Drug-induced Amnesia Is a Separate Phenomenon from Sedation: Electrophysiologic Evidence

Click on the links below to access all the ArticlePlus for this article.

Please note that ArticlePlus files may launch a viewer application outside of your web browser.

• https://links.lww.com/ALN/A106

THE sedative and amnesic effects of sedative–hypnotic drugs are closely related, as sedation itself produces impairment in memory performance, and both effects vary in the same direction as serum concentration changes. 1,2There is evidence that separate neuroanatomic regions mediate arousal or attention versus  memory processes in humans who have not received any drug, 3–6and these processes can be indexed by specific components of event-related potentials (ERPs). 7In particular, the P3, the longest and most studied ERP component in relation to memory processes, has been repeatedly shown to relate to subsequent memory. In general, a larger P3 is associated with subsequent remembering. 7–9This finding can be interpreted as the central nervous system (CNS) attending more to distinct aspects of the stimuli and remembering these in preference to other, less distinctive stimuli. 9As with memory processes, these ERP components may be localized to certain neuroanatomic structures. 10,11This provides a neuroanatomic basis for the concept that drug effects on memory and sedation are specific. 12Curran et al.  13used these properties of the ERP to dissociate the memory versus  sedative effects of lorazepam, diphenhydramine, and scopolamine. The sedation-independent effect of a drug on memory is often not appreciated. If a drug is given to induce unresponsiveness (hypnosis), no explicit memory can be formed. 14Consequences of this fact include the current perception of some, for example, that propofol has few or no amnesic properties independent of its hypnotic effects. 15Few investigations of the memory effects of drug are conducted with the ability to differentiate sedative from amnesic effects while participants are still responsive.

The best method to demonstrate dissociation of sedative from amnesic drug effects is the use of different drugs to produce equivalent sedative, yet differing memory effects. 1,2,12,16Using this approach, we previously showed that the memory effects of propofol are similar to those of midazolam and are independent of a subjective measure of sedation, the Norris visual analog scale. 16To index CNS effect, that analysis used measured serum drug concentrations, which can only be obtained retrospectively. The current report presents a new analysis of these memory effects using the ERPs obtained in that study. The auditory ERP is a concurrent, real-time measure of CNS effect of the drugs administered.

The late components of the ERP are visibly affected to a larger extent by midazolam and propofol when compared with thiopental or fentanyl (fig. 1). We hypothesized that this difference is a result of the specific effects of midazolam or propofol on memory versus  sedation. Although the ERPs were obtained using a go–no-go attention task, which does not specifically target memory processes, there is evidence that some memory processes are engaged and that various ERP components from this paradigm index memory function. 7–9To explore this effect, we used the method of Smith et al.  17and Dutton et al.  18to relate auditory ERP components to either memory or sedative effects by means of the predictive probability parameter (P^k). Although the previously used Norris visual analog scale is an accepted method of assessing sedation, it is by necessity subjective. 19In the current analysis we chose to use another, possibly more objective measure of sedation, the logarithm of the reaction time (logRT) to an auditory stimulus in a go–no-go class of task. 20As before, the memory effect was measured by the ability of volunteers to recognize words previously presented at different target drug concentrations at the end of the study day. 16 

This investigation was approved by the institutional review board of Memorial Sloan-Kettering Cancer Center, New York, New York, before accrual of volunteer participants took place. All volunteers were given detailed information about the study, and informed consent was obtained before participation.

After telephone interview, volunteers were assessed in person for inclusion in the study. This assessment included a brief physical examination. Potential volunteers were excluded if there was evidence of medical, neurologic, or psychiatric problems, substance abuse, or medication use that might interact with the study drugs. Volunteers participated in an orientation session in which the study requirements were fully explained, and all experimental tasks were practiced.

