Anesthesia requires the administration of several drugs to achieve the required end points of hypnosis, immobility, and suppression of reflexes to noxious stimulation. It is usually given as the combination of a hypnotic and an opioid, with the anesthesiologist manually titrating doses or infusion rates of the 2 drugs to provide the best balance of hypnosis, analgesia, heart rate, and arterial blood pressure control during surgery. With knowledge and experience, clinical judgment of the right doses that account for the effects of the individual drugs, their interactions in combination, and patient factors such as size and age is often good. However, target-controlled infusions (TCIs), which aim to keep the concentration of drug in the blood or at the effect-site constant, make this judgment of anesthetic drug dose easier because they incorporate the initial bolus doses and decreasing infusion rates over time to keep predicted concentrations constant. Some algorithms also account for the changes in pharmacokinetics (PK) that occur with patient age and body size. However, significant changes with age also occur in pharmacodynamic (PD) responses, and anesthetic drugs interact to varying degrees at relevant clinical end points such as hypnosis and response to noxious stimulation. Incorporating more of this information into the computer algorithm controlling the drug infusion rate is an attractive proposition that should both increase the accuracy of anesthetic drug administration and simplify anesthetic drug dosing.
For this study, we developed a TCI algorithm that also incorporates information on the PD of propofol and remifentanil. The software algorithm allows the anesthesiologist to choose the desired degree of hypnosis using the Bispectral Index (BIS) as a measure of hypnotic depth and the ratio of the commonly used IV anesthetic drugs propofol and remifentanil to achieve the required BIS target. The algorithm then continuously calculates the doses of each drug required to keep the combined drug effect on BIS constant. In theory, this approach should allow more precise targeting of anesthetic depth and make for easier adjustment of dosage to consider the varying degree of analgesia or suppression of responses to noxious stimulation required during the time course of surgery. The algorithm uses published population models of the PK and PD of the 2 drugs. These models include the effects of the patient factors age, sex, weight, and height (i.e., obesity) on propofol and remifentanil PK, the effect of age on their PD, and the interaction between the 2 drugs on the targeted effect (BIS). Thus, the system is a dual-pump, PD TCI (PTCI) algorithm for the delivery of total IV anesthesia (TIVA).1–4Figure 1 illustrates the aspects of the drug pharmacology incorporated into the device and how it compares to TCI current practice.
The algorithm uses similar models to those incorporated in the Navigator (Navigator Applications Suite, GE Healthcare, Little Chalfont, UK) and Smartpilot View (Intelligent Display, Drager Medical AG & Co. KG, Lubeck, Germany) software packages. These displays passively process the infusion rates of propofol and remifentanil given intraoperatively and display their predicted effects at various end points. The model also incorporates a number of the features proposed in an editorial concerning the need for increased real-time PK, PD, and drug interaction data to be made available to anesthesiologists to improve the accuracy of drug delivery during anesthesia.5 The purpose of the current study was to prospectively test the feasibility and accuracy of the model for a PTCI approach to IV anesthesia.
The study received approval from our local research ethics committee (URB/11/06/014) and was registered with the Australian and New Zealand Trials Registry (Ref: ACTRN12611000428965). Written informed consent to participate was given by each participant.
Adult patients aged 18 to 80 years with an ASA physical status of 1 to 3 and who were scheduled for elective surgery were studied. Only those patients for whom TIVA with neuromuscular blockade was appropriate were approached. Patients who had an allergy or contraindication to the study drugs, a history of drug abuse, and neurologic dysfunction or who required premedication were not studied. This was an observational study in which a convenience sample of 50 patients was chosen, on the basis of the typical size of past studies of TCI.6–8
Participants received propofol and remifentanil anesthesia. Propofol was given as a 10-mg/mL solution, and remifentanil was given as a 40-μg/mL solution. Anesthesiologists were asked to refrain from administering other opioids, IV or volatile anesthetics, and from giving midazolam, ketamine, nitrous oxide, clonidine, or dexmedetomidine. Analgesics for postoperative pain were administered on emergence from anesthesia, defined as return to BIS >90.
