EEG-controlled closed-loop dosing of propofol in rats

A. Tzabazis, H. Ihmsen*, M. Schywalsky and H. Schwilden

Klinik für Anästhesiologie, Universität Erlangen-Nürnberg, Krankenhausstrasse 12,D-91054 Erlangen, Germany

*Corresponding author. E-mail: Harald.Ihmsen@kfa.imed.uni-erlangen.de

Accepted for publication: November 11, 2003


    Abstract
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Appendix: Feedback system
 References
 
Background. Based on previous pharmacokinetic and pharmacodynamic studies, we have developed an EEG-controlled closed-loop system for the i.v. hypnotic agent propofol in rats.

Methods. Seven adult male Sprague–Dawley rats (weight 423–584 g) were included in the study. EEG was recorded with occipito–occipital needle electrodes and the EEG power spectrum was estimated. The median frequency (MEF) was extracted from the power spectrum and was modified MEF (mMEF) to account for the occurrence of spikes and burst suppression patterns in the EEG. Propofol infusion was controlled by a model-based adaptive control algorithm to maintain a set point of mMEF=3.0 (SD 0.5) Hz. The performance of the feedback system was characterized by the median performance error MDPE=median{(mMEF–set point)/set point} and the median absolute performance error (MDAPE). The effective therapeutic infusion (ETI) to maintain the set point was determined from the resulting infusion rates.

Results. In all rats a feedback period of 90 min could be performed. Mean MDPE was 1.2 (SE 0.4)% and MDAPE was 13.9 (0.3)%. The ETI was 0.73 (SD 0.20) mg kg–1 min–1. Mean arterial pressure before propofol infusion was 148 (14) mm Hg, with the lowest value during closed-loop infusion being 110 (20) mm Hg.

Conclusions. The feedback system presented here may be a useful tool not only for automatic drug control to maintain a defined hypnotic effect but may also be a powerful device in pharmacological studies such as the determination of dose requirements or the assessment of drug–drug interactions.

Br J Anaesth 2004; 92: 564–9

Keywords: anaesthetic techniques, i.v. closed-loop controlled infusion; anaesthetics i.v., propofol; brain, electroencephalography; rat


    Introduction
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Appendix: Feedback system
 References
 
Anaesthesia is a state where the conscious response of the treated subject is not available to optimize drug dosing. It is therefore useful to measure surrogate parameters like the EEG to adjust drug administration. These objective surrogate parameters are suitable to serve as an input for a feedback-controlled drug delivery system. It has been proposed that feedback models can be used in preclinical and clinical pharmacokinetic and pharmacodynamic studies for determining characteristic drug effects at an early stage and thus accelerate the testing and approval of new drugs.1 2

Feedback models are also powerful devices in pharmacological studies and for the assessment of drug–drug interactions. In circumstances where clinical or ethical problems occur in volunteer or patient studies, feedback-controlled animal models might at least indicate a trend for the type of interaction in humans and can lead to a distinct reduction in the number of animals needed to obtain significant results. The aim of this study was to establish an EEG-controlled closed-loop system for the i.v. hypnotic agent propofol in rats, based on previous pharmacokinetic and pharmacodynamic findings.3


    Methods
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Appendix: Feedback system
 References
 
Animals
After approval by the local German legislative control (AZ 621-2531.31-4/98, Reg. V. Mfr.) seven adult male Sprague–Dawley rats (Crl:CD), mean weight 488 (SD 51) g (range 423–584 g), were included in the study. Animals were delivered by Charles River Wiga GmbH, Sulzfeld, Germany at least 7 days before the experiments to allow for quarantine and acclimatization. Animals were healthy with respect to serology, bacteriology, parasitology and pathology. The rats were housed in pairs in polycarbonate cages type III (Uno BV, the Netherlands) on standard research bedding (soft wood fibre, Altromin) at 21.0 (0.5)°C, 60% humidity, 12 h light/dark cycle, with pelleted standard rodent diet (No. 1320, Altromin) and tap water ad libitum.

