INTRODUCTION
Amnestic mild cognitive impairment (aMCI) is characterized by an isolated memory decline in the absence of other cognitive or functional impairments, and is a high-risk factor for Alzheimer's disease (AD; Petersen, 2004). Not surprisingly, neuropathological changes in aMCI (Masdeu et al., 2005) predominate in the hippocampus and related structures; there are secondary changes in the anterior cingulate cortex, a brain structure with strong reciprocal connections to prefrontal regions. Consequently, one might expect that individuals with aMCI would have particular difficulty on memory tests with high executive demands. We examined whether prospective memory (PM) is impaired in aMCI, because PM tasks make considerable executive demands.
PM refers to the ability to remember to carry out future intended actions, such as meeting a friend for lunch at noon or stopping at the store on the way home. PM is believed to require executive ability because it involves self-initiation of a response, inhibition of the ongoing task, and switching to another action. Indeed, clinical studies (e.g., Cockburn, 1995) and neuroimaging studies (e.g., West & Ross-Munroe, 2002) show that frontal brain structures are critically involved in PM.
PM can be sub-classified as time based (e.g., requiring a response at a specific time of day) or event based (i.e., requiring a response following a specific cue; Einstein et al., 1995). Time-based tasks are believed to require more self-initiated retrieval than event-based tasks because no external cues are present (Einstein et al., 1995). Time-based tasks also require greater attention switching and inhibition: they are more likely to require interruption of an on-going task with an incompatible response (e.g., stop eating lunch to make a telephone call at 12:30), whereas event-based tasks are more likely to be compatible with the on-going task (e.g., take a medication with lunch). Consistent with this idea, a single-case study of a patient with bilateral frontal lesions showed more errors on time- than event-based PM tasks (Cockburn, 1995).
There is only one previously published study of PM in aMCI (Kazui et al., 2005), and this showed impaired performance on two single-trial event-based tasks. Event-based PM tasks are also sensitive to cognitively normal individuals who later develop AD (Jones et al., 2006), and to individuals whose apolipoprotein E genotype puts them at high risk for AD (Driscoll et al., 2005). Time-based PM has not yet been examined in populations at risk for developing AD. In individuals diagnosed with AD, both time- and event-based PM is impaired (e.g., Huppert et al., 2000; Maylor et al., 2002).
We developed time-based and event-based tasks that would be appropriate for aMCI. We minimized demands on retrospective memory (by utilizing the same behavioural response on each trial) so that participants would be able to hold the intention in memory long enough to carry it out. Our tasks required natural responses in order to increase face validity and consisted of multiple trials in order to increase reliability. Our primary hypothesis was that time-based PM would be more sensitive to aMCI than event-based PM, because the former makes more executive demands and because of early neuroanatomical changes in frontal structures in aMCI.
METHODS
Participants
Participants were 42 healthy older adults, 45 individuals with aMCI, and 24 individuals diagnosed with AD (see Table 1). All were screened by interview and questionnaires for health factors affecting cognition, including medical and psychiatric disorders, mood (Yesavage et al., 1983), medications affecting cognition, and substance abuse. Methods were approved by the local research ethics board.
Demographic and descriptive variables
Healthy older adults were recruited from the community. These participants did not have a memory complaint, and all scored within the normal range for their age on tests of general cognition (i.e., Mini-Mental State Examination, MMSE; Folstein et al., 2000) and memory (i.e., Hopkins Verbal Learning Test-Revised, HVLT; Brandt & Benedict, 2001; Brief Visuospatial Memory Test-Revised, BVMT; Benedict, 1997).
Individuals with aMCI were recruited from physician referrals (n = 36) and newspaper advertisements (n = 9). aMCI classification criteria (Petersen, 2004) were determined by consensus of two neuropsychologists. To determine presence of a new memory complaint and normal daily activities, we (i.e., AT or KM) interviewed the participants and, whenever possible, family members. Evidence of an objective memory impairment was obtained using HVLT and BVMT. Consistent with Petersen's recommendations, memory impairment was indicated by memory scores judged to be lower than expected for age, education, and intellectual function. Each individual performed below expected levels on at least one of the memory tests. On average, memory scores in this group were 1 to 1.5 standard deviations below age norms (Table 1). Normal general cognitive functioning was confirmed with the MMSE (Folstein et al., 2000), Digit Span (Wechsler, 1997), Boston Naming Test (Kaplan et al., 1983), Rey-Osterreith Complex Figure copy, and Trail Making Test (Spreen & Strauss, 1998). These scores were well within the normal range (Table 1). Lack of dementia was determined by considering all of the previous criteria and hinged on the degree of functional independence.