The study drugs were infused using 2 Graseby 3400 (Sims Graseby Ltd., Watford, UK) syringe pumps connected to a dedicated laptop computer via their RS232C serial interfaces.a The pumps were controlled via a program that delivers effect-site TCI of propofol and remifentanil using the PK models described by Schnider et al.3,4 for propofol and Minto et al.2 for remifentanil. The TCI rates of propofol and remifentanil were calculated to achieve effect-site concentrations of each drug that together cause the targeted effect selected by the user. These effect-site concentrations were calculated using the response surface model described by Minto et al.9 and the parameter estimates for propofol and remifentanil effects on BIS described by Bouillon et al.1 (Table 1). In this model, the interaction between propofol and remifentanil on BIS is additive. The accuracy of this software has been validated against existing TCI pump software including Stanpumpb for the effect-site PK algorithm and using an explicit solution written on a spreadsheet for the response surface component of the software.
The software allows the user to choose a target BIS level and the desired ratio of propofol to remifentanil. The following equations for the combined effect of the 2 drugs are then used to provide estimates of the predicted effect-site concentration of each drug required to achieve the targeted effect. The desired ratio (θ) of propofol to remifentanil was determined as:
UP and UR are the predicted concentrations of propofol [P] and remifentanil [R], respectively, normalized to units of common potency using the predicted effect-site concentration of each drug that results in 50% of its maximal effect on BIS (C50). A ratio of 0 is just remifentanil, while a ratio of 1 is just propofol. An interaction term (β) can be applied to the parameters of the sigmoidal combined response model,9 as indicated below for the combined units of propofol and remifentanil (U):
In the case of the propofol–remifentanil interactions on BIS, the interaction is additive (β = 0)1 so that U = UP + UR. The predicted BIS from the combination of propofol and remifentanil using the equation for a standard sigmoid dose–response curve then becomes:
where γ is a slope parameter describing the shape of the response surface, E0 is baseline BIS, and Emax is the maximal drug effect.
The attending anesthesiologists, who were briefed on the project and trained in the use of the device before commencement of the study, operated the software. The software displays current and future predictions of plasma and effect-site concentrations for each drug, as well as both individual drug contribution and the overall BIS effect, in real time. An icon on the main display allows users to navigate directly to the BIS and ratio targets. Anesthesiologists adjusted the BIS target and desired drug ratio as needed throughout each anesthetic according to their clinical judgment, how well pump predictions aligned with observed BIS, and with expected changes in analgesic or sedative requirements (i.e., during incision, recovery phase). Examples of the predicted propofol and remifentanil effect-site concentrations that would result from chosen BIS targets and propofol to remifentanil ratios are given in an online supplement (Figure, Supplemental Digital Content 1, http://links.lww.com/AA/A939). An example that illustrates how the PTCI pump works in practice with model predictions of the effect-site TCI propofol and remifentanil infusions from the desired combined effect is given in Figure 2.
Electroencephalography (EEG) was monitored using a BIS DSC-XP (software version 4.0.2) monitor and BIS Quattro Sensors (Covidien, Dublin, Ireland) using a 15-second epoch. All monitoring data, including BIS, were recorded electronically at 30-second intervals using the SaferSLEEP electronic record-keeping system (Safer Sleep LLC, Nashville, TN).
The predicted and observed BIS data were reduced to 1-minute epochs for analysis. Only BIS values associated with a Signal Quality Index ≥50 were used. Model performance measures as described by Varvel et al.7 were used to assess how well the model predicted the observed BIS. Performance error (PE) was calculated at each time point using weighted residuals and expressed as a percentage:
where Y is the measured BIS value and Ŷ is the model predicted BIS value.7 The median PE (MDPE) was calculated as a measure of bias for each individual i as:
The median absolute PE (MDAPE) was used as a measure of model inaccuracy:
Wobble, which measures within-individual variation in PE, was calculated as:
The above measures were calculated for all time points and for defined time intervals to assess the performance of the model during the induction (0–20 minutes), maintenance (20 minutes to end of infusion), and emergence (end of infusion until eye opening to verbal command) phases of anesthesia.
