Performance change while learning novel cognitive tasks as a potential identifier of preclinical Alzheimer's disease

Abstract

Background

In healthy older adults (OA), the effects of amyloid-β (Aβ) deposition on cognitive functions involved in learning are unclear.

Objective

This study aimed to determine how age, practice, and neuropsychological test performance are associated with performance change during the learning of three cognitive tasks, and if Aβ deposition impacts performance change in OA.

Methods

Fifty-five OA and 28 young adults completed neuropsychological tests, and Aβ deposition was assessed in OA. Participants learned three cognitive tasks: stop-go normal task (SGNT), stop-go reverse task (SGRT), and n-back task (NBT). Performance change was analyzed as change in accuracy and reaction time from the first to second and the first to third practice trials using linear mixed effect models. The basic model included age group and performance change, with neuropsychological test covariates. The second basic model mimicked the first but included Aβ deposition, instead of age group.

Results

In the basic model, more practice resulted in a larger performance change for SGNT and SGRT, but not NBT. In the second basic model, after two NBT practice trials, performance change increased with greater amounts of Aβ deposition and worse information processing speed but, after three practice trials, decreased with greater amounts of Aβ deposition and worse information processing speed. Across the three tasks, greater Aβ deposition tended (non-significant trend) to be associated with smaller improvements after more practice.

Conclusions

These results suggest that the ability to learn a cognitive task is maintained with age but is negatively impacted by Aβ deposition.

Introduction

Alzheimer's disease (AD) is the fifth leading cause of death in Americans aged 65 and older. ¹ Dementia is typically diagnosed, and the AD etiology confirmed, after those with AD or their caregivers report the onset of symptoms, but significant amyloid-β (Aβ) deposition begins in the brain 15–20 years before the identification of these symptoms. ² The 15–20 years between the onset of elevated Aβ deposition and neurodegeneration and when patients or their caregivers notice the resultant functional declines, known as the preclinical period, would be the time in which implementing interventions could be the most effective in slowing the progression of the disease and thus increasing a patient's quality of life. ³ Currently, cerebrospinal fluid and imaging analyses, such as positron emission tomography scans (PET), to quantify Aβ deposition, are the only definitive ways to detect preclinical AD. ⁴ However, these methods are costly and not readily available in under-resourced communities. Although newer, more accessible methods of quantifying Aβ deposition and identifying risk of developing AD are being developed, the relationship between neuropathological changes, including Aβ deposition, and the subtle cognitive declines that occur in the preclinical period are not clearly understood. Therefore, there is a need for examining the relationship between the risk for developing AD and subtle cognitive declines in the preclinical period, so that the risk and subtle cognitive deficits can both be identified and earlier treatment to slow or prevent the progression of dementia due to AD can be implemented.

The assessment of individual cognitive domains using traditional neuropsychological tests is an accessible means of detecting subtle cognitive changes, but currently, no single administration of a traditional neuropsychological test is sensitive enough to distinguish typical age-related declines from more malevolent cognitive dysfunction associated with elevated risk of developing AD. Olsen et al. determined that a single administration of a neuropsychological testing battery that incorporated assessments of episodic memory, working memory, and executive function could differentiate older adults (OA) without cognitive impairment, OA with probable AD, and OA with mild cognitive impairment. ⁵ This study provides evidence that dysfunction of episodic memory, working memory, and/or executive function is present in early AD and distinct from dysfunction present in those with mild cognitive impairment (MCI). However, the neuropsychological tests used in the study could only assess probable AD and were not able to assess the risk of developing AD, or preclinical AD. Van Havre et al., developing a more sensitive assessment, provided evidence that performance on neuropsychological tests over a 54-month period could distinguish healthy OA from those with preclinical AD. ⁶ Specifically, this study provided evidence that integrating the composite scores of assessments of visuo-spatial function, attention, information processing speed, and visual memory was more sensitive to detection of risk of developing AD than performance on a single neuropsychological test. However, the significant length of time needed to establish this composite score is inaccessible in a clinical setting, and therefore, a method to quantify the subtle cognitive changes associated with an elevated risk of developing AD in a single visit is still needed. Whitson et al. developed a more sensitive single-visit assessment, determining that cognitive-motor dual-task testing, utilizing a test of executive function for the cognitive task while participants walked, is associated with an elevated risk of developing AD based on the presence of the ε4 allele of apolipoprotein E (APOE4). ⁷ However, although those with genetic risk are at an elevated risk for developing AD, some individuals develop AD without genetic risk and some individuals with genetic risk never develop AD. In contrast, Aβ deposition is present in all individuals diagnosed with dementia due to AD. ⁸ Therefore, relative to a cognitive measure correlate of APOE4, a cognitive measure correlate of Aβ deposition might be more specifically associated with preclinical AD and should still be identified and developed.

One such potential cognitive measure that may be associated with elevated risk of developing AD is the change in performance while learning and practicing a new cognitive task, which is a measure of episodic learning. Episodic learning involves three distinct yet related cognitive processes: episodic memory, the memory of personal experiences, ⁹ working memory, the short-term storage of memories that includes analyzing and processing the influx of sensory information and connecting this information to long-term memories, ⁹ and executive function, a group of cognitive processes that are needed for active concentration and attention. ¹⁰ Although dysfunction of episodic memory, working memory, and executive function have been identified to negatively impact the acquisition and encoding stages of learning and as preclinical characteristics of AD, ¹¹ distinct tests for episodic memory, working memory, and executive dysfunction are not sensitive enough to detect subtle deficits associated with elevated risk of developing AD. ¹² An assessment of the performance change during episodic learning, a measure of the behavioral changes across multiple trials as a result of acquiring knowledge while simultaneously utilizing episodic memory, working memory, and executive function, ⁹ however, might be more sensitive and thus more specifically associated with elevated risk of developing AD. ¹¹ Indeed, episodic memory, working memory, and executive function must interact and function reasonably well for learning to occur, so the performance change while learning could indicate small deficits in one or more of these cognitive processes that are compounded over time in the learning process, which cannot be detected with distinct assessments of episodic memory, working memory, or executive function. Thus, subtle acquisition and encoding dysfunction, due to small episodic memory, working memory, and executive function deficits, could potentially impact performance change during episodic learning and indicate elevated risk of developing AD. ¹¹

