The authors evaluated the effects of midazolam, propofol, thiopental, and fentanyl on volunteer participants' memory for words and pictures at equisedative concentrations.
Sixty-seven healthy volunteers were randomized to receive intravenous infusions of midazolam (n = 11), propofol (n = 11), thiopental (n = 10), fentanyl with ondansetron pretreatment (n = 11), ondansetron alone (n = 8), or placebo (n = 16) in a double-blind design. Three increasing and then two decreasing sedative concentrations were achieved by computer-controlled infusion in each volunteer. Measures of sedation, memory, and drug concentration were obtained at each target concentration. Drug concentrations were normalized to equisedative effects using both Emax and logistic regression methods of pharmacodynamic modeling. The serum concentrations at 50% memory effect (Cp50s) were determined using four different memory end points. The relative potencies compared with midazolam for memory impairment were determined.
Equisedative concentrations were midazolam, 64.5 +/- 9.4 ng/ml; propofol, 0.7 +/- 0.2 microg/ml; thiopental, 2.9 /- 1.0 microg/ml; and fentanyl, 0.9 +/- 0.2 ng/ml. The Cp50s for 50% loss of memory for words were midazolam, 56 +/- 4 ng/ml; propofol, 0.62 +/- 0.04 microg/ml; thiopental, 4.5 +/- 0.3 microg/ml; and fentanyl, 3.2 +/- 0.4 ng/ml. Compared with midazolam, relative potencies (with 95% confidence intervals) were propofol, 0.96 (0.44-1.78); thiopental, 0.76 (0.52-0.94); and fentanyl, 0.34 (0.05-0.76). Large effects on memory were only produced by propofol and midazolam.
At equal sedation, propofol produces the same degree of memory impairment as midazolam. Thiopental has mild memory effects whereas fentanyl has none. Ondansetron alone has no sedative or amnesic effects.
One of the most common uses of intravenous sedative agents today is to produce amnesia during minor surgical procedures. Ideally, amnesia can occur while the patient is still awake and cooperative. The benzodiazepines and propofol are used because of their intense amnesic effects. [1–4] Although the literature indicates that propofol has substantial amnesic effects, [5–13] a common clinical perception is that propofol may not be a good amnesic agent. This perception may be based on some anecdotal reports of awareness during propofol anesthesia (for example, see Kelly and Roy  and Rupreht ). Because of its short context-sensitive half-life, propofol can be administered to produce profound sedation while still permitting rapid recovery.  Sedation itself can produce “amnesia” due to inattention to the stimuli presented for later recall. [17,18] It is unclear how much the sedative effects of propofol contribute to its clinically apparent amnesic effects. To clearly define the amnesic effect of any drug, we must control for its sedative effects. Another reason to distinguish sedative from amnesic effects is the recent availability of neuroimaging techniques that can define functional neuroanatomy. Recent positron emission tomography studies have indicated different neural networks subserving various types of memory and attention. [19–23] These neuroimaging techniques can also be used successfully to identify neural systems affected by these drugs. [24–26] In future neuroimaging studies of sedative agents, it will be important to quantitate the sedative versus the amnesic effects of these drugs.
We determined the amnesic effects of midazolam and propofol in this study. We used thiopental and fentanyl to control for the effect of sedation on memory. Thiopental has definite sedative effects, but there is some evidence indicating that it may also have some amnesic properties.  Because thiopental is active at gamma-aminobutyric acid A receptors, as are midazolam and propofol,  we included fentanyl as an additional sedative control, because opioids do not seem to have effects at the gamma-aminobutyric acid A receptor. Previously we found that fentanyl produced substantial sedation with some mild memory effects.  In that study, participants experienced a high incidence of nausea. In this study, we pretreated volunteers receiving fentanyl with the antiemetic drug ondansetron. A control group received ondansetron alone. To quantify the effect of sedation on memory, we normalized actual serum concentrations to an equivalent sedative effect for all drugs. We determined serum concentrations at 50% memory effect (Cp50s) for amnesia using these normalized concentrations. Relative potencies were calculated for each pair of drugs using the normalized concentration scales.
Materials and Methods
This study was reviewed by the hospital's institutional review board and human subjects research committee, and informed consent was obtained from all volunteers before the beginning of the study. Sixty-seven healthy volunteers (40 men, 27 women; age range, 21–45 yr; mean age, 27.8 yr) were recruited for this study and were paid for their participation. Participants were recruited by flyer and given a telephone screening interview, brief medical examination, and preliminary training session that included practice on all cognitive tests. To reduce pharmacokinetic variability, volunteers were within +/- 30% of their ideal body weight.* Three quarters of the volunteers had a college education or higher; of the rest, all but two had some education after high school. English was the native language of 61% of the volunteers, and the rest were fluent in English as a second language learned in childhood. All subjects practiced the word test used in the study during a familiarization session, and a delayed recall score equal to at least 75% of the list words was required before inclusion into the study. Participants reported a median of 6 h of sleep the night before the study, and 85% said they had slept normally. Volunteers were nil per os from midnight the day before and were allowed no caffeine on the day of the study.