Instrumentation
For instrumentation, animals were anaesthetized with ketamine (Ketavet® 100 mg ml–1, Pharmacia, Germany) 158 (19) mg kg–1 i.p. This dose ensured adequate anaesthesia and analgesia for at least 60 min without any reaction to surgical procedures. Incision sites were infiltrated with lidocaine 2%. The right jugular vein and the left femoral artery were cannulated for administration of propofol 10 mg ml–1 (Diprivan®, AstraZeneca, Germany) and blood gas analysis, respectively. Stainless steel EEG needle electrodes were placed occipito–occipitally. The trachea was intubated for artificial ventilation to maintain stable blood gas status. During artificial ventilation the rats were paralysed with repetitive doses of pancuronium. Temperature was maintained at 36.5 (0.5)°C with a heating pad. Arterial pressure was measured continuously via the arterial cannula.

EEG processing and pharmacodynamic analysis
One-channel EEG was continuously recorded with an Aspect A1000 monitor (Aspect medical systemsTM, Natick, MA, USA). The digitized EEG signal was processed online with internal EEG analysis software (sampling rate 128 Hz, epoch length 8 s) and the median frequency (MEF) of the power spectrum (0.5–32 Hz) was determined using a fast Fourier transformation. In previous studies3 4 we found that the EEG of rats under propofol anaesthesia shows burst suppressions (BS) and spike-like patterns with high frequency components so that the MEF first decreases with increasing propofol concentration and then paradoxically increases. We therefore introduced a modified MEF (mMEF), which decreases continuously with increasing propofol concentration; this mMEF was also used in the present study. The mMEF algorithm uses pattern recognition to identify spikes and modifies the MEF in a manner similar to that which has been used for BS.3 5

Drug administration
When the mMEF was at least 8 Hz, propofol administration was started, targeting constant propofol blood concentrations by means of a pharmacokinetic model (target-controlled infusion, TCI; see Appendix). As an EEG set point we chose an mMEF of 3 (0.5) Hz, based on previous experience with propofol. At this level, we have a relatively deep anaesthesia, the EEG is characterized by spike-like patterns but the incidence of BS is low and the propofol-induced arterial pressure decrease is not too profound. As the mMEF can further decrease until a minimum value of 0 Hz, which will be reached if the EEG is completely suppressed, a set point of 3 Hz will avoid the ceiling effect where the mMEF is virtually independent of drug concentration. When mMEF was close to the chosen EEG set point of 3 (0.5) Hz, the EEG-controlled closed-loop administration (feedback mode) was started and maintained for 90 min. Propofol infusion was controlled by a model-based adaptive control algorithm, as explained in detail in the Appendix.

Analysis of the EEG and control of the propofol infusion were performed on a notebook computer with software written by the authors. Infusion rates, cumulative doses and derived EEG parameters were stored on disk for further investigation. The effective therapeutic infusion rate (ETI), as proposed some time ago by Schwilden and colleagues,6 was calculated as the slope of the cumulative dose curves during closed-loop controlled infusion for each individual animal. A normal distribution function was fitted to the distribution of the individual ETI values, and the 50%, 75% and 95% points (ED50, ED75, ED95 respectively) as well as the interquartile range (ED25–ED75) were determined from the fitted function. To provide arousal stimuli and avoid natural sleep during feedback-controlled drug administration, rats received nociceptive stimuli (tail squeeze) which were randomized with respect to time and intensity.

