The Trail Making Test elucidates neural substrates of specific post-stroke executive dysfunctions

Abstract

Background and Purpose

Post-stroke cognitive impairment (PSCI) is typified by prominent deficits in processing speed and executive function. However, the underlying neuroanatomical substrates of executive deficits are not well understood and further elucidation is needed. There may be utility in fractionating executive functions to delineate neural substrates.

Methods

One test amenable to fine delineation is the Trail Making Test (TMT), which emphasizes processing speed (TMT-A) and set-shifting (TMT-B-A difference, proportion, quotient scores and TMT-B set-shifting errors). The TMT was administered to two overt ischemic stroke cohorts from a multinational study: (i) a chronic stroke cohort (N=61) and (ii) an acute-sub-acute stroke cohort (N=45). Volumetric quantification of ischemic stroke and White Matter HyperIntensities (WMH) was done on MRI, along with ratings of involvement of cholinergic projections, using the previously published Cholinergic Hyperintensities Projections Scale (CHIPS). Damage to the superior longitudinal fasciculus (SLF), which co-localizes with some cholinergic projections, was also documented.

Results

Multiple linear regression analyses were completed. While larger infarcts (β=0.37, p<0.0001) were associated with slower processing speed, CHIPS severity (β=0.39, p<0.0001) was associated with all metrics of set shifting. Left SLF damage, however, was only associated with the difference score (β=0.17, p=0.03). These findings were replicated in both cohorts. Patients with ≥2 TMT-B set shifting errors also had greater CHIPS severity.

Conclusions

In this multinational stroke cohort study, damage to lateral cholinergic pathways and the SLF emerged as significant neuroanatomical correlates for executive deficits in set shifting.

INTRODUCTION

Post-stroke cognitive impairment (PSCI) is typified by prominent deficits in attention, information processing speed, and executive functions.^1,2 As executive dysfunction is a main determinant of disability³ and mortality,⁴ treating it is a major goal. Longitudinal case-cohort studies in Alzheimer’s Disease (AD) suggest that cholinesterase inhibitors (CHEIs) may preferentially attenuate decline in attention and executive functions, which may secondarily influence memory decline.^5,6,7 CHEIs may also play a role in PSCI where attention and executive functions are prominently impaired, though trials have been inconclusive.^8,9 In part, this may be due to the conflicting and omnibus manner in which cognitive outcomes are defined and whether neuroanatomic modulators are accounted for. The use of composite cognitive measures is one popular approach to address these issues. However, the use of neuropsychological tests capable of honing in on specific constructs, along with accounting for potential neuroanatomic modulators, remains relatively unexplored.

One test amenable to fine delineation is the Trail Making Test (TMT). Initially developed and used for military screening in 1944, TMT has become one of the most commonly used neuropsychological tests, owing to its ease of use and ability to simultaneously measure multiple executive functions: (1) attention, (2) processing speed, (3) set-shifting; in addition to visuospatial ability and working memory.¹⁰ Importantly, a variety of TMT metrics can be used to hone in on specific executive functions depending on the construct chosen.^10,11 The TMT is also clinically relevant, predicting driving ability¹² and conversion to Alzheimer’s Disease (AD) from amnestic Mild Cognitive Impairment (MCI).¹³ Not surprisingly, it also serves as a common outcome measure in clinical trials of CHEIs in vascular cognitive impairment.^14,15

One important candidate neuroanatomic modulator is the cholinergic system, different components of which may subserve different cognitive constructs. Rodent models suggest that global basal forebrain cholinergic supply is important for attention, mental flexibility, reversal learning, and memory.¹⁶ However, distinct projections from the basal forebrain arise from distinct neuronal populations.¹⁷ Projections from the septal nuclei and vertical limb of the diagonal band of Broca project primarily to the hippocampus where acetylcholine likely modulates encoding and memory.¹⁸ By contrast, the lateral and medial projections arising from the nucleus basalis of Meynert (NbM) supply vast areas of neocortex and the cingulate respectively,¹⁷ and may relate more to executive functions.^19,20,21 Clinically, cerebrovascular pathology in lateral cholinergic projections relates to executive dysfunction^19,20,21 and post-stroke dementia,²² while degeneration of basal forebrain nuclei observed in those likely to convert from MCI to AD, may relate to memory retrieval.²³ The role of lateral cholinergic integrity in PSCI might, however, be affected by other factors such as infarct location, infarct volume, WMH volume, and infarcts affecting major association pathways. The superior longitudinal fasciculus (SLF) is an extensive association tract that travels in proximity to the lateral cholinergic pathways and is itself related to executive functions.^24,25

