Target-controlled drug delivery using computer-assisted infusion devices has been described for quite some time [1,2]. These devices control infusion of a drug based on known pharmacokinetic parameters and are different from commonly available infusion pumps that infuse drugs based on set infusion rates based on weights . Computer-assisted infusion devices consist of a laptop computer or similar hardware linked via a communications port to an infusion pump able to receive and transmit digital data. The computer runs a software program that calculates the necessary infusion rates to obtain and maintain a target drug concentration. In this study, the software program used was computer-assisted continuous infusion (CACI). The software transmits infusion instructions to the infusion pump and receives data about current pump settings. It can handle real-life events such as periods during which the infusion must be interrupted (e.g., intravenous catheter becomes blocked or infusion solution runs out). These calculations of infusion rate are based not only on the published pharmacokinetic parameters of the drug but also incorporate individual physiologic variables that may alter the pharmacokinetic parameters. In the case of CACI, beside choosing a drug with its associated pharmacokinetic parameter set,1 age, weight, and gender are also entered as variables. The software can target either peripheral serum concentration or effect site concentration. The infusion rate necessary to achieve a target concentration is calculated for a brief time period, and in the case of CACI, is 12 seconds. The program will then query the pump for the actual infusion rate (to correct for pump malfunctions, etc.), and a simulation will be performed to predict the serum and effect site concentrations from these data. This predicted concentration will be compared with the desired target concentration, and from this, the infusion rate required for the next set time interval will be calculated and transmitted to the infusion pump.
(1) The pharmacokinetic parameter sets used in CACI have been chosen based on the ability to perform with acceptable accuracy with fentanyl , propofol (Glass PSA, Goodman DK, Ginsberg B, Reves JG, Jacobs JR. Accuracy of pharmacokinetic model-driven infusion of propofol [abstract]. Anesthesiology 1989;711:A277), thiopental, and midazolam (Glass PSA, data not published).
These devices are not generally available for clinical use, and their primary usefulness still remains in the conduct of research . One group of research applications that can make excellent use of these devices are those that require the maintenance of a constant serum concentration of a study drug during prolonged periods of testing. The need for constant serum concentrations is particularly acute for drugs with short apparent half-lives [6,7]. Many drugs of interest to anesthesiologists fall in this category. Many tests of central nervous system function take significant time, and accurate determinations of these tests cannot be performed when serum and brain concentrations change significantly between the beginning and the end of the test, as occurs when these drugs are given as a bolus injection. The continually changing drug concentration over time is even more problematic when two tests need to be correlated (e.g., auditoryverbal memory and sedation effect). Although various infusion algorithms that an average clinician can find useful have been described, the accuracy of these methods may not be sufficient for the research purposes described above.
Although studies have been performed to assess the accuracy of various target-controlled drug infusion devices [8-10], few studies have assessed the performance and utility of these devices when they are used to maintain constant serum concentrations during testing of pharmacodynamic effect . In addition, the serum concentrations targeted in these studies have often been in the range required for the purposes of general anesthesia. Few have assessed the performance of these devices at the lower range of clinical application (i.e., sedative concentrations) . There are a number of reasons why performance of targetcontrolled infusion devices could be different at low drug concentrations. First, there is evidence that the pharmacokinetics of propofol are not linear as concentration changes . Second, since performance measures are reported as a percentage value, when a difference in target and predicted concentration that is actually small occurs at low serum concentrations, it will be represented as a larger error than if the same difference had occurred at higher concentrations.
The performance and temporal characteristics of CACI used to target low serum concentrations for four different sedative drugs are reported herein. We hypothesized that CACI would maintain constant serum concentrations of drugs over approximately one hour, the length of time a particular target concentration was maintained in this study. In this study, some subjects had arterial sampling, while others had venous sampling, based on the ability to place an arterial catheter with reasonable effort. As all blood sampling was performed at predicted pseudo-equilibration between serum and effect site concentrations and a previous study had demonstrated little, if any, differences in fentanyl concentrations based on sampling site at pseudo-steady state , we hypothesized that the differences between actual serum concentrations obtained from venous or arterial sampling would be small. The infusion algorithms implemented in CACI are based on published pharmacokinetic data sets for fentanyl, propofol, midazolam, and thiopental [4,13-15].
