Abstract
Background:
Long-term post-stroke cognitive impairment (PSCI) has often been overlooked, especially among patients with minor stroke or transient ischemic attack (TIA).
Objective:
To assess 6-year domain-specific cognitive trajectories among survivors of minor stroke or TIA and to identify possible indicators associated with cognitive trajectories, as well as long-term and incident PSCI.
Methods:
Eligible participants completed cognitive and clinical assessments at baseline (2 weeks after stroke) and up to 5 follow-up visits in 6 years. Mixed linear models and generalized estimating equations were adopted to analyze longitudinal data and survival analysis to explore incident PSCI, controlling for demographic, clinical, and vascular indicators.
Results:
The prevalence of PSCI and mortality rate ranged from 34.6% to 53.7%, and 0 to 7.7% respectively, among 244 patients. Incidence of PSCI was 21.9%. While visual memory demonstrated a significant improvement (p < 0.05), other cognitive domains showed a fluctuating yet stable pattern across visits (all ps > 0.05). Besides age, baseline IQCODE (attention: –0.218 SD/y, executive function: –0.238 SD/y, visual memory: –0.266 SD/y), and MoCA improvement within 1 year (visuoconstruction: 0.007 SD/y, verbal memory: 0.012 SD/y) were associated with longitudinal cognitive changes. Baseline MoCA (OR = 0.66, 95% CI = [0.59–0.74]), MoCA improvement within 3–6 months (OR = 0.79, 95% CI = [0.71–0.89], and within 1 year (OR = 0.86, 95% CI = [0.76–0.96]) were associated with long-term PSCI, while baseline MoCA (OR = 0.76, 95% CI = [0.61–0.96]) was also associated with incident PSCI.
Conclusion:
While most domains remained stable across-time, visual memory demonstrated an overall improvement. Short-term cognitive improvement could be an early indicator of long-term cognitive trajectory to identify individuals who may be resilient to PSCI.
Keywords
INTRODUCTION
Stroke is one the most common causes of cognitive impairment [1]. The prevalence of post-stroke cognitive impairment (PSCI) is reported in 70% of middle-aged and older patients, and is associated with hospitalization, poorer quality of life, and increased mortality [2–4]. Although having a major effect on patients’ and caregivers’ lives, PSCI has often been overlooked in the long-term management of stroke or transient ischemic attack (TIA), until progression to dementia occurs [5, 6].
Differences in cognitive trajectories after stroke have been observed and reported in a few studies looking at the longitudinal effect of stroke on cognition. While some patients recover rapidly from PSCI and maintain good cognitive functioning [7, 8], others showed persistent cognitive impairment or even further decline 10 years after stroke [9, 10]. Two recent population-based prospective studies have provided insights into the overall trend of post-stroke cognitive decline over a period of 6–10 years [9, 10], using different measures of cognition: while one study adopted a face-to-face assessment combining memory, language, and orientation into a global cognition evaluation [9], the other used a telephone-based assessment including word list learning, word list delayed recall, and animal fluency test and then a Six-Item Screener (SIS) to assess global cognition [10].
As the above two studies assessed the stroke information by medical record and did not collect data on stroke severity, limited insight into differential cognitive profiles stratified by stroke severities was rendered. Minor stroke or TIA account for up to around 70% of all acute cerebrovascular events [11]. Although their risk of cognitive impairment may be lower than those who have more severe stroke, but the risk is still higher comparing with their age-matched population [11, 12]. However, very few studies on the cognitive trajectories after minor stroke or TIA have been reported [6, 11]. Hence, studying cognitive trajectories among these patients over a long period of time is crucial, as it allows a continuous observation of the speed and magnitude of patients’ cognitive changes, which will enable the optimal windows for prompting management and intervention.
In addition, severity and magnitude of cognitive trajectory may not load evenly on cognitive domains, such as memory, executive function, and information-processing [13]. While a previous study using a comprehensive neuropsychological tests battery showed an improvement in memory and overall recovery among younger stroke survivors (between 18 to 65) [14], studies in the elderly mostly used a global cognitive assessment, providing little evidence on domain-specific cognitive trajectories. Therefore, it is crucial to examine domain-specific cognitive trajectory patterns among elderly participants, in order to optimize cognitive intervention strategies.
