Memory Function during Propofol and Alfentanil Anesthesia: Predictive Value of Individual Differences

After approval from the human investigations committee at the Academic Hospital Rotterdam, The Netherlands, 65 healthy (American Society of Anesthesiologists [ASA] physical status I or II) outpatients scheduled for surgery with general anesthesia gave written informed consent to participate and were enrolled in the study. Patients were aged 18–60 yr, fluent in Dutch, and reported intact hearing and no history of drug abuse or current use of psychoactive medication.

The study comprised three phases. Several weeks before surgery, speed of information processing and memory function were assessed (preoperative phase). During anesthesia and before surgery commenced—to avoid the potentially confounding effect of noxious stimulation—patients listened to a series of words and verbal commands while anesthesia was maintained at BIS, 60–70 (perioperative phase). After surgery, memory for presented words and commands was tested (postoperative phase).

Speed of information processing (SIP) was assessed with a choice reaction time task in which an arrow (black against white background, Geneva font, size 72) appeared in the center of a Macintosh PowerBook 3400 computer screen (Apple Computers, Cupertino CA) pointing either left or right. Participants were instructed to indicate the orientation of the arrow as quickly as possible by pressing a particular key. Stimuli remained visible until a response was generated with a maximum of 3 s. A new stimulus appeared after 1 s. Ten practice trials preceded 50 test trials. From the test trials, an average reaction time was derived for each person, after removing incorrect responses and outliers (reaction time above or below 3 SD from the mean). The number of included test trials ranged from 43 to 50.

To assess individual memory function, patients were shown 10 printed words introduced as illustrative examples of what could be played during anesthesia. Half of the patients were shown list A; the other half, list B. Patients were instructed to study and memorize all words for 1 min. Attention was then diverted (SIP test), and approximately 5 min after study, memory was tested using a word stem completion task (MEM test). Twenty printed word stems corresponding to the words of list A and B were shown. Participants were told that some of the stems could be completed with words from the study list. They were instructed to use stems as cue to recall studied words and to complete the stems accordingly, or to complete stems with the first word coming to mind in case recall failed. After completing all stems, patients were asked to indicate the words they remembered from the study list. For each patient, we established the number of study words used for stem completion (cued recall hits ) and the number of study words pointed out correctly (recognition hits ).

Electroencephalographic activity was measured using an A1000 monitor (Aspect Medical Systems Inc., Newton, MA) with a two-referential montage. Four self-prepping electrodes (Zipprep; Aspect Medical Systems Inc.) were attached to the following sites: channel 1 and 2 to the left and right outer malar bone (At1 and At2), a referential electrode midway the forehead (Fpz), and a ground electrode approximately 2 cm right from the reference electrode (Fp2). Electrode impedance was below 5 kΩ during EEG recording. Recording of BIS (algorithm 3.2) started before induction of anesthesia and continued throughout the presurgical anesthetic period.

The anesthetic protocol consisted of target-controlled infusion (TCI) with propofol using an IVAC infusion pump (Alaris Medical Systems, San Diego, CA) targeting propofol plasma concentrations with Diprifusor software (AstraZeneca Macclesfield, Cheshire, UK). Patients received no premedication, and anesthesia was induced with propofol, 6 μg/ml, TCI. After loss of the eyelash reflex, the lungs were ventilated with 100% oxygen. A bolus dose of alfentanil, 20 μg/kg, was administered intravenously when BIS decreased below 80, followed by suxamethonium, 1 mg/kg, when BIS decreased below 60. The trachea was then intubated, and the lungs were mechanically ventilated with a mixture of air and oxygen (ratio, 60:40%). When the train-of-four indicated return of muscular activity, a cuff was inflated to 250 mmHg around the forearm of the dominant hand, occluding blood flow to the hand. This IFT 16,17excluded the hand from neuromuscular block when vecuronium bromide, 0.1 mg/kg, was injected. The IFT enabled patients to indicate awareness either spontaneously or on request (explained below). Propofol plasma concentration was targeted to BIS, 60–70, for the remainder of the presurgical study period, and word presentation started as soon as BIS was above 60.