After informed consent and preliminary medical screening, 41 healthy volunteers were randomized to receive midazolam (n = 11; 5 female [F]), 6 male [M]), propofol (n = 10; 2 F, 8 M), thiopental (n = 10; 4 F, 6 M), or fentanyl with ondansetron pretreatment (n = 10; 4 F, 6 M), which was administered to prevent nausea. Volunteers were screened by a physician for neurologic and psychiatric abnormalities and underwent a physical examination. No subject had coexisting disease or gave a history of ongoing drug use, other than social consumption of alcohol and/or tobacco. Exclusion criteria included subjects whose body mass index2 was greater than the 85th percentile (27.8 for M, 27.3 for F aged 20-29 yr), age less than 18 yr, use of chronic medication, possibility of pregnancy, history of hiatal hernia, vasospastic disease, or carpal tunnel syndrome. Volunteers participated on a single occasion and were paid for their participation in this double-blind, institutional review board-approved study.
2 Body mass index (BMI) was computed as follows: BMI = weight in kg/(height in meters) .
All drugs were administered by CACI to achieve various constant target concentrations that did not result in loss of consciousness. The target concentration chosen was the effect site (i.e., brain). An initial bolus was infused to rapidly establish the effect site target concentrations, producing very high initial serum concentrations. Three increasing concentrations, followed by two decreasing concentrations, were used (Figure 1). Each target concentration was maintained at a constant concentration for 45-70 min to allow sufficient time for data collection. At each target concentration, a battery of psychomotor tests was administered, including a verbal learning task and perceptual and reaction time tasks, as well as recording of the electroencephalogram and event-related potentials. Data from the psychomotor test battery will be reported in a separate publication. All testing and blood sampling commenced following predicted equilibration of plasma and effect site (brain) concentrations, which varied from 3 min for propofol to 6 min for fentanyl. Target concentrations were somewhat variable (approximately 10%-20%) between subjects and were based on the response of the subject to previous concentrations. Three subjects did not achieve all five target concentrations. One fentanyl subject had arterial sampling for the first three concentrations only when the arterial catheter stopped functioning. This subject declined a reinsertion of the arterial catheter, and venous sampling was performed for the last two target concentrations. Venous sampling data from this subject are not included in the fentanyl venous sampling group. One subject receiving thiopental became nauseated at the third target concentration and could not complete this step. Data were only obtained for four target concentrations. Subjects were monitored with electrocardiogram, blood pressure, and pulse oximetry and received saline infusion with dextrose 5% at 125 mL/h in addition to the study medication.
An experienced anesthesiologist placed a catheter for blood sampling. In approximately one third of the subjects, the anesthesiologist encountered difficulty in placing an arterial line due to persistent vasospasm, and in these subjects, a venous catheter was inserted. Fifteen subjects had venous sampling (fentanyl 3 F; propofol 4 M; midazolam 3 F, 2 M; thiopental 3 M) and 26 had arterial sampling (fentanyl 1 F, 6 M; propofol 2 F, 4 M; midazolam 2 F, 4 M; thiopental 4 F, 3 M). Venous blood was obtained from the antecubital vein without the use of a tourniquet through a 16-gauge catheter from the arm opposite the one in which drug was being infused. No subject had simultaneous venous and arterial sampling. Blood samples were taken after predicted effect site equilibration. Thus, predicted concentrations were the same as target concentrations. At each target concentration, blood samples were taken at the beginning, middle, and end of the testing battery to verify constant serum concentrations during testing.
All blood samples were collected in heparinized (green top) tubes and placed immediately in a cooler with foam ice packs. This maintained samples at approximately 4 degrees C without freezing or hemolysis. Within 4 h of collection, the plasma supernatant was separated using a centrifuge (25,000 rpm for 10 min) and then stored at -70 degrees C until thawed for serum concentration assay. Drugs were assayed using high-performance liquid chromatography, except in the case of fentanyl, which was assayed using radioimmunoassay (see Appendix).
