We examined differences in the rate of learning and intertrial performance on an item memory test in contrast with an associative memory test between people with aMCI and healthy controls and the ability of these tests to distinguish between aMCI and normal aging.
Rate of learning
People with aMCI obtained significantly lower scores in terms of memory span (first trial) and rate of learning on both tests. These results are in line with those of previous studies [
14,
24]. As pointed out in previous studies [
12,
22], however, the interpretation of this deficit in learning rate needs to be conducted with caution. We cannot rule out the possibility that the interaction was due to scale effects [
35,
36] since there were clear differences in score in Trial 1 of both tests. In the AVLT, fortunately, although the performance of aMCI in Trial 2 was similar to the performance of NC in Trial 1, the learning curve of aMCI from trials 2–3 was still shallower than the curve of NC from trials 1–2, which provided evidence to some extent that people with aMCI learn slower than healthy controls.
The distinction between memory span and learning rate may be regarded as reflecting two different aspects of the learning process. For instance, subjective memory in multiple sclerosis patients is associated with initial-trial learning performance (i.e., memory span) but unrelated to recall performance on subsequent learning trials or aggregate learning scores [
37], while patients with Asperger’s syndrome exhibited impairment only in the later trials but not in the initial trial of a multi-trial recall [
38]. In the present study, the non-significant correlations between memory span and learning rate supports the notion that they reflect two different aspects of the learning process. In addition, although both memory span and learning rate are impaired in aMCI patients, there is no correlation between them in either test, indicating that memory span and learning rate are impaired independently.
Notably, in the current study, we found that people with aMCI had a slower learning rate for associative memory than for item memory, which provided additional evidence that associative memory was more impaired than item memory in aMCI, as what has been reported by investigation of memory span and total score in multiple learning trials. The results did not change even when the digit span forward, digit span backward, verbal fluency, and block design test were controlled as covariates, which suggested that the more severe impairment of learning rate in associative memory in aMCI patient is not because of their lower performance in executive functions.
Learning increment in multiple repeated trials reflect the ability of organizing information efficiently and of the development of higher-order memory units [
8]. Subjective organization refers to a person’s ability to associate seemingly unrelated events in memory [
39]. The ability of subjective organization is predictive of performance of multiple trial recall [
40]. Therefore, the slower learning rate in aMCI patients indicates their impairment in organizing the repeated material into appropriate higher-order memory units, resulting in a failure to facilitate, or even in retardation, in retrieval processing. Furthermore, as revealed in this study, this organization of repeated materials seems to be even more impaired for item-item bindings than for separate items in aMCI patients. This is in line with the idea that the ability to bind items relies heavily on the hippocampus and entorhinal cortex [
13], which are affected the earliest in aMCI [
41‐
43]. Note, however, that aMCI patients’ performance on memory tests increased statistically with increasing trials; thus, the deficit in learning rate in aMCI patients is not as serious as that in AD patients [
12].
Intertrial acquisition and consolidation
Our finding that GA and LA were uncorrelated supports that they are relatively independent processes [
12,
22]. We separated the rate of learning into intertrial GA and LA in order to investigate the separate contributions of failure in acquisition and failure in consolidation to the shallower learning curve in aMCI patients. We found that aMCI patients exhibited significant damage in both acquisition and consolidation between adjacent trials in both the PALT and AVLT. The finding suggested that the substantial decline in GA and more rapid forgetting underlie aMCI’s shallower learning curve not only for item learning test as previously studies reported, but also for associative learning test. Previous studies have found that this deficit in intertrial acquisition and consolidation in aMCI patients is similar to that found in AD patients, although the extent of deficits was not as extreme compared to in AD [
12,
22]. Therefore, our results, together with those of previous studies, contribute to a deeper understanding of memory function impairment in aMCI and indicate areas that can be targeted for intervention.
Discriminative power for aMCI
Both ROC curve and logistical regression analysis revealed that the third learning trial of both tests were able to differentiate aMCI from normal aging better than the first trials. Our results extend the results of previous studies, which found that the discriminative power of item list learning tests increases with increasing number of trials (>3 trials) [
4,
9], by showing the same pattern for item-item association list learning tests (even < 3 trial). Our results also correspond with Petersen et al.’s observation that the inability to acquire information in MCI becomes more apparent over several learning trials [
44]. This phenomenon is related to the difference in learning rate between aMCI patients and healthy controls, which becomes larger as learning progresses, resulting in the higher discriminative power in the later trials than in the former trials. Not surprisingly, the total scores (i.e., across all trials) on both the AVLT and PALT were able to differentiate the two groups better than the scores of individual trials because the total score accumulated all the group differences in each trial.