To identify if episodic learning can assess risk of developing AD and can differentiate elevated risk of developing AD from typical age-related declines, the first aim of this study was to determine if age group (young adults [YA] versus OA), amount of practice, and performance on traditional neuropsychological tests are associated with performance change during the learning of three cognitive tasks that utilize executive function, episodic memory, and working memory. Based on typical age-related declines in these cognitive domains,^13,14 we hypothesize that being a YA, increased amounts of practice, and better performance on neuropsychological tests will be associated with better episodic learning performance (i.e., larger performance change). The second aim of this study was to determine if Aβ deposition, amount of practice, and performance on neuropsychological tests impact performance change during the same three tasks in OA. Based on the deficits in executive function, episodic memory, and working memory function identified in early AD^15,16 and that increasing age is associated with cognitive declines, ¹⁴ we hypothesize that less Aβ deposition, increased amounts of practice, and better performance on neuropsychological tests will be predictors of better episodic learning performance (i.e., larger performance change).

Methods

Study sample

This is a secondary analysis of data collected from a supplemental visit to a longitudinal, observational trial (Brain Networks and Mobility Study [B-NET, NCT03430427]) that recruited participants from Forsyth County, NC and the surrounding regions. Individuals who participated in the supplemental visit were already enrolled in the parent study. The inclusion criteria for the parent study were: being a community-dwelling adult aged ≥70 years of age or 25–35 years of age, a willingness to provide informed consent, and an ability to communicate with study personnel. In addition to these criteria, the inclusion criteria of the current study were: completing an Aβ PET scan as part of the parent study (for OA participants only) and an ability to communicate in English. The exclusion criteria for the parent study included: serious or uncontrolled chronic disease, evidence of impaired global cognition as assessed by the study neuropsychologist and detailed in our previous, related work, ¹⁷ prior traumatic brain injury with residual deficits, an unwillingness or inability to have a magnetic resonance imaging (MRI) brain scan, dependence on a walker or another person to ambulate, plans to relocate within the next 2–3 years and an unwillingness to return for follow up visits, having a lower limb amputation, musculoskeletal impairments severe enough to preclude functional testing, active participation in an exercise or cognitive enhancing intervention, a history of a brain and/or a spinal cord tumor, any seizures in the last year, a significant and uncorrectable hearing and/or vision problem, and any other cause of concern to the study investigators or physician that the participant would not adhere to the protocol. In addition to these criteria, the exclusion criteria for the current study were: recent surgery or hospitalization within the last 6 months and a clinically manifested neurologic disease affecting mobility. All parent study visits, and supplemental visits were approved by their respective Institutional Review Boards. All participants gave written, informed consent before participating in the parent study and the supplemental visit.

The current study recruited OA and YA from the parent study to participate in the current study visit. Of the subsample of OA who completed the Aβ PET scan in the parent study, 22 OA at elevated risk of developing Alzheimer's disease (i.e., Aβ+) and 32 OA at low risk of developing Alzheimer's disease (i.e., Aβ−) were enrolled in the supplemental visit. One additional OA was enrolled in the supplemental visit but did not complete the Aβ PET scan and was thus not classified as Aβ+/−. OA were classified as being Aβ+/− based on an assessment of global brain Aβ deposition from the Pittsburgh compound-B (PiB) PET imaging conducted during the baseline visit of the parent study, which has been described in our associated work. ¹⁸ The threshold for the Aβ+ group was ≥1.21 global PiB standardized uptake value ratio (SUVR), a previously determined cutoff value. ¹⁹ Additionally, the scans of participants with a global PiB SUVR of intermediate range (1.21 ≤ PiB SUVR ≤ 1.4) were individually assessed by a trained visual rater to determine Aβ+/- status, since these values are considered borderline Aβ positive. ²⁰ Participants with a global PiB SUVR < 1.21 were classified as Aβ−. Additionally, to distinguish pathological changes due to Aβ deposition from typical age-related changes, 29 YA from the parent study were also enrolled in the supplemental visit, but one YA did not complete any of the supplemental visit procedures in the current study due to technical difficulties. A total of 84 participants, including 55 OA and 29 YA, were enrolled in the current study. A single group of OA was analyzed for both the first (n = 55) and second (n = 54) aims of this study, with Aβ deposition included as a continuous variable rather than a dichotomous Aβ+/Aβ− variable in the second aim. The dichotomous variable was only used to ensure that there were adequate numbers of OA with and without elevated Aβ deposition included in the study and that the testing order was counterbalanced in the OA based on Aβ deposition.

The current study utilized a double-blind experimental design. Baseline visits for the parent study and participant recruitment for the current study were performed by researchers associated with the parent study. All researchers involved with the current study were blinded to the group assignments of the OA until completion of the supplemental visits, using blinded group codes for the Aβ+ and Aβ− groups. OA were also not informed of their group assignment.

Baseline visit

Participants completed one baseline visit as part of the parent study. The cognitive tests completed during these visits included Trail Making Tests Trails A (TMT-A) and B (TMT-B), the Wechsler Adult Intelligence Scale-III Digit Symbol-Coding Subtest (DSC), the Flanker Task, and the Rey Auditory Verbal Learning Test (RAVLT). Demographics were also recorded, including years of education, sex, race/ethnicity, and age.

Trail making tests trails A and B

The TMT-A and TMT-B is a paper cognitive test that is comprised of two separate tests. TMT-A assesses information processing speed, including visual scanning, graphomotor speed, and visuomotor processing speed.^21,22 TMT-B assesses information processing speed, working memory, fluid cognitive ability, and executive function, specifically set-switching and response inhibition.^21–23 The time to complete each trial (in seconds) was converted to an age- and race-adjusted percentile score and used for analysis. ²⁴

Wechsler adult intelligence scale-III digit symbol-coding subtest

The DSC is a paper cognitive test that assesses information processing speed. ²⁵ There is also a small but statistically significant correlation between performance on this test and memory functions. ²⁶ Performance was reported as an age-adjusted percentile score, as per the WAIS-III manual scoring instructions, and used for analysis.

