Implicit Memory for Words Played during Isoflurane- or Propofol-based Anesthesia: The Lexical Decision Task

With approval from the local institutional review board (Medical School of Hanover, Hanover, Germany) and written informed consent of each participant, 64 patients undergoing lumbar disk surgery and 10 controls without anesthesia were enrolled in the study between December 1998 and April 2000 (table 1). The data from four patients were excluded from the results because the data were not correctly saved 3or the patient refused to participate in postoperative testing. 1Therefore, the results are based on 30 patients in the isoflurane group (isoflurane was the main anesthetic agent), 30 patients in the propofol group (propofol was the main anesthetic agent), and 10 controls. On the day before surgery, the patients were informed that an audio tape with word lists would be played to them during the operation and that they would undergo a short interview and a psychologic test several hours after the operation. Patients older than 45 and with a native language other than German were excluded, as were those with neurologic diseases, memory dysfunction, psychiatric diseases, head trauma, hearing impairment, or psychoactive medication. A control group consisted of 10 volunteer medical students who listened to the tapes without anesthesia and completed the implicit and explicit memory tests after a comparable delay.

Two questions were asked by the current study 1: is there a priming effect for material presented during anesthesia using the lexical decision task? 2and is there a difference between isoflurane- and propofol-based anesthesia techniques? For the first question, each group (anesthesia technique) could be viewed as an independent study with a within group comparison of two means (primed and unprimed words). Repetition priming studies the lexical decision task have yielded a size of the priming effect between 50 and 100 ms with reaction times of approximately 600 ms (SD 80 ms). 16Based on these data, the sample size was calculated as 18 (power 80%). #

For the second question, sample size estimation was based on an anticipated SD of the size of the priming effect of 50 ms and a difference of the priming effect between the two groups of 40 ms. This was done because smaller differences in priming effects between the two anesthetic regimens would not have been of clinical interest. This yielded an estimated sample size of 26 (power of 80%, α= 5%). 17 

Anesthesia was administered by two of the authors (M. S. and E. L.). Patients were randomly assigned to one of two anesthetic regimens: isoflurane–alfentanil–nitrous oxide or propofol–alfentanil–nitrous oxide. All patients were premedicated with 7.5 mg oral midazolam 0.5–1 h before anesthesia.

After administration of 0.5 mg intravenous atropine, anesthesia was induced with 2–2.5 mg/kg propofol and 10 μg · kg⁻¹· h⁻¹alfentanil. After the patients lost consciousness, they underwent ventilation with 100% oxygen, and 0.5 mg/kg atracurium was administered to facilitate tracheal intubation. Immediately after induction, an alfentanil infusion of 30 μg · kg⁻¹· h⁻¹was started, and isoflurane up to 1.5% was added to oxygen. Anesthesia was maintained with isoflurane at an end-expiratory concentration of 1.06 ± 0.3% combined with 60% nitrous oxide. The inhaled anesthetics and alfentanil infusion were kept constant until the end of surgery. Additional propofol (0.5–1 mg/kg) was administered whenever heart rate or arterial blood pressure levels increased by more than 10% of preoperative values. Atracurium (5–10 mg) was administered when more than two twitches of the train-of-four were present. Electric activity of the heart, noninvasive blood pressure, ventilation, end expiratory concentrations of carbon dioxide and inhaled anesthetics, peripheral oxygen saturation, and neuromuscular transmission were continuously monitored.

The anesthesia technique differed from the previously described only in that instead of boluses of propofol and isoflurane, target-controlled infusion (3500 TCI Graseby; Graseby Medizintechnik GmbH, Kirchseeon, Germany) of propofol was used for induction and maintenance of anesthesia. The initial target concentration during the intubation was 10 μg/ml and was reduced to a target concentration of approximately 3.3 μg/ml during the tape presentation (table 1).

