Cognitive recovery trajectories 3 months following stroke in Mexican American and non-Hispanic white adults

Abstract

Objectives:

We examined whether cognitive trajectories from 0–3 months after stroke differ between Mexican Americans (MAs) and non-Hispanic white (NHW) adults.

Materials and Methods:

The sample included 701 participants with ischemic stroke (62% MA; 38% NHW) from the population-based stroke surveillance study, the Brain Attack Surveillance in Corpus Christi (BASIC) Project, between 2008–2013. The outcome was the modified Mini Mental State Examination (3MSE, range 0–100 lower scores worse). Linear mixed effects models were utilized to examine the association between ethnicity and cognitive trajectories from 0–3 months following stroke, adjusting for confounders.

Results:

MAs were younger, had lower educational attainment, and fewer had health insurance than NHWs (all p< 0.01). A smaller proportion of MAs were rated by informants as exhibiting pre-stroke cognitive decline than NHW (p < .0.05). After accounting for confounders, MAs demonstrated lower cognitive performance at post-stroke baseline and at 3-months following stroke (−2.00; 95% CI =−3.92, −0.07). Cognitive trajectories from 0–3 months following stroke were indicative of modest cognitive recovery (increase of 0.034/day, 95% CI =0.030–0.036) and did not differ between MAs and NHWs (p = 0.68).

Conclusion:

We found no evidence that cognitive trajectories in the first three months following stroke differed between MAs and NHWs. MAs demonstrated lower cognitive performance shortly after stroke and at three months following stroke compared to NHWs. Further research is needed to identify factors contributing to ethnic disparities in cognitive outcomes after stroke.

Introduction

Cognitive impairment is common after stroke (1), is associated with increased mortality, cost of care and disability (2), and negatively impacts post-stroke quality of life(3). Mexican Americans (MAs) demonstrate worse cognitive, neurologic, and functional outcomes 90 days following stroke compared to non-Hispanic white (NHWs) (4). Although it is established that MAs demonstrate poorer 90-day cognitive outcomes following stroke than NHW, it is unclear whether MA disparities in cognitive outcomes are present immediately after stroke or whether the 90-day cognitive outcome disparities reflect a less favorable recovery trajectory in MAs. This is important so that targeted interventions can be properly timed. On average, a substantial proportion of cognitive recovery occurs in the first three months following stroke, followed by a plateau in recovery(5) and longer-term risk for accelerated cognitive decline (6). As such, the first three months following stroke represent a critical window to examine the trajectory of cognitive recovery and compare these trajectories between MAs and NHWs.

The present study aimed to evaluate ethnic differences in cognitive trajectories from post-stroke baseline to 90 days following stroke in a population-based stroke surveillance study. We hypothesized that compared to NHWs, MAs would demonstrate less robust cognitive recovery from post-stroke baseline to 90 days after stroke.

Methods

Participants and study setting:

The sample included participants from the Brain Attack Surveillance in Corpus Christi (BASIC) Project who had ischemic stroke between 2008–2013, when we stopped collecting baseline cognitive information. Details regarding study methodology are available elsewhere(7). Briefly, the BASIC project is a longitudinal, population-based stroke surveillance study in Nueces County, Texas, USA, a community that is approximately 60% MA and predominantly non-immigrant. Case ascertainment occurs through active and passive surveillance.

Procedure:

Baseline interviews were typically completed within 1–2 weeks after initial hospital presentation for stroke. Proxy interviews were performed for those that could not participate in the cognitive assessment (e.g., death, coma, aphasia, severe cognitive impairment, etc). All participants or their surrogates provided informed consent.

Cognitive assessment:

Cognition was assessed with the Modified Mini Mental State Examination (3MSE(8)) at two time points: post-stroke baseline (time 0) and at the 90-day follow-up (time 1). 3MSE scores range from 0–100, with higher scores indicative of better cognitive performance. Assessments were provided in English or Spanish(9), according to participant preference. Participants with motor impairments that could participate in all components of the cognitive assessment except for items requiring writing (pentagon copy; sentence writing) were included (n = 29 at baseline; n = 28 at 90 days) and scores were prorated (number correct/number attempted × 100)(10).

