Neurological, Functional, and Cognitive Stroke Outcomes in Mexican Americans
Background and Purpose—Our objective was to compare neurological, functional, and cognitive stroke outcomes in Mexican Americans (MAs) and non-Hispanic whites using data from a population-based study.
Methods—Ischemic strokes (2008–2012) were identified from the Brain Attack Surveillance in Corpus Christi (BASIC) Project. Data were collected from patient or proxy interviews (conducted at baseline and 90 days poststroke) and medical records. Ethnic differences in neurological (National Institutes of Health Stroke Scale: range, 0–44; higher scores worse), functional (activities of daily living/instrumental activities of daily living score: range, 1–4; higher scores worse), and cognitive (Modified Mini-Mental State Examination: range, 0–100; lower scores worse) outcomes were assessed with Tobit or linear regression adjusted for demographics and clinical factors.
Results—A total of 513, 510, and 415 subjects had complete data for neurological, functional, and cognitive outcomes and covariates, respectively. Median age was 66 (interquartile range, 57–78); 64% were MAs. In MAs, median National Institutes of Health Stroke Scale, activities of daily living/instrumental activities of daily living, and Modified Mini-Mental State Examination score were 3 (interquartile range, 1–6), 2.5 (interquartile range, 1.6–3.5), and 88 (interquartile range, 76–94), respectively. MAs scored 48% worse (95% CI, 23%–78%) on National Institutes of Health Stroke Scale, 0.36 points worse (95% CI, 0.16–0.57) on activities of daily living/instrumental activities of daily living score, and 3.39 points worse (95% CI, 0.35–6.43) on Modified Mini-Mental State Examination than non-Hispanic whites after multivariable adjustment.
Conclusions—MAs scored worse than non-Hispanic whites on all outcomes after adjustment for confounding factors; differences were only partially explained by ethnic differences in survival. These findings in combination with the increased stroke risk in MAs suggest that the public health burden of stroke in this growing population is substantial.
Mexican Americans (MAs), the largest subgroup of Hispanic Americans, have an increased stroke risk compared with non-Hispanic whites (NHWs) but experience less case fatality and longer poststroke survival.1,2 The finding of improved survival in MAs may give the false sense that MA stroke is less burdensome than stroke in NHWs, but it is possible that the tradeoff for decreased poststroke mortality is increased poststroke disability. Limited data support the hypothesis that Hispanics have poorer stroke outcomes than NHWs,3–10 but these studies were either not focused on MAs specifically,3–9 were limited to specialized populations,4,6,7,9 or did not include younger ages where the greatest ethnic difference in stroke incidence for MAs occurs.10 The objective of this study was to test whether MAs have poorer stroke outcomes than NHWs after adjustment for confounding factors in a population-based study. Secondarily, we sought to understand the extent to which any observed ethnic differences in stroke outcomes are attributed to ethnic differences in mortality.
Data are from the Brain Attack Surveillance in Corpus Christi (BASIC) Project, November 2008 through June 2012, for which methods have been published.11–13 Briefly, the BASIC Project is a population-based stroke surveillance study in a biethnic community in south Texas (Nueces County). The population of the County was 340 223 in 2010, with 61% of residents being MAs.14 This is a nonimmigrant population with most MAs being second- and third-generation US-born citizens limiting loss to follow-up because of return migration to Mexico.
Case ascertainment includes active and passive surveillance. Active surveillance involves identification of cases through daily screening of hospital admission logs, medical wards, and intensive care units. Passive surveillance involves identifying cases by searching hospital and emergency department discharge diagnoses, using International Classification of Diseases, Ninth Revision codes (430–438).15 All possible strokes undergo validation by a stroke neurologist blinded to race–ethnicity and age using source documentation. Only ischemic stroke cases were included using a standard clinical definition.16
Interview and Data Collection Procedures
Patients with stroke were invited to participate in a structured, in-person baseline interview and outcome interview conducted ≈90 days after stroke. If the patient was unable to complete an interview, a proxy interview was completed. Interviews were conducted in English or Spanish depending on patient preference. Patients who died before the outcome interview were excluded from the primary analysis but included in the secondary analysis which investigated the influence of mortality on ethnic differences in outcome. If an individual had >1 event, only the first was included. Patients with race–ethnicity other than MA or NHW were excluded because of small numbers (n=75).