This study was done in a double-blind design, and volunteers were randomized to receive midazolam (n = 11), propofol (n = 11), thiopental (n = 10), fentanyl with ondansetron pretreatment (n = 11), placebo “fentanyl” with active ondansetron pretreatment (n = 8), or placebo (n = 16). Ondansetron was administered as a 4-mg intravenous bolus dose approximately 30 min before infusion of the study drug. Two volunteers were dropped from the analysis. Immediately after the infusion began, one volunteer in the fentanyl group experienced severe nausea and the study was terminated. Another person in the placebo group could not follow instructions on several tasks and his data were omitted. Many volunteers receiving placebo (24% of the total) were required because they received placebo “midazolam”(n = 4), “propofol”(n = 4), “thiopental”(n = 4), or “fentanyl”(n = 4) to allow data collection following a similar time course as each active drug infusion. The “fentanyl” placebo volunteers included in the placebo group received placebo ondansetron (saline) pretreatment. Placebo solutions were prepared to have the same appearance as the active drug and were administered in exactly the same way as active drugs. For placebo propofol, intralipid was used, which was shown previously to have no cognitive effects.  Thus the placebo group included only volunteers receiving saline or intralipid.
All drugs** were administered by computer-controlled intravenous infusion (CACI) to achieve constant effect-site concentrations at low, moderate, and high degrees of sedation sequentially in a single session.  The volunteer arrived at 8:00 A.M., and drug infusion commenced at approximately 10:00 A.M. after baseline testing. Three increasing target concentrations (TC) followed by two decreasing concentrations were used. During each TC, volunteers were kept at a constant effect-site concentration for 45–60 min to allow completion of all testing. Transitions between increasing TCs were accomplished using a bolus method to target effect-site concentrations, which initially produced a high serum concentration. Testing started after twice the predicted equilibration time to ensure equilibration between blood and brain (effect-site) concentrations (Figure 1). The highest target concentration (third target, or TC3) is designated as MaxTarget. Participants were allowed a brief nap after the conclusion of testing at TC3 while drug concentrations declined, and they were given a light lunch after the testing was completed at TC4. All blood sampling and cognitive testing was performed after predicted equilibration of intravenous and effect-site (brain) concentrations. Volunteers were monitored with electrocardiograph, blood pressure, and pulse oximetry, and they received saline infusions with 5% dextrose at 125 ml/h in addition to the study medication.
Arterial or venous blood samples were taken to verify constant serum concentrations during testing at any given target concentration. Forty-one volunteers underwent arterial sampling; in 26 it was not possible to insert an arterial catheter. Because all blood sampling was performed after predicted equilibration between brain and serum concentrations, venous catheters were placed in these persons. Venous blood was obtained from the antecubital vein without use of a tourniquet through a 16-gauge catheter from the arm opposite the one in which drug was being infused. There was a trend for arterial concentrations to be approximately 10% higher than venous samples for fentanyl and propofol, but none of these differences were statistically significant.  No subject had simultaneous venous and arterial sampling. Three blood samples were taken at each target concentration and averaged to provide a mean value for further analysis. Arterial and venous values were combined for statistical analysis. Thiopental, propofol, and midazolam assays were performed by high-pressure liquid chromatography, and fentanyl assays were done by radioimmunoassay (see Veselis et al.  for details of assay procedures).
Sixteen bipolar visual analog scales of sedation were obtained repeatedly at each serum concentration. [33,34] Four scales were presented on each of four pages. Each was 10 cm long and the volunteer would make a mark along the length of this scale. Completion of the 16 scales took approximately 2 or 3 min. The order of presentation of pages was randomized between tests, and similar poles alternated between left- and right-hand ends of the scale (e.g., weak on right, clumsy on left) to avoid bias in responding. The individual scales were summed into four measures with possible values from 0 to 40 each: physical sedation (weak, clumsy, lazy, incompetent), mental sedation (drowsy, slow, dreamy, fuzzy), tranquilization (calm, contented, peaceful, relaxed), and “other” feelings (bored, hostile, sad, withdrawn). Because there were no significant differences between physical and mental sedation, these were averaged together to provide one measure of sedation, average sedation, also with a maximal value of 40.