Sixty-five volunteers (39 men, 26 women) aged 21–45 yr participated in a placebo-controlled study of mild, moderate, and maximal sedation levels in a single session. Volunteers were randomized to receive either midazolam (n = 11), propofol (n = 11), thiopental (n = 10), fentanyl with ondansetron pretreatment (n = 10), ondansetron alone (n = 8), or placebo (n = 15). All drugs were administered by computer-controlled continuous intravenous infusion (see Veselis et al.  21for details regarding kinetic data sets and pump performance) to target constant brain concentrations at three increasing and two decreasing levels of target concentration, with the highest concentration achieved designated as “maximum concentration” (see fig. 1 in Veselis et al.  16for more details regarding infusion profile). The highest concentration was chosen based on previous experience to produce deep sedation without unresponsiveness. Blood samples were drawn for later assay of serum concentrations by high-performance liquid chromatography. Serum concentrations were not determined for the ondansetron group. All volunteers fasted from midnight of the previous night. A 5% dextrose 1/2 normal saline solution was administered intravenously at approximately 125 ml/h to prevent stress from prolonged glucose restriction.

Nineteen channels of unipolar electroencephalogram were recorded using the International 10-20 System montage, referenced to linked mastoids, using NeuroScan SynAmps and software (Scan 3.0, NeuroScan Inc., Herndon, VA). Electroencephalographic data were acquired at a sampling rate of 200 Hz with a lowpass filter set at 40 Hz. A notch filter was used to remove 60-Hz artifact. Artifact rejection was conducted offline using various criteria, including a maximum amplitude criterion of ±100 μV, vertical electrooculogram, and electromyogram signals indicative of movement, and notes regarding artifacts and events were stored with the electroencephalographic data. Pure tone stimuli were presented via  headphones and were indicated in the electroencephalographic data file by timing marks for post hoc  ERP analysis.

Auditory stimuli were presented using an “oddball” paradigm, with frequent (80%: 1,000 Hz) and rare (20%: 2,500 Hz) tones presented at intervals of 1.1 s. Rare tones at 2,500 Hz were designated as “targets,” and participants pressed a button when they heard one of these tones. Frequent tones at 1,000 Hz (80%) were designated as nontargets, and no response was required. Five-minute trials included 33–43 target tones, from which the ERPs used in this report were derived (some target tones were lost after editing out artifacts from the ERP files). ERPs from target and nontarget tones were analyzed in 1-s epochs surrounding the target tone stimulus (−100 to +900 ms poststimulus). The latency, peak amplitude, and average amplitude of the P1, N1, P2, N2, P3, and negative slow wave ERP components from the target tones only were determined for the Cz electrode using Readpeak software (Christopher Nielsen, Copenhagen, Denmark). Although the Pz electrode is most commonly used in studies of the P3 component, we chose to present data from the more anterior electrode, Cz. We previously demonstrated that drug-induced changes in the electroencephalogram occur primarily in the frontal region with these drugs. 22The electrode Cz is close to most of the maximal ERP peaks (figs. 2 and 3; a color version of fig. 3is available in the Web Enhancement [fig. 3A]).

Following conventional procedures, amplitude was determined for each peak as the point with greatest deviation from baseline, and latency was measured as the value in milliseconds corresponding to maximal amplitude. The average amplitude of each peak was determined by averaging the voltage in time windows as follows: P1, 50–80 ms; N1, 80–120 ms; P2, 150–220 ms; N2, 160–350 ms; P3, 250–500 ms; N3, 400–700 ms. Average values for waveforms were used rather than actual peak values (except for N2P3, see below), because when drug is present, it is frequently difficult to identify which peak is present–absent in a given individual. In this situation, Readpeak software can provide a less biased estimate of ERP activity in the given time window than can be obtained by visual identification of a given peak in a given individual. The time windows were selected based on previous literature and examination of grand average waveforms obtained by pooling data across participants for the appropriate groups and conditions.

We found a substantial number of missing data points for the P3 peak amplitude and latency. This was a result of various factors, including the participant being too sedated to perform the tone discrimination task. In many cases the P3 was simply not recognizable because of the persistent α or β activity in the background electroencephalogram. In all of these cases, the P3 amplitude was determined as the average amplitude in the P3 time window (250–500 ms). This value was considered a less biased estimate of the P3 amplitude than a value of zero would have been. However, a latency value could not be accurately determined and was left as a missing value. Thus, there are more missing data for latencies than for amplitudes. This substitution of values was necessary for approximately 10–15% of the data points for the ondansetron, propofol, and thiopental groups, and for approximately one quarter of the data points in fentanyl and midazolam groups. No datapoints were missing from the placebo group.

