondary objective was to identify effect-site target concentrations for the models that are associated with adequate anaesthesia.

Methods
This investigator-initiated trial was conducted in the Department of Anesthesiology of the University Medical Center Groningen (UMCG), Groningen, The Netherlands. The Medical Ethics Committee of the UMCG approved the study (METc2018/ 216). The study was registered in The Netherlands Trial Register (NTR7146). All medical devices used in this study are approved for use in clinical research, and all drugs and routes of administration are approved for clinical use.

Subjects
Four groups of 25 subjects each, undergoing an elective surgical procedure, were included. Eligibility criteria were, age of 3 yr or more, an ASA physical status between 1 and 4, and an expected surgery duration of at least 1 h, with a surgery-or patient-related indication for arterial cannulation. Exclusion criteria were use of benzodiazepines preoperatively (chronically or as premedication), contraindications to use of propofol, inclusion in other perioperative interventional studies, or patients admitted to the ICU before surgery, had received propofol in the preceding 24 h, or both. Written informed consent was obtained from all subjects who were !16 yr of age. For subjects between 12 and 15 yr of age, consent was obtained from their parents/caregivers, and for subjects <12 yr of age informed consent was obtained from their parents/ caregivers.

Study design
We enrolled 25 subjects in each of the following four groups according to age, weight, or both: children (three age <18 yr), non-obese adults (18 age <70 yr; BMI < 30 kg m À2 ), older patients (age !70 yr), and obese adults (18 age <70 yr; BMI!30 kg m À2 ). Obese children and older subjects were not excluded.
Arterial blood samples were collected at 5, 10, 20, 30, 40, and 60 min after the start of propofol TCI, and every 30 min thereafter until end of surgery or 10 samples in total had been collected. If the target concentration had been recently increased, sampling was delayed by 3e5 min to reduce the possible influence of possible front-end kinetics on observed concentrations, and sampling times were recorded. All study procedures stopped after the last sample was collected. The TCI system remained connected and in use for the on-going delivery of propofol until the end of the procedure.

Study procedures
All procedures took place in the operation centre of the UMCG. All subjects were fasted according to local adult and paediatric protocols. No sedative premedication was allowed. Upon arrival in the operation theatre, an i.v. cannula was inserted and used for fluid (Ringer's lactate, Baxter, Deerfield, IL, USA) and drug administration, including Propofol (Fresenuis Kabi, Bad Homburg, Germany). Subjects were connected to a Philips IntelliVue MP50 vital signs monitor (Philips Medical Systems, Eindhoven, The Netherlands) using standard monitoring (ECG, noninvasive BP monitoring, pulse oximetry [SpO 2 ]). A BIS monitor (BIS Vista, Medtronic, Boulder, CO, USA) with bilateral BIS sensor used to record BIS. An arterial catheter was inserted into a radial artery under local anaesthesia before induction, in all subjects except in children and subjects refusing placement before induction. In these subjects, cannulation was performed as soon as possible after the induction of anaesthesia. The arterial line was used for the collection of blood samples to measure the propofol plasma concentration and for arterial BP monitoring. After induction of anaesthesia, the subject's trachea was intubated. Further perioperative care was according to standard care protocols and at the discretion of the anaesthetist. For analysis, recorded data were sampled at 10 s intervals.
An anaesthetic team, consisting of a qualified anaesthetist and a nurse anaesthetist, was responsible for perioperative care as per clinical practice. A research nurse was responsible for drawing the blood samples, and processing and storage of the plasma samples. A research student or PhD student was responsible for controlling RUGLOOP (Rugloop, DEMED, Temse, Belgium) and for other study procedures.

Drug administration
Patients received propofol delivered by a syringe pump (Ala-ris™ GH syringe pump, Becton Dickinson (BD), Franklin Lakes, NJ, USA) controlled by a computer running RUGLOOP II software (Demed, Temse, Belgium) for Windows (Microsoft, Redmond, WA, USA). The RUGLOOP II software was programmed to deliver propofol by TCI using the Eleveld PK-PD model. As opioids were used in all procedures, 'concomitant opioid use' was selected when subject covariates were entered into RUGLOOP.
Anaesthetists were asked to set the initial target concentrations equal to the age-specific effect-site concentration corresponding with 50% of maximal effect (Ce 50, corresponding with a predicted BIS of 47) for the induction of anaesthesia (Fig. 1). This was predicted to result in induction doses close to the recommendations in the propofol product label. 7 Adjustments to the target concentration during the procedure were at the discretion of the anaesthetist, with the advice to maintain BIS values between 40 and 60.
Remifentanil (TCI), sufentanil (TCI), or fentanyl (manual boluses, in some children, <10 yr only) were used for analgesia during the procedure. Inhaled anaesthetics, benzodiazepines and ketamine were not allowed during the study. If the anaesthetist deemed the use of these drugs necessary, then they were administered later in the procedure (at least after 1 h, preferably after the end of the study, 3 h after the start of propofol). PK and PD data from after the time of administration of these drugs were excluded from further analysis.