Individuals diagnosed with AD were recruited from physician referrals and diagnosed using criteria for probable AD (McKhann et al., 1984). Patients diagnosed with AD showed a mild level of dementia (i.e., MMSE scores ranged from 23 to 29), and memory was mildly impaired (i.e., 1.5–2 standard deviations below the mean for their age; Table 1).
There were no significant group differences in age, F(2,105) = 2.13, p = .12, education, F(2,105) = 1.58, p = .21, or sex, χ2(2, N=111) = 2.38, p = .30. There were significant group differences on MMSE scores, F(2,105) = 31.67, p < .001, which were higher in the control than aMCI group, p = .02, and higher in the aMCI than AD group, p < .001. Expressive vocabulary showed significant group differences, F(2,105) = 10.17, p < .001, with the AD group scoring lower than the other groups, p < .001. Because all groups were equivalent in years of education, this likely reflects the effects of AD on expressive language rather than pre-existing group differences in intelligence.
Procedures
Participants underwent a two-hour evaluation including a clinical interview, neuropsychological assessment, and administration of PM tests. PM trials were administered throughout the evaluation, and participants engaged in neuropsychological testing between trials. Detailed administration and scoring instructions are provided in a manual available from the authors and are briefly summarized here.
On the time-based (PM-time) task, participants were asked to monitor and report the time to the examiner every 30 minutes (e.g., 10:30, 11:00, 11:30, etc.). A wall clock was visible at all times. On the event-based (PM-event) task, participants were instructed to use a pen of a particular color whenever they engaged in tasks requiring a writing instrument. The target pen was placed in a pencil holder along with several other writing instruments, and a distractor pen was placed on the table directly below the pencil holder.
Each PM task consisted of four trials. Instructions for each trial were provided 20 to 30 minutes before each target time or event and were repeated as necessary using spaced repetitions until the participant could state the instructions following a 20-second delay. To score PM-time, two points were awarded for times reported within two minutes of the target time, and one point was awarded for times reported 3 to 10 minutes from the target time. On PM-event, two points were awarded for picking up the target pen before commencing the writing task, and one point was awarded for picking up the target pen after beginning to write but before the end of the writing task. If a time was not reported within 10 minutes or the target pen was not used by the end of the writing task, no points were awarded. On these trials, the examiner immediately asked questions to determine whether the instructions were remembered. Scores on the PM-time and PM-event tasks each ranged from 0 to 8.
Summed PM scores within the patient groups (n = 69) were moderately related to retrospective memory (HVLT and BVMT immediate recall, r's = .48 and .34) and executive ability (Trails B, r = −.47), all p's < .005. Several scores were negatively skewed (i.e., both PM scores in the control group and PM-event scores in the aMCI group), and group variances were unequal (i.e., lower in the control group). Thus, nonparametric statistics (i.e., Kruskal-Wallis χ2, Mann-Whitney U) were used to test the absence or presence (rather than the magnitude) of any differences. We used one-tailed Bonferroni-corrected post-hoc tests to compare the aMCI group with the other groups. Our primary analysis used Wilcoxon Signed-Ranks tests to compare the number of participants in each group that obtained lower time than event scores versus higher time than event scores. This analysis helped counteract ceiling effects, because participants with equivalent scores (because of ceiling effects on both tasks or equivalent off-ceiling scores on both tasks) are considered “ties” and are not analyzed. Because one cannot determine effect sizes or confidence intervals for nonparametric statistics, we present a number of summary statistics to provide substantive information. To confirm our findings, we examined the empirical distribution function using the Kaplan-Meier estimate to allow observations falling at ceiling to provide restricted—or censored—information. This allows examination of the score distribution up to the ceiling, and scores at ceiling are considered a minimum estimate of the actual score. A log-rank chi-square tested the equality of PM-score distributions for the different groups.
RESULTS
PM-time score means (and SDs) for the control, aMCI, and AD groups were 7.3 (1.0), 5.2 (2.5), and 3.7 (2.9), respectively. Mean ranks were 77.3, 48.3, and 33.2, respectively, χ2(2, N=111) = 34.33, p < .001. The control group ranked higher than the aMCI group, U = 432, z = −4.51, p < .001; the aMCI group ranked higher than the AD group, U = 371, z = −2.15, p = .016. Similarly, PM-event task score means (and SDs) were 7.4 (1.0), 6.1 (2.5), and 3.5 (3.5), respectively. Mean ranks were 68.4, 56.4, and 33.4, respectively, χ2(2, N=111) = 20.93, p < .001. The control group ranked higher than the aMCI group, U = 734, z = −2.01, p = .022; the aMCI group ranked higher than the AD group, U = 310, z = −3.05, p = .001.