Assessment of the performance of the feedback system
The feedback system was characterized by several performance measures as proposed by Varvel and colleagues.7 The performance error (PE) was determined as relative deviation from the set point:


where mMEFij is the jth measured mMEF in the ith animal. As a quantitative measure of bias we determined in each animal the median PE:


and as a quantitative measure of inaccuracy the median absolute PE:


whereby Ni is the number of performance errors in the ith animal. Furthermore, we performed in each animal a linear regression of |PEij| versus time and defined the divergence Di as the slope of the regression line, which is a measure for a time-related trend of the performance. The wobble Wi=median {|PEij – MDPEi|, j =1,...Ni} was calculated for each animal as a measure for the intra-individual variability of the PE. The performance in the population was characterized by the unweighted mean of MDPEi, MDAPEi, Di and Wi and SEM, as defined by Varvel and colleagues.7 All other data are presented as mean (SD), unless stated.


    Results
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Appendix: Feedback system
 References
 
Propofol administration was started 107 (34) min after ketamine. User-defined targets of propofol concentration (TCI mode) to get close to the chosen EEG set point of 3.0 (0.5) Hz were 5.6 (1.6) µg ml–1. Feedback-controlled infusion was started 35 (23) min after the start of TCI. In all rats a feedback period of 90 min could be performed. Figure 2 shows the time course of the performance quality for each animal, expressed as mMEF/set point. MDPE and MDAPE were 1.2 (SE 0.4)% and 13.9 (0.3)%, respectively, divergence was –3.5 (0.5)% h–1 and wobble was 13.4 (0.3)%. The mean mMEF for all animals during closed-loop infusion was 3.0 (0.7) Hz. Figure 3 shows the worst and the best case of the study, as indicated by the values of MDPE and MDAPE. Cumulative doses increased linearly and were 46 (11) mg in total and 36 (14) mg during closed-loop infusion (Fig. 4). The ETI rate was 0.73 (0.20) mg kg–1 min–1. Figure 5 shows the distribution of the ETI values to achieve the set point of 3.0 (0.5) Hz. From the probability curve, ED50, ED75 and ED95 were determined to be 0.73, 0.87 and 1.07 mg kg–1 min–1, respectively. The interquartile range ED25–ED75 was 0.60–0.87 mg kg–1 min–1.



View larger version (25K):
[in this window]
[in a new window]
 
Fig 2 Performance of the feedback control expressed as the ratio of the measured effect and the set point. Each line represents one case. Time is given in minutes since the start of closed-loop control.

 


View larger version (28K):
[in this window]
[in a new window]
 
Fig 3 Worst case and best case in this study. Time is given in minutes since the start of closed-loop control. The grey line shows the time course of the modified median frequency (mMEF); the black line depicts the cumulative dose of propofol. The set point of 3.0 (0.5) Hz is indicated by the dashed lines. MDPE, median performance error (PE); MDAPE, median absolute PE; D, divergence of the absolute PE; W, wobble of the PE.

 


View larger version (14K):
[in this window]
[in a new window]
 
Fig 4 Cumulative doses for each animal during closed-loop infusion of propofol. The bold line represents the mean for all rats.

 


View larger version (12K):
[in this window]
[in a new window]
 
Fig 5 Distribution of the individual effective therapeutic infusion rates (dots) for an effect of modified median frequency (mMEF)=3.0 (0.5) Hz and fitted to a normal distribution function (line). ED50, ED75 and ED95 (see text) can easily be derived from the plot.

 
As described previously, EEG spike patterns were observed shortly after the start of the target-controlled propofol administration and before the occurrence of EEG BS (Fig. 6). The BS ratio (BSR), defined as the fraction of the EEG epoch where the EEG is suppressed, was 6.1 (5.7)% for all animals. The number of spikes increased with increasing propofol concentrations.



View larger version (33K):
[in this window]
[in a new window]
 
Fig 6 Typical EEG signals (one epoch of 8 s each) observed in rats: (1) before propofol administration the EEG was characterized by high activity in the {alpha} and ß band; (2) at a propofol infusion rate of 0.3 mg kg–1 min–1 the EEG activity was shifted to lower frequencies; (3) at a propofol infusion rate of 0.6 mg kg–1 min–1 propofol spike patterns occurred.