Hence, the present study aimed to: 1) assess the 6-year domain-specific patterns of cognitive trajectory in the whole sample of stroke survivors in a multi-ethnic clinical cohort in Singapore; 2) identify demographic, clinical, and vascular indicators of different individual trajectory patterns; 3) identify the associations of baseline indicators with long-term and incident PSCI.
METHODS
Study design
This study was performed using a prospective, cohort study design with 6 years of follow-up. There were 6 timepoints: baseline (within 14 days after stroke onset), 3–6 months (visit 1), 1 year (visit 2), 3/4 years (visit 3), 5 years (visit 4), and 6 years (visit 5) after stroke.
Participants
284 patients with a recent minor ischemic stroke or TIA (National Institute of Health Stroke Score (NIHSS) < = 4) admitted to the stroke neurology service at the National University Health System of Singapore were recruited. No information regarding each individual’s cognitive medication was reported among participants over the 6 years of follow-up. Patients completed comprehensive clinical and investigational assessments at baseline (with Montreal Cognitive Assessment (MoCA) but without comprehensive neuropsychological assessments), 3–6 months, 1 year, 3/4 years, 5 years, and 6 years after the event. Exclusion criteria are: 1) major physical disability (modified Rankin Scale (mRS) > 4); 2) significant aphasia/dysarthria (NIHSS, best language (Aphasia) and dysarthria score > 1); 3) significant sensory impairment, e.g., visual impairment and hearing impairment, etc.; and 4) only completing baseline with no subsequent follow-ups and research diagnosis.
Standard protocol approvals and patient consent
This study was approved by National Healthcare Group Domain Specific Review Board (DSRB), and written informed consent was obtained from all participants or their legally acceptable representatives.
Clinical assessments
Data collected at baseline included demographics (age, sex, race, and education level), vascular risk factors such as hyperlipidemia, hypertension and diabetes mellitus, smoking status, and previous vascular events (ischemic heart disease, atrial fibrillation, peripheral artery disease, and previous stroke or TIA). Pre-stroke cognition was assessed by the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) to compare participants’ self-perceived cognitive capacity against that of 10 years ago. Neurological status including mRS and NIHSS were evaluated.
Cognitive assessment
Brief cognitive assessment (MoCA) was performed at all time points.
Formal neuropsychological assessments were conducted at all subsequent timepoints (3–6 months, 1 year, 3/4 years, 5 years, and 6 years) except for the baseline (within 2 weeks after stroke onset). The formal neuropsychological battery was locally validated for older Singaporeans [15] and administrated by trained research psychologists. The domains of the formal neuropsychological battery included: attention (digit span [16], visual span [17], and auditory detection [18]); visuomotor speed (symbol digit modalities [19], digit cancellation [20] and maze [21]); visuoconstruction (visual reproduction subtest of the Wechsler Memory Scale-Revised copy task [22], clock drawing [23], and the block design subtest of the Wechsler Adult Intelligence Scale-Revised [24]); executive function (frontal assessment battery [25]); verbal memory (word list and story recall [26]); visual memory (picture recall and the visual reproduction subtest of the Wechsler Memory Scale-Revised [22]).
Cognitive impairment was diagnosed by consensus in formal meetings of the research team utilizing the test results listed above. Scores of individual tests below education-adjusted cut-offs of 1.5 SD were categorized as test failure. Impairment in a domain was defined as failure in at least half of the tests in that domain. Dementia was diagnosed according to the Diagnostic and Statistical Manual of Mental Disease, 4th edition [27].
We used long-term PSCI to represent our outcome which demonstrating the mean odds of PSCI between each visit. Incident PSCI was defined as newly diagnosed cognitive impairment at subsequent visits among no cognitive impairment (NCI) participants at visit 1 (3–6 months).
In order to directly compare and analyze the cognitive outcome as a continuous variable, global and domain-specific cognitive performance at 3–6 months (visit 1) were set as the reference to calculate z-scores. The same approach has been applied elsewhere in the literature [9]. Domain-specific z-scores were generated by averaging the z-scores of individual tests in specific domains and standardized using the composite mean and SD of the reference group. Therefore, a z-score of 1 would represent 1 SD above the mean score of 3–6 months after stroke. In regression analyses, the z-scores were used for comparisons of regression coefficients across different domains.