Flanker task

The version of the Flanker Task utilized in this study was a computerized arrow version administered using E-Prime 3.0 software (Psychology Software Tools, Pittsburgh, PA). ²⁷ This task primarily assesses executive function, including response inhibition. ²⁸ The mean reaction time (ms) for accurate responses on both the congruent and incongruent trials (Flanker Total), accurate responses on the congruent trials (Flanker Congruent), and accurate responses on the incongruent trials (Flanker Incongruent) were used for analysis. Patients with AD and MCI typically exhibit longer mean reaction times and decreased accuracy compared to individuals without cognitive impairment. ²⁹

Rey auditory verbal learning test

The RAVLT is an auditory cognitive test that assesses learning and memory functions, including working memory. The Learning Over Time percentile score (LOT), an overall assessment of learning and semantic memory, the List B percentile score, a test of short-term memory, and the Total Correct percentile score, an overall assessment of memory, were used for analysis.^30,31 These sub-scores were age- and race-adjusted. The LOT was previously found to have a significant interaction with elevated levels of Aβ in middle-aged adults. ³² This test was only administered to OA.

Supplemental visit

Participants completed two additional cognitive paper tests at the supplemental visit, including the Comprehensive Trail-Making Test 2^nd Edition (CTMT) and the Wechsler Adult Intelligence Scale 4^th Edition Coding Subtest (Coding). The timing of the supplemental visit, relative to the baseline visit, was variable due to testing difficulties during the COVID-19 pandemic.

Comprehensive trail making test 2^nd edition

The CTMT is a paper cognitive test that is comprised of five different tests, which are similar to TMT-A and TMT-B. The CTMT is a test of executive function and was administered in addition to TMT-A and TMT-B because the CTMT can generate sub-scores that specifically assess inhibition control (CTMT-ICI) and set shifting ability (CTMT-SSI), in addition to general executive function (total composite index, CTMT-TCI). ³³ The age-adjusted percentile scores for the CTMT-ICI, CTMT-SSI, and CTMT-TCI were used in the analysis.

Wechsler adult intelligence scale-IV coding subtest

Coding is a paper cognitive test that is an updated version of the DSC, with updated age-adjusted normative values. Coding primarily measures information processing speed and memory functions. ³⁴ The age-adjusted, scaled score, with a maximum score of 19, was used in the analysis.

After completion of the paper cognitive tests, participants were taught and practiced three different auditory cognitive tasks that incorporate cognitive processes known to be affected in early AD: executive function and working memory. During the learning of each of these tasks, episodic memory is additionally used to recall experience and learning from the previous trials. The act of learning each of these tasks requires participants to integrate these three cognitive processes together, which may be more impacted by subtle cognitive dysfunction associated with the elevated risk of developing AD than in the performance of distinct neuropsychological tests for each of those three cognitive processes. The three auditory cognitive tasks taught to participants were the stop-go normal task (SGNT), stop-go reverse task (SGRT), and n-back task (NBT). The three tasks were performed in a counterbalanced order within each participant group (Aβ+ OA, Aβ− OA, and YA), with a parallel design across groups. All three tasks utilized auditory stimuli delivered by DirectRT software (v2020, Empirisoft Corporation, New York, NY) installed on a dedicated laptop via a wireless headset (H800, Logitech, Newark, CA) with a built-in microphone. The verbal responses and the timing of those responses were recorded by the headset microphone.

SGNT

The SGNT is an auditory reaction time task that assesses simple reaction time and executive function, ³⁵ but learning the task incorporates episodic memory, executive function, and working memory. During the SGNT, participants heard the color word “red” or “green” and were asked to respond “stop” when they heard “red” and to respond “go” when they heard “green”. Participants completed two practice trials of 30 responses (P1 and P2) to become familiarized to the task and one final trial of two minutes (P3). Before the trials, participants were instructed to respond as quickly and as accurately as possible to each auditory stimuli. For each of the trials, the response audio and individual response reaction times (ms) were recorded, and then average response accuracy (%) and average response reaction time (ms) were calculated.

From these two measures, a cognitive throughput (CT) was calculated to account for the speed-accuracy trade-off in units of number of correct responses per minute of responding: ³⁶

$$CT={\displaystyle \frac{AverageResponseAccuracy}{AverageResponseReactionTime}\times 600}$$

The ratio of the average response accuracy to average response reaction time was multiplied by a scaling factor of 600, to convert the percentage to a number and to convert the reaction time from milliseconds to minutes (600 = 60,000/100). With this calculation, a higher CT value, resulting from greater accuracy and/or faster reaction times, indicates better performance (i.e., a better speed-accuracy trade-off). The SGNT performance change (PC_SGNT) values were then calculated as:

$$TwoPracticeTrials:P{C}_{SGNT2}={\displaystyle \frac{C{T}_{P2}-C{T}_{P1}}{C{T}_{P1}}\times 100\text{\%}}$$

$$ThreePracticeTrials:P{C}_{SGNT3}={\displaystyle \frac{C{T}_{P3}-C{T}_{P1}}{C{T}_{P1}}\times 100\text{\%}}$$

SGRT

The SGRT is an auditory reaction time task that assesses executive function, specifically response inhibition, ⁷ but learning the task incorporates episodic memory, executive function, and working memory. During the SGRT, participants again heard the color word “red” or “green” and were asked to respond “go” when they heard “red” and to respond “stop” when they heard “green”. Participants completed two practice trials of 30 responses (P1 and P2) and one final trial of two minutes (P3). Before each trial, participants were instructed to respond as quickly and as accurately as possible to each auditory stimuli. For each of the trials, the response audio and individual response reaction times (ms) were recorded, and then average accuracy (%), average response reaction time (ms), and CT (# of correct responses per minute) were calculated. The SGRT performance change (PC_SGRT) values were then calculated as:

$$TwoPracticeTrials:P{C}_{SGRT2}={\displaystyle \frac{C{T}_{P2}-C{T}_{P1}}{C{T}_{P1}}\times 100\text{\%}}$$

$$ThreePracticeTrials:P{C}_{SGRT3}={\displaystyle \frac{C{T}_{P3}-C{T}_{P1}}{C{T}_{P1}}\times 100\text{\%}}$$