 
Mean arterial pressure (MAP) before the start of propofol infusion was 148 (14) mm Hg and lowest MAP during closed-loop infusion was 110 (20) mm Hg. After stopping the closed-loop infusion MAP reached baseline values within 3–5 min.


    Discussion
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Appendix: Feedback system
 References
 
We have developed a system for EEG-controlled feedback administration of propofol in rats as a research tool for pharmacokinetic/dynamic studies. In a traditional pharmacological approach, the dose is manually adjusted and the resulting effect is observed. With feedback systems, a defined pharmacodynamic effect is set and the dose necessary to achieve and maintain this effect will be found by the control algorithm. One application for a feedback-controlled dosage system is the speedy determination of dose requirements from which the ED50 or higher values such as ED75 or ED95, which have more practical relevance, can be determined easily (Fig. 5). Furthermore, feedback systems are useful devices not only to accelerate the testing of new drugs, but also for the evaluation of interactions. Drug–drug interactions are usually investigated by the construction of isoeffective curves (isoboles) on a graph with concentrations or infusion rates of drug A and drug B on the axes. A common problem with this method is the assessment of the end-points of the isobole.8 In volunteer or patient studies there are often ethical or clinical problems in studying isoeffective drug combinations with very low concentrations or infusion rates of one drug. These end-points could easily be assessed in animal models and the resulting isobole could at least indicate a trend for the type of interaction in humans. Usually, isoeffective drug combinations are found by manual adjustment of drug dosing to maintain the desired effect. With a feedback system, one can choose different doses of drug A and the feedback system will then automatically adjust the infusion rate of drug B to achieve the pharmacodynamic set point. This approach will accelerate the evaluation of drug–drug interactions and can help to reduce the number of animals used.

However, a condition precedent for a functioning feedback device is that a drug effect can be measured continuously and reliably, as this effect serves as an input for the control algorithm. As characteristic patterns occurred in the rat EEG with increasing propofol concentrations, we could not use ‘standard’ EEG parameters for feedback control, for example spectral edge frequency or MEF. The high-frequency components of these patterns make such simple EEG parameters unsuitable for closed-loop control. Therefore, mMEF, which revealed a good correlation with hypnotic drug concentrations in previous studies3 4 served as an input for our closed-loop system.

There are only a few publications regarding automatic drug control of anaesthetics in rats. Angel and colleagues9 have described a system that uses somatosensory evoked potentials (SSEP) for closed-loop control of propofol in rats. The mean PE in their experiments ranged from 3.8 to 90.5%. However, as SSEP show large spontaneous fluctuations in responsiveness, this PE cannot be directly compared with the PE in our study. Vijn and colleagues2 have published a feedback system to maintain different BSR with propofol and etomidate in rats for 60 min. BSRs maintained in their study were 30, 50, 70 and 90%. They assessed the performance of their feedback system by the SD of the BSR distribution around the BSR target during the 10–50 min time interval. In their experiments, SD never exceeded 5%, which seems to be a better performance than we achieved (MDAPE=13.9%). However, as their performance parameter is not directly comparable with the parameters for performance assessment used in our study, one can only roughly compare the performances of the two systems. Another problem is that EEG parameters show large statistical noise and thus an averaging process is necessary. This averaging process strongly affects the smoothness of the time curves of the measured effects and thereby the PE. In addition, when comparing different closed-loop systems, one has to consider the different set points. We tried to stabilize the pharmacodynamic effect on the steep part of the concentration–effect curve, whereas Vijn and colleagues maintained different BSR, which might indicate ceiling effects. Therefore, in our system small changes in propofol concentrations led to relatively large changes in the observed pharmacodynamic effect and thus to a relatively poorer performance. Another point when choosing the appropriate set point is that if one intends to establish a feedback device for preclinical evaluation of drug–drug interactions, BSR as a pharmacodynamic end-point might be too high to assess relevant effects of a second hypnotic drug. Whereas BS indicate deep or very deep states of general anaesthesia, it may be beneficial to target a pharmacodynamic parameter that is measurable and reliable over the whole range from light sedation to deep general anaesthesia.