Volunteers completed two sets of memory tests at each target concentration. The verbal task consisted of learning a list of 16 words, and the visual task consisted of viewing six pictures for 30 s each, during which time the volunteer was prompted to describe all details they saw in the picture. At the end of the study day, volunteers were required to recognize any words that were presented on any of the study lists by circling them on an answer sheet that contained presented and nonpresented words. They were also asked to verbally recall any pictures that had been presented. If recall was unsuccessful, the volunteers were asked to recognize any of the previously presented pictures among other pictures they had not seen. Further details of memory testing are included in Appendix 1.
Definition of “Amnesia”
Determination of the Cp50 for amnesia used logistic regression analysis, as described below. This requires a “yes/no” definition of effect. We chose two different cutoff points to represent the presence of amnesia on the basis of fitting an inhibitory Emax model to data obtained with midazolam and propofol. [35,36] The “mild” cutoff point was defined as the number of words or pictures recognized at 50% of maximal effect of propofol and midazolam, and it was equal to 7.5 words and 2.5 pictures. Thus we defined mild amnesia as present when fewer than eight words or three pictures were recognized. The “maximal” amnesia points were chosen based on the basement effect of midazolam and propofol, which occurred when volunteers recognized two words or one picture. In recognition testing, our experience indicates that participants will guess a word that has not been presented at a rate of approximately 3–4%; therefore we did not require the total absence of recall as a criterion for maximal amnesia.  For the picture memory test, we required sufficient description of the picture to be certain that the volunteer indeed remembered the specific picture presented.
Determination of Equivalent Sedative Concentrations: The Cp50 sub SED. The serum concentrations for all drugs at equivalent sedation were determined. Then all the actual serum concentrations for each drug were converted into normalized concentrations, where a value of 5 represented the Cp50 for sedation for each drug (Cp50SED). We determined the Cp50SEDusing two standard pharmacodynamic models, the Emax model [35,36] and a logistic regression model [36,38] as detailed in the methods section of Appendix 1. These models provide somewhat different results based on the fact that the Emax model assumes that the maximal effect of each drug is equivalent, whereas the logistic regression model uses a fixed criterion for the presence of an effect. We used two different methods of analysis to increase the reliability of any conclusions regarding equisedative amnesic effects. Two sets of normalized concentrations thus were derived, one using the Emax (Emax method normalized) and the other using the logistic regression model (LR method normalized).
Determination of Cp50AMNand Relative Potencies at Equivalent Sedation. We determined the Cp50s for amnesia (Cp50AMN) using a logistic regression analysis that applied the amnesia present/absent criteria described before against the Emax and LR sets of normalized serum concentrations, thus allowing direct comparison of different drugs at a common sedation effect.
To determine if differences of Cp50AMNat an equivalent sedative effect were significant, we performed two statistical tests. First, the Cp50AMNvalues were considered to be significantly different (P < 0.05) between any pair of drugs if there was no overlap between the 95% confidence intervals of the Cp50AMNvalues. This method compares differences in drugs only at the Cp50AMNvalue. The second method, which is considerably more complex, estimates the relative potency of any drug pair. The advantage of this method is that all points are compared, not just the Cp50AMNvalue. To do this, a new logistic regression analysis is performed to compare each pair of drugs. The regression parameter for drug effect, common slope for the natural log of the normalized serum concentration effect, and co-variance matrix are used to compute the relative potency (M), and the 95% confidence interval for M is determined using Fieller's theorem.  In this analysis, a relative potency of 1.0 indicates that there is no difference between drugs. If the confidence interval of M includes 1.0, then the two drugs compared cannot be considered different in potency. Midazolam was the standard in this analysis, with its relative potency being 1.0 by definition.
Other Statistical Tests. Absolute change in average sedation and memory scores between baseline and the MaxTarget was compared for each active drug and placebo using the Wilcoxon rank-sum test with Bonferroni correction (P < 0.01 for significance). In addition, change from baseline to MaxTarget was compared within participants using paired t tests for each drug administered (P < 0.01 for significance after Bonferroni correction). All statistical analyses, unless otherwise indicated, were performed using SAS, version 6.01 (Cary, NC).
There were no differences in participant characteristics among groups (age, 27.7 +/- 5.7 yr; weight, 68.7 +/- 12.5 kg; body mass index, 23.2 +/- 3.2). The distribution of non-native English speakers did not vary significantly between the experimental groups. By repeated measures analysis of variance, native language (English vs. other) was not a significant predictor for either baseline or MaxTarget memory scores, and the interaction of drug group and native language was not significant.