Measurement of drug effect
The cerebral effect of propofol was measured using bilateral frontal processed EEG monitoring (BIS®, BIS Vista, Medtronic, Dublin, Ireland). A BIS value of 100 represents the awake state and a BIS of 0 defines no detectable electrical brain activity. BIS values between 40 and 60 are considered to be appropriate for general anaesthesia. 8

Analysis of blood samples
Blood was collected in ethylenediaminetetraacetic acid tubes (EDTA) Becton Dickinson (BD) and stored at room temperature for a maximum of 60 min. Afterwards, samples were centrifuged (Labofuge 400R, Heraeus Holding GmBh, Hanau, Germany) for 5 min at 3200 rpm at 20 C. Plasma was transferred into cryovials and stored at À80 o C until analysis. Propofol plasma concentration analysis was performed by the Department of Clinical Pharmacy and Pharmacology of the UMCG using ultra-high-performance liquid chromatographyemass spectrometry (TSQ-Quantiva, Thermo Scientific, Waltham, MA, USA) in atmospheric pressure chemical ionization mode.
The lower and upper limits of quantification were 0.1 and 25 mg L À1 , respectively. The method was validated according to European Medicines Agency 9 and Food and Drug Administration 10 regulations and complied with requirements set in these regulations.

Data handling
Recorded BIS values were used for analysis when the signal quality index was >50. BIS signals recorded from the subjects' left and right sides were collected into a single signal and median filtered over 30 s periods to obtain the observed BIS value. Propofol infusion rates and cumulative dose were calculated from the (difference in) infused volume at each 10 s sampled time interval. Model predictions were simulated in NONMEM 7.4 (ICON Development Solutions, Ellicott City, MD, USA) and analysed using R (version 2.14.1; R Foundation for Statistical Computing, Vienna, Austria).

PK-PD predictive performance
The predictive performance of the PK model predicting the time course of the arterial propofol concentrations was evaluated using the Varvel criteria 11 with performance error for observation j in individual i calculated as: These are summarised for each individual as MdPE i and MdAPE i as: These quantify prediction bias, precision, and the variability within an individual, respectively, and we report mean (range) values across the population. An MdPE<20% and MdAPE<40% have been suggested to indicate clinically acceptable performance. 12,13 Other PK and PK-PD models for comparison (and patient groups) considered were the Schnider model 14 (adults, older subjects, obese) the Marsh model 15 (adults, older subjects, obese), the Marsh model with correction by Servin 16 (obese), the Cortinez model 17 (adults, older subjects, obese) and Cortinez 'obese' model 18 (obese), the Kataria model 19 (children), the Paedfusor model 20,21 (children), and the Araú jo model 22 (adults, older subjects, obese). We compared model differences in MdAPE i and Wobble i using a one-sided (other model performs better) Wilcoxon signed rank test. We considered P<0.01 as significant to compensate for multiple testing as we compared the Eleveld PK-PD model with two PK models for children, four PK models for adults, four PK models for the older subjects, and five PK models for the obese.
Monte Carlo simulations of the study design were performed including random observation error to obtain an a priori estimate of statistical power. With 25 subjects in each subgroup, the simulation median (5e95% percentile) for MdAPE was 19% (15e22) across all groups.
The PD predictive performance was calculated in a similar manner to the PK predictive performance except we used a formula for PE ij more suitable for additive error models: Differences in BIS values <5 were considered clinically irrelevant. 23