Two audio tapes, each consisting of a list of 30 different low-frequency German nouns (all < 50 occurrences per 1,000,000, mean frequency 8.9/1,000,000, frequency matched between lists) 18were prepared. One half of the participants from each group (15 each from the isoflurane and propofol groups and 5 from the control group) were exposed to the first list during the study phase, and the other half listened to the second tape. Presentation was started at the time of skin incision. The same format was used for each control participant without anesthesia. The words were repeated in varying sequences 10 times during a 15-min period at a rate of one word every 3 s by a male voice. The control participants were asked to count how many times they heard a certain word to ensure that they were listening and to prevent them from rehearsing the words. The investigator (M. M.) who conducted the postanesthetic memory tests and interviews was blinded to the tape contents.

Implicit and explicit memory for the words and explicit memory for intraoperative events and dreams was assessed 6–8 h after exposure to the tapes. Implicit memory was tested first using a lexical decision paradigm. Subjects were asked to decide as quickly as possible whether the stimulus presented was a real word or a nonword. It has been shown numerous times that previous exposure to test stimuli speeds up this decision, leading to shorter response latencies of primed words relative to unprimed words. A practice trial using 20 items that were not included in the experimental tapes was performed before the test session.

For the test session, words from both tapes plus 60 nonwords were spoken by the same male voice that had been used for the study tapes. These words were AD converted at a rate of 22,700 Hz, digitally edited to determine word onset for trigger purposes, and stored on magnetic disks. Primed and unprimed words and nonwords were presented in random order by a special purpose computer program at an interstimulus interval of 2,700 ms (stimulus duration 500–800 ms). The participants were instructed to listen to the stimuli and to press a button held in the right hand whenever a stimulus qualified as a legal German word, and to press a button held in the left hand whenever a stimulus was not a German word. The reaction times for the different events were recorded by the computer from the onset of the stimulus to the patient's response.

Explicit memory for intraoperative events was assessed using a structured interview pertaining to (1) the final memories before falling asleep, (2) the initial memory on waking, and (3) to dreams or other experiences during surgery. Finally, a recognition test for the word lists was administered. Patients and control participants were given a page containing a random list of the 30 primed and the 30 nonprimed words. The participants were asked to indicate which words had been played during the test phase and how certain they were about their decision (yes–certain, yes–uncertain, no–certain, no–uncertain).

Mean reaction times for all correctly classified primed words and all unprimed words were determined for each subject separately and entered into statistical analysis. The statistical comparisons were done by analysis of variance with group (isoflurane vs.  propofol) as a between-subjects factor and repetition (primed vs.  unprimed words) as a within-subjects factor. Because there was concern that the postanesthetic state in the patient group might have led to an increased number of very long reaction times, therefore somewhat corrupting the utility of the individual mean reaction times, several additional analyses were conducted on (1) the median reaction times, (2) the mean reaction time after exclusion of outliers in the individual data sets using the Grubbs test, **and (3) the mean reaction time after rejection of all values exceeding 2.5 SDs over the individual mean. Finally, error rates were determined and expressed as percentages.

The principal question was whether response latencies would be affected by previous exposure to the words. The results are presented in table 2. It can be seen that the lexical decision latencies for the control group showed faster reactions to primed words than to unprimed words (t  test, P < 0.0001). Although patients of both study groups were slightly faster for the primed words, the main effect of repetition just failed to reach the significance level (analysis of variance, F^1,59= 3.89;P = 0.0537) when the individual subjects’ means were used. Post hoc  comparisons indicated a marginally significant effect of repetition in the isoflurane group (P < 0.09), whereas the effect in the propofol group was not significant (P > 0.3). There was no group by word type interaction (F^1,59= 1.03). Faster reaction times for primed words were found in 22 of 30 patients of the isoflurane group, 18 of 30 patients of the propofol group, and in all 10 control subjects.

For the analyses on the medians and means (with outliers removed), the results were similar in that the effect of repetition was marginally significant (median F^1,59= 3.61;P = 0.062; outlier removed F^1,59= 4.05;P = 0.048 ††). For the analysis on the means (after rejection of all values above the 2.5-SD limit, table 3), the effect of repetition was significant (F^1,59= 4.66;P = 0.035). No group by word type interaction was present for this measure (F^1,59= 1.33). Again, post hoc  tests indicated a marginally significant effect of repetition in the isoflurane (P < 0.06) but not in the propofol group (P > 0.3). For this measure, the effect sizes 19were 0.28 for the isoflurane group (power 0.44) and 0.07 for the propofol group (power 0.10). An analysis of variance on the errors did not reveal any significant effects of repetition or anesthetic technique (all F^1,59< 1.5).