The 44 words used in this study were selected after a pilot WS test in a comparable group of patients (n = 30). The pilot established the probability of correct word stem completion without previous presentation of the corresponding word (i.e. , base rate). Base rates of selected words ranged from 0.17 to 0.33 (mean, 0.25), and words were assigned to four lists so that comparable base rates occurred. Each patient was presented with two lists during anesthesia (22 targets ) and tested on all 44 words, including the two lists that had not been presented during anesthesia (22 distractors ). Lists were counterbalanced so that all words appeared equally often as a target and as a distractor. Words were digitally recorded in a female voice (first author), saved as 16-bit sound samples and played to patients via  closed headphones connected to a Macintosh PowerBook 3400 (Apple Computers, Cupertino, CA). A computer program generated a different random word order for each patient and repeated each target 10 times consecutively with a 1-s delay in between repetitions. The ten repetitions of each target took approximately 35 s. Using automatic data logging, the average BIS during the 35-s word presentation was recorded. Observers were blinded to the contents of word presentation.

Between word presentations, i.e. , after 10 repetitions of the same word, response to verbal command (“awareness”) was assessed. The computer program for word presentation was halted, headphones were partially lifted, and the observer, who continuously held hands with the patient, spoke close to the patient's ear. Each command was preceded by the patient's name and followed by stroking the palm of the nonparalyzed hand. Patients were first asked to squeeze the observer's hand once. If no squeeze occurred within approximately 10 s, responsiveness was scored absent (nonresponse ), and word presentation continued. Patients who did squeeze once were then asked to squeeze twice. Failure to squeeze twice was considered an inadequate (equivocal ) response, and word presentation continued. Squeezing twice on command was considered an adequate (unequivocal ) response indicating awareness, and we then asked patients to squeeze twice if they felt all right or to stretch their fingers if not. In the latter case, propofol infusion was increased with 0.2 or 0.5 μg/ml TCI, depending on the going target concentration (below or above 3 μg/ml, respectively). If necessary, additional alfentanil (1 mg) was given. Before patients entered the operating room and surgery commenced, the anesthetic was increased until BIS, 45, and the cuff, headphones, and electrodes were removed.

Within hours after surgery, patients were tested on the ward just before hospital discharge. We assessed conscious recall with a short structured interview consisting of the following four questions: What is the last thing you remember before falling asleep? What is the first thing you remember after waking up? Do you remember anything in between? Did you dream? 18 

The WS test was then administered in combination with the PDP. PDP requires participants to complete a memory test during two different experimental conditions. In both conditions, subjects are presented with cues (e.g. , word stem) serving recall of information (e.g. , words) presented earlier (e.g. , during anesthesia). In the inclusion condition, participants are instructed to complete stems with a word presented earlier or with the first word coming to mind in case recall fails. Because explicit and implicit memory function lead to the same response (i.e. , completion with the presented word), the inclusion condition is a measure of general memory performance. To separate explicit and implicit memory effects, the exclusion condition is required in which the instruction is given not  to use presented words for stem completion but to use another word instead. Contrary to the inclusion condition, an explicit memory effect in the exclusion condition results in less frequent responding with presented words. Comparison of test scores in the two conditions indicates whether memory test performance should be attributed to implicit or explicit memory function, and relative contributions of both may be estimated. 4–6 

In line with the PDP, the WS test consisted of an inclusion and exclusion part, the order of which randomly varied between patients. Instructions were thoroughly explained before each part and illustrated with an example. Word stems were presented auditorily via  closed headphones and visually on a computer screen, using the same headphones and computer as during word presentation. The digital sound sample of the word stem had been derived from the sound sample of the word presented during anesthesia, which had been copied and cut so that only the first few letters remained audible. For each patient, the computer program selected two word lists (one target and one distractor) for the inclusion condition and exclusion condition each and presented the corresponding word stems in a different random order. Overall, each of the 44 words appeared equally often as a target and as a distractor in the inclusion and exclusion test condition. In each condition, patients completed 11 word stems corresponding to target words and 11 corresponding to distractors. Responses were entered into the computer by the observer.

As a measure of cued recall, patients were instructed to report each word they recalled hearing during anesthesia. They were reminded to do so once during and once after the inclusion and exclusion test. Finally, all 44 words were read to patients in a fixed order, and for each word, patients decided whether it had been played to them during anesthesia (recognition test).

The first part of data analysis focused on observed hit rates in the two conditions of the WS test, a “hit” representing response with a study word. The distractor hit rate establishes the probability of response with a study word without it being presented previously (“base rate”). The hit rate for targets, in contrast, establishes the probability of response with a study word when it has been presented during anesthesia (“hit rate”). Paired sample t  tests were used to compare the hit rate with the base rate in the inclusion and exclusion condition. Implicit memory function would lead to hit rates higher than base rate in both parts of the test. Explicit memory function, on the other hand, would result in a hit rate higher than base rate in the inclusion condition and a hit rate lower than base rate in the exclusion condition. This result shows response control, a feature associated with explicit memory function only. Response control, however, may be observed in the absence of conscious recall (unconscious–controlled memory) or in its presence (conscious–controlled memory). Absence of memory is indicated by base rate performance in the inclusion and exclusion parts of the WS test.