Computation of Performance Parameters
Performance parameters for CACI were determined as described previously by Varvel et al. .
Calculations. For each sample, the performance error (PE) of the predicted concentration in plasma was calculated as: Equation 1
Note that in this particular study, target and predicted concentrations are the same, as all samples were obtained after the equilibration of target with predicted concentrations.
Another measure, constancy error (CE), was calculated to represent the variation of actual serum concentration around the mean value obtained at any constant target concentration: Equation 2 where mean is the mean value of the three actual serum concentrations obtained during any given target concentration. In this particular study, the primary goal was to maintain a constant serum concentration regardless of actual concentration.
Intrasubject Data Analysis
The following parameters were calculated as standard performance parameters.
Median Prediction Error. The percentage median prediction error (MDPE) for the ith subject represents the bias of the infusion device: Equation 3 where N is the number of target concentrations, j is the number of blood samples obtained, and each j represents the average of three blood samples obtained at any particular target concentration.
Median Absolute Prediction Error. The inaccuracy of the infusion device is expressed as the median absolute prediction error (MDAPE) and is calculated as: Equation 4
Median Absolute Constancy Error. The variability of actual serum concentrations versus the mean value of those concentrations during the time that serum concentration is to remain constant (i.e., a particular target concentration) is expressed as the median absolute constancy error (MDACE) and is given by: Equation 5
Divergence. Divergence is a time-related parameter that indicates how the inaccuracy of the infusion device changes as time increases. A non-zero value indicates a widening gap between actual serum and target concentrations, while a zero value means that performance accuracy does not change over time. Divergence is obtained by linear regression of PE values against time for each subject and is the slope of that regression expressed as percent change per hour.
Wobble. Wobble is a time-related index of changes in performance accuracy and measures the intrasubject variability in performance errors. It is calculated as: Equation 6
All analysis was performed using SPSS for Windows (version 6.1, SPSS, Inc., Chicago IL). Comparisons between values obtained using arterial sampling versus venous sampling were performed using nonparametric tests (Mann-Whitney U-test, Wilcoxon's test). Multivariate analysis of variance was used to give an estimate of the power to reject a false null hypothesis, the null hypothesis being that there are no differences between arterial and venous samples. Note that no subject had simultaneous arterial and venous sampling. Comparisons of performance parameters among drugs were performed using a Kruskal-Wallis one-way analysis of variance (a nonparametric test).
There were no differences in subject demographics among groups, including educational level. The demographics were as follows (mean +/- SEM): age 27.7 +/- 0.9 yr, weight 67.9 +/- 1.9 kg, body mass index 23.0 +/- 0.5, last intake of food 10.4 +/- 0.2 h before study. Almost all volunteers had completed education beyond high school. The actual measured serum concentrations for each drug at the highest target concentration were as follows: fentanyl 2.33 +/- 0.42 ng/mL, midazolam 125.6 +/- 28.7 ng/mL, propofol 1.40 +/- 0.41 micro g/mL, and thiopental 4.5 +/- 1.4 micro g/mL (mean +/- SD; individual serum concentrations at highest target concentration were averaged to one value for each subject; arterial and venous samples were combined; data were normally distributed). Median serum concentrations (range 10%-90%) of all samples taken are as follows: fentanyl, arterial 1.1 (0.5-2.5), venous 0.9 (0.4-2.1) ng/mL; propofol, arterial 0.6 (0.1-1.4), venous 0.7 (0.2-1.6) micro g/mL; thiopental, arterial 2.1 (0.8-4.9), venous 2.3 (1.2-4.4) micro g/mL; midazolam, arterial 75 (31-158), venous 69 (38-122) ng/mL. There were no significant differences between the average arterial and venous drug concentrations for groups of individuals for any drug at any target concentration.
The performance accuracies of CACI for the four drugs tested are illustrated in Figure 2 and Figure 3 (MDPE and MDAPE). The temporal performance characteristics of CACI for the four drugs tested are illustrated in Figure 4, Figure 5 and Figure 6 (divergence, wobble, and MDACE). The PE of CACI was significantly different for different drugs, with thiopental and midazolam displaying the greatest inaccuracy. Temporal characteristics were different among drugs as well, with the lowest CE obtained with thiopental. Wobble was consistently larger than MDACE. Differences between arterial and venous sampling were few. When these occurred, arterial sampling indicated larger errors. Significant differences between arterial and venous sampling occurred for the MDAPE for fentanyl, the divergence for propofol and midazolam, and the wobble for thiopental.