Regarding relative performance on the two PM tasks, most participants in each group were off ceiling (i.e., 60% of the control group, 87% of the aMCI group, 89% of the AD group) and difference scores were thus not forced to zero. Figure 1 shows these data for participants who obtained total scores less than 16. The number of control participants obtaining numerically lower time than event scores (n = 10) versus higher time than event scores (n = 9) was not significantly different, z = −0.46, p = .65. In contrast, more aMCI participants obtained lower time than event scores (n = 25) versus higher time than event scores (n = 10), z = −2.16, p = .017. In the AD group, these numbers (n's = 8) were not different, z = −0.26, p = .80. The incidental exclusion of ceiling-level performers from the analysis did not result in inclusion of only lower-functioning controls: off-ceiling participants did not obtain lower scores than on-ceiling participants on any age-corrected cognitive variable shown in Table 1, all t's < 1.6, p's > .11.)
Number of participants in each group obtaining numerically higher time- than event-based prospective memory scores, equivalent time and event scores, and lower time than event scores. Only data from participants off ceiling on the combined tasks are shown here. Control n = 25/42; amnestic mild cognitive impairment (aMCI) n = 39/45; Alzheimer's disease (AD) n = 23/24.
These findings were supported by empirical distribution analyses that forced ceiling-level scores to provide restricted information. PM-time score distributions differed between groups, χ2(2, N = 111) = 33.6, p < .001; scores were significantly higher in the control than aMCI groups, χ2(1, N = 87) = 18.8, p < .001, and higher in the aMCI than AD groups, χ2(1, N = 69) = 4.0, p = .046. PM-event score distributions also differed between groups, χ2(2, N=111) = 19.6, p < .001. Scores did not differ between the control and aMCI groups, χ2(1, N = 87) = 2.7, p = .10, but were higher in the aMCI than AD groups, χ2(1, N = 35) = 7.7, p = .005. Thus, aMCI was impaired relative to controls on PM-time but not PM-event.
DISCUSSION
Our primary finding was a pattern of relatively poorer time-based than event-based PM in aMCI, with converging evidence from nonparametric and empirical-distribution analyses. Because time-based PM makes higher executive demands, this pattern likely reflects subtle involvement of the frontal system that exacerbates hippocampal-related memory problems in aMCI. By its nature, time-based PM requires substantial self-initiation, including time-monitoring and internal cueing, to perform the response at the correct time. Time-based PM also requires more extensive attention switching and inhibition: performing our time-based PM task was likely less compatible with the ongoing task (e.g., reporting the time while listening to instructions for a vocabulary test), whereas performing our event-based task was more compatible with the ongoing task (e.g., picking up the pen while listening to instructions for a writing task). The AD group showed equivalent time/event scores, with significantly lower scores on both tasks in comparison to controls. This suggests that the more extensive AD-associated neuropathology in medial-temporal and frontal regions produces more global memory problems, interfering with tasks making even relatively low executive demands.
An alternate interpretation of the difference in time- and event-based PM scores in aMCI is the simple effect of difficulty. That is, perhaps the time-based task is generally more difficult than the event-based task, and ceiling-level performance precluded obtaining this pattern in the controls. However, most of the control participants (i.e., 25/42) obtained off-ceiling total scores across tasks. Task-difference scores for these individuals were not forced to equal zero but were free to vary in either direction; nevertheless, an equal number of participants showed relatively more difficulty on the time-based task (n = 9) and on the event-based task (n = 10). The remaining off-ceiling control participants had equal difficulty on both tasks (n = 6). Thus, the two tasks were equally difficult for the control group, as well as the AD group (see Fig. 1), arguing against a simple difference in difficulty.
This study provides several contributions to the literature. Most of what is known about memory in aMCI focuses on retrospective memory. The single previous study of PM in aMCI (Kazui et al., 2005) examined only event-based PM. We have shown an additional impairment in time-based PM, and, importantly, we demonstrated greater time- than event-based impairments. This informs us about the nature of the memory deficit in aMCI (i.e., exacerbated when tasks require greater self-initiation, switching, and/or inhibition) and the likely brain regions involved (i.e., frontal-system in addition to medial-temporal-lobe structures).
Regarding clinical applications, there has been recent interest in developing PM tests that are valid for clinical assessment (e.g., Raskin & Buckheit, 2001). PM testing could provide information about practical memory ability for memory-impaired patients and their families. Given the large variability in PM performance by our aMCI group, our task may be a promising tool for predicting dementia onset, because PM difficulty may reflect a more advanced neurodegenerative process. Future research will help determine the clinical usefulness of our PM tasks.
ACKNOWLEDGMENTS
Supported by Alzheimer's Society of Canada and Desjardins Financial. The authors thank Malcolm Binns, Triti Namiranian, Brent Hayman-Abello, Larry Leach, and Morris Moscovitch.