The actual measured serum concentrations for each drug at MaxTarget were as follows: fentanyl, 2.33 +/- 0.42 ng/ml; midazolam, 125.6 +/- 28.7 ng/ml; propofol, 1.40 +/- 0.41 micro gram/ml; and thiopental, 4.5 +/- 1.4 micro gram/ml. We have demonstrated in a separate manuscript that serum concentrations during testing were constant. 
The rate of increase in sedation as serum concentration increased was the same for all drugs (i.e., all the slopes were parallel; see Figure 2(B) and the Logistic Regression section in Appendix 1). Sedation was maximal at MaxTarget and decreased as serum concentrations declined, returning to baseline levels by the end of the study (Figure 3(A)). Participants receiving placebo showed a significant increase over baseline (P < 0.01) at the time corresponding to MaxTarget and showed a similar decline during the rest of the day. Volunteers receiving propofol showed a significant increase in sedation over baseline levels (P < 0.001) but were less sedated than volunteers receiving midazolam (P = 0.022) and fentanyl (P = 0.022). At MaxTarget, the propofol and thiopental groups did not differ from the placebo group at the designated level of significance, although the sedation effect approached significance in the volunteers who received thiopental (P <0.02). Table 1shows the values determined for Cp50SEDfor each drug.
Results of the various memory tests are presented in Figure 3(B-D).
Memory for Words. As shown in Figure 3(B), the percentage of words recognized from the lists presented was significantly reduced by midazolam, propofol, and thiopental (P < 0.001), compared with baseline and with placebo at MaxTarget. “False alarms” on the word recognition test were relatively infrequent. Of the 64 nonpresented distractor words, the mean percentage incorrectly identified by participants was 5% and 5.7% in the propofol and midazolam groups, 7% and 11.4% in the fentanyl and thiopental groups, and 6.5% and 10.4% in the placebo and ondansetron groups, respectively. The rate of false alarms was not significantly different between groups. Thus the participants experiencing amnesia were not more likely to make random guesses on the distractor items.
Mild Impairment (Fewer than 8 of 16 Words). As shown in Figure 4(A and B) and Table 2, all drugs produced mild impairment of word memory. The relation of midazolam and propofol is somewhat different, depending on whether sedation was normalized using the Emax or the logistic regression method. However, both midazolam and propofol produced equivalent amnesia at lesser sedative effects than did thiopental. Propofol could be differentiated from midazolam and thiopental only in the LR model of sedation. Both models found that midazolam and thiopental produced more amnesia than fentanyl. Fentanyl had almost no memory effect, even at concentrations that produced equivalent sedation to the other drugs.
Severe Impairment (Fewer than 3 of 16 Words). As expected, the normalized Cp50AMNvalues for mild memory impairment are much lower than the Cp50AMNvalues for maximum memory impairment (see Table 2and Figure 5). This degree of memory impairment was achieved only by midazolam and propofol.
Memory for Pictures. The percentage of pictures recalled and recognized are shown in Figure 3(C and D) for all groups. Midazolam and propofol severely impaired recall and recognition of pictures presented at MaxTarget. The Cp50AMNusing the picture recall criterion was much lower than that required to block memory on the picture recognition test (see Table 2and Figure 5). The Cp50AMNvalues for loss of picture recall are in fact similar to the Cp50 sub AMN for verbal amnesia using the maximum impairment criterion. For picture recognition, only midazolam and propofol produced sufficient memory impairment to meet the criteria for definition of either mild or maximum amnesia. Comparing picture recall to the more sensitive recognition test, retrieval from memory increased from 1.7% on the cued recall test to recognition of 45% of the pictures shown for propofol, from 4.5% to 54.5% for midazolam, and from 29.6% to 87% for thiopental. Although several volunteers in the placebo, ondansetron, and fentanyl groups could not recall previously presented pictures, these persons could recognize all pictures previously shown (100% retrieval from memory). “False alarms” on the picture recognition test, with a volunteer responding “yes” to a picture that had not been previously presented, were rare.
Relative Potencies for Amnesia. Table 3details the relative potencies of each drug pair using various memory criteria at equisedative effects, where midazolam has a potency of 1.0. Propofol has equivalent amnesic effects to midazolam when the Emax model is used to control for sedation (relative potencies, 0.96 to 1.13). On some criteria, propofol is more potent than midazolam when the LR model is used to control for sedation (relative potencies, 1.76 to 2.13). The reasons for these differences are outlined in the discussion and in Appendix 1. Thiopental is less potent than midazolam (relative potencies, 0.76 to 1.01), but results were obtained using “mild” criteria only. Thiopental has no amnesic effects using “maximal” criteria. Similar results are found for fentanyl, with a relative potency of 0.54 only obtained using “mild” criteria. Figure 6illustrates the relative potency relations determined using both the Emax and LR models with the “mild” verbal memory impairment criterion.