Results
We screened 167 patients for inclusion in the study; the consort diagram is shown in Supplementary Figure S1. Informed consent was obtained from 112 patients, their caregivers, or both. Ten patients were excluded before any study measurements were made for various reasons (i.e. preoperative benzodiazepine usage, not able to obtain i.v. access preinduction in children). The data from a further two subjects were excluded from analysis. In one subject this was because of a brief leakage from the propofol administration system leading to uncertainty in administered propofol dose, and in the other an event occurred during induction which led to a suspicion of preoperative recreational drug use. Excluded patients were replaced and the data from 100 subjects, 25 in each subgroup, were used for analysis. The characteristics of the population are shown in Table 1. Age, weight, height, and BMI are presented as median (range). The children and older groups include one and three obese patients, respectively. The PK and PD observations are included as Supplementary Digital Content. Table 2 shows the predictive performance of the Eleveld PK-PD model and other models. The Eleveld model showed bias (MdPE) <20% in children, adults, and obese adults (À4.4%, À14.1%, À14.1%, respectively) but there was a greater bias (À27%) in older subjects. Precision (MdAPE) was <30% for all groups. Some of the other models showed lower average PK MdAPE than the Eleveld PK-PD model, but the differences were not significant at a level of P<0.01. PK predictive performance vs time is shown in Supplementary Figure S2. For BIS predictions, MdPE and wobble were <5 BIS units, which is considered to be clinically irrelevant. 23 MdAPE for BIS was close to 10 BIS units for all groups, suggesting that considerable interindividual variability exists in overall propofol BIS drug effect. BIS predictive performance vs time is shown in Supplementary Figure S3.
For Figures 2e4, distributions of study data are shown as percentiles (5e95% light grey; 25e75% dark grey; median black line) for each interval. Smoothing was achieved by including immediately adjacent intervals in each percentile calculation. Green shaded induction doses and infusion rates indicate recommended drug administration for anaesthesia from the propofol product label. 7 These were: children (3 yr < age < 16 yr: induction 2.5e3.5 mg kg À1 , maintenance 12e18 mg kg À1 h À1 for the first 30 min, followed by 7.5e9 mg kg À1 h À1 ), adults (18 < age < 55 yr: induction 2e2.5 mg kg À1 , maintenance 9e12 mg kg À1 h À1 for 15 min, followed by a reduction of 40% until 30 min, followed by 3e6 mg kg À1 h À1 ), and older subjects (age >55 yr: induction 1e1.5 mg kg À1 , maintenance 3e6 mg kg À1 h À1 ). Green shaded BIS values indicate the range 40e60 which is considered to be adequate for intraoperative anaesthesia.
Smoothed distributions of the effect-site target concentration selected by the anaesthetists, infusion rates, and achieved BIS values over 1 min intervals for each group are shown in Figure 2. Target concentrations and infusion rates were higher (per kg) in children compared with adults, whereas these were lower in older subjects and obese adults. Despite these differences in target concentration and drug administration between groups, similar BIS values of 40e60 were observed for all groups. Figure 3 shows the relationship with age for effect-site target concentration, induction dose (cumulative dose over the first 2 min), and BIS values during TCI. The target concentrations selected by the anaesthetist follow a similar decrease with age as predicted by the age-adjusted Ce 50 of the model. The resulting induction dose matched with the product label recommended doses across the age range. 7 The achieved BIS values were similar for all age groups and showed good agreement with the intended BIS values of 40e60. The median (5e95% percentile) observed BIS during TCI was 46 (33e63).
To gain insight into the selection of effect-site target concentrations by the anaesthetists, we plotted these relative to the Ce 50 (%Ce 50 ) vs age, time, and observed BIS values, shown in Figure 4. When the target concentration was expressed relative to the Ce 50 , then was no longer a relationship with age, suggesting that the incorporation of the age adjustment to the Ce 50 resulted in an acceptable depth of anaesthesia during clinical use. Target concentrations are often increased in the early phase of anaesthesia, but tend to be lowered more towards the Ce 50 value after~45 min. Figure 4 also shows dose individualisation or 'titration-toeffect' performed by the anaesthetists. Before the initial onset of drug effect, BIS values of about 90e100 were observed and anaesthetists tended to choose target concentrations close to the Ce 50 . In this period, the predicted effect-site concentration is still increasing towards the target. When observed BIS values were in the 60e80 range, then target concentrations were typically increased to~110% of Ce 50 . In contrast, when observed drug effects were greater than intended, with BIS values <40, target concentrations tended to be lowered, reaching 85% of Ce 50 for a BIS value of 20. At BIS values around 40, anaesthetists typically selected target concentrations matching the Ce 50 .