Ninety-Day Outcome Measures
Neurological deficits were measured by the National Institutes of Health Stroke Scale (NIHSS) administered by a certified study coordinator. Functional outcome was assessed using scales that measure activities of daily living (ADLs) and instrumental activities of daily living (IADLs). The ADL scale measures 7 basic functional abilities (walking, bathing, grooming, eating, dressing, moving, toileting), and the IADL scale includes 15 questions related to daily functioning. Respondents self-reported level of difficulty performing each ADL/IADL task by themselves with responses including 1 (no difficulty), 2 (some difficultly), 3 (a lot of difficulty), and 4 (can only do with help). Responses were summed for each of the 22 ADL/IADL items and divided by the number of items resulting in an average score ranging from 1 to 4, in keeping with the original Likert scale for ease of interpretation.17 Global cognitive function was assessed using the Modified Mini-mental State Examination (3MSE),18 with a cut point of 80 used to classify individuals as having dementia.19 To determine ability to participate in cognitive testing, we administered a series of questions to assess language dysfunction. Patients who failed this screening were excluded from cognitive testing. Mortality was ascertained from the social security death index, Texas Department of Health, and next of kin reports.
Confounding factors were ascertained from baseline interviews and medical records. Baseline interview data included race–ethnicity using Census-defined categories, marital status, educational attainment, prestroke function, and prestroke cognitive status. Prestroke function was measured by the modified Rankin scale ascertained by asking a series of structured questions in reference to the prestroke period and categorized as 0 to 1, 2 to 3, and ≥4 (higher scores represent worse function). Prestroke cognitive status was measured using the validated 16-item Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) completed by an informant who knew the patient well and scored as the average of the individual questions resulting in a scale ranging from 1 to 5 (higher scores represent worse cognitive function).20
Medical record data included age, sex, insurance (yes/no), risk factors (history of stroke/transient ischemic attack [TIA], hypertension, diabetes mellitus, coronary artery disease, atrial fibrillation, high cholesterol, smoking, excessive alcohol), comorbidities (myocardial infarction, cancer, chronic obstructive pulmonary disease, dementia, Alzheimer disease, epilepsy, congestive heart failure, Parkinson disease, end-stage renal disease), body mass index (BMI), initial stroke severity, treatment with tissue-type plasminogen activator (tPA), nursing home residence before stroke, and do not resuscitate orders written during the hospitalization. A comorbidity index was created by summing the above individual risk factors and comorbidities (range, 0–17). Initial stroke severity was abstracted from the medical record or calculated using previously validated methods.21
Descriptive statistics were calculated for all variables by ethnicity, and differences were assessed using χ2 and Kruskal–Wallis tests. Linear regression was used to obtain age-adjusted ethnic differences for ADL and IADL subscores and for individual ADLs/IADLs and items of the NIHSS. Tobit regression was used to assess associations between ethnicity and 90-day outcomes using data from participants with complete data. Tobit regression was used to minimize bias that might result because the primary outcomes are constrained by lower and upper bounds.22 Because of skewness, NIHSS was modeled as natural logarithm of NIHSS plus 1. Because of the lack of normality of the residuals for the NIHSS model (violation of assumptions for Tobit model), we used ordinary linear regression with robust standard errors for this outcome only. Parameter estimates were back transformed to represent the percent difference in NIHSS between ethnic groups. Models were run unadjusted including only ethnicity and then adjusted for prespecified potential confounders including age, sex, education, insurance status, marital status, nursing home residence before stroke, prestroke modified Rankin scale, prestroke IQCODE, initial NIHSS, risk factors, BMI, and the comorbidity index. Functional forms of continuous covariates were checked by testing whether higher order polynomial terms (eg, quadratic) were significant. It was determined that age, IQCODE, and BMI were appropriately modeled using a linear term and that initial NIHSS required a quadratic term.
To investigate the impact of differential mortality on the associations between ethnicity and the outcomes, we re-estimated outcome models using weights such that patients who completed the 90-day interview but had a low probability of being alive at 90 days received higher weight. Weights were constructed as the ratio of the model-predicted probability of remaining alive at 90 days as predicted by the variables included in the fully adjusted outcome models (see list above) divided by the model-predicted probability of remaining alive at 90 days as predicted by these same factors in addition to do not resuscitate status. Weights ranged from 0.27 to 7.1. We constructed 95% bootstrap confidence intervals (CIs) for all regression coefficients.