Statistical analyses
Demographic characteristics are presented as mean±SD, number with/without number of cases (%) as appropriate. Missing values of individual test among participants with cognitive diagnosis were imputed by using the mean of the specific test at that timepoint.
To describe global and domain-specific cognitive trajectories in a whole sample level, we graphed the overall means of standardized z-scores of domain-specific cognitive performances at each time point (Fig. 3). Repeated measures ANOVA was conducted to measure any difference of the means of each domain between five visits (from 3–6 months to 6 years after stroke onset). The Tukey test was used to assess any difference between two timepoints if statistical significance was found in an ANOVA test.
Mixed linear models with a random intercept and slope for each participant were used to identify possible baseline indicators of domain-specific cognitive trajectories. We fitted the first model to examine the associations of baseline indicators with domain-specific cognitive performance (domain-specific standardized z-score) by including age, education years, sex, race, baseline NIHSS, baseline IQCODE, baseline MoCA, MoCA improvement within 3–6 months and within 1 year, hypertension, hyperlipidemia, diabetes, smoke history, and previous vascular events. We then fitted the second model to identify baseline indicators associated with longitudinal change in domain-specific cognition (rate of change in domain-specific standardized z-score) across the study duration with indicators which were significant in the first model (baseline NIHSS was included due to its clinical relevance, despite statistical non-significance) and the interactions between each indicator and time [28].
Generalized estimated equation (GEE) model for a binary outcome was used to estimate the associations of baseline indicators with the odds of PSCI across the study duration (PSCI at 3–6 months, 1 year, 3/4 years, 5 years, and 6 years after stroke). Kaplan-Meier survival curve was assessed by survival analysis to calculate the rate of incident PSCI among NCI participants. We then performed multiple logistic regressions to examine associations of baseline indicators with the odds of incident PSCI. Binary outcomes (long-term PSCI versus no PSCI, incident PSCI versus no PSCI) were analyzed by logistic models.
All statistical analysis was performed using SAS 9.4, setting statistical significance level at < 0.05.
RESULTS
Baseline characteristics
Figure 1 shows the study recruitment and retention summary. Baseline characteristics are shown in Table 1. There was no significant difference in baseline characteristics between patients lost to follow-up and those who were not (Supplementary Material). Of the initial 284 patients, 23 died during follow-up. Mortality rate ranged from 0 to 7.7% at varying timepoints (Fig. 2).
Study Recruitment and Retention Summary.
Baseline Sample Characteristics
NIHSS, National Institute of Health Stroke Score; IQCODE, Informant Questionnaire on Cognitive Decline in the Elderly; MoCA, Montreal Cognitive Assessment.
Prevalence of Post-stroke cognitive impairment (PSCI) and Mortality over the 6-Year Follow-up.
Domain-Specific and Global Cognitive Trajectory among Stroke Survivors from 3 Months up to 6 Years after the Event. Values are presented as means of global and domain-specific standardized z-scores of each visit. Visit 1 = 3–6 months follow-up (n = 227); Visit 2 = 1 year follow-up (n = 198); Visit 3 = 3/4 years follow-up (n = 100); Visit 4 = 5 years follow-up (n = 104); Visit 5 = 6 years follow-up (n = 116).
A total of 244 patients who completed at least one neuropsychological assessment were included in the analysis. Prevalence of PSCI was 53.7% at 3–6 months, 43.4% at 1 year, 41.0% at 3/4 years, 34.6% at 5 years, and 38.8% at 6 years (Fig. 2).
Overall domain-specific cognitive trajectories
Results from repeated measures ANOVA showed that there was no difference between the means of standardized z-scores of cognitive performances at each visit in attention, visuomotor speed, visuoconstruction, executive function, and verbal memory (all ps > 0.05), demonstrating fluctuating yet stable trajectory patterns as a whole group in these domains across the 6 years of follow-up. On the other hand, visual memory showed a significant improvement across the 5 visits (p for trend < 0.05). Tukey tests showed that the improvement mainly occurred between 3–6 months to 1 year after stroke and remained stable in the subsequent visits. Figure 3 shows the overall cognitive trajectories across the study follow-ups.