Although P3 peaks were difficult to identify, N2 peaks were more easily identifiable. The peak-to-peak measurements for N2P3 amplitude demonstrated relatively high P^kfor memory effect (see below), and we thus include the N2P3 amplitude in this report as a peak-to-peak measure. If a P3 peak could not be identified, the average P3 value was substituted as the P3 peak value, as explained above. N2 and P3 amplitude average values are referenced to a 100-ms prestimulus baseline. The peak-to-peak measure is independent of the 100-ms prestimulus baseline value. The P^kvalues calculated using averaged N2 and P3 amplitudes were similar but not as large as peak-to-peak N2P3 values, possibly because of this consideration.

Volunteers completed a modified Rey Auditory-Verbal Learning Test at each target concentration, approximately 30 min before ERP recording at the same target concentration. Different word lists were used at each concentration in randomized fashion. The task consisted of learning a list of 16 words sequentially presented four times for memorization. Several hours later, after the drug had worn off, participants were required to recognize at the end of the study day any words that were presented on any of the study lists by circling them on an answer sheet that contained presented and nonpresented words (n = 112 and n = 64, respectively; more details are presented in the appendix of Veselis et al.  16).

Reaction times (RTs) to correctly detected target tones were calculated using the Neuroscan Respwin utility with an accuracy of 5 ms. As is customary, RTs were log-transformed (base 10) before statistical analysis to more closely approximate the normal distribution.

Sixteen bipolar visual analog scales of sedation were obtained repeatedly at each serum concentration. 19Four scales were presented on each of four pages. Each was 10 cm in length, and the participant would make a mark along the length of this scale. Completion of the 16 scales took approximately 2–3 min. The order of presentation of pages was randomized between tests, and similar poles alternated between left and right 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). As 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.

The P^k, a nonparametric test of the ability of one variable to predict another, was used to evaluate the strength of relation between ERP components from target tones and behavioral measures. 18This test is appropriate even if the data fail to approximate the normal distribution. This parameter accurately measures the predictive ability of one parameter for another regardless of scale units or underlying distribution of data. If perfect prediction is obtained, then the P^khas a value 1 (or 0). If there is no relation between parameters, then the P^kis 0.5. The P^kwill be between a value of 0.5 and 1 if one indicator increases as the other parameter does. P^kvalues between 0.5 and 0 indicate that as one parameter increases, the other decreases. The key statistical test is to determine if the P^kvalue is significantly different from 0.5, and this is performed using a standard t  test for proportions. P < 0.05 was taken as the level of statistical significance.

PkMACRO, obtained from Dr. Warren Smith (California State University, Sacramento, CA), was used to relate various ERP components to two predicted measures: word recognition by the participant at the end of the study day as a measure of memory (P^kmemory), and the logRT to target tones as a measure of sedation(P^ksedation). All observations from different dosages of the drug were included in the analysis. After this initial analysis, it was found that the N2 latency was the most predictive of sedation and that the N2P3 amplitude was the most predictive of memory when drug was present.

A combined ERP parameter was calculated that was to represent the “sedation-independent” amplitude of N2P3. In general, sedation will decrease the amplitude of any ERP component, as well as lengthening the latency of the ERP. Thus, the amplitude of N2P3 would be larger in the absence of the sedative effect, and the correction was empirically calculated by factoring in the increase in latency present with sedation: N2P3 amplitude (microvolts) + N2 latency (milliseconds)/25. The derivation of this parameter was based on similar considerations to those that were used to derive the “single index parameter” recently described by Dutton et al.  18The N2P3 peak-to-peak amplitude was chosen as the best indicator of memory and the N2 latency as the best indicator of sedation. Dutton et al.  used a corrective factor of 10 for the latency, working with an ERP component of latency approximately 100 ms. Thus, we chose 25, as we were working with an ERP component of latency of approximately 250 ms.