The power to reject a false null hypothesis (the null hypothesis being that there are no differences between arterial and venous sampling) at the P = 0.05 level for performance indices is 0.6 for fentanyl, 0.4 for propofol, 0.5 for thiopental, and 0.5 for midazolam. Similarly, for actual concentrations obtained from arterial and venous sampling, the power at P = 0.05 is 0.2 for fentanyl, 0.2 for propofol, 0.1 for thiopental, and 0.2 for midazolam.
Target-controlled infusion devices use a single set of pharmacokinetic parameters that is usually derived from a small group of patients given a single bolus dose or short infusion of the drug. Thus, achieving perfect accuracy in a general population is impossible. The differences in conditions between studies in which performance parameters are tested and those in which the original pharmacokinetic parameters were obtained may contribute to performance errors . Variables that may contribute to this effect include the numbers and demographics of the subjects studied, the method by which the drug is administered, the total dose of the drug, the method of blood sampling, the method of drug assay, and the statistical analysis used to determine pharmacokinetic parameters. In the case of CACI and the drugs studied, pharmacokinetic parameters were originally obtained from 12-20 subjects, and except for propofol, were obtained using a single bolus administration of drug in volunteer subjects whose age and weight were similar to the subjects in the current study. The pharmacokinetics of propofol were determined in somewhat older surgical patients who received regional anesthesia during infusion . Venous sampling was used in the pharmacokinetic studies of midazolam and thiopental [14,15], and arterial sampling was used in the studies of propofol and fentanyl [4,13]. No study utilized the same drug assay methodology as was used in this study. This may account for the large bias in the midazolam results, as we used an high-performane liquid chromatography method to assay midazolam, whereas Greenblatt et al.  used a much more sensitive assay. Despite the age differences among subjects in the propofol pharmacokinetic study and the fact that they were undergoing surgery, the performance of CACI with propofol was very good, which raises the possibility that pharmacokinetic parameters obtained during constant infusions with arterial blood sampling may perform the best in target-controlled drug delivery devices that operate using a continuous infusion.
Despite these performance inaccuracies, however, CACI performed very well at maintaining constant serum concentrations (Figure 6). In studies evaluating the concentration-effect relationship at pseudo-steady state, this property is more important than accurately achieving a particular target concentration. To quantitate the property of maintenance of a constant concentration, an additional parameter, the MDACE, is introduced. This CE excludes any assessment of accuracy, as its value is independent of the relationship between predicted and actual concentrations and thus provides no information on the accuracy of the pharmacokinetic parameters used in the software controlling the infusion pump. This is in contrast to other indices of performance, which are based on the relationship between actual and predicted serum concentrations. Although wobble and divergence are variables that provide information about time-related performance, they implicitly assess the accuracy of the device as the relationship between predicted and actual concentrations is used in the derivation of these parameters. In addition, wobble and divergence are conceptually somewhat difficult to understand, whereas the CE described in this study is conceptually very easy to understand. The CE gives valuable information regarding the question of the variability of serum concentrations over the time of testing. Wobble was consistently greater than CE. This may result from the fact that the performance accuracy of an infusion device at the same predicted concentrations may be quite different when concentrations are increasing versus decreasing, and the performance inaccuracies will be reflected in the wobble and divergence calculations. However, this effect may be of little consequence to the investigator as long as actual serum concentrations are obtained.
Another important consideration is the accuracy of drug assays at these low concentrations. For instance, the lower limits of detectability for thiopental and midazolam were close to the lowest concentrations achieved in this study. Thus, some of the performance inaccuracy may be related not only to the device itself but also to difficulties with the drug assay method.