Using the feedback system, the ETI rate in our study was 0.73 (0.20) mg kg–1 min–1. This is comparable to the rate of 0.64 (0.03) mg kg–1 min–1 found by Vijn and colleagues2 to maintain 30% BSR. For maintaining a BSR of 90%, they found a mean infusion rate of 1.70 (0.05) mg kg–1 min–1. In a study of Akrawi and colleagues,10 anaesthetic drug infusion was manually adjusted to maintain a state in which suppressions were present 80% of the time. Mean infusion rates of propofol to maintain this effect were reported to be 1.6–2.0 mg kg–1 h–1, which is approximately 60-fold less than in our study and that of Vijn and colleagues. It is unlikely that these large differences are caused by differences in animals with respect to strain and age (Vijn and colleagues used Wistar rats weighing 220–300 g; Akrawi and colleagues used Sprague–Dawley rats weighing 375–450 g) or by differences in BS detection. Nevertheless, one needs to consider that different methods for detection of BS might influence the derived pharmacodynamic parameters and thereby the resulting infusion rates.

In conclusion, the closed-loop device presented here seems to be a promising research tool with reasonable performance to conduct pharmacological studies such as determination of dose requirements or drug–drug-interactions at preset levels of effect.


    Appendix: Feedback system
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Appendix: Feedback system
 References
 
Propofol infusion was controlled by a model-based adaptive control algorithm to maintain a chosen set point.11 The feedback system combined a pharmacokinetic and a pharmacodynamic model based on a previous study3 to relate the dose with effect. The pharmacokinetic model was a one-compartment model with a central volume of distribution (Vc) of 0.68 litres and a clearance (CL) of 45 ml min–1. An additional effect compartment (biophase) was assumed as a site of drug action to assess the delay between propofol plasma concentration and effect (hysteresis).12 The equilibration between plasma concentration and effect-site concentration (cE) was characterized by the rate constant ke0=0.4 min–1. In this model, the plasma concentration after a bolus dose D is c(t) = A·e{alpha}·t and the corresponding cE is


with the coefficient A=D/Vc and the exponent {alpha}=CL/Vc. For any arbitrary infusion, the concentrations are obtained by convolution of the disposition function with the infusion. The pharmacodynamic model for the concentration–effect relationship was a sigmoid Hill equation:12


with the mMEF as pharmacodynamic effect E and the following parameters: baseline effect E0=10 Hz, concentration at half maximum effect EC50=4 µg ml–1 and Hill coefficient {gamma}=2.5.

Based on the pharmacokinetic model, the controller determined the propofol infusion rate to achieve a defined propofol concentration (target). This target concentration could either be set by the user (TCI mode) or it was automatically adjusted by the feedback algorithm (closed-loop control mode). If the measured effect E(c1) at the target concentration c1 was different from the set point, a new target concentration c2 was defined. For this purpose we applied a first-order Taylor series approximation of the concentration–effect relationship about the concentration c1:


where dE/dc is the first derivative of the Hill function at concentration c1. From this equation, a new target concentration c2 was obtained as:


where {Delta}E was the average deviation of the mMEF from the set point during the last 20 epochs. To avoid overshooting, the maximum increase of the target was limited to 10% of the current target, and a new target was not set until the calculated effect-site concentration had approximated the current target (difference less than 10%). The target was not changed if abs({Delta}E) was less than 0.25 Hz.