The relative potency analysis gives slightly different results than Cp50AMNcomparisons, finding more significant differences between drug pairs than were found with comparisons of Cp50 sub AMN values. For instance, using the mild word memory criterion, relative potency analysis indicates that midazolam is significantly more potent at producing amnesia than thiopental and fentanyl (Figure 6and Table 3), whereas when compared only at the Cp50AMNvalue, midazolam is not significantly different from thiopental (Figure 4(A) and Table 2).
Many previous studies have examined the amnesic effects of sedative drugs. [39–43] The effect of a drug on memory is frequently defined as a decrement from baseline or maximal possible performance. An alternative approach is to use quantal analysis to determine the probability of a defined response being present, analogous to the concept of MAC. It allows direct comparison between drugs with differing maximal effects at an equivalent, submaximal end point. Amnesia is not an “all or none” phenomenon, but rather presents a spectrum of effects depending on the intensity of stimulus, the method used to present it, and the methods used for later retrieval.  This is similar to the concept of anesthetic depth.  Whether an animal or human is “asleep” depends on the stimulus administered and the participant's response, and is evident in the current concept of MAC, which now includes MACawake, MACincision, MACintubation, and MAC-BAR. This study examined amnesic effects at various memory end points, which were designated “mild” and “maximal,” representing approximately the 50% and 100% memory effects of midazolam or propofol.
Few studies have addressed the issue of sedation produced by drugs and how it interacts with any observed amnesic effects. [39,40] It was found that the sedation versus serum concentration relation is parallel for the drugs we studied (as seen in Figure 2(B)). Thus, by normalizing actual serum concentrations at a given point, a concentration scale representing equivalent sedative effects at all points could be obtained. To increase the reliability of any conclusions, the equisedative concentrations were determined using two different methods of pharmacodynamic modeling, the Emax and quantal response logistic regression methods. Details of these methods are provided in Appendix 1.
We found that propofol has equivalent amnesic effects to those of midazolam at equal sedation. This result is consistent using different memory criteria with both normalized concentration scales. The Cp50AMNvalues found for “mild” verbal memory impairment correlate well with other investigators. Persson et al. found the EC50for midazolam for “partial amnesia” to be 64–81 ng/ml, which is close to our Cp50AMNof 56 ng/ml.  Leslie et al. found that propofol had a Cp50 of 0.66 +/- 0.1 micro gram/ml for suppression of learning during regional anesthesia, and this agrees well with our Cp50AMNvalue of 0.62 micro gram/ml.  The results of our study indicate unequivocally that the amnesic effect of propofol is not solely related to any sedative properties it may possess. This and other studies [5–12] indicate that previous reports stating that propofol is not a true amnesic agent are probably related to low concentrations of propofol being present at the time of intraoperative recall, as Glass has discussed. 
The logistic regression method of determining equivalent sedative concentrations is sensitive to the lesser sedative effect of propofol, and thus the relative amnesic potency of propofol using these normalized concentrations is greater. Two volunteers in the propofol group consistently scored themselves at lower sedation values than other volunteers in the propofol and other drug groups; the reason for this bias is unknown. There is no reason to exclude these persons from the data analysis, and their lower sedation scores account for the smaller sedation effect found in the propofol group, which, however, was of borderline statistical significance. Both propofol and midazolam result in similar amnesic effects when concentrations are high enough. The results of this study do not support statements such as “propofol is a better amnesic agent than midazolam.”
A comparison of the effects on verbal memory of propofol and midazolam has been done before.  This study in 35 healthy volunteers (five persons in each study group) found that midazolam had a greater amnesic effect than propofol, although both impaired recognition memory to a greater extent than did placebo. There are several explanations for the differences in our results. Drugs in the study by Polster et al.  were titrated to a reaction time task that was used as a measure of sedative effect. Whether this task is equivalent to the measure of sedation used in this study is unknown. No information is provided about the amount of drugs given, and no serum concentrations were obtained. The serum concentrations of drugs present in the Polster study were probably not equivalent to our study, at least in terms of the midazolam:propofol ratio. The memory stimuli and the way they were presented in the study by Polster et al. were also different than those in our study. In Polster et al.'s study, participants were not specifically instructed to remember the stimuli, whereas in our study participants were aware of the nature of the memory task after the familiarization session. The stimuli in the Polster study consisted of visually presented words at different degrees of fragmentation. Participants viewed sequentially more intact words until recognition occurred, and no learning task was involved. In our study, participants were repeatedly presented words verbally and were instructed to memorize them. It has been shown that lorazepam, a benzodiazepine, impairs visual processing,  and this may account for the greater degree of impaired performance Polster et al. found with midazolam on visually presented fragmented material.