The second part of data analysis addressed individual differences in memory function during anesthesia using a linear multiple regression model with general memory performance as a dependent variable. Therefore, this part of the analysis was based on the data obtained in the inclusion condition of the WS test. Given the limited sample size, a restricted number of variables were studied as predictors. 19Before surgery, patients’ speed of information processing and memory function were investigated because good learning ability has been associated with quicker sensory processing 20,21and better memory function. 22,23During surgery, we assessed response to command because postoperative memory is often attributed to awareness. From these assessments, we derived the (a) mean reaction time in the SIP test, (b) number of cued recall hits and (c) recognition hits in the MEM test, (d) percentages of equivocal and (e) unequivocal response to IFT command, and entered (a) to (e) as predictor variables in the regression analysis (SPSS 9.0, SPSS Inc., Chicago, IL). Only cases with observations for all five variables were included in the analysis. SPSS tested correlations between memory performance and predictor variables for one-tailed significance, which is common for statistical testing of a priori  (directional) hypotheses.

Third, we performed an item analysis. Despite careful memory test construction, not all words and word stems (i.e. , items) in this study were expected to measure memory equally well. As malfunctioning items elicit unwanted variance thereby reducing statistical power, item analyses were performed on the word stem completion data to identify and remove such items from the test. The aim was to improve the tests’ overall quality. Using (SPSS) Scale Reliability Analysis, we identified 15 items that correlated poorly with other items regardless of a memory effect, i.e. , based on the distractor data in the inclusion and exclusion conditions. These items also correlated poorly with other items in the target inclusion condition, which measures memory. Removing these items from the test increased its internal consistency, as measured by Cronbach α, 24from 0.12 (44 items) to 0.43 (29 items) to 0.67 (24 items) after removing another five. Based on this improved version of the WS test, we calculated new hit scores for each patient and reanalyzed the memory effect and its relation to predictor variables as described previously. P < 0.05 was considered statistically significant. Data are presented as mean ± SD.

Nine patients were excluded from the data set, either because median BIS remained below 60 during word presentation (n = 5) or because of failure to complete memory testing (n = 4). Included patients (25 women, 31 men) were aged 37 ± 10 yr (range, 19–58) and underwent orthopedic (n = 45), general (n = 7), or plastic surgery (n = 4). They were anesthetized for 39 ± 11 min in the presurgical study period, followed by 45 ± 17 min of general anesthesia for surgery. The IFT was implemented for 24 ± 3 min during which words and commands were presented during a period of 19 ± 2 min. BIS during word presentation was 64.0 ± 3, and 1,082 commands were given overall (average, 19 ± 3 per patient). No response was observed to 887 commands (82%), equivocal response to 56 commands (5%), and unequivocal response to 139 commands (13%). Fifteen patients (27%) did not respond to command at any time (309 nonresponses), whereas 37 patients (66%) responded unequivocally to command at some point during their anesthetic (range, 1–16 times). Memory was assessed 3.8 ± 0.5 h after word presentation and 2.8 ± 0.5 h after surgery.

All patients were interviewed about conscious recall of perianesthetic events: nine (16%) reported partial yet accurate recall of verbal commands, but no patient remembered any of the words presented between commands. Presentation of word stems (cued recall test) resulted in 8 correct recollections versus  37 incorrect recollections overall. When memory was prompted further by showing the target words mixed with distractor words (recognition test), patients overall recognized 143 words correctly versus  140 incorrectly. These observations indicate absence of unprompted and prompted conscious recall for presented words.

Observed hit rates for target and distractor words in the WS test are shown in table 1. In the original WS test, the hit rate tended to be higher than base rate in the inclusion condition and lower than base rate in exclusion condition, but neither difference was statistically significant (P = 0.25, inclusion condition;P = 0.13, exclusion condition). When comparing the hit rates of the two test conditions, significantly more hits were scored in the inclusion part than in the exclusion part (P < 0.05). This indicates that patients did have memory for presented words. In the improved version of the WS test, results were basically the same: hit rates differed significantly between test conditions (P < 0.05) but not from base rate within test conditions (P = 0.09, inclusion condition;P = 0.33, exclusion condition).