Population data were summarized as the mean, standard deviation, and range of these individual medians. Model performance in each individual was also graphed. Poor performance of the algorithm was defined as a deviation of >50% from the population mean. This value was chosen by combining the expected variability in the individual drugs at the effect site and the combined-effect model.10 Patient demographic data are presented as mean and range. The influence of patient age and body mass index (BMI) on mean MDPE (as a measure of model performance) was examined using a 2-tailed Spearman ρ test to test linear correlations.
Fifty patients participated in the study. Five participants were excluded from our analysis: 4 received the excluded drugs midazolam and morphine, and infusion data were lost for 1 participant because of a computer failure. Patient demographic data are summarized in Table 2. The final analysis included 28 female and 17 male participants. Three participants did not receive neuromuscular blockade, which may have affected the quality of the BIS trace because of electromyography interference during their anesthetic, but were included in the analysis.
Measures of performance of the model for each phase of anesthesia, and for the total study time, are summarized in Table 3 and Figure 3. Overall performance was described by MDPE of 8% (SD 24%), MDAPE of 25% (SD 13%), and wobble of 15% (SD 7%). Performance was least accurate during the early induction phase of anesthesia compared to the maintenance and recovery phases. There were outliers at all time points with 7 patients tracking with a median bias >50% of model predictions. The predicted concentrations of propofol and remifentanil chosen are shown in Figure 4 and the predicted and observed BIS targets in Figure 5. The administered ratios of propofol to remifentanil are shown in Figure 6. The individual tracking of participants is available as a graph for each participant in an online supplement (Supplemental Digital Content 2, Supplemental Figure, http://links.lww.com/AA/A940).
Twelve participants were older than 60 years, and 15 participants had a BMI >30. Examination of these covariates showed nonsignificant bias (age: ρ –0.119 for N = 45, P = 0.44; BMI: ρ –0.09 for N = 45, P = 0.56), with the model overestimating drug requirements in both the young and the slim by about 10% compared to the elderly or obese (Figs. 7 and 8). Examination of performance versus target BIS and target drug ratio found no significant bias, but performance was frequently poor due to outliers (Figs. 9 and 10).
This is the first study to use a combined effects PD model to control drug administration during TIVA. By targeting the effects of the drug combination, the model incorporates significantly more population pharmacology than existing infusion algorithms where either drug infusion rates or blood concentration targets are chosen by the anesthesiologist. The PTCI model also has a theoretical advantage over current approaches in that by combining the effects of the drugs it allows users to easily adjust the ratio of opioid to anesthetic drug without altering anesthetic depth, enabling improved ability to independently control arterial blood pressure or analgesia while maintaining precise depth targeting.9 Although the PTCI device had a large bias (>50%) in some patients, this was easily compensated for by appropriate adjustment of the BIS target.
We chose the most comprehensive models available for propofol and remifentanil and their combined effects. The Schnider and Minto models include an assessment of both the PK and PD effects of propofol and remifentanil, respectively, and include the important covariates of age and BMI.2–4,11 They have also been prospectively validated and are used in commercially available infusion pumps.12–17 A weakness of using these models is that the Keo was derived from EEG spectral edge frequency (SEF) rather than the BIS. However, Billard et al.18 estimated similar Keo values for BIS and SEF for propofol EEG changes in adults, so the effect of this should be small. Using these models also allows the age-related changes in Keo for remifentanil to be incorporated into the infusion model. The Bouillon et al.1 model was chosen similarly for the combined effect PD response surface model for BIS because of its thorough assessment (i.e., the wide range of concentrations for both drugs over which the study was conducted, which covered the full dose–response surface).10,19 Notably, this model was created using a homogeneous group of healthy adults,1 and so did not include an analysis of the covariates of age or illness for their influence on the response surface model. It did, however, include a comprehensive assessment of the drug ratios required at sedative (tolerance to shouting and shaking) and noxious (response to laryngoscopy) end points. One limitation of these models is that they were derived in volunteers in independent studies. There is a possibility that the drugs interact at a PK level and the PD model could be altered by the sustained noxious stimulus of surgery. However, the overall bias in our study was not particularly large at 8%.