Baseline indicators of domain-specific cognitive performance
Baseline indicators of domain-specific cognitive performance across the 6 years of follow-up are shown in Table 2. Participants with higher education were more likely to have better cognitive performances in visuoconstruction, visuomotor speed, and verbal memory across the 6 years of follow-up. No significant association was found between baseline NIHSS and domain-specific cognitive performance. Higher baseline MoCA was associated with better cognitive performances in all domains across the 6-year follow-up period. Additionally, both MoCA improvement within 3–6 months and within 1 year were significant associated with better domain-specific cognitive performances, except for visuomotor speed (Table 2).
Associations of Baseline Indicators with Domain-specific Standardized Z-scores during 6 Years of Follow-up
NIHSS, National Institute of Health Stroke Score; IQCODE, Informant Questionnaire on Cognitive Decline in the Elderly; MoCA, Montreal Cognitive Assessment. Values are presented as estimate (95% CI) on domain-based neurocognitive standardized z-scores. Bonferroni correction was applied. For domain-based cognitive performance, significance was determined as *p < 0.05/6≈0.008, †p < 0.01/6≈0.002, ‡p < 0.001/6≈0.0002.
Baseline indicators of longitudinal change in domain-specific cognition
Results from mixed linear models showed that with per year increase in baseline age, the standardized z-scores declined 0.003 SD/y, 0.002 SD/y, 0.004 SD/y, and 0.008 SD/y in attention, visuoconstruction, verbal memory, and visual memory, respectively. One unit increase in baseline IQCODE was associated with 0.218 SD/y, 0.238 SD/y, and 0.266 SD/y decline of standardized z-scores in attention, executive function, and visual memory respectively while one unit increase in MoCA improvement within 1 year was associated with 0.007 SD/y and 0.012 SD/y increase of standardized z-scores in visuoconstruction and verbal memory (Table 3).
Baseline Indicators Associated with Rate of Changes in Domain-specific Standardized Z-scores (SD/year) during 6 Years of Follow-up
IQCODE, Informant Questionnaire on Cognitive Decline in the Elderly. Values are presented as estimate (95% CI) on domain-specific standardized z-scores. Model was adjusted for time, baseline age, baseline age x time, baseline education, baseline education x time, sex, race, baseline NIHSS, baseline NIHSS x time, baseline IQCODE, baseline MoCA, baseline MoCA x time, MoCA improvement from baseline to 3-month, MoCA improvement from baseline to 3-month x time. Other indicators that did not reach statistical significance (defined as p < 0.05) in Table 2 were removed from this model; these were hypertension, hyperlipidemia, diabetes, ever smoked, and previous vascular events. Interactions which are significant are shown in the table. *p < 0.05, †p < 0.01, ‡p < 0.001.
Baseline indicators of long-term PSCI
The associations between baseline indicators of long-term PSCI within 6 years of follow-up were demonstrated in Table 4. Increasing age was associated with higher odds of long-term PSCI (OR = 1.12, 95% CI = [1.08–1.16]), while Chinese (OR = 0.38, 95% CI = [0.20–0.73]), baseline MoCA (OR = 0.66, 95% CI = [0.59–0.74]), MoCA improvement within 3–6 months (OR = 0.80, 95% CI = [0.71–0.89]) and within 1 year (OR = 0.86, 95% CI = [0.76–0.96]) were associated with lower risk of long-term PSCI across the study duration.
Indicators of long-term PSCI and Incident PSCI within 6-year Follow-ups
PSCI, Post-stroke cognitive impairment; NIHSS, National Institute of Health Stroke Score; IQCODE, Informant Questionnaire on Cognitive Decline in the Elderly; MoCA, Montreal Cognitive Assessment. Generalized estimated equation (GEE) model was used to analyze the association between baseline indicators and long-term PSCI among all patients. Multivariate logistic regression was used to analyze the association between baseline indicators and incident PSCI among no cognitive impairment (NCI) patients at the 3–6 months after stroke (visit 1). *p < 0.05, †p < 0.01, ‡p < 0.001.
