Race/Ethnic Disparities in Mild Cognitive Impairment and Dementia: The Northern Manhattan Study

Study population

The NOMAS sample enrolled adults aged 40 and older (range 40–94) who had resided in Northern Manhattan for more than three months at the time of recruitment and had never been diagnosed with a stroke. Participants were recruited between 1993 and 2001 using random-digit dialing to participate in a baseline interview and assessment. The enrollment response rate was 75%, yielding participants (N = 3,298) of diverse race/ethnic backgrounds (Hispanic: 61% Dominican, 14% Puerto Rican, 12% Cuban, and 13% other; Black 22%, and White 32%). Annual telephone and in-person follow-up continues (loss to follow up < 5%). In 2003, the MRI sub-study began recruitment during annual telephone follow-up and included NOMAS participants who were clinically stroke-free, age 50 and older, and had no contraindications to MRI. An additional 199 household members were recruited to yield a final sample size of 1,290 by 2008. The study is approved by the IRBs of Columbia University Medical Center and the University of Miami School of Medicine and all subjects provided written informed consent.

Baseline evaluation

Participants had a thorough evaluation of vascular risk factors and medical history at the time of enrollment, including a physical/neurological examination by study physicians. Race/ethnicity was self-identified based on questions adapted from the 2000 US census and classified into four categories (White non-Hispanic, Black non-Hispanic, Hispanic, and non-Hispanic other race). Years of educational attainment, including degree achieved, were self-reported. Participants were asked about a history of hypertension and high blood pressure, diabetes mellitus and high blood sugar, high cholesterol, congestive heart failure, angina, coronary artery disease, atrial fibrillation, valvular heart disease, and peripheral vascular disease based on standardized questions adapted from the Behavioral Risk Factor Surveillance System of the Centers for Disease Control. Questionnaires were used to identify medical history, medication use, and health behaviors [6]. Medications for seizures, psychiatric indications, and pain were also documented. Smoking was categorized as current (within the past year), former, or never smoker of cigarettes, cigars, or pipes. Leisure-time physical activity was assessed by self-report using a questionnaire adapted from the National Health Interview Survey and moderate to heavy physical activity was defined as participation in at least one of several rigorous physical activities in a typical 14-day period [7]. Moderate alcohol use was defined as current drinking of one drink per month up to two drinks per day. Fasting blood specimens were analyzed at the Core Laboratory of the Irving Center for Clinical Research to determine blood glucose and lipoprotein levels using published methods [8]. Plasma levels of cholesterol subfractions were measured using standardized enzymatic procedures with a Hitachi 705 automated spectrophotometer (Boehringer Mannheim, Mannheim, Germany). Waist and hip circumferences were measured in inches with a flexible tape measure while participants were standing and wearing no heavy outer garments. Waist circumference (WC) was measured at the level of the umbilicus, and hip circumference was measured at the level of the bilateral greater trochanters, as previously described [9]. Systolic and diastolic blood pressures were calculated by averaging two measurements (before and after the physical examination) from the right brachial artery after a 10-minute rest in a seated position (Dinamap Pro100, Critikon Inc).

Brain MRI

Imaging was performed on a 1.5-Tesla MRI system (Philips Medical Systems, Best, The Netherlands) at the Columbia University Hatch Research Center. To segment white matter hyperintensities, semiautomated measurements of pixel distributions using mathematical modeling of pixel-intensity histograms for cerebrospinal fluid and brain white and gray matter were used to identify the optimal pixel-intensity threshold to distinguish cerebrospinal fluid from brain matter, using a custom-designed image analysis package (QUANTA 6.2 using a Sun Microsystems Ultra 5 workstation) [10]. For subclinical brain infarct readings, methods to identify and classify MRI-defined subclinical infarcts (SBI) have been published [11]. Two independent raters used a superimposed image of the subtraction, fluid attenuated inversion recovery (FLAIR), proton density, and T2-weighted images at 3×magnified view for interpretation of lesion characteristics. Agreement among raters has been good (published kappa values, 0.73–0.90) [12]. Processing of MRI scans to calculate total intracranial volumes (ICV), total cerebral volumes, and white matter hyperintensity volumes (WMHV), has been previously described [13]. To correct for head size, WMHV was calculated as percent total ICV and log-transformed to a normal distribution (log-WMHV). Proportion of total cerebral volume to total ICV was examined as brain parenchymal fraction. SBI were rated as present or absent.