Our primary objective in this multi-centre, image-guided study was to determine what distinct brain-behaviour relationships could be delineated within PSCI from a single cognitive test (TMT), by using (1) timed, (2) derived, and (3) error metrics and their relation to (1) White Matter HyperIntensity (WMH) volume, (2) infarct volume, (3) infarct location, (4) strokes compromising the SLF, and (5) lateral cholinergic pathway integrity. Our secondary objective was to examine how these refined relationships might expand our understanding of specific post stroke dysexecutive symptoms and suggest avenues for future clinical trials.

METHODS

Participants

Two ischemic stroke cohorts were included: a chronic stroke cohort (N=61) from a National Institute of Health (NIH) (057514) study at Sunnybrook Health Sciences Centre, Toronto, Canada and the University of Illinois Medical Centre, USA; and an acute-sub-acute stroke cohort (N=45) from Hallym University Sacred Heart Hospital, Anyang, South Korea. Both complied with the National Institute for Neurological Disorders and Stroke – Canadian Stroke Network Vascular Cognitive Impairment Harmonization Standards (NINDS-CSN-VCIHS) for neuropsychological testing.²⁶ Studies were approved by research ethics boards at their institutions. See supplementary Table I for inclusion criteria.

Magnetic Resonance Imaging

MRIs in the North American cohort were harmoniously acquired on 3.0-Tesla General Electric Healthcare scanners (Waukesha, Wisconsin). The South Korean cohort underwent MRI within 4 days of stroke onset on a 1.5-Tesla Philips scanner (Philips, Best, the Netherlands). See supplementary Table II for MRI acquisition protocols.

Image Processing

Strokes were manually traced and all were confirmed by a research neuroradiologist (FQG). Chronic infarct volumes were calculated from tracings of hypointensity on T1 MRI using ANALYZE 8.0 software (Biomedical Imaging Resource, Mayo Foundation, Rochester, MN, USA). Traced acute infarcts were those that appeared hyperintense on Diffusion Weighted Imaging. Stroke locations on native MRI were rated by a trained analyst and certified by the gold standard of an experienced research neuroradiologist. Strokes were categorized as either absent or present (0 or 1) in (i) executive network structures²⁷ and (ii) SLF²⁴ (online-only data supplement). Strokes involving the executive network were those that included the dorsolateral frontal lobe, ventrolateral prefrontal cortex, dorsomedial prefrontal cortex, lateral parietal lobe, anterior thalamus, or caudate nucleus.^27,28

A validated semi-automatic fuzzy lesion extractor (FLEX) pipeline quantified WMH volumes on FLAIR images.²⁹ Semi-automated tissue segmentation classified brain into 4 tissue classes: normal appearing gray matter (NAGM), normal appearing white matter (NAWM), ventricular cerebrospinal fluid (vCSF) and sulcal cerebrospinal fluid (sCSF).³⁰ Total intracranial volume (TIV) values were calculated as the sum of NAGM, NAWM, WMH, infarction volume, sCSF, and vCSF. For head-size correction, all tissue volumes used for analyses were expressed as a percentage of TIV. The Cholinergic Hyperintensities Projections Scale (CHIPS)²⁰ was used to quantify the degree of WMH and stroke involvement within lateral cholinergic projections. CHIPS is a weighted scale that rates the degree of lateral cholinergic involvement on a scale from 0 – 100. Levels with greatest cholinergic fiber density in the low external capsule are weighted more heavily compared to the dispersed lateral cholinergic projections in the centrum semiovale.²⁰

Neuropsychological Testing

The Trail Making Test (TMT) is a validated timed measure of processing speed (TMT-A) and mental flexibility, a key executive function involving set shifting (TMT-B).¹⁰ TMT-B also measures visuospatial, speed of processing, and working memory abilities. In order to minimize their contributions to TMT-B time we used three TMT-B derived constructs of set-shifting.^10,11 The difference score (TMT[B – A]) may only partially control for non-set shifting elements by subtracting their contribution as measured by the TMT-A. On the other hand, the proportion (TMT[(B –A)/A]) and quotient scores (TMT[B/A]) may better control for non set-shifting elements by dividing out TMT-A, thereby emphasizing set-shifting ability.¹¹ We also documented the frequency of errors made on TMT-A and TMT-B. Errors made on TMT-B were classified as errors of (i) set-shifting (failure to alternate between numbers and letters) or (ii) sequencing (failure to connect numbers or letters following a set-shift in ascending order).