We found that the discriminative power of the associative memory test was greater than for the item memory test in Trial 1, Trial 3, and total score, confirming the results of Troyer et al. (2008) [
14] that associative memory test had advantage for identifying aMCI from normal aging in comparison to item memory. More importantly, beyond that, the present study provided evidence that the differences in discriminative power for aMCI between the PALT and AVLT tended to be larger in later trials than in earlier trials. The pattern was consistent when using both the ROC curve and the logistical regression analysis. Thus, taken together, these results support our hypothesis that associative memory tasks tended to be better for discriminating aMCI from normal aging with the number of trials increasing. Therefore, associative memory tasks with multiple learning-test trials may be particularly useful instruments for the detection of potential aMCI individuals.
The early detection of aMCI is very important because intervention might be most valuable at the earliest stages for those who are either recently diagnosed or at risk for developing it [
45‐
47]. To aid the early detection of aMCI, clinicians and researchers are increasingly relying on neuropsychological assessment, genetic testing, and the use of neuroimaging [
48]. However, genetic and neuroimaging methods are not appropriate for larger-scale screening in community-dwelling elderly because of their high cost and inconvenience. Therefore, for people with aMCI who are characterized by memory impairment, neuropsychological assessments with high discriminative power for screening or detecting memory decline at a very early stage would be more practical and helpful. However, the advantage of associative memory and multiple-trial learning tests suggested in previous studies [
4,
14] were seemingly ignored. The present study indicates further that an associative memory test with multiple learning-test trials is more powerful in amplifying the memory difference between aMCI and normal elderly. Therefore, an associative memory test with multiple learning-test trials would be a very useful instrument for the detection of potential “early aMCI,” a condition wherein elderly individuals’ memory performance is between the normal stage and conventional aMCI [
45]. Further research is needed to confirm and develop this finding.
Limitations
There are several limitations in the present study, especially regarding the assumption of the homogeneity of the PALT and AVLT that should be acknowledged. Firstly, the AVLT is a free recall test. In the PALT, however, participants were required to recall the second word in the pair only after the first word was given. Thus, the recall in the PALT resembles a cued recall rather than a free recall. On the other hand, there is some advantage to this “half-free” recall in the PALT. As performance on a free recall associative memory test is somewhat dependent on item scores (participants would be unable to report the associations if they could not recall the items) [
14], the “half-free” recall in PALT reduces this effect by providing the first word in each associative pair. Secondly, the number of items or associations was not equal in these two tests. Previous studies have argued that the sensitivity of word list learning tests may be enhanced by increasing the number of items to be remembered (>10 items) [
4,
9]. The current study, however, revealed that the PALT, which has 12 items (association), had more discriminative power than the AVLT, which has 15 items, regardless of how the scores were analyzed. This finding provides further evidence for the idea that associative memory tests may be particularly sensitive to aMCI compared to item memory tests [
14]. Thirdly, the presenting order of the two tests was not balanced, as the same CRF (Case Report Form) was applied to all participants. The PALT was always presented first, followed by a series of neuropsychological tests (all non-memory), and then the AVLT. But if the AVLT was influenced by the PALT due to proactive inhibition effect, then the impairments in PALT would have been underestimated. Even this, we still found more deficits in PALT than in AVLT. Therefore, we think the presentation order of the tests would not have led to negative effects for our findings.
Though all of the above factors differed between the two tests, they all favored PALT. Thus, they would not affect our findings of more impairment in PALT in principle. But we have to acknowledge the following two factors which may have confounded our findings. One concern is there were time restrictions on recall during the associative task that were not present in the item task; the other is that stimuli on the item task were presented in the same order on each trial, whereas they were presented in a different order on each trial of the associative task. As it would be easier to learn information over repeated trials when stimuli are presented in the same order each time, and easier to recall learned items when there are no time restrictions, these two factors may have confounded our conclusions to some extent.
In addition, although there is no significant difference in word frequency and age of acquisition between the two tests, Almond et al. (2013) [
20] found that high-frequency words are learned at a higher rate than low-frequency words in older adults compared to younger adults, which indicates that the memory deficits may be modulated by word frequency. However, the number of words used in the present study is too limited to be separated into high and low frequency words for further statistical analysis. Future research on the learning rate of aMCI with a consideration of word frequency is needed.
Lastly, as mentioned in the Methods section, aMCI patients were not able to further classify into single- or multi-domains due to large sample based norms unavailable. Thus, we are unsure whether the results would differ between single- and multi-domain aMCI subtypes.