The acquisition of material from the picture memory task was consistently less affected than the word memorization task by the presence of drug. This can be explained by differences in the salience of stimuli, the procedures used for encoding them, and the level of cues used to retrieve material. Picture stimuli were colorful and visually detailed in an attempt to realistically simulate the experiential qualities of normal episodic memory. To observe a picture and to describe it ensured that the memory was doubly encoded using both imagery and a verbal description.  This double encoding makes the picture stimuli more salient than words, which were encoded using verbal but no explicit visual representation. The increased memory strength of dual representation would account for the lesser degree of amnesia seen with the picture recognition memory test. In the picture recall task, when only one representation of the picture is provided as a cue, the amnesic effects of drugs are similar to the word recognition task. Picture recall involved only verbal cues and would only activate the auditory-verbal representation of the picture material. In fact, some persons who could not recall pictures after being given associative cues remarked that they could remember having said those words but could not remember in what context. Memory performance was much better when the visual representation of the memory was elicited by viewing the picture stimulus. Recognition of the stimulus in the same modality as it was originally presented is a powerful cue for retrieval,  and the results of this study are consistent with previous research that finds that recognition testing is more sensitive in detecting memories than recall testing. 
Thiopental had significant effects on memory using the “mild” word criterion, and these were greater than those of fentanyl. Thus thiopental's memory effects are not entirely related to sedation, and this study corroborates previous reports of thiopental's amnesic effects.  Note that if a more stringent criterion of memory loss is used, then neither thiopental nor fentanyl demonstrate any effects on memory. Several drugs have mild effects on memory, [42,52] and it is difficult to discern these when nonspecific sedation is present concurrently. The methods presented in this study will be useful in discriminating these separate effects.
We found that fentanyl has little effect on memory, consistent with other studies or case reports involving opioids. [53–57] A previous report from this laboratory found that fentanyl had a greater memory effect than we found in this study, even though similar maximum serum concentrations were achieved (approximately 2–2.5 ng/ml).  The most likely explanation for this difference is that participants receiving fentanyl in this study also received ondansetron. It is unlikely that 4 mg of ondansetron given intravenously has an effect on memory.*** However, the serotonergic system plays a role in memory function, [58–60] and ondansetron and other substances acting on the serotonin receptor system have been shown to enhance memory performance in healthy and demented elderly persons. [61,62] Thus it is possible that ondansetron antagonized the weak memory effects of fentanyl. Another possible explanation for the difference between this and our previous study is the presence of nausea. Despite the use of ondansetron, participants were still nauseated. As we had no formal measure of nausea we cannot assess if participants were more nauseated in this study than in the previous one. Nausea may have increased arousal, thereby improving memory performance. However, nausea did not seem to affect the ratings of sedation by the volunteers, and nausea during encoding of a stimulus would tend to distract the volunteer from a memory/learning task and thus impair memory.
We do not believe that differences in English fluency affected the memory performance of participants in this study. Persons who speak Spanish and English equally fluently ("balanced" bilinguals) do not differ from monolingual English speakers on tests of verbal long-term memory.  Persons whose English is less fluent ("non-balanced" bilinguals) are at a disadvantage in memory testing using the English language. We ensured that bilingual speakers performed at a level similar to monolingual speakers before inclusion in our study.
The placebo group also demonstrated a sedation effect, which may be a result of intensive testing resulting in fatigue. This effect was not maintained and sedation decreased as time progressed throughout the day. A substantial placebo effect seems to have occurred, because sedation mirrored the drug concentrations in the active drug groups. Thus sedation in persons receiving active drug may have been greater than can be attributed to the drug effect alone. Correction for this placebo effect is not simple, because the placebo effect is time-related, and our analysis of memory effect is based on serum concentrations rather than time. Although serum concentration and time are highly related in the study design, some participants had lower serum concentrations at the third target concentration than others at the second target concentration.
We have described methods by which the separate influence of sedation on the memory effects of drugs can be determined. These methods are particularly important when drugs with weak effects on memory are studied. This investigation shows unequivocally that propofol has similar effects on memory to those of midazolam, and these effects are not due to sedation. Previous reports of the lack of amnesic effects of propofol are most likely related to pharmacokinetic issues. Thiopental has a relatively weak effect on memory that is revealed only using mild criteria of memory impairment. This effect is not entirely mediated by sedation because participants with an equal degree of sedation (fentanyl/ondansetron) showed much smaller effects on memory. We found that ondansetron has neither sedative nor amnesic effects itself, but there is a suggestion that it may antagonize some of the effects of fentanyl on memory. Future studies assessing amnesic effects of sedative drugs need to standardize the definition of “amnesia,” account for the salience and intensity of stimuli presented, the cues used for assessing recall of stimuli, and control for strong confounding factors such as sedation and the presence of concurrent arousing stimuli (i.e., pain and nausea).