Statistical Analysis

Inter-rater reliabilities for absolute agreement with 10 gold standard datasets produced by neuroradiologist FQG were examined using the Intra-Class Correlation Coefficient (ICC) and Kappa statistic (κ). High inter-rater reliability was observed among authors responsible for tissue segmentation, stroke tracing, CHIPS, SLF involvement, and executive network strokes in relation to the gold standard (ICC>0.94; κ >0.83).

Descriptive statistics were calculated for all variables of interest. Continuous measures were summarized using means and standard deviations, while categorical measures were summarized using counts and percentages. Correlations between TMT, demographic data, and neuroimaging data were assessed using Pearson’s r for normally distributed variables and Spearman’s rho for non-normal data. In addition, Student’s t-test and Chi-squared tests were used to compare means and proportions for continuous and categorical variables of interest between the North American and South Korean cohorts. Non-normally distributed variables were assessed using the Mann Whitney U test.

Multiple linear regression models were completed in a two block approach. In the first block, age and education were entered. In the second block, sex; infarction volume; global WMH volume; CHIPS scale; and stroke location within the executive network and SLF were considered in the stepwise elimination of non-significant variables. We also submitted study site and post-stroke delay to regression models when they significantly correlated with TMT measures. In subsequent regression models we assessed both the laterality of CHIPS and SLF strokes. Prior to modeling, variables were assessed for multicollinearity (tolerance statistic value <0.4), and if multicollinearity was found only one member of a correlated set of variables was retained in the model. The final model was assessed for any potential violations to linear regression modeling using residual plots. No violations were found. All analyses were carried out using SAS Version 9.1 (SAS Institute, Cary, NC, USA).

RESULTS

Characteristics of participants

Table 1 shows demographic and neuroimaging data and Table 2 shows the correlates of processing speed and set-shifting.

Table 1.

Demographic and Neuroimaging data.

Variable	North American Ischemic Stroke Cohort (N=61)	South Korean Ischemic Stroke Cohort (N=45)	p-value
Men/Women	34/27	21/24	0.4
Age (years)	64.3 ± 12.4	63.2 ± 14.0	0.7
Years of education	14.7 ± 3.1	9.3 ± 4.7	<0.0001
Days between stroke onset and neuropsychological testing	422.6 ± 246.1	101.7 ± 19.1	<0.0001
MMSE	27.8 ± 2.3	24.6 ± 4.6	<0.0001
NIHSS at neuropsychological testing	1.7 ± 2.2	1.2 ± 2.1	0.04
TMT-A (s)	55.4 ± 32.2	58.0 ± 59.2	0.2
TMT-B (s)	144.0 ± 81.3	140.4 ± 115.5	0.1
TMT-B-A difference score	88.0 ± 63.4	82.4 ± 83.6	0.08
TMT-B-A proportion score	1.69 ± 1.05	1.65 ± 1.71	0.06
TMT-B-A quotient score	2.69 ± 1.05	2.65 ± 1.71	0.06
CHIPS	22.4 ± 15.5	13.4 ± 11.2	0.001
SLF stroke involvement	53% (n=32)	24% (n=11)	0.004
Executive network stroke involvement	69% (n=43)	67% (n=30)	0.7
Total Infarction (%)†	2.3 ± 3.7	0.71 ± 1.5	0.004
WMH (%)†	0.84 ± 1.08	0.36 ± 0.44	<0.0001
NAGM (%)†	42.8 ± 3.3	45.9 ± 5.9	0.001
NAWM (%)†	32.3 ± 3.8	33.7 ± 5.0	0.1
vCSF (%)†	3.2 ± 1.6	2.3 ± 1.3	0.001
sCSF (%)†	18.5 ± 3.9	17.0 ± 2.8	0.03

Mean ± SD

^†Volumes expressed as a percentage of Total Intracranial Volume for head size correction

Abbreviations:

National Institute of Health Stroke Scale (NIHSS);Mini-mental State Examination (MMSE); Trail Making Test (TMT); White Matter Hyperintensities (WMH); Normal appearing grey matter (NAGM); Normal appearing white matter (NAWM); ventricular cerebrospinal fluid volume (vCSF); sulcal cerebrospinal fluid volume (sCSF); Cholinergic Hyperintensities Projections Scale (CHIPS); Superior Longitudinal Fasciculus (SLF)

Table 2.