NBT

The NBT is an auditory reaction time task that assesses working memory, ³⁷ but learning the task incorporates episodic memory, executive function, and working memory. Three versions of the test were prepared, corresponding to three levels of difficulty: 1-back (easiest), 2-back, and 3-back (hardest). In each version of the test, participants listened to a continuous sequence of letters and were asked to recall whether the letter they heard was the same or different from the letter they heard n letters ago, answering “yes” or “no” after each letter stimulus. For example, the 2-back required the participant to respond “yes” if the letter they heard was the same as the letter they heard two letters ago, and to respond “no” if the letter they heard was different from the letter they heard two letters ago. To ensure that all participants experienced a similar level of difficulty during testing, participants sometimes learned and practiced multiple versions of the task, as needed, until they achieved between 80–99% accuracy on one version of the test. Specifically, all participants were first taught the 2-back version of the test and then were regressed to the 1-back or progressed to the 3-back, as necessary. Different versions of the test were used to ensure each participant was similarly challenged with a higher working memory load but that the task was not so difficult that they resorted to guessing. Participants completed at least two practice trials of the version of the task (P1 and P2) ultimately utilized for the final trial (P3). The first 2 trials required 30 “yes” or “no” responses. The final trial was two minutes in duration. Before each of the trials, participants were instructed to respond as quickly and as accurately as possible to each auditory stimuli. The response audio and individual response reaction times (ms) were recorded during each trial, and then average accuracy (%), average response reaction time (ms), and CT (# of correct responses per minute) were calculated. The NBT performance change (PC_NBT) values were then calculated as:

$$TwoPracticeTrials:P{C}_{NBT2}={\displaystyle \frac{C{T}_{P2}-C{T}_{P1}}{C{T}_{P1}}\times 100\text{\%}}$$

$$ThreePracticeTrials:P{C}_{NBT3}={\displaystyle \frac{C{T}_{P3}-C{T}_{P1}}{C{T}_{P1}}\times 100\text{\%}}$$

Statistical analysis

Performance changes (P1 to P2 and P1 to P3) were analyzed using linear mixed effect models for each of the three cognitive tasks (PC_SGNT, PC_SGRT, PC_NBT). For analysis of the overall sample, a basic model (model 1) was fit with age group (YA versus OA), the amount of practice (2 versus 3 practice trials completed), and their interaction term. Model 1 for PC_NBT also included the NBT version. Least square means for each age group and practice combination were obtained and contrasts were used to test the difference between YA and OA for each practice. The basic model was expanded to adjust for demographic variables including years of education and sex (model 2). Finally, each neuropsychological test predictor (i.e., from the CTMT sub-scores, TMT-A, TMT-B, the Flanker task sub-scores, DSC subtest, and Coding subtest) was added to model 2 separately to evaluate its association with performance change. For analysis of the OA sample, the basic model (model 1) included Aβ deposition (as a continuous variable) the amount of practice, and their interaction term. Model 1 for PC_NBT also included the NBT version. Additionally, contrasts were used to estimate the slopes for Aβ deposition at each practice and the mean performance change at each practice for fixed levels of Aβ deposition (0.2 increments). The basic model was expanded to adjust for years of education, sex, the time difference between the supplemental visit and PiB PET scan, and age (model 2). The time difference between the supplemental visit and the PiB PET scan was included as a covariate since it was variable due to testing difficulties during the COVID-19 pandemic. Similar to the analysis of the overall sample, each neuropsychological test predictor (i.e., from the CTMT sub-scores, TMT-A, TMT-B, the Flanker task sub-scores, DSC subtest, Coding subtest, and RAVLT sub-scores) was added to model 2 separately to evaluate its association with performance change. Sensitivity analyses were conducted using cube root transformation for performance changes to improve normality assumption. Results were similar and not presented in the paper. We note that this analysis was exploratory in nature and no adjustment for multiple comparisons were performed to control potential type I errors. Nominal p values of less than 0.05 were considered statistically significant.

Results

Fifty-five OA without cognitive impairment and 28 YA without cognitive impairment participated in the supplemental visit study procedures. As previously mentioned, one OA was excluded from the second aim of the study because they did not complete the Aβ PET scan and were thus not classified as Aβ+/-. Therefore 55 OA and 28 YA were included in the aim 1 analysis, and 54 OA and 28 YA were included in the aim 2 analysis. Certain participants were excluded from portions of the analyses from each of the aims as follows: Two OA from the Aβ+ group and three OA from the Aβ− group were excluded from the NBT analysis due to minor hearing impairments that made distinguishing letters in the NBT task difficult. One OA from the Aβ− group and one YA were excluded from the SGNT analysis and one OA from the Aβ− group was excluded from the SGRT analysis due to technical difficulties. Additionally, two OA in the Aβ+ group did not complete the Coding subtest because of a non-neurologically based hand tremor that would have invalidated the test score. Two YA did not complete the Flanker test due to technical difficulties. Age, sex, years of education completed, Aβ deposition (PiB SUVR), and the time difference between the supplemental visit and PiB PET scan are summarized in Table 1, and performance scores on each of the baseline and supplemental visit cognitive tests are summarized in Table 2.

Table 1.

Characteristics of participants in the YA and OA groups.

	YA (n = 28)	OA (n = 55)
Demographics
Age (y)	30.4 (27.3–33.2)	75.1 (72.7–79.7)
Sex	11 males, 17 females	30 males, 25 females
Race/Ethnicity (n)
Caucasian/White	18	48
African American/Black	3	6
Asian	3	1
Multiracial/Other	4	0
Years of Education	16 (16–18)	16 (14–18)
Aβ deposition (PiB SUVR)	–	1.20 (1.14–1.60)
Time Difference Between PET Scan and Supplemental Visit (months)	–	18 (7.5–19)
Traditional Neuropsychological Tests
DSC (age-adjusted %)	75 (50–84)	75 (50–84)
Coding (age-adjusted scaled score, max 19)	12 (11–13)	12 (11–13)
TMT-A (age- and race-adjusted %)	28 (11–39)	36 (17–72)
TMT-B (age- and race-adjusted %)	21 (16–38)	54 (37–72)
CTMT-TCI (age-adjusted %)	58 (21–70)	58 (40–77)
CTMT-ICI (age-adjusted %)	53 (18–67)	55 (37–70)
CTMT-SSI (age-adjusted %)	61 (31–80)	63 (40–82)
Flanker Total (ms)	455.7 (413.6–488.9)	630.0 (568.0–707.3)
Flanker Congruent (ms)	432.9 (381.7–478.1)	602.5 (542.1–657.0)
Flanker Incongruent (ms)	479.1 (439.5–504.7)	655.5 (590.4–732.1)
RAVLT Total Correct (%)	–	63 (44–84)
RAVLT List B (%)	–	75 (50–84)
RAVLT LOT (%)	–	63 (50–84)

Values are Median (Q1 – Q3).