A limitation of this study is that concentration data were not collected. This means that performance of the PK component is unknown for each model and, therefore, the true performance of the novel PD component of the PTCI system could not be independently assessed. PK TCI models have been considered to be within acceptable limits of performance when the mean bias is ≤20% and inaccuracy is ≤30%.7,20,21 However, it is commonplace to report global values of performance assessments that may mask model inaccuracies occurring at distinct stages of anesthesia and in individual patients. For example, Glen and Servin assessed the inaccuracy of propofol PK models in patients with normal renal and hepatic function, splitting their analysis into distinct stages of anesthesia similar to those used here.12 This revealed a bias of –20% for Schnider’s model during the 21 minutes after infusion start, meaning concentrations predicted by the model were larger than the observed concentrations of propofol. Overprediction of drug concentration may in turn lead to overprediction of drug response and might have contributed to the high response predictions seen in the induction phase of this study (MDPE 14%). Misspecification arising from the Keo parameter also likely contributed to model performance error during this period.
A definition of clinically acceptable performance has yet to be established for PD models. However, the overall performance of the model was similar to past assessments of TCI, being best during the maintenance phase, when the average wobble was 13%. Minto et al.11 reported a bias of 7% and inaccuracy of 19% for their PD model describing remifentanil effects on SEF, while Schnider et al.4 reported a bias of 6% and inaccuracy of 17% for a similar model for propofol. Our overall bias of 8% and inaccuracy of 25% is a small deterioration on this performance. This inaccuracy is largely a reflection of the expected intrinsic subject-to-subject variability. Given the heterogeneity of the patients we studied and the varying surgical stimuli the patients received, we judged this as acceptable. Population means may, of course, obscure significant deviations in individual patients. The wobble of 13% was also similar to past studies of TCI where wobble of 10% to 19% was found, depending on the heterogeneity of the patients studied.12,22
We did not attempt retrospective fitting of other models to assess whether they may perform better than ours. The only other suitable assessment of this drug combination is that of Bouillon et al.,1 who in the PK component of model development found propofol to increase remifentanil concentrations because of a 40% reduction in central volume and distributional clearance and 21% reduction in elimination clearance.19 However, the model was derived in a comparatively homogeneous group of patients aged 20 to 43 years and did not include a Keo for either propofol or remifentanil because of limitations in the experimental design. Their conclusions were that the PK interaction was only important over the induction phase of anesthesia and that this interaction may account for the poor performance during induction with our chosen model.
Our PTCI algorithm is, in some respects, similar to the dual drug closed-loop control algorithms that have been successfully tested for propofol and remifentanil.23,24 A manual control algorithm has the advantage of not requiring the complexity and safety systems required for a closed-loop controller and the attendant regulatory requirements. It also has the benefit of providing future predictions of likely drug effect. The algorithm could also be used as the basis of a dual infusion closed-loop anesthesia system and could be improved by incorporating a model of noxious stimulus as well as hypnosis into the display. It is not possible to incorporate both the hypnotic and the immobilizing effects of the combination into the same infusion algorithm, so anesthetic judgment is still required.
Our study has several limitations. Multiple factors that may impact on model performance were not controlled for. All patients experienced surgical stimulus, and no attempt was made to limit the type of surgery included in the study. Pain types and intensities were therefore different among participants, influencing each individual’s requirements for opioids. The PD model describing propofol and remifentanil effects used was developed in a group of volunteers under carefully controlled circumstances1 and so may not reflect changes in BIS in surgical patients experiencing painful stimuli. Patient demographics spanned a wide range of ages and BMIs. Age, in particular, influences the PK and PD of both drugs, but only the PK and Keo components of the algorithm contained an adjustment for age.2–4 The lack of significant bias in our heterogeneous population provides reassurance that there was not an important age effect missing in the model and makes the study more relevant to the typical diversity of patients encountered in everyday anesthesia.