Discussion
We prospectively evaluated the predictive performance of the Eleveld PK-PD model in a TCI system for children, adults, older subjects, and obese adults with regard to propofol concentrations (the PK component) and BIS values (the PD component). For propofol plasma concentrations, precision (MdAPE) <30% was found in all groups. Bias (MdPE) <20% was observed in children, adults, and obese adults, but a greater bias (À27%) was observed in older subjects. Our results are similar to those found by Hü ppe and colleagues 5 in adults (bias À18%; precision 22%). Yi and colleagues 6 prospectively validated the propofol PK model from Park and colleagues 24 in underweight Korean patients and also evaluated the Eleveld PK-PD model and found similar results (bias À18.6%; precision 23.1%). This suggests that the PK properties of this special group 25 are similar to the non-obese adults tested here. Only the Eleveld PK-PD model was applicable over all groups considered: children, adults, older subjects, and obese adults. While some other PK models had a lower MdAPE (i.e. better precision) than the Eleveld PK-PD model, there was considerable interindividual variability in MdAPE i and the difference did not reach significance. This suggests that the observed lower MdAPE may not be reliably repeatable across investigations.
For BIS observations, the Eleveld PK-PD model showed negligible bias (<5 BIS units) for all groups, whereas precision was about 8e10 BIS units for all groups. This suggests that dose individualisation (titration-to-effect) by adjusting target concentrations will remain an essential part of propofol TCI drug administration. These adjustments are needed to compensate for the biological interindividual PK and PD variability so that an adequate drug effect can be achieved in each patient.
During the study procedures, effect-site target concentrations chosen by the anaesthetist varied between 85% and 140% of the age-adjusted Ce 50 . Target concentrations exceeded the Ce 50 in the early phase of anaesthesia and when observed BIS values were in the 60e80 range. These adjustments are possibly in anticipation of stimulation as a result of laryngoscopy and intubation, noxious surgical stimulation, or, in case of a BIS value between 60 and 80, individuals showing insensitivity to propofol drug effects. Target concentrations were sometimes decreased below the Ce 50 when BIS values were in the 20e40 range. These patients may have greater pharmacodynamic sensitivity to propofol, or variability in pharmacokinetics that resulted in higher than predicted plasma concentrations.
The anaesthetists titrated propofol target concentrations to achieve clinically adequate depth of anaesthesia as assessed by BIS and other clinical monitoring. The practice of defining target concentrations relative to the Ce 50 is also found in clinical use of inhaled anaesthetics (minimum alveolar concentration) and in drug-interaction research for developing response surface models to predict interactions between two or more drugs on drug effect. 26 As Eleveld and colleagues 2 indicated in the publication of their PK-PD model, selecting TCI effect-site target concentrations equal to the model age-adjusted Ce 50 would result in predicted propofol induction doses and maintenance infusion rates close to those recommended in the propofol product label. 7 This was confirmed in the current study as the anaesthetists were able to obtain and maintain adequate anaesthesia using targets close to Ce 50 in all groups. This is the first study showing that TCI with a single PK-PD model can result in clinically adequate anaesthesia for a diverse group of patients, achieving a BIS between 40 and 60, while drug dosing remains close to product label recommendations. 7 Dose individualisation or 'titration-to-effect' is likely more difficult when drug response shows intraindividual variability (wobble) compared with predicted values. The Eleveld PK-PD model showed lower PK wobble than other models and wobble for BIS was <5. This suggests that dose individualisation with the Eleveld PK-PD model will be at least as successful compared with other models.
We found that the Eleveld model showed high PK bias in older subjects but a low bias was observed in all groups for prediction of BIS values. It seems likely that the overprediction of arterial concentrations in older subjects is compensated by the age-adjustment in the Ce 50 , resulting in BIS predictions and clinical effects that are adequate for anaesthesia. The origin of the PK bias in older subjects is not clear. It may be caused by differences in the individuals and studies used to develop the Eleveld PK-PD model compared with the individuals studied here. PK bias may be an issue for future scientific studies using the Eleveld model in older subjects. However, it has little impact for TCI in clinical practice, since during this study clinicians were not aware of PK bias during the execution of the study, and it only became apparent after sample analysis.
This study required an arterial line for sampling and this may bias the study population based on their medical history or surgery indication. We cannot be certain that the subjects included in this study are representative of other groups with different procedures or concomitant drug administration. Also, the Varvel performance measure for arterial samples is asymmetric, having a lower bound of À100% and an upper bound of infinity, and this may influence the comparisons of predictive performance between models.
In conclusion, the Eleveld PK-PD model showed predictive precision <30% for arterial propofol plasma concentrations and BIS predictions with a low (population) bias. None of the other PK models considered were applicable across all of the patient groups tested. The Eleveld model was not inferior in its predictive performance compared with the other models.
Anaesthetists used the Eleveld model in TCI to achieve clinically relevant intraoperative BIS values for children, adults, older subjects, and obese adults using effect-site target concentrations of about 85e140% of the age-adjusted Ce 50 .