|Home | About | Journals | Submit | Contact Us | Français|
Objectives: to examine the association of language (English vs Spanish), and commonly used measures of memory and word fluency among older adults.
Setting: community-based settings in New York City, including senior centres and residential complexes.
Subjects: four hundred and twenty independently living adults aged 60 or older (mean 73.8 years).
Methods: participants completed the Mini-Mental State Exam (MMSE), animal naming test (ANT) and Wechsler Memory Scale III (WMS) Story A immediate and delayed subtests. Scores were examined by strata of language, age or education and for different thresholds of the MMSE. We tested the association of language and cognitive test performance using multivariable linear regression.
Results: twenty-one per cent of subjects were interviewed in Spanish and 16.2% reported poor-fair English proficiency. The mean WMS scores were not statistically different between English and Spanish groups (immediate recall, 9.9 vs 9.5, P = 0.44; delayed recall, 8.0 vs 7.6, P = 0.36, respectively), whereas ANT scores did differ (16.6 vs 14.3, P < 0.0001). These associations were consistent across MMSE thresholds. The association of language and ANT score was not significant after accounting for education.
Conclusions: we found little difference in performance on the Story A subtests from the WMS suggesting that this test may be used for both English- and Spanish-speaking populations. Results suggest that variations in ANT performance may be accounted for by adjusting for the level of education. These results have important implications for the generalisability of test scores among diverse older populations.
Community-based research often seeks to examine the association between cognition and health outcomes. However, this is a major challenge when multiple languages are spoken within a given cohort. Test selection must account for demographic characteristics of the sample, such as age and education that are known to affect test performance [1–3]. While the effects of age and education are well established, less is known about language of assessment, although evidence suggests that it influences test performance [3–5]. Spanish is the second most commonly spoken language in the United States with prevalence estimates of 12% of the total population  and one-third of the older population. Yet there is little normative data for US Spanish-speaking cohorts that might inform the selection of cognitive tests.
There are many variables to consider when assessing Spanish speakers in the United States such as literacy in Spanish, country of origin, education, length of time in the United States and proficiency of English [4, 5]. In studies of Spanish-speaking adults, previous investigators have found a nominal impact of language on test performance when the research subjects were equally skilled in English and Spanish language, whereas Spanish-dominant speakers have tended to perform less well on tests of memory and word fluency, suggesting that acculturation may contribute to differences in performance [4, 5]. Since the association between cognition and health outcomes may be used to improve health care information and dissemination, it is important to select assessment tools that can be used among Spanish speakers.
In this report, we took a pragmatic approach to characterise performance of Spanish speakers on cognitive tests of memory and word fluency. Spanish speakers were defined as those who preferred to be tested in Spanish. We also assessed self-reported English proficiency. While this approach may not permit the assessment of the impact of all nuances of culture and language, it reflects a practical way to include Spanish speakers in community-based research and may provide the basis for developing additional sensitivity for clinical assessments of cognition. We conducted an analysis of data from a subset of tests from established neuropsychological batteries , specifically the first paragraph from the Wechsler Memory Scale III (WMS) logical memory I and II subtest  and the animal naming test (ANT) , from a diverse population of urban, independently living older adults. We examined performance on these assessments stratified by age and education, and present data on the effect of both testing language (English or Spanish) and self-reported English proficiency.
Data for these analyses were collected in a larger study that addressed health insurance decision-making among older adults. We recruited independently living adults aged 60 years and older from 30 community-based settings in New York City, including senior centres and residential complexes. Senior centres were identified through listings from the New York City Department for the Aging and residential complexes through a listing of federal Housing and Urban Development-supported low-income housing facilities. We selected sites in zip code areas with median household incomes below $50,000, and we over-sampled men because they are outnumbered by women in these communities and in this age group and in our experience are less likely to participate in survey interviews .
Individuals were recruited during site-sponsored meals or special events for a longitudinal study about ‘health, health care use and health insurance’ that provided $20 for the baseline interview and $10 for a follow-up interview. Interviews were conducted with only one member of a household and were conducted on-site in English and Spanish by trained bilingual interviewers. The interviewers were native Spanish speakers from Puerto Rico, the Dominican Republic, and Central America, congruent with the cultural backgrounds of much of the study population. To ensure the ability to complete the visual tasks of the Mini-Mental State Exam (MMSE), we only included individuals whose visual acuity was 20/50 or better as determined with a handheld Snellen chart, with or without use of corrective lenses. Written informed consent was obtained from all participants prior to screening and interviews. The study was approved by the Mount Sinai School of Medicine Institutional Review Board.