None of the participants reported conscious recollection of events during surgery. Two patients of the propofol group had had a pleasant dream that was not related to any intraoperative events. One patient reported that she had dreamt of having her wound drainage removed, which actually was a recall of an immediate postoperative event. In the word recognition test, half of each study group indicated that they were sure about not having heard any words during anesthesia. Another half indicated that they probably had heard some words. However, all subjects were at chance level with regard to the recognition decision (49.3 ± 3.9% of words were classified correctly), regardless of how sure the patients were that they had or had not heard those words. In the group of control subjects, however, the results were different, with 82.8% (SD 5.5) of the words classified correctly in the recognition test.

Because some theorists posit that recognition memory can be influenced by both explicit and implicit memory processes, 20we examined the question of whether forced choice recognition judgments and individual priming effects were correlated. This analysis was performed for the isoflurane group only because this group showed at least a marginal priming effect. It turned out that there was no correlation between the recognition performance (at chance for the entire group) and priming (Pearson r =−0.2026, P = 0.28).

We observed a robust reaction time effect in our nonanesthetized control group and a tendency toward quicker reaction times for intraoperatively presented words in the patients. Because the patients had no recall of the intraoperative period and the recognition of the words was at chance level, the finding should be interpreted in terms of a small effect of implicit memory. This dissociation between explicit and implicit memory after surgery and anesthesia was not observed in the awake controls, who performed above chance level in the word recognition test. The slow reaction times of the patients (compared with the controls) suggest that both patient groups experienced a general cognitive slowing due to the anesthesia. In this situation, lapses of attention might lead to lapses of attention in the trials with very long reaction times. Consequently, any priming effect, if present, might be masked. In the literature, different methods to deal with such skewed reaction time distributions have been suggested. 21Simulation studies and studies with real data, however, have not resolved the question as to which of these methods is superior. Therefore, we conducted several additional analyses with different methods of outlier correction. All of these analyses suggested the same thing, i.e. , the effect of repetition (implicit memory) was significant at approximately the 5% level. The inspection of the means suggest that this main effect of repetition was mainly attributable to the isoflurane group. The group by repetition interaction, however, was not significant in any of the analyses. The effect size (Cohen d) 19in the isoflurane group was 0.28 and thus within the same range found by Merikle and Daneman 6in their meta-analysis of implicit priming studies and anesthesia. These authors determined that effect sizes decrease systematically as the interval between surgery and test increases. For the first 12 h after surgery, an average effect size of 0.22 was found in the meta-analysis.

The reason to adopt the lexical decision task was its principal ability to quantify even weak memory effects in terms of millisecond changes in decision times. It has been shown to be a sensitive index of implicit memory in both normal subjects and amnesic persons. 10–15To our knowledge, it has not been previously used in anesthetized patients. Methodologic aspects, such as the type of verbal information and anesthetic aspects, which may account for the reduced memory performance in our experimental groups, are discussed in the following section.

In the current study, the lexical decision task was used in the same format as in most studies of amnesic and normal persons. The same study test modality, low-frequency words, and several repetitions of the study words during anesthesia were obtained because these factors have been shown to enhance the performance on test session. However, a review of the implicit memory studies conducted in anesthetized patients reveals that these factors may differ during anesthesia conditions. For example, the type of words used in this study may have reduced the memory effect. A robust implicit memory effect has been demonstrated with the word stem completion task if familiar words were used as stimuli. 22,23The effect was small or no more detectable if unfamiliar words were presented intraoperatively. 24,28The same pattern of results has been observed in studies using the category generation task. 24–27It seems that during anesthesia conditions, a familiar and meaningful item leaves a stronger and more accessible trace in memory than a less familiar one. The results of the current study are in line with this suggestion. However, this observation in anesthetized patients contrasts sharply with the findings with healthy participants and with amnesic patients, in which priming in lexical decision is generally larger for less familiar words than for familiar words. 14 