Unadjusted correlates of processing speed and set-shifting.

Variable	North American and South Korean Ischemic Stroke cohorts (N= 106)
	TMT- A Correlation (p-level)	TMT-B-A Difference Score Correlation (p-level)	TMT-B-A Proportion Score† Correlation (p-level)	TMT-B-A Quotient Score† Correlation (p-level)
Age	0.34 (<0.0001)	0.48 (<0.0001)	0.24 (0.01)
Gender (Female)	0.29 (0.003)	0.30 (0.002)	0.13 (0.2)
Years of Education	-0.12 (0.2)	-0.21 (0.03)	-0.14 (0.1)
Global WMH Volume	0.52 (<0.0001)	0.53 (<0.0001)	0.31 (0.001)
CHIPS	0.43 (<0.0001)	0.52 (<0.0001)	0.37 (<0.0001)
Infarct Volume	0.31 (0.001)	0.19 (0.05)	0.05 (0.6)
SLF Stroke	0.28 (0.003)	0.26 (0.007)	0.14 (0.2)
Executive Network Stroke	0.08 (0.4)	0.15 (0.1)	0.12 (0.2)
Study Site	-0.2 (0.04)	-0.17 (0.08)	-0.19 (0.06)
Days between stroke onset and neuropsychological testing	0.2(0.04)	0.2 (0.02)	0.3(0.01)

^†The proportion and quotient scores were perfectly correlated with each other. Study Site was coded as: North America=0, South Korea=1.

Set-Shifting Measures

Correlational analyses between TMT-A, TMT-B time and derived metrics shows that only the proportion and quotient scores are not related to processing speed (TMT-A) (supplementary Table III). Table III also demonstrates that set-shifting errors are not correlated with processing speed (TMT-A) and are strongly associated with the proportion and quotient scores.

Predictors of processing speed and set-shifting

Multiple linear regression analyses are summarized in Table 3. Longer TMT-A times were associated with increasing age, fewer years of education and larger infarctions. With respect to set-shifting executive function: while an increase in the TMT-B-A difference score was related to increasing age, education and CHIPS severity, increasing TMT-B-A proportion and quotient scores were related to CHIPS severity. Given the differences between our cohorts in Table 1, we ran linear regression models separately in both the South Korean and North American stroke cohorts. We found the same associations noted above (see supplementary Tables IV and V).

Table 3.

Linear Regression analyses

	North American and South Korean Ischemic Stroke cohorts (N= 106)
Model	TMT- A β, (p-level), r2	TMT-B-A difference score β, (p-level), r2	TMT-B-A Proportion Score† β,(p-level), r2	TMT-B-A Quotient Score† β, (p-level), r2
Significant Variables*	Age*	Age*	Age
	0.29, (0.001), 0.08	0.33, (<0.0001), 0.09	0.12, (0.2), 0.01
	Education*	Education*	Education
	-0.29, (0.001), 0.08	-0.19, (0.01), 0.04	-0.13, (0.2), 0.01
	Infarct Volume *	CHIPS*	CHIPS*
	0.37, (<0.0001), 0.13	0.39, (<0.0001), 0.15	0.34,(<0.0001), 0.11

^†The proportion and quotient scores were perfectly correlated with each other

Set-shifting errors and lateral cholinergic integrity

In Table 4 we compared those who made ≥2 to those who made <2 set-shifting errors. There were no differences in age, gender, education, stroke severity, or speed of processing between these two groups. The groups only varied in their degree of set-shifting impairment measured by the difference, proportion and, quotient scores and by the degree of lateral cholinergic damage. No other types of TMT errors reflected differences in CHIPS severity (see supplementary Tables VI, VII, and VIII). In Table VIII, those who made sequencing errors had larger global WMH volumes.

Table 4.