Our cognitive assessments focused on memory and category fluency. We assessed memory using the immediate and delayed recall for Story A of the Logical Memory test from the WMS III . Story A is a brief passage read by an interviewer who asks the subject to recall as much of the story as possible immediately after it is read. The interviewer then informs the participant that they will be asked to recall it again ‘later’. The participant is asked to recall as much of the story as possible again 25–30 min later. Subjects are given credit for recalling core elements of the story. This method is comparable to those used in other large multicentre studies .
We assessed category fluency using the ANT . The ANT measures semantic verbal fluency, an ability that involves language and executive functions [9, 11–13]. The test requires subjects to name as many animals as they can in 1 min. Scores equal the total number of unique animals named within 1 min.
We also assessed levels of global cognitive function using the MMSE , a widely used dementia screening instrument. Interviewers were formally trained by a neuropsychologist in the Mount Sinai School of Medicine, Alzheimer's Disease Research Center, in the administration of all cognitive assessments. Their proficiency in test administration was established prior to data collection.
We used two variables to represent language. The first variable indicated whether the subject was assessed in Spanish. The study participants were given this option irrespective of their English-speaking ability. The second variable was a self-reported measure of English language proficiency, assessed using the question: ‘How would you describe your ability to speak and understand English?’ with six responses ranging from very poor to excellent.
We collected data on variables having established associations with cognitive function including age, education level, income and health and functional status [2, 15–19]. Health status measures included self-reported general health from the Short Form-36 , total number of chronic diseases and basic activities of daily living (ADL) and instrumental activities of daily living (IADL). We documented self-reported performance on six ADLs (bathing or showering, walking, getting in or out of bed or chairs, eating, dressing and using the toilet)  and five IADLs (managing money, preparing meals, doing light or heavy housework and shopping) . The study participants were considered to have impairment if they experienced ‘a lot of’ difficulty with or were unable to do one or more tasks. Additional variables included sex, race and self-identified ethnicity (non-Hispanic black, non-Hispanic white, Hispanic and other).
We tabulated means and standard deviations for scores on the immediate and delayed recall tasks and the ANT, stratified by language. In comparing test performance between English and Spanish speakers, we aimed to provide data that would inform future studies involving cognitive assessments in older, community-dwelling adults. Hence, we used three commonly applied MMSE score thresholds to identify clinically relevant subgroups often used in community-based research: MMSE of 22 or higher to ensure that we captured individuals with moderate to no cognitive impairments; MMSE of 26 or higher to limit the subgroup to individuals with possible mild cognitive impairment (MCI) to no cognitive impairments; and MMSE of 28 or higher to have a subgroup minimal or no cognitive impairments [22, 23]. Test scores were compared using t-tests. Because of the problem of small cell sizes with stratification, age strata were limited to 60–69 years, 70–79 and 80 and older, and education to high school or higher versus did not graduate high school.
We also conducted a series of linear regression analyses to examine the association of language (language of interview administration or self-reported English proficiency) with test performance. We first modelled test performance as a function of language alone. We then sequentially added age, education and then other covariates that had a statistically significant bivariate association (P < 0.05) with the outcome measure. The participants reporting poor or fair English-speaking ability were also likely to request interview administration in Spanish (kappa 0.73), and results for language of interview administration and self-reported language proficiency were qualitatively the same. We therefore only presented results for interview language (results for analyses using the language proficiency variable are available from the authors). The test results were considered statistically significant at the P < 0.05 level.
Most questions were missing data for fewer than 5% of study participants. However, 13.0% were missing data on income, consistent with self-reported income data from other studies [24, 25]. We used multiple hot-deck imputations to replace missing observations for income in logistic regression analyses . Imputed data sets were created with STATA version 10 using the hotdeck command (Stata Corporation, College Station, TX, USA). All analyses were conducted with SAS version 9.1 (SAS Institute, Inc., Cary, NC, USA).
Of the 453 individuals recruited, 420 (92.7%) had MMSE scores of 22 or higher and were included in our analyses. The study subjects who were administered the interview in Spanish were more likely to be among the 31 excluded individuals than English speakers (58.1% vs 41.9%, P < 0.0001).
Sixty-eight subjects (16.2%) reported having very poor to fair English language skills, and the interview (including cognitive tests) was administered in Spanish to 86 (20.5%) (Table (Table1).1). Among those reporting very poor to fair English language skills, 61 (89.7%) completed the Spanish-language interview. Those who completed the interview in Spanish had lower levels of education and income and had worse general health and greater use of emergency department care than those who completed it in English (Table (Table1).1). The mean age of the cohort was 73.8 years (Table (Table1).1). Two-thirds (63.1%) were women, one-quarter (27.2%) had not graduated high school and more than half (54.3%) had household incomes at or below $1,350 per month. The sample was approximately one-third black (30.5%), one-third white (31.9%) and one-third Hispanic (30.7%).