Another feature of the implicit memory that seems to differ during anesthesia conditions is the durability of the memory trace. For example, smaller word priming effects were found if the implicit memory tests were applied the day after surgery or even later. 24,28In studies of amnesic and normal controls, the durability of the memory trace has been shown to be better for familiar words than for unfamiliar words. 13It is unclear whether this is the case in anesthetized patients.

A further source of discrepancy is the fact that there are less experiments in auditory modality than in visual modality in the amnesic literature, and those results may be valid only for the particular task and study conditions. In visual modality, the priming effects in the lexical decision task have been shown up to 2 days after the first stimulus presentation. However, this finding may not account for auditory modality. 29 

The results of the current study are somewhat surprising if compared to our previous results. Using a postoperative reading speed task, we found an implicit memory effect for stories presented during propofol–alfentanil–nitrous oxide anesthesia but not in a subsequent study during isoflurane–alfentanil–nitrous oxide anesthesia. 8,9In the current study, we did not find a clear differential effect of the two anesthetic regimens on the lexical decision task. However, a stronger tendency toward an enhanced implicit memory performance was detected during isoflurane-based anesthesia. The lexical decision task, unlike the reading speed task, does not require information processing at a high cognitive level. Therefore, we predicted that using experimental conditions similar to those in the reading speed experiments, we should find a word priming effect at least in patients to whom propofol-based anesthesia was administered. Given our current results, the question arises of why the same anesthetic regimen abolished implicit memory effects in one task but not in the other. One possible explanation is that these two tasks tap different sources of implicit memory and are therefore not directly comparable. On the lexical decision task, perceptual features, such as phoneme structure, voice of the speaker, and so forth, facilitate processing on reexposure. The memory process in this type of task is called perceptual priming. The stories tested by reading speed task in our previous study required another type of memory processing, called conceptual priming. The term implies that for a reader to benefit from previous exposure to such stories, processing at a postperceptual, conceptual level is required. 30These two levels of processing—perceptual versus  conceptual—may affect the durability and strength of the memory trace. 13Memory traces that include the stimulus and its context may be stronger than those of the stimulus alone. 31For example, associations for stories presented during general anesthesia have been demonstrated 24 h to 5 days after surgery. 32,33 

The available literature suggests that single words can be processed under relatively high end-expiratory concentration (1.0 ± 0.5%) of isoflurane. 23If surgical stimulus is lacking, lower concentrations (up to 0.8%) are sufficient to suppress implicit memory. 34The isoflurane group of our study received an end-expiratory concentration of 1.06 (± 0.35)%. Propofol has been used less frequently in studies of implicit memory. Most of these studies, all using relatively low infusion rates, have reported positive implicit memory effects. 35–37In our previous study, implicit memory was found under propofol concentrations of 2.78 μg/ml, ‡‡whereas in the current study, the concentration was 3.3 (± 0.4) μg/ml. It might be that the plasma concentrations of the current study were better controlled because individual target controlled infusions were used. These concentrations of isoflurane and propofol may attenuate the strength of the memory trace to a degree that is barely detectable by the lexical decision task. In all three experiments, patients received oral premedication of 7.5 mg midazolam, which might in addition suppress implicit memory. However, the current literature implies that midazolam, even if exercising an amnesic effect on explicit memory, has a minor effect on implicit memory. 38 

Because many studies suggest that learning may occur in deeply anesthetized patients, we should determine which kind of implicit memory processing this might be. Because the performance in memory test depends greatly on the strength and durability of the memory trace left in memory by the first presentation, and the memory trace left during general anesthesia would presumably be weak, the indirect tests of memory used in amnesic studies need further validation and modification for anesthesia conditions.

The authors thank Jobst Kilian, Technician, Department of Neurology, Medical School of Hanover, Hanover, Germany, for his technical support.