Covariates:

Variables collected from the medical record included age, sex, insurance status (yes/no), stroke risk factors (history of stroke/TIA, hypertension, diabetes mellitus, coronary artery disease, atrial fibrillation, high cholesterol, body mass index [BMI], smoking, excessive alcohol use), initial stroke severity [NIHSS], and nursing home residence before stroke. A comorbidity index was created by summing several individual risk factors and comorbidities (cancer, COPD, Alzheimer’s disease/dementia, congestive heart failure, Parkinson’s disease, end stage renal disease, amyotrophic lateral sclerosis, epilepsy, myocardial infarction/coronary artery disease, hypertension, diabetes, atrial fibrillation, high cholesterol, and alcohol use; index range 0–14). Variables collected at the baseline interview included race-ethnicity (Census-defined categories), educational attainment, marital status, modified Rankin scale, and informant-rating of pre-stroke cognitive decline (Informant Questionnaire on Cognitive Decline in the Elderly, IQCODE(11); scale 1–5; higher scores represent worse pre-stroke cognitive decline). Depression history (binary variable indicating presence or absence of any depression history) was measured from combined information from the medical record and baseline interview. Participants were coded as positive for depression history if there was a current or prior history of depression and/or medication prescribed for depression reported during the baseline interview.

Analysis:

We used chi-square (for categorical variables) and independent sample t-tests (for continuous variables) to test for differences in clinical and demographic information across groups. A series of linear mixed-effects models were performed to examine ethnic differences in post-stroke 3MSE scores from Time 0 to Time 1. Model 1 included ethnicity and ethnicity*time (days since baseline). Model 2 added demographics (age, sex, education, marital status) and insurance status to Model 1. Model 3 added medical covariates (NIHSS, hypertension, diabetes, coronary artery disease, atrial fibrillation, high cholesterol, smoking, BMI, comorbidity score, depression history, IQCODE) to Model 2. Given that age, stroke severity, and pre-stroke cognitive impairment affect cognitive trajectories after stroke(12, 13), we also included interactions between age and time, NIHSS and time, education and time, and IQCODE and time. Non-significant (p >.05) covariates were removed from the fully adjusted model (Model 3) to create a parsimonious model. We retained ethnicity and ethnicity*time by design regardless of significance. Hypertension, diabetes, atrial fibrillation, smoking history, BMI, depression history, age*time, NIHSS*time, education*time, and IQCODE*time were non-significant and were removed from the final model. We repeated our analyses with complete cases as a sensitivity analysis. The primary pre-specified outcome was the main effect of ethnicity and ethnicity*time interaction for 3MSE in the fully adjusted model.

Missing values.

A higher proportion of cognitive data were missing for individuals with severe cognitive impairment, which could bias results. We addressed attrition at baseline and outcome using inverse probability weighting (IPW)(14). Weights for each source of attrition were calculated as the inverse of the predicted values of stepwise logistic regression, stabilized by their mean predicted probabilities. The weights were then trimmed at the 1^st and 99^th percentiles. Following stabilization and trimming, final weights were generated by the multiplication of the baseline and outcome weights.

Ten datasets were generated after the use of multiple imputation to generate values for missing 3MSE values (64 missing at baseline, 26 missing at outcome) and covariate values. Imputed covariates included education (n=1), insurance (n=27), marital status (n=1), IQCODE (n=78), BMI (n=1), and pre-stroke depression (n=63). Results from each imputed dataset were pooled with Rubin’s rules. Participants with missing data for both cognitive outcome time points were excluded (n = 147; Figure 1).

Figure 1.

Flow diagram of eligible and included participants.

Standard Protocol Approvals, Registrations, and Patient Consents:

Informed consent was obtained from all participants or their surrogates to participate in the study. The BASIC project was approved by the institutional review boards at the University of Michigan and the local hospital systems.

Data Availability:

Reasonable requests for de-identified data will be considered by the investigators based on existing data use agreements and consent documents. Interested parties should contact the corresponding author.

Results

Participants:

Sample derivation is available in Figure 1 and descriptive characteristics are available in Table 1. The sample included 701 participants (62.3% MA, 37.7% NHW). Participants with race/ethnicity other than NHW or MA were excluded due to low numbers. MAs were younger, had lower educational attainment, and fewer had health insurance than NHWs. A smaller proportion of MAs were rated by an informant as having pre-stroke cognitive impairment compared to NHWs (IQCODE: 45% of MA and 35% of NHW classified as normal pre-stroke cognition(15); Table 1). There was no ethnic difference in the time from stroke presentation to baseline interview (NHW median (IQR) = 9 (4, 21); MA median (IQR) = 9 (4, 21), p = 0.51). Most MA participants completed the 3MSE in English (86.0% at baseline interview; 88.1% at outcome interview). No participants in either group received acute intervention with tPA or mechanical thrombectomy (n = 36 were missing data).