RAVLT: Rey Auditory Verbal Learning Test.

Table 2.

Mean performance change after 2 and 3 practice trials.

	YA	OA	Overall
	mean (SE)	mean (SE)	mean (SE)
PCSGNT
2	12.32 (4.19)	8.86 (2.96)	10.59 (2.57)
3	22.78 (4.19)	16.02 (2.96)	19.40 (2.57)
Overall	17.55 (3.62)	12.44 (2.56)
PCSGRT
2	10.48 (4.14)	9.66 (2.98)	10.07 (2.55)
3	18.68 (4.14)	22.60 (3.00)	20.64 (2.56)
Overall	14.58 (3.80)	16.13 (2.74)
PCNBT
2	23.96 (9.67)	24.67 (7.95)	24.32 (7.75)
3	26.81 (9.67)	32.32 (7.93)	29.57 (7.74)
Overall	25.39 (9.30)	28.50 (7.67)

SGNT: Young versus older adults

In the basic model for the SGNT, there was a main effect of the amount of practice on performance change (PC_SGNT), with two practice trials resulting in a smaller PC_SGNT (β = 10.59%) than with three practice trials (β = 19.40%) (p = 0.001). There was no main effect of age group on PC_SGNT (YA: β = 17.55%, OA: β = 12.44%, p > 0.05) or an interaction between the amount of practice and age group (p > 0.05) (Table 2). In subsequent models for the SGNT that adjusted for years of education and sex, CTMT sub-scores, Coding, Flanker task sub-scores, DSC, and TMT-A and TMT-B were not significant predictors (all p > 0.05) (Table 3, Supplemental Table 1). Additionally, years of education and sex were not significant covariates (both p > 0.05) (Supplemental Table 1).

Table 3.

Neuropsychological test predictors for SGNT, SGRT, and NBT task across YA and OA.

	SGNT		SGRT		NBT
	β(SE)	p	β(SE)	p	β(SE)	p
CTMT-
TCI	0.10 (0.08)	0.2306	−0.26 (0.08)	0.0029*	−0.07 (0.14)	0.5936
ICI	0.15 (0.08)	0.0787	−0.26 (0.09)	0.0033*	−0.05 (0.14)	0.7388
SSI	0.02 (0.08)	0.8134	−0.21 (0.08)	0.0124*	−0.07 (0.13)	0.5703
TMT-
A	−0.02 (0.08)	0.7963	−0.05 (0.09)	0.5836	0.04 (0.14)	0.8002
B	0.03 (0.10)	0.8094	0.05 (0.12)	0.6902	−0.13 (0.18)	0.4505
Flanker Task-
Total	−0.02 (0.02)	0.2343	0.02 (0.02)	0.2231	0.04 (0.03)	0.2115
Congruent	−0.03 (0.02)	0.0769	0.03 (0.02)	0.1115	0.04 (0.03)	0.2460
Incongruent	−0.01 (0.02)	0.5836	0.01 (0.02)	0.4254	0.03 (0.03)	0.2028
DSC	0.11 (0.09)	0.2362	−0.20 (0.10)	0.0483*	−0.21 (0.15)	0.1590
Coding	1.28 (1.03)	0.2168	−2.59 (1.11)	0.0218*	−1.84 (1.70)	0.2817

*: designates significant at p < 0.05.

SGRT: Young versus older adults

In the basic model for the SGRT, there was a main effect of the amount of practice on performance change (PC_SGRT), with two practice trials resulting in a smaller PC_SGRT (β = 10.07%) than with three practice trials (β = 20.64%) (p < 0.001). There was no main effect of age group on PC_SGRT (YA: β = 14.58%, OA: β = 16.13%, p > 0.05) or an interaction between the amount of practice and age group (p > 0.05) (Table 2). In subsequent models for the SGRT that adjusted for years of education and sex, worse DSC scores (β = −0.20%, p = 0.0483) and Coding scores (β = −2.59%, p = 0.0218) were associated with increased PC_SGRT (Table 3). Additionally, worse CTMT-TCI scores (β = −0.26% p = 0.0029), CTMT-ICI scores (β = −0.26% p = 0.0033), and CTMT-SSI scores (β = −0.21% p = 0.0124) were associated with a larger PC_SGRT (Table 3). Flanker task sub-scores and TMT-A and TMT-B performance were not significant predictors (both p > 0.05) (Table 3, Supplemental Table 1). Sex and years of education were not significant covariates (both p > 0.05) (Supplemental Table 1).

NBT: Young versus older adults

In the basic model for the NBT, there was no main effect of NBT version, amount of practice (two practice trials β = 24.32%, three practice trials β = 29.57%, p = 0.12), or age group on PC_NBT (YA: β = 25.39%, OA: β = 28.50%, p > 0.05), or an interaction between the amount of practice and age group (p > 0.05) (Table 2). In the subsequent models for the NBT that adjusted for years of education, sex, and NBT version, years of education was a significant covariate, with decreasing years of education associated with increased PC_NBT (β = −3.2% p = 0.0228). CTMT sub-scores, Coding, Flanker task performance, DSC, and TMT-A and TMT-B were not significant predictors (all p > 0.05) (Table 3, Supplemental Table 1). Sex and NBT version were not significant covariates (both p > 0.05) (Supplemental Table 1).