When the likelihood of missing outcomes depends on the outcome itself (eg, patients with lower cognitive scores could be less likely to complete the outcome interview),23 using data from only complete cases may produce biased results. We conducted sensitivity analyses by modeling the probability of missing outcome values as dependent on the outcome itself.23 We assumed that those missing functional outcome data would have 0, 0.14, 0.43, or 0.75 points higher ADL/IADL scores (higher disability) after adjusting for covariates. Similarly, we assumed that those missing cognitive outcome would have 0, 2.6, 7.8, or 15.1 points lower 3MSE scores and that those missing NIHSS would have 0%, 11%, 37%, or 146% higher NIHSS. Zero difference is equivalent to data missing at random. We used multiple imputation to fill in missing values of covariates under the assumption that covariates were missing at random.
All patients provided written informed consent, and the study was approved by the institutional review boards at the University of Michigan and the local hospitals.
There were 1198 MA and NHW patients with ischemic stroke during the study period, with 842 (70.3%) agreeing to be interviewed. Mortality at 90 days was 14.5% (n=122), resulting in 720 patients eligible for the outcome interview. Of the 720 (461 MAs, 259 NHWs), 57 patients (7.9%) refused participation in the outcome interview and 69 (9/6%) patients (or their proxies) could not be located for the outcome interview (note 3 of the 69 patients completed the NIHSS but then requested a proxy to complete the interview but a proxy could not be located). Twenty-four (3.3%), 25 (3.5%), and 123 (17.1%) patients had missing or incomplete outcome data for neurological, functional, and cognitive outcomes, respectively. Thus, of the 720 eligible, 573 (79.6%), 569 (79.0%), and 471 (65.4%), had data for neurological, functional, and cognitive outcomes, respectively. Because of missing data on covariates, models with complete cases were estimated with sample sizes of 513 (89.5% of 573), 510 (89.6% of 569), and 415 (88.1% of 471) for neurological, functional, and cognitive outcomes, respectively. Twenty percent of baseline and 21% of outcome interviews were collected from a proxy.
Table 1 displays baseline characteristics by ethnicity for patients with neurological outcome data (n=573). MAs were younger, had lower educational attainment, were less likely to be treated with tPA, to be a former/current smoker, and to have atrial fibrillation than NHWs. MAs were more likely to have diabetes mellitus, hypertension, and higher BMI than NHWs. Patients with neurological outcome data were more likely to be MA, younger, have a higher BMI, and have a lower initial NIHSS, and less likely to have a history of stroke/TIA than patients who survived to 90 days but were not included in this analysis (Table I in the online-only Data Supplement).
Twenty percent of eligible MAs (n=416) and 21% of eligible NHWs (n=259) did not have an NIHSS at 90 days. Median NIHSS was 3 (interquartile range [IQR], 1–6) in MAs and 2 (IQR, 0–4) in NHWs. MAs on average scored higher (worse) on all NIHSS elements with the exception of extinction/inattention (Table II in the online-only Data Supplement). A significant age-adjusted ethnic difference was noted for level of consciousness items, visual, motor, language, and dysarthria. Among those with complete data (n=513), in unadjusted analysis MAs had a 42% (CI, 18%–71%) higher NIHSS than NHWs. The ethnic difference persisted and became stronger after multivariable adjustment (48%; CI, 23%–78%; Table 2). Prestroke function, initial NIHSS, and history of stroke/TIA were associated with worse NIHSS at 90 days, whereas tPA treatment was associated with lower NIHSS at 90 days. Prestroke cognitive status (IQCODE) demonstrated a borderline significant association with NIHSS at 90 days.
Twenty percent of eligible MAs (n=461) and 23% of eligible NHWs (n=259) did not have functional outcome data at 90 days. Median ADL/IADL score was 2.5 (IQR, 1.6–3.5) in MAs and 2.1 (IQR, 1.2–3.0) in NHWs. MAs on average scored higher (worse) and reported a greater frequency of “can only do with help” (4 on Likert scale) on all ADLs/IADLs (Table III in the online-only Data Supplement), with significant age-adjusted ethnic differences noted for all items. Among those with complete data (n=510), the unadjusted ethnic difference (MA versus NHW) in ADL/IADL score was 0.30 (95% CI, 0.08–0.52). After multivariable adjustment, the ethnic difference became stronger (0.36; 95% CI, 0.16–0.57) and remained statistically significant (Table 2). Age, female sex, prestroke function, prestroke cognitive status (IQCODE), comorbidity index, initial NIHSS, and history of stroke/TIA were associated with worse ADL/IADL score at 90 days, whereas tPA treatment was associated with better ADL/IADL score at 90 days.