Table 1.

Sample characteristics

		NHW (N = 264) N or Median (% or Q1, Q3)	MA (N = 437) N or Median (% or Q1, Q3)	p value
Age		70 (59, 80)	64 (56, 73)	<.0001
Sex (female)		131 (49.62)	201 (46)	0.35
Education1	Less than high school	34 (12.88)	207 (47.48)	<.0001
	High school	89 (33.71)	119 (27.29)
	Vocational/some college	89 (33.71)	74 (16.97)
	College or more	52 (19.7)	36 (8.26)
Insurance (insured)		228 (88.37)	332 (79.81)	<0.01
Marital status	Single/Never Married	13 (4.92)	30 (6.88)	0.19
	Married/Living with Someone	132 (50)	228 (52.29)
	Widowed	68 (25.76)	84 (19.27)
	Divorced/Separated	51 (19.32)	94 (21.56)
Nursing home resident		3 (1.14)	3 (0.69)	0.68
MRS	no disability	138 (52.27)	233 (53.32)	0.96
	slight to moderate disability	110 (41.67)	179 (40.96)
	severe disability	16 (6.06)	25 (5.72)
IQCODE	Missing	33 (12.5)	45 (10.3)	<0.05
	Normal	92 (34.85)	196 (44.85)
	cognitive impairment no dementia	90 (34.09)	137 (31.35)
	Dementia	49 (18.56)	59 (13.5)
NIHSS	Mild	181 (68.56)	307 (70.25)	0.64
	moderate/severe	83 (31.44)	130 (29.75)
History of stroke		70 (26.52)	124 (28.38)	0.59
Hypertension		190 (71.97)	366 (83.75)	<0.001
Diabetes		76 (28.79)	257 (58.81)	<.0001
Coronary Artery Disease		73 (27.65)	136 (31.12)	0.33
Atrial fibrillation		50 (18.94)	38 (8.7)	<0.0001
High cholesterol		147 (55.68)	217 (49.66)	0.12
Smoking	Never	140 (53.03)	282 (64.53)	<0.01
	Current	71 (26.89)	96 (21.97)
	Former	53 (20.08)	59 (13.5)
Excessive Alcohol		25 (9.47)	38 (8.7)	0.73
Body Mass Index	underweight	4 (1.52)	4 (0.92)	<.0001
	normal	76 (28.79)	66 (15.14)
	overweight	98 (37.12)	154 (35.32)
	obese	86 (32.58)	212 (48.62)
Comorbidity score		3 (1, 4)	3 (2, 4)	0.22
Depression history	no history and never took medication	156 (65.55)	268 (67)
	history of depression or currently taking medication	82 (34.45)	132 (33)	0.71
Days from presentation date to baseline interview		9 (4, 21)	9 (4, 21)	0.51
Baseline 3MSE score		90 (83, 95)	86 (77, 92)	<.0001
3-Month 3MSE score		93 (84, 97)	88 (79, 94)	0.0036

Note:

¹Education data was missing for n = 1 MA participant. IQCODE = Informant Questionnaire on Cognitive Decline in the Elderly. MRS = Modified Rankin Scale. NIHSS = National Institute of Health Stroke Scale. We used chi-square (for categorical variables) and independent sample t-tests (for continuous variables) to test for differences across groups.

Post-stroke cognitive functioning:

NHW participants had higher 3MSE scores at post-stroke baseline (NHW median (IQR) = 90 (83, 95); MA median (IQR) = 86 (77, 92), p < 0.0001) and at 3 months following stroke (NHW median (IQR) = 93 (84, 97), MA median (IQR) = 88 (79, 94), p = 0.0036; Table 1). Overall, modest cognitive improvement was observed from post-stroke baseline to the 90-day assessment (3.03 points improvement over 90 days; 95% CI 1.42 to 4.64; Model 3). There were no ethnic differences in the slope of cognitive recovery in unadjusted (ethnicity*time, p =0.61) or fully-adjusted models (ethnicity*time, p = 0.68), the pre-specified primary outcome. Age, education, stroke severity, and pre-stroke cognitive function were not associated with differences in cognitive recovery slopes. Table 2 displays the results from the linear mixed-effects models. MAs had poorer cognitive function scores than NHWs at post-stroke baseline (Model 1). This ethnic difference in cognition was attenuated after accounting for sociodemographic differences (Model 2) but was significant in the fully adjusted model that included pre-stroke cognitive function (IQCODE) and medical comorbidities as covariates (Model 3). In the fully adjusted model, the ethnic difference was significant at baseline (−2.00, 95% CI −3.93 to −0.07, p = .04) but not at 3-month follow-up (at median 77 days from baseline: −1.64, 95% CI −0.37 to 0.37, p = .11). This ethnic difference in cognitive scores did not meet a suggested threshold for a clinically meaningful difference on the 3MSE (5 points(16)). Older age (2.95 points lower per decade of age; 95% CI 2.88 to 3.02), lower education (5.93 points lower for those with less than high school education compared to those with high school education; 95% CI 3.92 to 7.9 points), and greater pre-stroke cognitive decline (4.96 points lower for 1-point increase in IQCODE score; 95% CI 3.02 to 6.9) were associated with lower cognitive function scores.

Table 2.

Linear mixed effects models for the association between ethnicity and cognitive performance

		Model 1 Estimate (SE)	Model 2 Estimate (SE)	Model 3 Estimate (SE)
Intercept		85.46 (0.79)***	85.95 (1.12)***	90.87 (1.24)***
MA1		−3.57 (0.99)***	−1.38 (0.97)	−2.00 (0.98)*
Days since baseline		0.03 (0.01)***	0.04 (0.01)***	0.03 (0.01)***
Days since baseline* MA		0.01 (0.01)		0.005 (0.01)
Age (centered)			−0.35 (0.04)***	−0.30 (0.04)***
Sex2	Female		−0.42 (0.90)
Education3	College		4.07 (1.46)**	3.91 (1.36)**
	Some college/vocational		2.99 (1.18)*	2.63 (1.09)*
	Less than high school		−6.75 (1.10)***	−5.93 (1.02)***
Insurance (no)4			−0.62 (1.25)
Marital status5	Never married		−1.38 (1.84)
	Widowed		0.11 (1.20)
	Divorced/separated		−0.61 (1.13)
IQCODE (centered; baseline)				−4.96 (0.99)***
NIHSS				−0.48 (0.08)***
Coronary artery disease				2.65 (1.01)**
High cholesterol				2.99 (0.96)**
Comorbidity score				−1.58 (0.37)***

^***p < .001

^**p < .01

^*p < .05.

¹Reference category is NHW.

²Reference category is male.

³Reference category is high school educational attainment.

⁴Reference category is has insurance.

⁵Reference category is married.

Sensitivity analyses:

We repeated our analyses with participants with complete data on both cognitive outcomes. Supplemental Table 1 displays demographic characteristics of this sample and Supplemental Table 2 displays model results. In this model, the main effect of ethnicity in the fully adjusted model (Model 3) was weaker in magnitude (p = 0.06). All other main effects and interactions remained consistent with the primary models.

Discussion

We examined cognitive recovery trajectories from post-stroke baseline through 90 days post-stroke in MA and NHW adults in a population-based stroke surveillance study. MA participants demonstrated poorer cognitive outcomes soon after stroke compared to NHWs. This difference persisted at 90 days after stroke. We found no evidence of ethnic differences in cognitive trajectories from post-stroke baseline to 90 days after stroke.

The fact that MAs demonstrate worse cognitive outcomes following stroke than NHW and that these disparities are evident as soon as 1–2 weeks after stroke may suggest that MAs are in need of more acute stroke, stroke hospitalization and early intensive rehabilitation services than NHWs, although evidence suggests that MAs receive lower quality stroke care(17) and less intensive rehabilitation services than NHW(18). MAs may not currently receive sufficient services to narrow and/or eliminate post-stroke cognitive disparities through rehabilitation. Another possible explanation for lower cognitive outcomes in MAs compared to NHWs is that these differences at least partially reflect pre-stroke cognitive test performance differences rather than differences fully attributable to stroke. The informant provided measure of pre-stroke cognitive function, the IQCODE, actually found better pre-stroke cognitive function in MAs compared with NHWs. Although we asked informants to rate whether participants experienced pre-stroke cognitive decline, we did not measure cognitive test performance pre-stroke, so we cannot completely rule out the possibility of worse pre-stroke cognitive test performance in MAs compared with NHWs. MAs are disproportionately impacted by several risk factors that adversely impact cognitive test performance across the life course, such as disparities in access to quality education and health care, and adverse brain health consequences of chronic stress related to exposure to discrimination(14). In addition, existing tools for measuring cognitive health have been under-validated in diverse communities, and educational, cultural and linguistic factors may impact precision of measurement of cognition for MA individuals(19). Supporting the possibility that the cognitive test performance may have been influenced by pre-stroke factors, we found that education and informant-rated pre-stroke cognitive decline were the strongest factors influencing cognitive test scores, and of greater magnitude than factors such as stroke severity and medical risk factors.