Routine arterial sampling for drug concentrations has been advocated on the basis that systematic bias is introduced by venous sampling [19,20]. This consideration is most important when pharmacokinetic parameters are being determined or when pharmacodynamic effects are being assessed with rapidly changing serum concentrations, such as following bolus injections of rapidly redistributed drugs [6,21]. Drugs that are not dependent on typical metabolic pathways, such as those in the liver and kidney, will also exhibit large arterial-venous differences, a good example being remifentanil. However, when equilibration of the serum concentration and the concentration at the effect site of interest has occurred and the drug is metabolized/excreted by the liver or kidney, venous sampling may be sufficiently accurate depending on the particular drug being studied. Although no subject had simultaneous venous and arterial sampling, some inferences can be made from data obtained from this study. There were few differences between subjects with arterial versus those with venous sampling (A-V difference). Propofol has been shown to exhibit the largest A-V difference among the drugs used in this study [6,12,22]. Both Coetzee et al.  and Wang et al.  demonstrated significant A-V differences, but most differences occurred at serum concentrations greater than 2.0 micro g/mL, and this may explain the lack of systematic A-V differences found in the current study. Also, blood samples in the current study were only taken once a pseudosteady state had been achieved, and, thus, distribution into tissues was small, thereby tending to minimize A-V differences compared with those after a bolus dose. Another possible reason that few differences were found was the small number of subjects in this study. This is quantitated by determining the power of statistical testing, which determines the probability that a real A-V difference can be detected given the sample size and variability of the data. The probability to detect a real difference in performance parameters derived from arterial versus venous samples ranged from 0.4 for propofol to 0.6 for fentanyl. The probability to detect a real difference in actual serum concentrations between arterial and venous sampling was much lower, approximately 0.2 for all drugs. The higher probability of detecting a real difference in performance parameters is indicated by the fact that significant A-V differences were demonstrated in some performance parameters but not in actual serum concentrations. Another way of interpreting these results is that for a small number of subjects, the variability of measured parameters from factors other than arterial versus venous sampling is large enough to obscure any differences based on sampling site. When differences did occur in this study, however, the values obtained by arterial sampling always demonstrated larger performance errors. Also, the variability in performance parameters was generally higher with arterial sampling than venous sampling even when significant differences did not occur.
The results of this study indicate that to assess the accuracy of an infusion device, arterial sampling should be used. The main purpose of this study, however, was to correlate cognitive effects with low serum concentrations of drugs, and the results of this study indicate that venous sampling provides a sufficiently accurate determination of the actual serum concentration for this purpose. In the case of propofol, the data from this study indicate that venous sampling is adequate when serum concentrations are less than 2.0 micro g/mL and predicted equilibration between serum and the effect site concentration has occurred. Although arterial sampling is the "gold standard," depending on the objectives of the study, one may question the use of vigorous attempts at placement of arterial catheters in volunteer subjects in whom this may be difficult.
In conclusion, CACI is able to maintain very stable serum concentrations for periods of 45-70 minutes at low concentrations for four sedative drugs. Although performance accuracy may be poor for some drugs, CACI was still able to maintain a constant serum concentration very accurately. It is recommended that future studies examining the pharmacodynamic effects of sedative-hypnotic drugs utilize a device similar to CACI when drugs with rapid redistribution properties are being studied. This will ensure a constant drug effect during testing and accurate comparisons between different time points and serum concentrations. However, the inaccuracies in target concentrations indicate the need to obtain actual serum concentrations to accurately correlate with pharmacodynamic effects. Arterial sampling is preferable when the accuracy of infusion devices is being assessed. However, venous sampling may be an acceptable alternative when the primary goal of the study is to determine pharmacodynamic effects at constant serum concentrations.
Methodology to Assay Drug Concentrations
Plasma fentanyl was measured by a radioimmunoassay procedure with reagents from Janssen Biotech Research Products, Beerse, Belgium . Precision of the assay was determined by replicate analysis of quality-controlled specimens at three different concentrations. The intraassay coefficients of variation were 4.0%, 2.0%, and 2.6% for quality controls equivalent to 0.5, 1.0, and 2.0 ng/mL plasma fentanyl, respectively. The corresponding interassay coefficients of variation were 10.0%, 6.3%, and 5.2%, respectively. In each case, serial specimens from individual patients were quantified in duplicate within the same run to ensure precision. The sensitivity of the fentanyl assay (minimum detectable plasma fentanyl concentration) was 0.1 ng/mL.