To adapt the pharmacokinetic/dynamic model to the individual animal, the controller performed an online analysis of the measured effect (mMEF). For a number of Np epochs, the hypothetical effect-site concentrations were calculated from the measured effects by inverting the Hill equation:


From the known infusion rates, the ratio A/EC50 for the individual animal was determined online, where A is the coefficient of the pharmacokinetic disposition function. If no measured concentrations are available, one cannot determine both A and EC50 from the measured effect but only the ratio A/EC50.13 If, for example, the effect is greater than predicted by the default model, this can be caused by a higher coefficient A and, consequently, a higher concentration than predicted, or it can be caused by a decreased EC50 (i.e. a greater sensitivity to the drug). The parameter adaptation was performed for the first time 10 min after the start of closed-loop control and subsequently at intervals of 5 min. The number Np of analysed epochs was equal to the total number of elapsed epochs (Ne) if Ne<=200, and was Np=200 if Ne>200.



View larger version (30K):
[in this window]
[in a new window]
 
Fig 1 Experimental setup. A surrogate parameter for anaesthetic depth is derived from the measured EEG and serves as an input for the control algorithm for propofol administration. The control algorithm consists of an adaptive pharmacokinetic/dynamic (PK/PD) model. The infusion rate is determined from the measured effect and the set point as defined by the user.

 

    References
 Top
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Appendix: Feedback system
 References
 
1 Schwilden H, Stoeckel H, Schuttler J. Closed-loop feedback control of propofol anaesthesia by quantitative EEG analysis in humans. Br J Anaesth 1989; 62: 290–6[Abstract]

2 Vijn PC, Sneyd JR. I.V. anaesthesia and EEG burst suppression in rats: bolus injections and closed-loop infusions. Br J Anaesth 1998; 81: 415–21[CrossRef][ISI][Medline]

3 Ihmsen H, Tzabazis A, Schywalsky M, Schwilden H. Propofol in rats: testing for nonlinear pharmacokinetics and modelling acute tolerance to EEG effects. Eur J Anaesthesiol 2002; 19: 177–88[ISI][Medline]

4 Schywalsky M, Ihmsen H, Tzabazis A, et al. Pharmacokinetics and pharmacodynamics of the new propofol prodrug GPI 15715 in rats. Eur J Anaesthesiol 2003; 20: 182–90[ISI][Medline]

5 Rampil IJ. A primer for EEG signal processing in anesthesia. Anesthesiology 1998; 89: 980–1002[ISI][Medline]

6 Schwilden H, Schuttler J. [The determination of an effective therapeutic infusion rate for intravenous anesthetics using feedback-controlled dosages]. Anaesthesist 1990; 39: 603–6[ISI][Medline]

7 Varvel JR, Donoho DL, Shafer SL. Measuring the predictive performance of computer-controlled infusion pumps. J Pharmacokinet Biopharm 1992; 20: 63–94[ISI][Medline]

8 Schwilden H, Fechner J, Albrecht S, Hering W, Ihmsen H, Schuttler J. Testing and modelling the interaction of alfentanil and propofol on the EEG. Eur J Anaesthesiol 2003; 20: 363–72[ISI][Medline]

9 Angel A, Arnott RH, Linkens DA, Ting CH. Somatosensory evoked potentials for closed-loop control of anaesthetic depth using propofol in the urethane-anaesthetized rat. Br J Anaesth 2000; 85: 431–9[Abstract/Free Full Text]

10 Akrawi WP, Drummond JC, Kalkman CJ, Patel PM. A comparison of the electrophysiologic characteristics of EEG burst-suppression as produced by isoflurane, thiopental, etomidate, and propofol. J Neurosurg Anesthesiol 1996; 8: 40–6[ISI][Medline]

11 Stoeckel H, Schwilden H. [Methods of automatic feedback regulation for anesthesia. Concepts and clinical use]. Anasth Intensivther Notfallmed 1986; 21: 60–7[Medline]

12 Holford NH, Sheiner LB. Kinetics of pharmacologic response. Pharmacol Ther 1982; 16: 143–66[CrossRef][ISI][Medline]

13 Schwilden H, Schuttler J, Stoeckel H. Quantitation of the EEG and pharmacodynamic modelling of hypnotic drugs: etomidate as an example. Eur J Anaesthesiol 1985; 2: 121–31[Medline]