Incident PSCI among NCI participants
A total of 105 participants were diagnosed as NCI at visit 1. Among whom, 23 (21.9%) developed incident PSCI during 6-years of follow-up (Fig. 4). Multiple logistic regressions for incident PSCI (Table 4) showed that higher baseline MoCA was significantly associated with lower odds (OR = 0.76, 95% CI = [0.61–0.96]) of incident PSCI, while presence of previous history of vascular events (ischemic heart disease, atrial fibrillation, peripheral artery disease, and previous stroke or TIA) was associated with 8.5 times (95% CI = [2.08–32.97]) higher odds of developing incident PSCI, adjusting for all other covariates.
Kaplan-Meier Survival Curve of Incident Post-stroke cognitive impairment (PSCI) among Stroke Survivors with No cognitive impairment (NCI) at 3–6 Months Visit over the 6-Year Follow-ups.
Sensitivity analyses
Among a total of 244 participants in the analysis, the data loss rate was 14.1% during the study course. We conducted simple linear regression imputation to adjust for missing values. Cognitive trajectories showed high consistency with Fig. 3. Additionally, there was no difference in baseline indicators of cognitive trajectories and long-term and incident PSCI (all ps > 0.05, data not shown), indicating the robustness of the models and results in the present study.
DISCUSSION
This study examined the distinct domain-specific cognitive trajectories among patients after minor stroke or TIA, with consecutive annual or biennial follow-ups for up to 6 years. We found that in the sample as a whole, while most domains showed a fluctuating yet stable pattern, visual memory domain showed an improvement over six years. Among all indicators, while lower pre-stroke cognition, as measured on the IQCODE, was associated with longitudinal cognitive decline, short-term improvement within 1-year post-stroke as assessed by MoCA was associated with longitudinal cognitive improvement. Higher baseline MoCA was a significant protector against long-term and incident PSCI.
Although previous researchers have acknowledged the need for long-term monitoring of cognition after stroke [6], most focused on the magnitude of cognitive changes over a long interval post-stroke. To our knowledge, only one study has measured domain-specific cognition repeatedly using comprehensive cognitive assessment among younger post-stroke survivors. Moreover, instead of the two previous studies which reported a persistent cognitive decline after stroke [9, 10], our results demonstrated a fluctuating yet stable pattern in domain-specific cognitive trajectories such as executive function, attention, and visuomotor speed, except for visual memory. However, as our study only included patients with minor stroke or TIA, they may have less severe brain structural damage caused by stroke, hence the patients may have been more likely to maintain their cognition or recover.
In this study, visual memory showed a significant improvement over 6 years of follow-up. Our finding is consistent with a study which reported visual memory recovery being common among stroke patients [29]. Visual memory impairment after stroke was mostly found among patients with right-hemisphere damage [29], and such patients generally showed more cognitive improvement over time, as compared to patients with left-hemisphere damage [30]. In right-hemisphere stroke patients, visual memory impairment after stroke is more likely to be caused by a lack of organizational strategy which is linked with disruption in the encoding function rather than storage or retrieval [31]. As a result, the improvement in visual memory in our cohort can be potentially explained by a rapid and sustained recovery in encoding function of visual information. Future studies could look at the varying visual impairment severities, and their impact on long-term cognitive and clinical outcomes.
Stroke severity has been reported to be an important risk factor of PSCI [11]. Previous studies found 25–50% patients with minor stroke or TIA have some PSCI during early follow-up [32] and that is consistent with our results (53.7% and 43.4% at 3–6 months and one year after stroke respectively). However, few studies of long-term PSCI among patients with minor stroke or TIA have been done [32]. We found the prevalence of PSCI was highest at the 3–6 months after stroke and then remained stable over the 6-year follow-ups, demonstrating that patients with minor stroke or TIA may have a cognitive trajectory pattern that is different from patients with more severe stroke. In our study, NIHSS was not a significant indicator of long-term PSCI. This may be due to the study focusing on patients with minor stroke or TIA, with a lower NIHSS score (from 0 to 4) which may have hampered its predictive value on long-term PSCI. Therefore, our findings suggest a stratified analysis by stroke severity when studying the longitudinal effect of stroke on cognition.