Plasma thiopental concentrations were measured by reverse-phase high-pressure liquid chromatography with ultraviolet detection . Plasma was prepared for chromatography by the addition of internal standard and precipitation of plasma proteins with acetonitrile, as described previously . The column was eluted isocratically with acetonitrile-water (32:68 v/v) at a flow rate of 0.9 +/- 0.1 mL/min. The detector was set at 254 nm and 0.05 absorbance units, full scale. The precision of the assay was determined by replicate analysis of quality-controlled specimens at three different concentrations. The intraassay coefficients of variation were 3.1%, 0.2%, and 3.6% for quality controls equivalent to 12.5, 25, and 50 micro g/mL plasma thiopental, respectively. Interassay coefficients of variation were 3.7%, 2.1%, and 4.7%, respectively. The minimum detectable concentration of plasma thiopental was 1.6 micro g/mL.
Plasma propofol was measured by highperformance liquid chromatography with fluorescence detection . Plasma was prepared for chromatography by precipitation of plasma proteins with acetonitrile. Propofol was detected by a fluorescent detector at an excitation wavelength of 275 nm and an emission wavelength of 310 nm. The minimum detectable concentration of plasma propofol was estimated as 0.1 micro g/mL. The intraassay coefficients of variation determined by replicate analysis of quality control specimens at three different concentrations (0.63, 2.5 and 10 micro g/mL) were 2.5%, 2.0%, and 2.0%, respectively. The interassay coefficients of variation for propofol were 3.6% at 0.63 micro g/mL, 3.1% at 2.5 micro g/mL, and 2.2% at 10 micro g/mL.
Plasma midazolam concentrations were measured by reverse-phase high-performance liquid chromatography with ultraviolet detection . Plasma (0.5 mL) was prepared for chromatography with the addition of internal standard and isolation of midazolam on 3-M Empore[TM] extraction disk cartridges from Varian (Harbor City, CA). Midazolam was collected from the preparatory column with 250 micro L acetonitrile after a preelution with 1 mL 20% methanol. The extract (100 micro L) was injected onto the chromatographic column and eluted isocratically, as previously described . Midazolam was detected with the ultraviolet detector set at 220 nm at a sensitivity of 0.001 absorbance units, full scale. The intraassay coefficients of variation for midazolam were 1.6% at 100 ng/mL and 4.8% at 200 ng/mL. The interassay coefficients of variation at two different midazolam concentrations (100 and 200 ng/mL) were 7.8% and 8.3%, respectively.
We thank Jim Jacobs, PhD, from Duke University for technical assistance in the operation of CACI, and Dick Smith, PhD, from Duke University for helpful suggestions regarding statistical analysis.
1. Glass PSA, Jacobs JR, Quill TJ. Intravenous drug delivery systems. In: Fragen RJ, ed. Drug infusions in anesthesiology. New York: Raven Press, 1991:23-61.
2. Glass PSA, Shafer SL, Jacobs JR, Reves JG. Intravenous drug delivery systems. In: Miller R, ed. Anesthesia. New York: Churchill Livingstone, 1994:389-416.
3. Egan TD. Intravenous drug delivery systems: toward an intravenous "vaporizer". J Clin Anesth 1996;8:8S-14S.
4. McClain DA, Hug CJ. Intravenous fentanyl kinetics. Clin Pharmacol Ther 1980;28:106-14.
5. White PF. Intravenous anesthesia and analgesia: what is the role of target-controlled infusion? J Clin Anesth 1996;8:26S-28S.
6. Peacock JE, Blackburn A, Sherry KM, Reilly CS. Arterial and jugular venous bulb blood propofol concentrations during induction of anesthesia. Anesth Analg 1995;80:1002-6.
7. Hughes MA, Glass PS, Jacobs JR. Context-sensitive half-time in multicompartment pharmacokinetic models for intravenous anesthetic drugs. Anesthesiology 1992;76:334-41.