This study of PTCI found that the chosen model performed adequately in a clinical setting and could undoubtedly be refined further. Assessment of the utility of PTCI model–controlled anesthesia in comparison to standard drug dosing decisions made by the attending anesthesiologist is warranted. There is some evidence to suggest that overly deep anesthesia may be harmful.25,26 If this were to be confirmed, then use of PD models may be an effective method for improving the titration of depth of anesthesia without the complexity and safety issues of closed-loop anesthetic systems.
Name: Timothy G. Short, MBChB, MD, FANZCA.
Contribution: This author contributed to study design, data collection, analysis, manuscript preparation, and is the archival author.
Attestation: Timothy G. Short attests to the integrity of the original data and the analysis reported in this manuscript.
Name: Jacqueline A. Hannam, PhD.
Contribution: This author contributed to study design, data collection, and analysis; wrote the first draft of the manuscript, and edited and revised the final manuscript.
Attestation: Jacqueline A. Hannam attests to the integrity of the original data and the analysis.
Name: Stephen Laurent, MBChB, FANZCA.
Contribution: This author contributed to data gathering and writing.
Name: Douglas Campbell, BM, FRCA, FANZCA.
Contribution: This author contributed to data gathering and writing.
Name: Martin Misur, BHB, MBChB, FANZCA.
Contribution: This author contributed to data gathering and writing.
Name: Alan F. Merry, MBChB, FANZCA.
Contribution: This author contributed to manuscript preparation and provided project supervision.
Name: Yuk Ho Tam, BSc, MPhil, CEng, MIET, MHKIE.
Contribution: This author contributed to software pro gram ming.
This manuscript was handled by: Ken B. Johnson, MD.
a CCIP v2.4, written by YH Tam and available at: www.cuhk.edu.hk/med/ans/softwares.htm.
b Stanpump, open source code available at http://opentci.org.
1. Bouillon TW, Bruhn J, Radulescu L, Andresen C, Shafer TJ, Cohane C, Shafer SL. Pharmacodynamic interaction between propofol and remifentanil regarding hypnosis, tolerance of laryngoscopy, bispectral index, and electroencephalographic approximate entropy. Anesthesiology. 2004;100:1353–72
2. Minto CF, Schnider TW, Egan TD, Youngs E, Lemmens HJ, Gambus PL, Billard V, Hoke JF, Moore KH, Hermann DJ, Muir KT, Mandema JW, Shafer SL. Influence of age and gender on the pharmacokinetics and pharmacodynamics of remifentanil. I. Model development. Anesthesiology. 1997;86:10–23
3. Schnider TW, Minto CF, Gambus PL, Andresen C, Goodale DB, Shafer SL, Youngs EJ. The influence of method of administration and covariates on the pharmacokinetics of propofol in adult volunteers. Anesthesiology. 1998;88:1170–82
4. Schnider TW, Minto CF, Shafer SL, Gambus PL, Andresen C, Goodale DB, Youngs EJ. The influence of age on propofol pharmacodynamics. Anesthesiology. 1999;90:1502–16
5. Forrest FC, Tooley MA, Saunders PR, Prys-Roberts C. Propofol infusion and the suppression of consciousness: the EEG and dose requirements. Br J Anaesth. 1994;72:35–41
6. Marsh B, White M, Morton N, Kenny GN. Pharmacokinetic model driven infusion of propofol in children. Br J Anaesth. 1991;67:41–8
7. Varvel JR, Donoho DL, Shafer SL. Measuring the predictive performance of computer-controlled infusion pumps. J Pharmacokinet Biopharm. 1992;20:63–94
8. Syroid ND, Johnson KB, Pace NL, Westenskow DR, Tyler D, Brühschwein F, Albert RW, Roalstad S, Costy-Bennett S, Egan TD. Response surface model predictions of emergence and response to pain in the recovery room: an evaluation of patients emerging from an isoflurane and fentanyl anesthetic. Anesth Analg. 2010;111:380–6