Analysis of TMT-B set-shifting errors: where abnormal is classified as ≥2 errors

Variable	TMT-B set errors ≤1 (n=86) Mean ± SD	TMT-B set errors ≥2 (n=14) Mean ± SD	Statistic
Age	63.2 ± 12.9	65.2 ± 14.4	p=0.6
Education	12.7 ±4.4	11.7±4.7	p=0.4
Gender	45% female	50% female	p=0.5
NIHSS	1.41 ± 2.22	1.29 ± 1.27	p=0.6
TMT-A (seconds)	51.2±42.5	54.0±22.2	p=0.1
TMT-B (seconds)	119.8±87.0	214.2±75.8	p<0.0001
TMT-B-A difference score (seconds)	68.3±63.1	159.21±66.8	p<0.0001
TMT-B-A proportion score	1.42±1.15	3.21±1.42	p<0.0001
TMT-B-A Quotient score	2.42±1.15	4.21±1.42	p<0.0001
CHIPS	16.60±13.45	27.43±18.28	p=0.03
WMH %TIV	0.53±0.58	1.23±1.93	p=0.2
Infarct Volume %TIV	1.57±3.24	1.52±2.03	p=1.0
Executive Network Strokes	66% yes	86% yes	p=0.1
SLF Stroke	38% yes	43% yes	p=0.7

SLF stroke involvement

We subsequently assessed laterality of SLF and CHIPS. While age, education and CHIPS held the same strength of associations seen in Table 3, stroke involvement of the left SLF (β=0.17, p=0.03, r²=0.02) became moderately associated with difference score, but not the proportion or quotient scores. When both left and right CHIPS severity were entered into this regression model, no association was found between stroke involvement in the left SLF and the TMT-B-A difference score. The TMT-B-A proportion and quotient scores were only associated with the left CHIPS score (β=0.37, p<0.0001, r²=0.13). However, the TMT-B-A difference score was associated with both left (β=0.38, p<0.0001, r²=0.13) and right (β=0.20, p=0.007, r²=0.08) CHIPS scores.

DISCUSSION

In this combined multicentre stroke cohort study, we employed a unique approach to fractionate executive functions and study their relations to neural substrates. Using a widely used and easily administered neuropsychological test, the Trail Making Test, coupled with structural MRI, we suggest that lateral cholinergic integrity might relate to some aspects of executive dysfunction post-stroke. We found that speed of processing, sequencing, and set shifting were differentially correlated with different anatomical measurements in the setting of stroke. Slower processing speed was associated with increasing global infarct volume. Set-shifting deficit, on the other hand, was related to CHIPS severity, but not global WMH or infarct volume. It is notable that we replicated this finding in acute-subacute and chronic stroke cohorts.

We also noted that those who made ≥2 set-shifting errors on TMT-B differed only by having a greater CHIPS score. Furthermore, those who made sequencing errors on TMT-B had a greater volume of WMH, but not CHIPS severity (Table VIII). Correct sequencing may involve working memory and other executive functions such as planning and task-setting, which may account for its differential relationship to global WMH volume in this study. Furthermore, the finding that strokes in the executive network were not associated with TMT derived or error metrics of set-shifting may be related to the relationship between TMT-B and frontal lobe functions.³¹ The executive network, in addition to frontal lobe structures, includes the lateral parietal lobe, anterior thalamus, and caudate nucleus.^27,28 Overall, this study supports the role of lateral cholinergic pathways in PSCI, where attentional and executive deficits are prominent, but also suggests that there is further specificity in this relationship. Set-shifting is one crucial executive function that may be impaired by damage to lateral cholinergic pathways in the settings of acute and chronic stroke.

However, CHIPS may not reflect lateral cholinergic integrity exclusively as in some CHIPS regions (see supplementary Figure I and Figure II), the lateral cholinergic pathways run adjacently to white matter tracts of the Superior Longitudinal Fasciculus (SLF), an association tract which may also be related to executive functions.²⁵ Evidence from Diffusion Tensor Imaging (DTI) suggests that the SLF may have a role in set-shifting as measured by TMT-B,²⁴ although the authors of this work did not use TMT-B derived or error metrics to account for other cognitive processes involved in TMT-B time as we did. As CHIPS does not rate the possible influence of the SLF, we therefore explored whether stroke involvement in the SLF might be associated with set-shifting deficit. To date, no study assessing lateral cholinergic pathways has attempted to account for adjacent white matter tracts in this manner. When stroke involvement of the SLF was entered into regression models, the left SLF, along with age, education and CHIPS severity, was associated with set-shifting as measured by the difference score, but not the proportion or quotient scores. Furthermore, there were no observed differences in the proportion of patients with SLF strokes in those who made ≥2 set shifting-errors. These differing neuroanatomical correlates between different metrics of the TMT, summarized in Table IX, speaks to the utility of assessing specific cognitive constructs.