SGNT: Aβ deposition

In the basic model for the SGNT, a greater amount of Aβ deposition was associated with a smaller performance change after two and three practice trials, with slopes β = −0.97% and −1.96% for two and three practice trials, respectively (Table 4, Figure 1) The decline in performance change associated with greater Aβ deposition after three practice trials appeared to be steeper compared to that after two practice trials (β = −0.99%), although the difference did not reach statistical significance (p = 0.86) (Table 4). Additionally, we observed a larger performance change at P3 compared to at P2 (Table 4). However, the improvement after three practice trials, relative to two practice trials, decreased as the Aβ deposition level increased (Table 4). For example, using contrasts, we found that in individuals with a PiB SUVR of 1.4, the performance change after three practice trials was 7.21% higher than after two practice trials (p = 0.0014) (Table 4). This difference was only 6.62% in individuals with a PiB SUVR of 2.0 (i.e., greater Aβ deposition) (p = 0.10) (Table 4).

Table 4.

Associations of Aβ deposition and performance change with learning in OA.

SGNT
	After two practice trials	After three practice trials	Difference between two and three practice trials
Aβ	β(SE)	β(SE)	β(SE)	p value for difference
1.20	9.04 (2.81)	16.45 (2.81)	7.41 (2.44)	0.0038*
1.40	8.85 (2.47)	16.06 (2.47)	7.21 (2.14)	0.0014*
1.60	8.65 (2.76)	15.67 (2.76)	7.01 (2.39)	0.0050*
1.80	8.46 (3.53)	15.27 (3.53)	6.82 (3.06)	0.0305*
2.00	8.26 (4.55)	14.88 (4.55)	6.62 (3.94)	0.0994
2.20	8.07 (5.68)	14.49 (5.68)	6.42 (4.92)	0.1981
slope	−0.97 (6.46)	−1.96 (6.46)	−0.99 (5.61)	0.8605
SGRT
1.20	10.02 (3.63)	24.68 (3.65)	14.66 (2.90)	<0.0001*
1.40	10.02 (3.19)	23.02 (3.21)	12.99 (2.55)	<0.0001*
1.60	10.03 (3.57)	21.35 (3.58)	11.33 (2.84)	0.0002*
1.80	10.03 (4.56)	19.69 (4.57)	9.66 (3.62)	0.0103*
2.00	10.04 (5.87)	18.03 (5.87)	8.00 (4.66)	0.0922
2.20	10.05 (8.86)	16.37 (7.33)	6.33 (5.81)	0.2811
slope	0.02 (8.33)	−8.30 (8.34)	−8.33 (6.61)	0.2137
NBT
1.20	24.59 (7.54)	34.84 (7.51)	10.25 (4.83)	0.0391*
1.40	26.71 (7.32)	35.23 (7.31)	8.52 (4.26)	0.0514
1.60	28.83 (7.78)	35.62 (7.77)	6.79 (4.71)	0.1564
1.80	30.95 (8.81)	36.01 (8.80)	5.06 (5.96)	0.4002
2.00	33.07 (10.23)	36.39 (10.22)	3.33 (7.61)	0.6642
2.20	35.19 (11.91)	36.78 (11.90)	1.59 (9.47)	0.8670
slope	10.60 (11.25)	1.94 (11.20)	−8.66 (10.73)	0.4238

*: designates significant at p < 0.05.

Figure 1.

Amount of Aβ deposition in OA in relation to the performance change for each task.

In subsequent models for the SGNT that included various cognitive test performances and covariates, better performance on the RAVLT LOT score was associated with a larger PC_SGNT (β = 0.25%, p = 0.0314), regardless of Aβ deposition in OA (Table 5). None of the CTMT sub-scores, Coding, Flanker task sub-scores, DSC, other RAVLT sub-scores, or TMT-A and TMT-B were significant predictors (p > 0.05, Table 5, Supplemental Table 2). Additionally, age, time difference between the PiB PET scan and the supplemental visit, years of education, and sex were not significant covariates (all p > 0.05) (Supplemental Table 2).

Table 5.

Neuropsychological test predictors for SGNT, SGRT, and NBT task in OA.

	SGNT		SGRT		NBT
	β(SE)	p	β(SE)	p	β(SE)	p
CTMT-
TCI	0.07 (0.10)	0.4813	−0.22 (0.13)	0.0850	0.04 (0.18)	0.8257
ICI	0.16 (0.10)	0.1007	−0.21 (0.13)	0.1134	−0.04 (0.19)	0.8264
SSI	−0.07 (0.09)	0.4252	−0.21 (0.13)	0.1028	0.13 (0.17)	0.4250
TMT-
A	−0.05 (0.08)	0.4975	0.00 (0.11)	0.9825	0.05 (0.15)	0.7362
B	0.01 (0.11)	0.9008	0.19 (0.15)	0.1946	−0.15 (0.21)	0.4877
Flanker Task-
Total	−0.02 (0.02)	0.2173	0.02 (0.02)	0.3891	0.01 (0.03)	0.7111
Congruent	−0.03 (0.02)	0.0872	0.03 (0.02)	0.2155	0.02 (0.03)	0.5214
Incongruent	−0.01 (0.02)	0.5098	0.01 (0.02)	0.6559	0.00 (0.03)	0.9175
DSC	0.03 (0.11)	0.7829	−0.10 (0.15)	0.4992	−0.06 (0.19)	0.7566
Coding	−0.81 (1.47)	0.5866	−1.67 (2.00)	0.4103	−1.55 (2.70)	0.5679
RAVLT-
Total Correct	0.10 (0.10)	0.3279	−0.23 (0.13)	0.0775	0.07 (0.18)	0.7219
List B	−0.08 (0.09)	0.3923	−0.17 (0.13)	0.2093	−0.10 (0.18)	0.5591
LOT	0.25 (0.11)	0.0314*	0.17 (0.17)	0.3060	0.27 (0.21)	0.2074

*: designates significant at p < 0.05.

RAVLT: Rey Auditory Verbal Learning Test.

SGRT: Aβ deposition

Dissimilar from the SGNT, in the basic model for the SGRT, a greater amount of Aβ deposition was associated with a greater performance change after two practice trials and with a smaller performance change after three practice trials, with slopes β = 0.02% and −8.30% for two and three practice trials, respectively (Table 4, Figure 1). However, similar to the SGNT, the difference in the performance change between P2 and P3 did not reach statistical difference (β = −8.33%, p = 0.2137) (Table 4). Again, similar to the SGNT, we observed a larger performance change at P3 compared to at P2, but the improvement decreased as the Aβ deposition level increased (Table 4). For example, using contrasts, we found that in individuals with a PiB SUVR of 1.4, performance change after three practice trials was 12.99% higher than that at two practice trials (p < 0.0001) (Table 4). This difference was only 8% when in individuals with a PiB SUVR of 2.0 (i.e., greater Aβ deposition) (p = 0.092) (Table 4).