8. Ausems ME, Stanski DR, Hug CC. An evaluation of the accuracy of pharmacokinetic data for the computer assisted infusion of alfentanil. Br J Anaesth 1985;57:1217-25.
9. Shafer SL, Varvel JR, Aziz N, Scott JC. Pharmacokinetics of fentanyl administered by computer-controlled infusion pump. Anesthesiology 1990;73:1091-102.
10. Glass PS, Jacobs JR, Smith LR, et al. Pharmacokinetic model-driven infusion of fentanyl: assessment of accuracy. Anesthesiology 1990;73:1082-90.
11. Buhrer M, Maitre PO, Hung OR, et al. Thiopental pharmacodynamics. I. Defining the pseudo steady state serum concentraion EEG effect relationship. Anesthesiology 1992;77:226-36.
12. Coetzee JF, Glen JB, Wium CA, Boshoff L. Pharmacokinetic model selection for target controlled infusions of propofol. Assessment of three parameter sets. Anesthesiology 1995;82:1328-45.
13. Gepts E, Camu F, Cockshott ID, Douglas EJ. Disposition of propofol administered as constant rate intravenous infusions in humans. Anesth Analg 1987;66:1256-63.
14. Greenblatt DJ, Abernathy DR, Lockniskar A, et al. Effect of age, gender, and obesity on midazolam kinetics. Anesthesiology 1984;61:27-35.
15. Ghoneim MM, Van HM. Pharmacokinetics of thiopentone: effects of enflurane and nitrous oxide anaesthesia and surgery. Br J Anaesth 1978;50:1237-42.
16. Rudman D. Assessment of nutritional status. In: Braunwald E, Isselbacher KJ, Petersdorf RG, et al., eds. Harrison's principles of internal medicine. 11 ed. Vol. 1. New York: McGraw-Hill, 1987:390-3.
17. Varvel JR, Donoho DL, Shafer SL. Measuring the predictive performance of computer-controlled infusion pumps. J Pharmacokinet Biopharm 1992;20:63-94.
18. Shafer SL. Constant versus optimal plasma concentrations [editorial; comment]. Anesth Analg 1993;76:467-9.
19. Chiou WL. The phenomenon and rationale of marked dependence of drug concentration on blood sampling site. Implications in pharmacokinetics, pharmacodynamics, toxicology and therapeutics (part II). Clin Pharmacokinet 1989;17:275-90.
20. Chiou WL. The phenomenon and rationale of marked dependence of drug concentration on blood sampling site. Implications in pharmacokinetics, pharmacodynamics, toxicology and therapeutics (part I). Clin Pharmacokinet 1989;17:175-99.
21. Scott JC, Stanski DR. Decreased fentanyl and alfentanil dose requirements with age. A simultaneous pharmacokinetic and pharmacodynamic evaluation. J Pharmacol Exp Ther 1987;240:159-66.
22. Wang YP, Cheng YJ, Fan SZ, Liu CC. Arteriovenous concentration differences of propofol during and after a stepdown infusion. Anesth Analg 1994;79:1148-50.
23. Woestenborghs RJ, Stanski DR, Scott JC, Heykants JJ. Assay methods for fentanyl in serum: gas-liquid chromatography versus radioimmunoassay. Anesthesiology 1987;67:85-90.
24. Celardo A, Bonati M. Determination of thiopental measured in human blood by reversed-phase high-performance liquid chromatography. J Chromatogr 1990;527:220-5.
25. Crankshaw DP, Boyd MD, Bjorksten AR. Plasma drug efflux-a new approach to optimization of drug infusion for constant blood concentration of thiopental and methohexital. Anesthesiology 1987;67:32-41.
26. Vree TB, Baars AM, de Grood PM. High-performance liquid chromatographic determination and preliminary pharmacokinetics of propofol and its metabolites in human plasma and urine. J Chromatogr 1987;417:458-64.
© 1997 International Anesthesia Research Society
27. Puglisi CV, Pao J, Ferrara FJ, de Silva JA. Determination of midazolam (Versed) and its metabolites in plasma by high-performance liquid chromatography. J Chromatogr 1985;344:199-209.