The SLF is an extensive association pathway connecting dorsolateral prefrontal areas to supplementary motor, superior temporal, parietal, and occipital areas. The finding that strokes in the left SLF were associated with the difference score, but not set shifting errors or the proportion/quotient scores, might suggest that the difference score measures efficiency in a broader network of cognitive processes mediated anatomically by lateral cholinergic pathways and the left SLF. This is also supported by the persisting association between the difference score and TMT-A, but not the quotient or proportion scores. In summary, the findings suggest that damage in lateral cholinergic pathways and perhaps the left SLF, along with regions of the frontal lobe demonstrated in previous studies, ¹¹ should be added to the list of structures implicated in set-shifting.

These results have direct relevance to patients with PSCI, where executive dysfunction can be an important clinical determinant of mortality.⁴ Cognitive neurorehabilitation targeting set-shifting may be possible. Cholinergic pharmacotherapy may also be useful in stroke patients with a high CHIPS score – a rating easily derived from clinically acquired MRI. However, it is unclear whether infarction of the external capsule region ablating lateral cholinergic pathways and a section of the left SLF may be less amenable to pharmacologic rescue. While cholinergic modulation of executive function has been demonstrated in AD⁷, trials assessing the use of CHEIs in vascular cognitive impairment have been inconclusive to date.⁸ Randomized clinical trials that incorporate image-guided patient stratification would test these hypotheses more directly.

Certain limitations of the present study suggest avenues for future research. CHIPS is a visual rating scale, though it does have good inter-rater reliability and topographic validation.^20,17 Although some regions rated by CHIPS are adjacent to associations tracts such as the SLF, the regions of greatest cholinergic density – validated by immunohistochemistry¹⁷ – contribute most to the CHIPS score. Though this is the first study to assess stroke involvement of the SLF, future studies using DTI or template based methods may be better able to account for the role of these association tracts. Furthermore, CHIPS primarily assesses the lateral cholinergic projections originating from the NbM and travelling to frontal and parietal neocortical areas, but not those lateral cholinergic projections to the temporal lobes.²⁰ Future investigations studying memory and other facets of cognition should assess these cholinergic pathways also along with projections from the septal nuclei and diagonal band nucleus – all of which may have unique roles in cognition.^18,23 Finally, using DWI to assess acute stroke volumes in the South Korean cohort may have overestimated patients’ final infarction volumes, although initial DWI volumes correlate strongly with final T2 volumes.³²

To our knowledge, this is the first study to combine international databases in accordance with the NINDS-CSN vascular cognitive impairment harmonization standards. ¹⁴ It is noteworthy that despite differences in stroke acuity and severity, years of education, language, MMSE, infarction volumes, and global WMH burden between South Korean and North American cohorts, consistent associations were found between: (i) CHIPS and set-shifting; and (ii) global infarction volume and processing speed. Replication of findings in both cohorts suggests that, despite the linguistic differences between the Korean and English versions of the TMT, both versions appear to be measuring very similar constructs of processing speed and set-shifting executive function. This supports the use of TMT as a supplementary outcome measure in international clinical trials in PSCI, where it can be used to fractionate both neuropsychological constructs commonly impaired after stroke and their differentially associated neural substrates. The TMT may also be an appropriate outcome measure to identify responsiveness to cholinergic intervention, which further underscores the utility of neuropsychological harmonization to facilitate comparison across cultural and linguistic barriers. This is also the first study to isolate set-shifting, an important executive function commonly impaired after acute and chronic stroke. We propose that set-shifting is associated with the integrity of the lateral cholinergic system and possibly the left SLF, though confirmation in additional samples is needed. This study emphasizes the importance of image-guided lesion localization and precisely defined neuropsychological parameters in the search for neuroanatomical substrates of post-stroke cognitive impairment.