In subsequent models that included the cognitive test performances and covariates, none of the cognitive test performances were significant predictors (all p > 0.05) (Table 5, Supplemental Table 2) and none of the covariates were statistically significant (all p > 0.05) (Supplemental Table 2).

NBT: Aβ deposition

Dissimilar from the SGNT and SGRT, in the basic model for the NBT, a greater amount of Aβ deposition was associated with a greater performance change after two and three practice trials, with slopes β = 10.6% and 1.94% after two practice trials and after three practice trials, respectively (Table 4, Figure 1). The increase in performance change after two practice trials appeared to be steeper compared to that after three practice trials (β = −8.66%); although, the difference did not reach statistical significance (p = 0.4238) (Table 4). Similar to both the SGNT and SGRT, we observed a larger performance change after three practice trials compared to after two practice trials, but the improvement decreased as the Aβ deposition level increased (Table 4). For example, using contrasts, we found that in individuals with a PiB SUVR of 1.4, performance change after three practice trials was 8.52% higher than after two practice trials (p = 0.0514) (Table 4). This difference was only 3.33% in individuals with a PiB SUVR of 2.0 (i.e., greater Aβ deposition) (p = 0.6642) (Table 4).

In a subsequent model that adjusted for information processing speed (i.e., Coding score) and included covariates, there was a significant interaction effect of amount of practice and Aβ (β = −22.55%, p = 0.0295) (Supplemental Table 2). Specifically, individuals with greater Aβ deposition exhibited a larger PC_NBT from practice 1 to practice 2, but they exhibited a smaller PC_NBT from practice trial 1 to practice trial 3, relative to individuals with less Aβ deposition. In the other subsequent models that included the cognitive test performances and covariates, none of the cognitive test performances were significant predictors (all p > 0.05) (Table 5, Supplemental Table 2) and none of the covariates were statistically significant (all p > 0.05) (Supplemental Table 2).

Discussion

This study aimed to determine if age group (YA versus OA), amount of practice, and performance on traditional neuropsychological tests impact performance change during the learning of three cognitive tasks that utilize executive function, episodic memory, and working memory and to determine if Aβ deposition, amount of practice, and performance on neuropsychological tests impact performance change during learning the same three tasks in OA. This study has three main findings. The first main finding is that, in both YA and OA, more practice resulted in better speed-accuracy trade-offs (i.e., larger performance changes) while learning the SGNT and SGRT. The second main finding is that in OA, after two practice trials of the NBT, the performance change increased with greater amounts of Aβ deposition and worse information processing speed; while, after three practice trials, the performance change decreased with greater amounts of Aβ deposition and worse information processing speed. The third main finding is that across the three cognitive tasks (NBT, SGNT, SGRT), greater Aβ deposition appears to be associated with smaller improvements after more practice trials.

In partial support of our hypothesis, our first main finding is that greater amounts of practice were associated with better episodic learning performance during the SGNT and SGRT, regardless of age group. More specifically, with an increased number of practice trials, participants generally exhibited a larger performance change. This improvement in speed-accuracy trade-off from a participant's first to subsequent practice trials is likely a practice effect, or an improvement in performance due to repeated trials of the same task. ³⁸ Practice effects were observed across both age groups and were larger after three practice trials, relative to after two practice trials. Consistent with the practice effects observed in our study, Collie et al. ³⁹ similarly observed that older adult participants exhibited an improved performance from their first to second practice trial and an even better performance after three practice trials, relative to two, of an information processing speed task. Inconsistent with our hypothesis, though, age group did not impact performance change while learning and practicing the SGNT and SGRT. ³⁹ We had hypothesized that YA would exhibit larger performance changes than OA because working memory and executive function typically decline with age, ¹³ however, no difference in performance change between YA and OA was observed, potentially because episodic memory remains relatively constant with aging. ¹³ Since learning SGNT and SGRT requires episodic memory, working memory, and executive function, perhaps the SGNT and SGRT are not sensitive enough to detect a decline in only working memory and executive function, with intact episodic memory masking subtle deficits in working memory and executive function. This main finding suggests that regardless of age more practice can increase performance on episodic learning tasks, and this practice effect is consistent with typical aging.

In contrast to our hypothesis, our second main finding is that when accounting for information processing speed, after two practice trials of the NBT, the performance change increased with greater amounts of Aβ, but after three practice trials of the NBT, the performance change decreased with greater amounts of Aβ. More specifically, in OA, with two practice trials of this working memory-load task, the performance change increased with greater amounts of Aβ deposition and worse information processing speed; while, with three practice trials, the performance change decreased with greater amounts of Aβ deposition and worse information processing speed. One possible explanation is that relative to OA with less Aβ deposition, OA with greater Aβ deposition exhibited worse P1 performance, allowing for greater improvements (i.e., increasing response accuracy and/or decreasing their reaction time) from P1 to P2, but then the greater Aβ deposition and worse information processing speed resulted in less improvements from P2 to P3. In the context of verbal working memory tasks, information processing speed and working memory are highly correlated, meaning worse information processing speed performance could indicate worse working memory abilities. ⁴⁰ The declining amounts of improvement with greater amounts of practice in individuals with greater Aβ deposition and worse information processing speed suggests that greater Aβ deposition and worse information processing speed may impact the ability to recruit additional cognitive domains needed to employ new strategies of improvement as a task becomes more challenging, as in the case of trying to continually improve speed and accuracy of responses. In a study by Chen et al., ⁴¹ healthy middle-aged adults who carried the APOE4 gene, increasing their risk of Aβ accumulation, exhibited additional recruitment of cognitive resources in comparison to participants who did not carry the gene during low-working memory-load tasks. However, in higher-working memory-load tasks, like the NBT, the participants who carried the APOE4 gene exhibited less activation of working memory circuitry, in the bilateral frontal and parietal brain regions, in comparison to the OA who were not carriers. This previous study and the results of the current study suggest that OA with greater Aβ deposition and worse information processing speed experience a decreased ability to activate the necessary cognitive resources to continue to improve with more practice. These findings suggest individuals at a higher risk of developing AD with slower information processing speed improve less after more trials while learning a working memory task, possibly due to less activation of working memory circuitry from the presence of Aβ deposition.