The predictive probabilities for each of the ERP components (latency and average amplitude) were compared as a set. To compare the P^kcoefficients with each other, the value of 0.50 was subtracted from each P^kvalue, and the result was expressed as the absolute value. This allows parameters that change in different directions to be directly compared. For instance, increased latency as well as decreased amplitude may predict sedation equally well with P^kvalues of 0.60 and 0.40. Absolute values were then ranked in order (rank 1 = largest P^k), with tied ranks expressed as the mean rank. The sedative drugs were ranked together (midazolam, propofol, thiopental, and fentanyl for sedation; fentanyl excluded for memory, as it had no memory effect). The rank order of ERP components in order of their predictive probabilities for word recognition (memory) and logRT (sedation) were determined. To test the hypothesis that, independent of drug used, a given ERP parameter was significantly better at predicting memory or sedation, group differences between the P^kvalues of all drugs combined (fentanyl excluded in the case of memory) of a given ERP component versus  the best ranked component were tested using z scores, with no correction for multiple comparisons using a PkDIFF macro, also obtained from Dr. Warren Smith (California State University).

Results are reported throughout as mean ± SD unless otherwise indicated. Traditional ERP components were analyzed by testing for differences between study groups by an analysis of variance with drug as a grouping factor. The behavioral variables (memory, Average Sedation, logRT) were tested at maximum concentration by one-way analysis of covariance with baseline scores as a covariate. Post hoc  comparisons of significant results were performed using the Dunnett one-tailed test, as the anticipated direction of change was for increased sedation, decreased memory, increased ERP latencies, and decreased ERP amplitudes with drug administration. In figures 1 and 4, change from baseline in recognition memory, reaction time, and N2 latency were analyzed by paired t  test for each group. P < 0.05 was taken as the criterion value for statistical significance in all tests.

Mean values for volunteer participants’ age and weight can be found in table 1for each drug group. All volunteers claimed to have slept normally the night before (mean sleep duration, 6.2 ± 0.9 h). No differences were found between drug groups in age, body weight (kilograms), body mass index, sleep, or right hand dominance.

All drugs impaired initial learning of the word lists (P < 0.0001 by analysis of covariance) compared with volunteers receiving placebo and ondansetron (table 1). Once learned, the material was well retained in the placebo, ondansetron, fentanyl, and thiopental groups, but participants receiving midazolam and propofol had further loss of information when tested by recognition at the end of the study day (P < 0.0001 by analysis of covariance). Recognition memory calculated as a change from baseline is shown in figure 1A, and absolute values are given in table 1.

The mean values of variables measuring sedation are shown in table 1. It should be noted that the placebo group also had significant increases in the visual analog scale measure of sedation, with the range of scores being somewhat similar to that in the groups receiving a sedative drug. RT was significantly increased for midazolam, propofol, and thiopental compared with placebo (P < 0.0001 by analysis of covariance;fig. 4). Accuracy of detecting target tones declined for all drugs compared with placebo (P < 0.001), but absolute performance levels were not impaired as much as might have been expected for the degree of sedation present (percent correct responses of total target tones: placebo, 100%; propofol, 87%; thiopental, 85%; fentanyl, 78%; midazolam, 66%). The midazolam group, in which volunteers missed one third of all target tones, had significantly worse (P < 0.05) performance than all other study groups. Because of high variability on this measure, only the midazolam and thiopental groups responded to significantly fewer tones than placebo. False alarms (responses to nontarget tones) averaged only 3% for the midazolam and propofol groups (P < 0.05 vs.  placebo) and were less than 1% for thiopental, fentanyl, and placebo.

Table 2presents mean latency and amplitude measures for traditional ERP components at maximum concentration and an analysis of variance analysis of drug effects on these. The corresponding grand average waveforms are shown in figures 1B–1F.