Of the subjects eligible for the cognitive testing (n=575), 104 subjects failed the language screening (MAs, n=78; NHWs, n=26). In total, 37% of eligible MAs (n=461) and 31% of eligible NHWs (n=259) did not have cognitive outcome data at 90 days. Among those without significant language dysfunction, median 3MSE score was 88 (IQR, 76–94) in MAs and 92 (IQR, 79–96) in NHWs. Thirty-one percent of MAs had poststroke dementia compared with 25% of NHWs. Among those with complete data (n=415), MAs had on average 1.9 (CI, −1.05 to 4.98) points lower (worse) 3MSE scores compared with NHWs in the unadjusted analysis. After multivariable adjustment (Table 2), the ethnic association became stronger and was statistically significant with MAs having on average 3.39 points lower 3MSE scores (CI, 0.35–6.43). Age, nursing home residence before stroke, prestroke cognitive status (IQCODE), comorbidity index, and initial NIHSS were associated with worse 3MSE scores, whereas increasing education, tPA treatment, and coronary artery disease were associated with better 3MSE scores.
After re-estimating the outcome models using weights to account for differential poststroke mortality by ethnicity, ethnic differences in all outcomes were attenuated but statistically significant. MAs had a 39% (CI, 30%–48%) higher NIHSS, 0.22 (CI, 0.15–0.30) higher ADL/IADL score, and 2.61 (CI, 1.44–3.44) points lower 3MSE scores compared with NHWs after accounting for mortality differences.
Ethnic differences were slightly attenuated after accounting for the possibility of bias because of missing data, but remained statistically significant across various assumed strengths of the association between the missing values for the outcomes and the likelihood of missing values (Figure).
MAs have increased stroke risk but lower case fatality and longer poststroke survival compared with NHWs.1,2,13 Our current results suggest that this prolonged survival is at the expense of poor outcomes because MA stroke survivors experienced poorer neurological, functional, and cognitive outcomes than NHWs even after adjustment for a comprehensive list of factors. Differential mortality by ethnicity explained some but not all of the observed ethnic differences in outcomes, suggesting that research on the causes of poorer outcomes in MAs compared with that of NHWs is warranted.
The results of this study provide information on the prognosis of MA stroke survivors. On average at 90 days poststroke, MAs experience mild neurological and cognitive impairment, but one third of MA stroke survivors were classified as having dementia, and MAs experienced more aphasia than NHWs. Levels of functional impairment were more substantial. MAs reported greater difficulty than NHWs with all ADLs and IADLs. Averages for several IADL items were close to or >3 indicating many MAs have difficulty doing these tasks on their own. These findings are of particular importance given that increasing ADL/IADL scores are highly predictive of nursing home admission and the need for informal care.24,25
Our findings for MAs regarding dependence in specific ADLs/IADLs are similar but somewhat lower (ie, better function) than those reported for elderly MAs with self-reported stroke in the Hispanic Established Populations for the Epidemiologic Study of the Elderly (EPESE), possibly attributable to the younger average age of our MA population.10 A limited number of studies have reported that Hispanics have poorer functional and cognitive outcomes than NHWs after stroke.3–10 Our results build on these studies by providing data on outcomes from a population-based stroke study specifically focused on MAs, the largest and fastest growing subgroup of Hispanic Americans in the United States.
Although many studies have measured stroke outcomes in NHWs, comparisons with our results are challenging attributable to the differing study populations, time frames for outcome assessments, and measures used. Our results regarding dependence in specific ADLs for NHWs are similar but slightly lower (ie, better function) than those reported in stroke survivors from Framingham, potentially caused by our NHWs being several years younger on average.26 Comparability of our results for NHWs to those in Framingham suggests that our findings regarding ethnic differences are not likely attributable to more favorable outcomes in our particular population of NHWs.
Given the rapidly growing MA population, increased stroke risk in this population in combination with prolonged survival and increased disability will result in an escalating number of MA stroke survivors requiring assistance. Studies have shown that MAs are less likely to be admitted to a nursing home, suggesting that informal care may be particularly important in this group; however, there are virtually no data available on the informal stroke caregiving experience of MAs.25,27,28 This topic should be a target of future research to understand the impact of informal stroke caregiving on both caregivers and patients.