Appendix 1 Methods
This computer-assisted intravenous infusion device consists of a laptop computer running CACI software (courtesy of Jim Jacobs, Ph.D., and Peter Glass, M.B., Duke University) linked via a communications port to an infusion pump (Lifecare model 4P; Abbott Laboratories, Chicago, IL) that can receive and transmit digital data. Details of operation and performance as used in this study are provided in a separate report. 
Auditory-Verbal Memory Testing (Words). A modified Rey Auditory-Verbal Learning Test was used.  At baseline and at each target concentration, participants were presented one of six different 16-word lists selected from lists previously prepared by Drs. Ghoneim and Block of the University of Iowa.  The lists were administered in counterbalanced order by Latin square such that each list occurred equally often at each target drug concentration. Word lists were read aloud to the participant four times by the same investigator. After each reading of the list, 45 seconds was given for verbal recall of all the words. The number of words recalled on the fourth repetition was taken as the score for the learning phase of the test. Immediate recall was obtained 2 min after the fourth reading, after an interval filled with an interference task (forward digit span). Delayed recall for each list was obtained approximately 45 min later, at the same serum concentration. Delayed recognition for all words from all lists was obtained at the end of the study day.
In this study we used only the measure of delayed recognition, the most sensitive measure of memory performance; results of analyses with the other measures have been reported separately.  For the recognition test, the participants were presented a two-page written answer sheet containing all 112 study words, including those from the training session, as well as 64 nonpresented (distractor) words. Participants circled the words that they recalled from the word lists learned earlier that day and in the prior training session. Stimulus and distractor words for the study lists were selected according to frequency criteria of word occurrence in the English language using the Thorndike-Lorge norms.  All lists were equivalent for meaningfulness, imagery, and concreteness and were drawn from normative compilations. [67,68]
Visual-Verbal Memory Testing (Pictures). Color pictures (8.5 x 11 inches) free of text material were taken from magazine photographs or advertisements. Picture stimuli were colorful and visually detailed, to realistically simulate the experiential qualities of normal episodic memory. The pictures “told a story” and were not merely landscapes. Pictures contained sufficient details to occupy the participant in verbal description for 30 s and were unique enough that there would be unequivocal evidence of the participant's memory (e.g., merely saying “tree” or “house” would not be sufficient evidence of recall). Six pictures were shown at each target concentration. Each picture was shown approximately equally often at each target concentration during the experiment. Participants viewed each picture for 30 s and described it aloud while the examiner took notes.
The recall test was delayed until the end of the study day, just before the volunteers were discharged home. The volunteers were asked to recall and give a brief verbal description of all 36 pictures seen earlier that day. Four minutes were allowed for spontaneous recall of all pictures. If the volunteer failed to recall any picture, a standardized hint, prepared in advance for each picture, was given and a further 20 s was allowed for recall. The same cue was given to all persons who failed to spontaneously recall a particular picture. For example, the hint for a picture of a hot air balloon might be “up, up, and away”; a hint for a picture of a caterpillar changing into a butterfly might be “a physical transformation.” If recall again failed, an associative phrase drawn from the participant's own comments given when originally viewing the picture was supplied, and a further 20 s was allowed for a response. We tried to select “associations” that were sufficiently unique to relate only to that picture and to no other, but not so evident that the participant could guess the picture without actual recall. This association cue necessarily varied among the participants, but this was considered a more powerful cue than the standardized hint.
If this “association” cue did not prompt recall, the picture in question was presented for recognition, along with several distractor pictures that had not been shown previously, in a ratio of 1:4. The pictures were placed face down in front of the participant. The experimenter turned over one picture at a time, in rapid succession, and the participant was asked to respond “yes”(recognized) or “no”(not recognized) to each picture. This response was recorded. When a participant claimed to recognize a picture presented in this way, he or she was asked whether they really remembered the picture or whether they were just guessing based on the cues the tester had given previously. If a person said he or she had no true memory of the picture, but was just responding based on the previous hints, the recognition was scored as a failure (this occurred in only two or three persons). The cumulative total proportion of pictures either recalled or later recognized from each drug concentration was taken as the score used in assessing amnesic effect.