We found relatively modest cognitive recovery overall from post-stroke baseline to the 90-day follow-up. Available evidence suggests that approximately 80% of the acute cognitive impairment recovers at three months after stroke(5). We considered whether our finding of modest recovery was impacted by inclusion of participants with pre-stroke cognitive decline, given that approximately 15% of our sample had informant-rated pre-stroke cognitive decline that exceeded screening cut-offs for dementia. However, we did not find an association between informant-rated pre-stroke cognitive decline and post-stroke cognitive trajectory. It is also possible that the 3MSE instrument is not optimally sensitive to measure acute cognitive changes following stroke. The 3MSE has shown substantial ceiling effects(20), suggesting that this instrument may not optimally capture subtle cognitive changes after stroke and associated subtle cognitive recovery. In addition, evidence suggests that post-stroke cognitive recovery varies by cognitive domain(21), suggesting that a more comprehensive neuropsychological assessment may be useful for future efforts to quantify ethnic differences post-stroke cognitive recovery.

Although cognitive function is less often measured acutely after stroke given that many factors can impact cognitive assessment engagement (e.g., fatigue, delirium), prior work has indicated that acute cognitive assessment following stroke is predictive of longer-term cognitive and functional outcomes (22, 23). Further research is needed to measure cognitive outcomes and their trajectories in the first three months following stroke, particularly given the sensitivity of this window to cognitive recovery. Future examination of acute cognitive recovery trajectories in MA and NHW may offer insights regarding factors contributing to ethnic disparities in long-term post-stroke cognitive outcomes, which may then identify strategies to optimize cognitive recovery and reduce cognitive outcome disparities following stroke. Future work is also needed to identify optimal tools to measure cognitive health following stroke that are sensitive to subtle cognitive impairment and change, and can be interpreted equitably across culturally, linguistically, and educationally diverse communities. These findings also underscore the need for stroke prevention and aggressive acute stroke treatment to reduce the impact of stroke on cognitive health.

Our study has limitations. Our sample included only participants who could participate in cognitive assessment either at post-stroke baseline or at 3-month outcomes. As such, individuals with the most severe cognitive impairment were not included as they were not able to participate in formal cognitive assessment. We used multiple imputation to estimate missing values and inverse probability weighting to minimize selection bias due to participation. Nonetheless, it is unclear whether these findings would generalize to those with the most severe cognitive sequelae following stroke.

In summary, our findings indicate that MA disparities in cognitive outcomes are present from post-stroke baseline and that cognitive recovery is modest and occurs at a similar course among MAs and NHWs during the three months after stroke on average. Education and pre-stroke cognitive function were stronger contributors to cognitive test scores than stroke severity or medical risk factors. Future research is needed to identify factors contributing to ethnic disparities in cognitive health, both prior to and following stroke, and for stroke prevention and aggressive acute and post-acute stroke care to reduce cognitive consequences of stroke. Future research is also needed to identify optimal methods for measurement of cognitive outcomes following stroke that are sensitive to subtle cognitive change and may be interpreted equitably across culturally, linguistically, and educationally diverse stroke survivors.

Figure 2.

Race/ethnicity-specific estimates of 3MSE over time with 95% CI, modeled by unadjusted (Panel A) and adjusted (Panel B) linear regression with continuous time and an interaction term between race/ethnicity and days since baseline. Results from multiple imputed datasets were combined using Rubin’s rules.

Highlights.

• Mexican Americans showed lower cognitive performance compared to non-Hispanic white adults at post-stroke baseline and 3 months following stroke.

• Cognitive trajectories from 0–3 months following stroke did not differ between Mexican Americans and non-Hispanic white adults.

• Further research is needed to identify factors contributing to ethnic disparities in cognitive outcomes after stroke.