In our study, domain-specific cognitive declines are greater among our participants who are older (in domains of attention, visuoconstruction, verbal memory, and visual memory) and have lower baseline IQCODE (in domains of attention, executive function, and visual memory). Age and pre-stroke cognitive status have been reported to be associated with cognitive deterioration after stroke [33]. Our findings provide more specific information about which domains are most likely to be affected by these two factors.
We found that baseline MoCA and short-term MoCA improvement are two independent protectors against long-term PSCI and MoCA improvement within 1 year is associated with longitudinal cognitive improvement. Our finding is not only consistent with previous studies in which early MoCA predicts long-term PSCI among stroke patients [27, 35]; moreover, short-term MoCA improvement after stroke onset exhibited a strong protective effect against long-term PSCI. A previous study found that MoCA change between 3–6 months to 1 year after stroke was associated with the transitional diagnosis of PSCI [36]. In addition, repeated measures of MoCA might be capable of detecting mild cognitive impairment [37]. The increased MoCA scores may be due to brain plasticity. Therefore, cognitive recovery over a short period of time (within 1 year) could serve as an early indicator of long-term cognitive stability. Additionally, MoCA improvement may be attributed to practice effect which itself could be an indicator of cognitive recovery/decline over time [38]. Recently, a comprehensive review reported that absence of practice effect could be a potential marker of long-term PSCI [39], implying the deterioration/loss of learning ability of new materials repeated over a short period of time. Since the MoCA has been commonly used in clinical settings as a concise and accurate measurement of cognition among stroke patients, further studies are warranted to investigate in a larger-scale clinical setting to explore the predictive value of short-term MoCA change combined with other indicators in identifying stroke patients with the highest risk of long-term PSCI or dementia.
We found that 21.9% patients with NCI at 3–6 months after stroke developed incident PSCI across 6 years of follow-ups. Our result was consistent with a previous 2-year clinic-based study in which the prevalence of incident PSCI was 19.2% among NCI patients within a week after stroke [40]. Incident cognitive impairment was reported 8.1% among stroke-free NCIs in the community [41], which is less than half compared to that in post-stroke NCIs. Hence, our finding highlighted that stroke is more likely to lead to long-term incident PSCI, even among subjects who are initially cognitively unimpaired after minor stroke or TIA.
Our study has several strengths. Use of a comprehensive neuropsychological battery allowed assessment of domain-specific cognitive performance among post-stroke patients. Additionally, the 6 years of follow-ups with consecutive cognitive performance assessments enable us to obtain temporal patterns of domain-specific cognitive trajectory in order to provide in-depth evidence of the magnitude and speed of differential cognitive changes overtime. Finally, our study had a good sample size, and our results are robust in sensitivity analysis after imputing missing values.
There are also limitations in our study. Firstly, our results may not be generalized to all stroke patients because we only included patients with minor stroke or TIA. Future studies should include stroke patients with a wider range of severity. Secondly, there were 40 drop-outs after the baseline and the number of participants varied between each visit, causing potential bias by attrition and data loss. However, in our sensitivity analysis, we did not find difference in baseline characteristics between drop-outs and those who participated. Additionally, we compared the last cognitive performance between drop-outs and those who participated before lost to follow-up, and no difference was found across the 6-year follow-ups except for 3/4 years to 5 years after stroke (Supplemental Material). Therefore, our results are robust and can still provide valid information on the prevalence of PSCI among minor stroke or TIA patients. Additionally, although we did not provide any cognitive medication to our participants, we did not have information about whether cognitive medication was given besides our study and that could be a potential confounder when study the cognitive trajectory after stroke. Lastly, we did not have imaging data on stroke size and location, which hindered us from further exploring the underlying mechanism of changes among temporal domain-specific trajectories, especially in the visual memory domain.
In conclusion, in the group as a whole, fluctuating yet stable cognitive trajectories patterns were found in most cognitive domains except for visual memory among patients after minor stroke or TIA. Pre-stroke cognition and short-term cognitive improvement were associated with long-term cognitive trajectories. Baseline MoCA and short-term MoCA improvement were associated with long-term PSCI. Future studies should investigate the long-term effect of stroke on cognition stratified by stroke severity and explore the predictive value of short-term MoCA improvement in identifying individuals who may have resilience to PSCI.