Pharmacodynamic Models for Estimating Cp50s
In comparing drugs, the relation of the actual serum concentrations present at equisedative effect will be different depending on the method used to model the concentration versus the sedation effect. The Emax method is well accepted as a model of a sigmoidal relation between drug effect and drug concentration, with comparisons between drugs usually made at 50% of maximal effect for each drug. The model is fitted to raw data, with the ceiling response occurring at a level defined by the data obtained. An implicit assumption in this model is that all drugs compared achieve the same maximal effect. In this study, all drugs had maximal sedation values of 25–27 (on a scale of 40) except propofol, which had values of about 19.
The other method we chose, binary quantal response logistic regression analysis, compares all drugs using a given criterion (in this study, a sedation score of 25), regardless of maximal effect achieved. This model also assumes a sigmoidal relation of drug effect, but it is fitted against the probability of achieving a given response rather than raw effect measures. The ceiling effect is determined by the criterion chosen, but in some cases this criterion may not be achieved. That is why in Figure 4and Figure 5only the initial increasing portion of the response curve is seen for fentanyl. Fentanyl could not achieve 50% probability of the memory effect used in these criteria. In this study, the criterion for sedation was one standard deviation above the mean for all values obtained from the placebo group. This value is substantially higher compared with the 50% of maximal effect determined by the Emax method, and thus the Cp50SEDvalues determined by logistic regression are higher than those obtained with the Emax model. Thus the Emax model compares the potency for production of amnesia at lower levels of sedation than the logistic regression model. Logistic regression analysis will represent the weaker sedative effect of propofol more distinctly, as can be seen in Figure 4(B and A).
Determination of Equisedative Effect Using the Emax Model. Actual average sedation scores were plotted against actual serum concentrations for each drug. Using a curve-fitting program (Table Curve 2D, version 3, Jandel Scientific, San Rafael, CA), the Cp50SEDwas determined using the following Equation 1: where D is the Cp50 value, A is the baseline sedation value, B is the maximal increase in sedation, and C represents the steepness of response.
Determination of Equisedative Effect Using the Logistic Regression Model. An average sedation score of 25 was chosen to represent the presence or absence of an effect. Logistic regressions for sedation present or absent were computed for each drug individually against log concentration, and the Cp50SEDwas calculated using the formula Cp50SED=-(intercept/slope). Further testing was performed to determine if the slopes of the sedation-concentration response for all drugs were parallel to each other by including all four sedative drugs in a single logistic regression, and examining the interaction term Drug by log(Concentration) in the logistic regression analysis. If this interaction term is significant, the slopes are not parallel. This interaction term was not significant (P = 0.11), and neither were the interactions among any of the drug pairs examined in the relative potency analysis. Thus, because the slopes are parallel, the relative sedative effects of the drugs studied are the same regardless of the point of comparison chosen (e.g., Cp10SED, Cp50SED, or Cp90 sub SED).
The authors thank Abbott Laboratories for providing the infusion pumps used in the study. The authors also thank the following persons for helpful suggestions and support: J. Jacobs, Ph.D., and P. Glass, M.B., Ch.B., from Duke University Medical Center for providing the CACI software; M. M. Ghoneim, M.D., and R. Block, Ph.D., of the University of Iowa for providing standardized word lists that were used for memory testing; P. Sebel, M.B., Ph.D., for helpful suggestions in the analyses; D. Leung, Ph.D., of MSKCC and J. Sigl, Ph.D., from Aspect Medical Systems for technical assistance on logistic regression analysis; A. Dnistrian, Ph.D., and Raia Stankievic for performance of drug assays; S. Dutcher, R.Ph., and E. Grivas-Mousouris, M.S., R.Ph., for preparation of study solutions; S. Miodownik, M.E.E., for technical assistance with equipment; and M. Duff, B.A., for assistance in data collection and analysis.
*Ideal body weight  was calculated as follows: for men, 48 kg for the first 152 cm of height and 1.1 kg for each centimeter over 152. For women, 45 kg for the first 152 cm and 0.9 kg for each centimeter over 152. In addition, body mass index was computed as weight in kilograms/(height in meters). Persons over the 85th percentile of body mass index (27.8 for men, 27.3 for women, ages 20–29 yr) were not recruited into the study.
**Propofol (Zeneca Pharmaceuticals, Wilmington, DE), midazolam (Hoffmann-LaRoche, Nutley, NJ), thiopental sodium (Abbott Laboratories, N. Chicago, IL), fentanyl citrate (Elkins-Finn, Cherry Hill, NJ), ondansetron HCl (Cerenex Pharmaceuticals, Research Triangle Park, NC).
***Participants receiving ondansetron alone showed an increase in word recognition between baseline and maximal target concentration (P = 0.06 by paired t test), but this is inconclusive in this small number of persons.