Because conscious influences affect the PDP results, we performed post hoc  analyses on the word stem completion data of patients with and without conscious recall of verbal commands, even though none of the participants remembered the words presented. As can be seen in table 2, the hit rate of patients with conscious recall was significantly higher in the inclusion condition than in the exclusion condition (P < 0.05). No other comparisons reached statistical significance. Characteristics of the two groups (table 3) indicated that patients with recall more often responded to IFT commands, although BIS was comparable during such responses. Preoperative and postoperative performance were comparable as well.

Thirty-eight patients were included in the regression analysis, which aimed at predicting WS test performance from preoperative and intraanesthetic variables. In the remaining 18 patients, preoperative assessments were not obtained. Because conscious recall turned out to be an important variable, it was added as predictor in the analysis. Results for the original and improved versions of the WS test were comparable, and, for informative purposes, the strongest correlations are reported here. No significant correlations were found between postoperative hit scores and preoperative recognition hits (r = 0.25), choice reaction time (r =−0.19), percentages of equivocal (r = 0.24) or unequivocal (r = 0.06) response to command, and conscious recall (r = 0.03). Hit scores did correlate significantly to the number of hits produced in the preoperative memory test (r = 0.35, P < 0.05). As can be seen in figure 1, this correlation was particularly strong in the recall group (r = 0.61, n = 6) compared with the no-recall group (r = 0.29, n = 32), but neither was statistically significant (P = 0.1 and P = 0.06, respectively). Taken together, the regression model explained 13–15% of the variance in postoperative hit scores (original vs.  improved WS test).

Memory function during anesthesia has been studied for different (physiologic, psychological, medicolegal) reasons and with mixed results. 25,26Much research was hampered by a lack of control for anesthetic adequacy, and for a long time, memory tests were thought to reflect either conscious (explicit) or unconscious (implicit) learning. Recent technologic and methodologic advances have improved reliable measurement of hypnotic adequacy and memory function during anesthesia. BIS was introduced as an online index of hypnotic state ranging from 100 (awake) to 0 (isoelectric brain), and PDP enabled separation of explicit from implicit memory effects within one test. 7,8 

Using BIS and PDP, we have demonstrated dependence of memory function on hypnotic state in acute trauma patients undergoing surgery at different levels of anesthesia, as well as preserved implicit memory function during apparently adequate hypnosis (BIS, 40–60). 4In a more recent study, memory function in women undergoing emergency cesarean section was investigated. 5The probability of awareness and memory function is increased in these patients because hypnotic state is generally light during surgery. Patients had no conscious recall of intraoperative events yet demonstrated unconscious–controlled memory for the words presented at consistently light levels of anesthesia (BIS, above 70). 5The current study assessed memory for words presented during BIS, 60–70, using a WS test in combination with PDP, like in previous studies. We also aimed to identify patient-related sources of variation in memory performance.

Conventionally, memory is evident when a group exposed to study material outperforms another group not exposed to the material. In a similar vein, we compared target to distractor hit rates in both conditions of the WS test. In the inclusion condition, patients were inclined to use a study word more often for stem completion when it had been presented during anesthesia than when it had not. In the exclusion condition, in contrast, they tended to use fewer presented words when instructed to avoid those. These observations suggest explicit memory for presented words, but a conventional memory effect was not found. This is in line with the absence of conscious recall and recognition of presented words seen in this study.

Patients did make correct inclusion–exclusion decisions, however, as evidenced by the higher hit rate in the inclusion condition compared with that in the exclusion condition. Thus, they displayed response control over words presented on average at BIS, 64. Because none of the words was consciously recalled or recognized, decisions to either include or exclude a word were made without conscious awareness. Hence, we found evidence of unconscious–controlled memory. This finding corroborates the results of Lubke et al. , 5who presented words at BIS levels consistently above 70 during emergency cesarean sections. As response control is typically associated with explicit memory function 7,8but was not accompanied by conscious recall, Lubke et al.  5concluded that unconscious–controlled memory is a weak form of explicit memory. Our current findings extend the evidence of weak explicit memory function to a near-adequate level of hypnosis. However, Lubke et al.  5demonstrated response control in addition to a conventional memory effect. Therefore, their evidence of weak explicit memory function was stronger than in the present study. This is also indicated by effect size estimates (h) for response control in the previous and present studies (h = 0.31 vs.  0.13). The stronger evidence found by Lubke et al. , 5in combination with an average higher BIS during cesarean section, supports dependence of memory function on hypnotic state as measured by BIS.