Strengths of this study include the population-based design, the nonimmigrant population limiting return migration, comprehensive outcome ascertainment, thorough adjustment for confounding factors, and sensitivity analysis to understand the impact of potential selection bias. There are some limitations that warrant discussion. We did not have data on the psychosocial impacts of stroke, such as depression, or on poststroke rehabilitation, both of which may impact outcomes and differ by ethnicity.6,7,9,29–31 These factors are important targets for future research. We did not have data on ischemic stroke subtype, although we have previously demonstrated no ethnic differences in stroke subtype in this community suggesting this factor does not explain ethnic differences.32 As in any prospective observational study, there was some loss to follow up and there were differences noted between patients included in our primary analysis and those who were not included with respect to age, ethnicity, history of stroke/TIA, initial stroke severity, and BMI. Importantly, we included these factors in our multivariable models, and our sensitivity analysis demonstrated that our results regarding ethnic differences were robust to missing data. Our outcome measures were broad measures of neurological, functional, and cognitive outcomes chosen for their validity and previous use in MAs or Hispanics. Given our findings of ethnic differences in these broad measures, future research should aim to unravel the more subtle differences that might be behind these disparities. Our measure of functional outcome was self-reported, and thus measurement error is possible. It is possible that our observed ethnic difference in cognitive outcome was attributable to potential cultural bias in the test or noncognitive factors.33 Educational attainment is a potent cognitive confounder but was included in our multivariable model. Our data on risk factors were collected from medical records only versus objective measurements which may have resulted in residual confounding by these factors. However, we have previously demonstrated that access to medical care in this community is high in both ethnic groups, suggesting that there are not large differences in the likelihood of diagnosis,34 although there may be differences in risk factor control that we did not account for in our analysis. Finally, the study is focused on 1 community in south Texas where the majority of MAs are second- and third-generation citizens, and therefore, results may not be generalizable to other populations or to immigrant MAs.
MA patients with stroke experienced moderate functional disability and nearly one third had poststroke dementia. In addition, MA patients with stroke experienced worse neurological, functional, and cognitive outcomes at 90 days than NHWs. Increased stroke risk, prolonged poststroke survival, and increased poststroke disability suggest that the future public health burden of stroke in the growing and aging MA population will be staggering.
Sources of Funding
Drs Lisabeth, Sánchez, Smith, and Morgenstern are funded by National Institutes of Health R01 NS38916 (grant funding this work).
Dr Lisabeth is funded by National Institutes of Health R01 NS062675, R01 HL098065, and R01 NS070941. Dr Brown is funded by National Institutes of Health R01 NS062675, R01 HL098065, R01 NS070941. The other authors report no conflicts.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.113.003912/-/DC1.
- Received October 21, 2013.
- Revision received January 6, 2014.
- Accepted January 28, 2014.
- © 2014 American Heart Association, Inc.
- Morgenstern LB,
- Smith MA,
- Lisabeth LD,
- Risser JM,
- Uchino K,
- Garcia N,
- et al
- Ottenbacher KJ,
- Campbell J,
- Kuo YF,
- Deutsch A,
- Ostir GV,
- Granger CV
- Dhamoon MS,
- Moon YP,
- Paik MC,
- Boden-Albala B,
- Rundek T,
- Sacco RL,
- et al
- Piriyawat P,
- Smajsová M,
- Smith MA,
- Pallegar S,
- Al-Wabil A,
- Garcia NM,
- et al
- 14.↵United States Census Bureau. State and County Quickfacts Nueces County, Texas. http://quickfacts.Census.Gov/qfd/states/48/48355.html. Accessed December 19, 2013.
- 15.↵World Health Organization. MONICA Manual, Part IV: Event Registration. Section 3: Event Registration Quality Assurance Methods. Geneva, Switzerland: World Health Organization; 1990. http://www.Thl.Fi/publications/monica/manual/part4/iv-3.htm. Accessed December 19, 2013.
- Williams LS,
- Yilmaz EY,
- Lopez-Yunez AM
- Little R,
- Rubin D
- Petrea RE,
- Beiser AS,
- Seshadri S,
- Kelly-Hayes M,
- Kase CS,
- Wolf PA
- House A,
- Knapp P,
- Bamford J,
- Vail A
- Willey JZ,
- Disla N,
- Moon YP,
- Paik MC,
- Sacco RL,
- Boden-Albala B,
- et al
- Black SA,
- Espino DV,
- Mahurin R,
- Lichtenstein MJ,
- Hazuda HP,
- Fabrizio D,
- et al
- Smith MA,
- Risser JM,
- Lisabeth LD,
- Moyé LA,
- Morgenstern LB