Table Table22 shows the mean and standard deviations for immediate recall scores, delayed recall scoresand ANT scores, stratified by age and education, for English- and Spanish-speaking subjects with MMSE scores of 22 or higher. As expected, performance on the assessments declined with increasing age and lower educational attainment, for both English and Spanish speakers.
Subjects interviewed in English and Spanish had similar scores on tests of memory across all three MMSE thresholds, whereas those who were administered the ANT in Spanish had poorer performance (see Appendix Table 1 in the supplementary data available at Age and Ageing online). Similarly, individuals who reported poor English-speaking ability had poorer performance on the ANT than better English speakers (data not shown). When test scores were stratified by the level of education, there were no significant differences by language for any of the tests, although the difference in WMS Story A immediate recall scores between Spanish and English speakers at or above the 12th grade level of education was of borderline statistical significance, with Spanish speakers having slightly higher performance on the assessment (11.7 ± 3.5 vs 10.3 ± 3.7, P = 0.05) (see Appendix Table 2 in the supplementary data available at Age and Ageing online).
Although the differences between Spanish and English language performance on memory assessment scores were not statistically significant on bivariate analyses, we conducted regression analyses adjusting for age and education since these are potent correlates of test performance, and Spanish- and English-speaking study participants differed markedly by these characteristics. Indeed, the association between Spanish language interview and immediate recall scores was statistically significant after adjusting for education and closely approached significance for delayed recall (Table (Table3).3). For immediate recall, language narrowly lost statistical significance when age was added to the model, in contrast to that for delayed recall.
In multivariable linear regression analysis (Table (Table3),3), Spanish language remained significantly associated with ANT scores after adjusting for age. However, the magnitude of the association between Spanish language and ANT scores was reduced by 62% when education was added to the model (Table (Table3,3, model 3) and the association lost its statistical significance. The magnitude of the association was reduced further by the addition of income, general health and functional limitations to the model (model 5). Analyses demonstrated that among these latter variables, income was responsible for the additional confounding (data not shown).
Assessing the cognitive functioning of older individuals can be particularly challenging in diverse urban areas where different languages are represented. Fortunately, for two commonly used neurocognitive assessments, we have found little difference in performance between English and Spanish speakers among a community-dwelling urban population. This population is notable for a high percentage of non-Hispanic black English speakers and is typical of other urban older communities. We administered the immediate and delayed recall assessments of the Story A subtest of the WMS III and a measure of word fluency, the ANT, to this diverse sample of urban, independently living older adults. We used easily defined and practical definitions of language—self-reported English language proficiency and preference for Spanish language assessments—and found similar patterns of test performance for each. Only with the ANT, a measure of word fluency, did we observe significantly lower scores by Spanish speakers. Stratified and multivariable analyses demonstrated that the poorer performance by Spanish speakers on the ANT were largely attributable to lower levels of education. Previous research has suggested that the ANT is relatively insensitive to variations in English language ability [11, 27]. Our findings somewhat contradict these prior observations, and suggest the need for education-based adjustments to ANT scores in multilingual samples. On the other hand, our findings indicate that education adjustments by language may not be needed when the WMS Story A memory assessment is administered in multilingual samples.
Independent of language effects, our data on the ANT are consistent with previously published data from other populations, in that age and education had an impact on performance [8, 11, 28, 29]. Scores on the WMS Story A delayed recall task in our study were higher than those observed in the Framingham Heart Study cohort , but they were normally distributed and followed the expected patterns of age and education associations [27, 28]. Specifically, we found that cognitive test performance declined with increasing age and lower levels of education, consistent with findings from studies across a host of populations.
Our study provides data for two commonly used neuropsychiatric assessments for independently living, older adults from demographically diverse urban communities. However, a number of factors should be considered before extrapolating these data to other populations. First, the sample size of 420 individuals restricted our analyses to large strata of age and education. Second, we employed convenience sampling rather than true random sampling which could result in selection bias. In addition, a greater proportion of Spanish speakers were excluded from our analyses because of low MMSE scores. Nonetheless, the impact of selection bias is probably nominal in the range of MMSE scores of 22 and greater as indicated by the balanced distribution of participant characteristics and the normal distribution of scores on all neurocognitive assessments. Third, we administered only one component of the logical memory subtests of the WMS III during subject interviews. Hence, we have not obtained comprehensive measures of memory.
In conclusion, our study provides descriptive data for Story A of the WMS III memory assessments and the ANT which suggests that these tests have relatively comparable ranges permitting research to be extended to Spanish language speakers.
This study was supported by a Paul B. Beeson Career Development Award in Aging from the National Institute on Aging (Dr. Federman, 1K23AG028955-01). Additional support was provided by the Mount Sinai School of Medicine Alzheimer's Disease Research Center (NIH AG0051318).
Supplementary data are available at Age and Ageing online.