Figures 5–8graphically demonstrate the ability of particular components from the auditory ERP to target tones to predict memory or sedation. For instance, in figure 5, the P^kmemory of the average amplitude of the N2 component (solid dark gray bar), is significantly greater than 0.5 for all drugs other than fentanyl, which had no significant memory effect. This indicates that memory is better when the amplitude of N2 is larger. On the other hand, the P^ksedation (fig. 6) of the average N2 amplitude is significantly less than 0.5 in the case of propofol and thiopental, indicating that as sedation increases, the average amplitude of N2 decreases. This relation between memory and sedative effects is evident for the majority of ERP components, indicating that there is a very close relation between memory and sedation measurable in various ERP components. To further quantify the data in figures 5–8, the results of the rank sum analysis are presented in table 3. In general, the group differences between various ERP components were small and not significant in many cases. It should be noted that the combined parameter was almost as predictive of memory as the N2P3 amplitude, but possibly less predictive of log RT than N2 latency in comparison with N2P3 amplitude.

In general, the N2 latency was the best predictor of logRT in all groups, including placebo and ondansetron. In our study, when volunteer participants were under the influence of drug, the N2P3 amplitude was the best predictor of subsequent recognition of material presented at that particular drug concentration. However, in the placebo group, amplitude of the early components of the ERP predicted subsequent memory performance, with P^kmemory for N1 average amplitude of 0.70 ± 0.04 (P < 0.001).

The predictive performance of the combined ERP parameter is presented in table 4, in comparison with the predictive ability of actual measured serum concentrations. It should be noted that, in many cases, the predictive ability of ERP values for memory and sedation are as good as serum concentration itself.

For volunteers receiving ondansetron, P^ksedation for the N2 latency was 0.60 ± 0.05 (P < 0.05), similar to the other groups. No ERP components related to memory performance for participants receiving ondansetron, even though the distribution of memory scores was similar to participants receiving placebo. For this reason, ondansetron data have been omitted from figures 5–8to save space. The lack of predictive relation between ERP variables and memory may be related to the fact that the fewest number of participants were in the ondansetron group (8 vs.  15 for placebo).

Midazolam and propofol, drugs with memory effects independent of sedation as measured using subjective ratings, 16seem to have a differential effect on the late positive components of the auditory ERP in relation to other drugs with sedative properties and lesser effects on memory, such as thiopental and fentanyl (fig. 1). 23These late components, including the P3, relate to novelty detection and memory processes. 7,24Subsequently remembered material is associated with a larger ERP component in this latency range and has been localized to prefrontal regions, close to those exhibiting dose-related decreases in regional cerebral blood flow with midazolam. 7,25–28Components of the ERP from 50–250 ms relate to general arousal or attention and may index the sedative effects of drugs. 18Curran et al.  13dissociated the amnesic from sedative effects of lorazepam by differential effects on early and late ERP components.

In general, the results from this study extend the findings of Curran et al. , 13with various components differentially predicting measures of sedation and memory (table 3). Some ERP components consistently predict memory performance or sedation for each drug and thus have a substantially better ranking than other components, despite the fact that, overall, the differences in P^kvalues may be small and not necessarily significant. These small and variably significant differences are confirmed by the analysis of variance results in table 2. As in the study by Curran et al. , 13latencies of P1 and N1 were unaffected, while the amplitudes of P3 and N2P3 were decreased significantly by midazolam and propofol. In contrast to the findings of the current study, Curran et al.  reported that lorazepam or scopolamine did not affect N2 latency and amplitude. Differences between this and the current study include how ERP amplitudes were measured (peak vs.  average amplitude), measures used to assess sedation (tapping and more complex psychomotor tasks vs.  reaction time), and memory measures (a continuous recognition task with simultaneous ERP recording vs.  a repeated presentation of word lists).

The N2 latency was increased by all drugs at the highest concentration (fig. 4), with P^kSEDATION values ranging from 0.59 for thiopental, which had the smallest change in N2 latency, to 0.71 for fentanyl. 29Close inspection of figure 1reveals that the maximal peak amplitude of N2 may be larger (i.e. , more negative) when drug is present, which is contrary to the generalized decrease in ERP amplitudes seen with sedative–hypnotic drugs. The increase in peak N2 amplitude with sedation may be related to the volunteer’s increased “effortful processing” to overcome the increased difficulty of the discrimination task when drug is present.