9. Minto CF, Schnider TW, Short TG, Gregg KM, Gentilini A, Shafer SL. Response surface model for anesthetic drug interactions. Anesthesiology. 2000;92:1603–16
10. Short TG, Ho TY, Minto CF, Schnider TW, Shafer SL. Efficient trial design for eliciting a pharmacokinetic-pharmacodynamic model-based response surface describing the interaction between two intravenous anesthetic drugs. Anesthesiology. 2002;96:400–8
11. Minto CF, Schnider TW, Shafer SL. Pharmacokinetics and pharmacodynamics of remifentanil. II. Model application. Anesthesiology. 1997;86:24–33
12. Glen JB, Servin F. Evaluation of the predictive performance of four pharmacokinetic models for propofol. Br J Anaesth. 2009;102:626–32
13. Coppens M, Van Limmen JG, Schnider T, Wyler B, Bonte S, Dewaele F, Struys MM, Vereecke HE. Study of the time course of the clinical effect of propofol compared with the time course of the predicted effect-site concentration: performance of three pharmacokinetic-dynamic models. Br J Anaesth. 2010;104:452–8
14. Doufas AG, Bakhshandeh M, Bjorksten AR, Shafer SL, Sessler DI. Induction speed is not a determinant of propofol pharmacodynamics. Anesthesiology. 2004;101:1112–21
15. Struys MM, Coppens MJ, De Neve N, Mortier EP, Doufas AG, Van Bocxlaer JF, Shafer SL. Influence of administration rate on propofol plasma-effect site equilibration. Anesthesiology. 2007;107:386–96
16. Diago MC, Amado JA, Otero M, Lopez-Cordovilla JJ. Anti-adrenal action of a subanaesthetic dose of etomidate. Anaesthesia. 1988;43:644–5
17. Fragen RJ, Weiss HW, Molteni A. The effect of propofol on adrenocortical steroidogenesis: a comparative study with etomidate and thiopental. Anesthesiology. 1987;66:839–42
18. Billard V, Gambus PL, Chamoun N, Stanski DR, Shafer SL. A comparison of spectral edge, delta power, and bispectral index as EEG measures of alfentanil, propofol, and midazolam drug effect. Clin Pharmacol Ther. 1997;61:45–58
19. Bouillon T, Bruhn J, Radu-Radulescu L, Bertaccini E, Park S, Shafer S. Non-steady state analysis of the pharmacokinetic interaction between propofol and remifentanil. Anesthesiology. 2002;97:1350–62
20. Glass PS, Shafer SL, Reves JGR. Miller. Intravenous drug delivery systems. In: Miller’s Anesthesia. 2005 Philadelphia, PA Elsevier (Churchill Livinstone):439–80
21. Masui K, Upton RN, Doufas AG, Coetzee JF, Kazama T, Mortier EP, Struys MM. The performance of compartmental and physiologically based recirculatory pharmacokinetic models for propofol: a comparison using bolus, continuous, and target-controlled infusion data. Anesth Analg. 2010;111:368–79
22. Coetzee J, Glen J, Boshoff L. Pharmacokinetic model selection for target controlled infusions of propofol: assessment of three parameter sets. Anesthesiology. 1995;82:1328–45
23. Payen JF, Dupuis C, Trouve-Buisson T, Vinclair M, Broux C, Bouzat P, Genty C, Monneret D, Faure P, Chabre O, Bosson JL. Corticosteroid after etomidate in critically ill patients: a randomized controlled trial. Crit Care Med. 2012;40:29–35
24. Liu N, Chazot T, Hamada S, Landais A, Boichut N, Dussaussoy C, Trillat B, Beydon L, Samain E, Sessler DI, Fischler M. Closed-loop coadministration of propofol and remifentanil guided by bispectral index: a randomized multicenter study. Anesth Analg. 2011;112:546–57
25. Absalom A, Pledger D, Kong A. Adrenocortical function in critically ill patients 24 h after a single dose of etomidate. Anaesthesia. 1999;54:861–7
26. Jameson P, Desborough JP, Bryant AE, Hall GM. The effect of cortisol suppression on interleukin-6 and white blood cell responses to surgery. Acta Anaesthesiol Scand. 1997;41:304–8