Consistent with our hypothesis, the third main finding is that greater Aβ deposition appeared to be associated with a smaller performance change after more practice trials across all three tasks, while those with less Aβ deposition exhibited a larger performance change after more practice trials. In other words, individuals with greater Aβ deposition tended to exhibit smaller improvements with more practice, in comparison to individuals with less Aβ deposition. More specifically, in the SGNT and SGRT, this trend was observed in individuals with a PiB SUVR in the range of 1.20–1.80, with no trend observed at 1.00 or ≥ 2.00 PiB SUVR because there were not enough participants in these PiB SUVR ranges within this study sample. Similar to the interpretation of the second main finding, the trend of greater Aβ deposition being associated with smaller improvements with more practice across the SGNT and SGRT may be the result of individuals with greater Aβ deposition having fewer or less flexible cognitive resources available to find new ways to improve their performance with subsequent practice trials, relative to those with less Aβ deposition. This interpretation is supported by the findings of several studies that determined that declines in specific cognitive domains, including executive function, episodic memory, and working memory, in older adults without cognitive impairment were associated with an Aβ+ PET scan or Aβ+ cerebrospinal fluid level,^42–44 and that, even more specifically, greater declines in episodic memory and executive function were associated with greater Aβ deposition/levels.^45,46 Because learning new tasks, like the SGNT and SGRT, requires the effective coordination of executive function and episodic memory and declines in these cognitive domains has been associated with greater Aβ deposition,^44,45 greater Aβ deposition may not only contribute to declines in these cognitive domains but also poorer coordination of executive function and episodic memory to efficiently learn new tasks and thus lead to a decreased ability to improve between subsequent trials (i.e., a smaller performance change). In the NBT, a more limited trend was observed in individuals with a PiB SUVR in the range of 1.20–1.40, with individuals with 1.40 PiB SUVR tending to exhibit smaller improvements with more practice, in comparison to individuals with 1.20 PiB SUVR. The NBT could be considered a more challenging task than the SGRT or the SGNT because in addition to requiring working memory during the learning of the task, as with the SGNT and SGRT, the NBT requires working memory during the performance of the task. Indeed, participants in the current study subjectively stated that they found the NBT to be much harder than the SGNT and the SGRT. Thus, the NBT may require more cognitive resources than the SGNT and SGRT to learn, perform, and then improve with practice, and individuals with PiB SUVR greater than 1.40 may not have been able to coordinate enough cognitive resources to significantly improve their learning after three practice trials, relative to two practice trials. This result suggests that Aβ deposition impacts an individual's ability to learn a new cognitive task, with learning being even more limited in individuals with greater Aβ deposition when learning a more challenging task, even among individuals without cognitive impairment or dementia.

Limitations

This study has several limitations. First, in comparison to the number of outcome measures and analyses performed, this study has a relatively small sample size, no adjustment for multiple comparisons is performed, and p values of less than 0.05 were considered statistically significant. Further, the study sample contained a relatively small number of older adults with elevated amounts of Aβ deposition. However, we believe that the uniqueness of the study sample (i.e., including OA without cognitive impairment at elevated risk of developing AD based on Aβ deposition) justifies the exploratory nature of this study. Nonetheless, the conclusions of our study should be interpreted with caution, and future follow-up studies with larger sample sizes and lower p value thresholds are warranted to confirm our results. Second, the baseline and supplemental cognitive testing, while extensive, was not an exhaustive battery of neuropsychological tests and did not fully overlap with the relevant cognitive domains needed to learn and practice the three cognitive tasks utilized in this study. As previously mentioned, this study is a secondary analysis of data collected from the supplemental visit and larger parent study. Thus, the neuropsychological tests administered at the baseline and supplemental visits were not planned with the current analysis in mind, and while there was significant overlap between the cognitive domains assessed during the neuropsychological testing and utilized to learn and practice the novel cognitive tasks, some cognitive domains, such as working memory, were not assessed during the baseline or supplemental neuropsychological tests. Additionally, the RAVLT was only administered to the OA group. Despite this, there was significant overlap between the cognitive domains of the novel cognitive tasks and the neuropsychological tests, ensuring interpretable results. Future research should include follow-up studies that assess an exhaustive battery of neuropsychological tests that fully overlap with the relevant cognitive domains needed to learn and practice the three cognitive tasks utilized in this study. Finally, the timing between the baseline visit with the PiB PET scan and the supplemental visit were variable due to the COVID-19 pandemic. To account for a potential impact of the timing difference between visits, the timing between visits was included as a covariate in the analyses. This covariate was not significant in any of the analyses, providing evidence that this variable timing did not impact the study results.

Conclusion

This study aimed to identify if episodic learning can differentiate between YA and OA and to determine if Aβ deposition, a marker of AD risk, is associated with performance change while learning a new cognitive task. First, our study found that more practice while learning cognitive tasks results in better performance, regardless of age, suggesting that an individual's ability to improve on a cognitive task using episodic learning is maintained with increasing age. Second, during a working memory task, individuals with slower information processing speed at a higher risk of developing AD exhibited less improvement while learning a working memory task compared to individuals with faster information processing speed at a lower risk of developing AD, possibly suggesting that greater Aβ deposition hinders the brain circuitry involved in working memory. Third, an individual's ability to learn a new cognitive task appears to be negatively impacted by Aβ deposition, with individuals with greater Aβ deposition tending to exhibit smaller improvements with more practice, in comparison to individuals with less Aβ deposition. However, these conclusions should be interpreted with caution, considering the relatively small sample size and potential multiplicity issues. Future studies with a larger sample size and more comprehensive battery of neuropsychological tests that fully cover the cognitive domains involved in episodic learning are needed to assess the potential of episodic learning more completely as a measure of subtle cognitive declines associated with an elevated risk of developing AD.