The N2 is automatically elicited by changes in stimulus properties, and its latency and the reaction time are closely related. 30,31Decreased arousal or attention will impair the ability to rapidly detect a deviant stimulus and can thus affect N2 latency. The N2 has various subtypes, and the one most commonly observed in the “oddball” paradigm is the N2b. 32,33Its frontocentral distribution is similar to our findings (fig. 3). The N2b–P3a complex is thought to represent evaluation of stimulus information relevant to attentional focus and response choice. 31 

Increases in reaction time may or may not be synonymous with “sedation.” There is currently no well-accepted direct measure of sedation, and different measures can yield different results. Sedation as measured by reaction time was little changed by fentanyl. However, visual analog scale measures showed similar effects of fentanyl to other drugs. Reaction time may have been enhanced by the arousing stimulus of the nausea experienced by 70% of volunteers receiving fentanyl. The opposite pattern was seen with propofol, where the visual analog scale of sedation changed little, but reaction time and N2 latency indicated substantial sedative effect. Other investigators have noted similar discrepancies between different measures of sedation, as well as imprecise perception of participants as to degree of objective impairment. 13,34A comparison of the best-ranked ERP components (table 3) to predict changes in logRT and Norris visual analog scale was made (table 5). In general, similar predictive abilities were obtained, except in the case of propofol. It may be possible that the subjective perception of sedation during propofol is inaccurate compared with other drugs. As the results from the propofol group are congruous with other drugs when logRT is used, one might argue this is a more objective measure of sedative effect.

We found that the peak-to-peak amplitude of N2P3 was the best predictor for memory impairment, ranging from 0.62 for propofol to 0.66 for thiopental (table 3). Because neither fentanyl nor ondansetron produced impairment of memory, there was no relation between the N2P3 amplitude and memory with these drugs. The of N2P3 amplitude is also predictive of sedation, except in the case of midazolam, where it seems to reflect a memory effect independent of sedation. As N2 latency was consistently predictive of sedation, it was used in an attempt to factor out the sedative effect on the amplitude of the N2P3. 18The predictive ability for a memory effect is retained for propofol, midazolam, and thiopental, whereas the ability to predict sedation is retained only for thiopental and fentanyl (figs. 5 and 6). The predictive ability of this combined parameter compared with N2P3 amplitude alone is presented in table 3. These findings, indicative of differential sedative and amnesic effects in the ERP components, are congruous with our previous ones derived from other independent measures. 16A thorough analysis of all possible linear combinations of various ERP components and P^kvalues might reveal a memory predictive parameter largely independent of sedation.

Changes in the state of arousal affect memory performance. The N1 component of the ERP, too early to reflect any memory processes, relates to arousal and automatic initial processing of stimuli. 3The predictive ability of N1 amplitude for memory performance in participants receiving placebo might be explained by higher levels of arousal compared with drug. More cognitive resources were available to perform both the discriminative attention and the memory tasks with less effort. Alternatively, the failure to find the expected relation between later ERP components such as N2 and P3 and memory performance in the placebo group may be a result of the limited range of scores in this group (mean recognition score: 14.0 + 1.5 words from a list of 16 words.)

The results of this study need to be interpreted while keeping in mind certain considerations. Although ERPs were obtained soon after the word memorization task at constant serum concentrations, they were not obtained simultaneously with word encoding. The oddball paradigm used examines the attention paid to a deviant stimulus. Thus, it measures more the ability of the brain to attend to and discriminate stimuli rather than memory processes per se . In fact, a dissociation between enhanced memory of distinctive stimuli and the amplitude of the P3 may exist. 9In future investigations, ERPs should be obtained concurrently with a longer list of stimulus words. The P^kvalues obtained in this study are less than the desirable values of greater than 0.8 (or < 0.2), such as those reported by Dutton et al.  18In that study, the P^kvalues derived from ERP components were similar to those from end-tidal anesthetic gas concentrations, which can be considered analogous to measured serum concentrations for intravenous agents. Despite our relatively lower P^kvalues, these are comparable with the P^kvalues obtained for measured serum concentrations. Explanations for this finding include the following: (1) the ERPs were not recorded simultaneously with the verbal learning task or with the delayed recognition test; (2) Dutton et al.  used a well-defined end point, loss of responsiveness, whereas no such similar measure exists for sedation or memory; and (3) the data in this study represent relatively small changes in serum concentration from baseline to maximal sedation, whereas Dutton et al.  were able to study the much larger ranges associated with full anesthesia (P^kvalues tend to be higher for larger data ranges).