It should be noted that only patients with conscious recall demonstrated response control, and, as such, the evidence of weak explicit memory function in the present study should be attributed to this particular group of patients. No reliable effect was found in patients not reporting recall. This difference between patients supports the PDP in that the presence or absence of conscious recall leads to differential response patterns. Unlike most PDP experiments performed in the laboratory with awake subjects, however, our studies show that the ability to exclude is not necessarily accompanied by conscious awareness and recollection. None of our patients, including those with recall, remembered or recognized the words presented during anesthesia. They were nevertheless able to include or exclude words. From a clinical point of view, the difference in word stem completion performance between patients with and without conscious recall is also relevant because it indicates unconscious consequences of intraanesthetic awareness. Those with recall unknowingly “act” differently. Our findings strongly suggest that such awareness affects postoperative memory, but the relation is not straightforward, however, as illustrated by the fact that only one of four patients responding to command during anesthesia reported conscious recall afterward.

Using PDP, Stapleton and Andrade 6also observed reliable explicit memory in the absence of conscious recall. Whereas they were not convinced that such findings should be treated as special, we tend to go along the lines of Lubke and Sebel, 27arguing that the dichotomy between conscious recall and implicit memory is too crude. Instead, a third expression of memory should be distinguished that incorporates implicit and explicit features. Unconscious–controlled memory is implicit in terms of conscious recollection, which is absent, but explicit in terms of response control, which is present. Unlike the assumption of the PDP, response control does not necessarily reflect conscious information processing and may arise from a sense of familiarity, an unconscious process, as well. As Stapleton and Andrade 6correctly observed, explicit memory may be overestimated at the expense of implicit memory when separate estimates are calculated using PDP 7,8or related models. 4,5 

Stapleton and Andrade 6did not measure BIS during word presentation but determined blood concentrations of propofol after presenting a word list during surgery and assessed response to command at the start and end of the list. Neither index of depth of anesthesia correlated with memory performance, and the authors expressed confusion over these findings as “explicit memory is typically very sensitive to manipulations of consciousness.”6We too found awareness during anesthesia to be of little predictive value to memory performance apart from recall. Our observations and those of Stapleton and Andrade 6could be well explained considering the experience of being awake during anesthesia. Some of our patients accurately recollected awareness, but none remembered the words presented within seconds before of after the commands. Although ethical considerations confine the scope of memory research to relatively unimportant words, the current findings suggest that painful experiences, conversations about life-threatening issues, or fearful events such as awakening paralyzed are a more memorable stimulus than simple (neutral) words. Our patients expressed concern over what had been happening while being awake rather than remembering or paying attention to the words. Hence, responsiveness may be expected to correlate poorly with memory performance for presented words.

Predicting variation in memory performance was partially successful. On the one hand, preoperative memory performance correlated with postoperative hit scores, a finding interesting in two respects. Whereas preoperative hit scores had predictive value, recognition memory did not. This may indicate that implicit rather than explicit memory function affects memory formation during anesthesia, although no attempts were made to separate the two in the preoperative memory test. Further, because this correlation could be demonstrated in a relatively small sample, the study of presumably stable (internal) patient characteristics, as opposed to transient (external) variables like depth of anesthesia, is a potentially promising field of research. On the other hand, preoperative hit scores explained 12% (R²) of the variance in postoperative hit scores, leaving a large proportion of variance unexplained. The predictive value was even weaker in patients without conscious recall, who constituted the main part of our study sample. Further, the difference between patients with and without conscious recall is unfortunately left unexplained by this study. Beside awareness during anesthesia, no other indicator discriminated between the groups.

In summary, it appears that decisions to use or avoid words presented during anesthesia in a postoperative WS test can be made without conscious recall or recognition of presented words. This corroborates a previous finding of unconscious-controlled memory for words presented at consistently light to moderate sedation as measured by BIS 5and extends the evidence of weak explicit memory function to a near-adequate level of hypnotic state. Our results are consistent with the hypothesis that memory performance depends on hypnotic state during word presentation and suggest unconscious influences of awareness during anesthesia. Memory function may be predicted by stable patient characteristics such as memory function before surgery. Finally, awareness seems an important predictor of conscious recall, but the relation between these two phenomena is not straightforward.

The authors thank the patients and staff from the ambulatory surgical unit (Erasmus MC, The Netherlands) for their cooperation; Aspect Medical Systems (Newton, MA) for providing the EEG equipment; Dick J. Bierman, Ph.D. (Department of Psychology, University of Amsterdam, The Netherlands), for the memory test software; and Ian F. Russell, M.D. (Department of Anaesthesia, Hull Royal Infirmary, Kingston upon Hull, UK), and Professor Peter S. Sebel, Ph.D. (Department of Anesthesiology, Emory University School of Medicine, Atlanta, GA), for their advice in designing this study.