Multiple comparisons were performed of P^kvalues against a value of 0.5. Based on previous results, we hypothesized that N2 latency would be the most sensitive to sedation effects and that the P3 or similar ERP component would be the most sensitive to memory effects. 29,35These hypotheses comprise approximately three to five a priori  comparisons. Actual P  values for a two-tailed t  test are reported in figures 5–8, and those significant at a level less than 0.01 are conservatively appropriate for the number of a priori  comparisons.

The auditory P3 is generated in distributed brain regions. 10,11,36The neuroimaging literature demonstrates that memory function is a distributed process that may correspond to the components of the ERP later than 250 ms. 5,37ERP components before 50–100 ms are produced by neural transmission through the brainstem and thalamus, sites that are prime candidates for the hypnotic (as defined by loss of responsiveness) effects of anesthetics. 38,39At concentrations of anesthetic associated with responsiveness, this hypnotic effect is manifested as “sedation,” measurable in given components of the ERP. At concentrations producing heavy sedation, but not complete unresponsiveness, both propofol and midazolam have the ability to produce oscillations in the 12–14-Hz range 40similar to previously described thalamocortical oscillations. 41Midazolam and propofol produce regional decreases in thalamic blood flow at these concentrations, 39,42and changes in thalamic metabolism have been correlated with changes in arousal with other benzodiazepines. 43These effects may represent the expression of sedative drug effects via  thalamic mechanisms at these concentrations.

In conclusion, the differential effect of CNS-active drugs on memory and arousal–attention–sedation can be demonstrated in the ERP. Using ERPs as a measure of CNS effect may be particularly advantageous as objective measures of both memory and sedative effects are obtained simultaneously and in real time. In addition, it is now possible to locate potential generators of particular ERP components with reasonable accuracy in the brain. 11The findings of this study are in agreement with the hypothesis that sedation, the “hypnotic” effect of anesthetics at lower concentrations than those causing unresponsiveness, may not act exclusively by generalized CNS depression. 38,39,44The fact that certain CNS structures relate both to arousal or attention and memory processing 45provides a neuroanatomic basis for the interaction of sedation with memory when drugs are administered. The ability to relate the electrophysiologic changes to the memory and sedative behavioral effects of drugs provides the ability to measure these effects independently without the potentially confounding factor of subjective perception. These identified electrophysiologic changes can then be related to brain images from positron emission tomography or functional magnetic resonance imaging to accurately localize and quantify the neuroanatomic regions mediating these drug effects.

The authors thank Ann M. Dnistrian, Ph.D. (Associate Clinical Chemist, Department of Clinical Laboratories, Memorial Sloan-Kettering Cancer Center, New York, NY), for performing the serum concentration assays, and Martin Duff, B.A., Diamanto Tsakanikas, B.A., and Obiageli Anyaogu, B.A. (Research Assistants, Department of Anesthesiology and Critical Care Medicine, Memorial Sloan-Kettering Cancer Center, New York, NY), for data collection and analysis. They thank Warren D. Smith, Ph.D. (Professor, Biomedical Engineering Program, California State University, Sacramento, CA), for sharing PKMACRO software and for helpful suggestions in implementing this program. Ray Johnson, Jr., Ph.D. (Professor, Department of Psychology, Queens College of CUNY, New York, NY), provided helpful comments on the analysis and interpretation of the ERP data. The authors thank the clinical members of their department (Memorial Sloan-Kettering Cancer Center, New York, NY), whose daily efforts help support research activities such as these.