Cognitive and functional assessment

Participants in the MRI sub-study were recruited between 2003 and 2008 and assessed in-person (defined as visit 1) and were interviewed by a trained research assistant, who administered structured questionnaires, a neuropsychological (NP) battery, and rated cognitive and functional status based on a structured interview. Adequacy of hearing and vision were assessed qualitatively by the research assistant and documented along with current medications. All tests were administered in English or Spanish based on participant preference. The Mini-Mental State Examination (MMSE) and Cognitive Failures Questionnaire (CFQ) were administered at the time of the NP battery in a designated quiet room. Most testing was on the day of MRI [14, 15]. Episodic memory was measured using three sub-scores derived from a 12-word five trial list-learning task: list learning total score, delayed recall score, and delayed recognition score. Executive function was assessed with two sub-scores: the difference in time to complete the Color Trails test Form 1 and Form 2, and the sum of the Odd-Man-Out subtests 2 and 4 [16]. Processing speed was assessed with the Grooved Pegboard task non-dominant hand time and the Color Trails test Form 1 [17, 18]. Working memory was assessed with the Digit Ordering and Letter Number Sequencing tests [19, 20]. Semantic memory was measured using three tests: picture naming (modified Boston Naming), category fluency (Animal Naming) and phonemic fluency (C, F, L in English speakers and P, S, V in Spanish speakers) [21, 22]. The Visual-Motor Integration test and the Symbol Digit Modalities tests were also administered [17, 23]. At the initial visit we estimated premorbid intelligence and literacy with the Peabody Picture Vocabulary Test (Test de Vocabulario en Imagenes Peabody for Spanish speakers), the Wide Range Achievement Test (English speakers) and the Word Accentuation Test (Spanish speakers) [24–26]. Depressive symptoms were quantified with the Center for Epidemiological Studies Depression scale (CES-D), a 20-item scale assessing depressive affect, somatic complaints, positive affect, and interpersonal relations [27].

A mean of 5.0±0.6 years after the initial neuropsychological assessment, a second in-person asse-ssment was conducted between 2008 and 2015 (visit 2). At the second visit, the CFQ, CES-D, MMSE, and neuropsychological battery were repeated with the addition of the Letter Number Sequencing test. The Informant Questionnaire of Cognitive Decline in the Elderly (IQCODE) was administered to a family member or friend of the participant as close in time to the neuropsychological assessment as possible [28]. The relationship of the informant to the participant was documented and aided adjudication.

Ascertainment of mild cognitive impairment and dementia

In 2015 NOMAS entered its fifth consecutive grant cycle and for the first time was funded to ascertain MCI and dementia status. A team of neuropsychologists and neurologists with dementia expertise used criteria from the National Institute on Aging–Alzheimer’s Association (NIA-AA) and the Diagnostic and Statistical Manual of Mental Disorders (5th Edition) for case ascertainment [29, 30]. MCI was based on 1) concern elicited from the participant (CFQ), informant (IQCODE), or other witness about a decline in cognition, 2) decline from a previous level of performance in at least one cognitive domain, 3) lack of significant functional impairment, and 4) lack of a psychiatric or other diagnosis that would explain the cognitive status. Dementia (major neurocognitive disorder) was based on 1) evidence of a decline from a previous level of performance in at least one cognitive domain, 2) functional impairment, and 3) lack of a psychiatric or other diagnosis that would explain the cognitive status. Each pair of adjudicators reviewed data from both visit 1 and visit 2 and completed a visit 2 grading form documenting cognitive status as healthy, MCI (with notation of amnestic and non-amnestic sub-types), dementia, other psychiatric disorder, or unable to classify.

To assess cognitive performance, we used individual neuropsychological test scores and cognitive domain-specific scores. Cognitive domains (episodic memory, semantic memory, processing speed and executive function) were created based on an exploratory factor analysis and prior findings. Because NOMAS participants have low educational attainment, literacy, and socio-economic status, and there are limited established norms that consider these factors, we constructed NOMAS-specific norms using neuropsychological test scores from visit 1. The normative values for each NP test were calculated from age (50–60, 61–70, 71–80 and > 80 years old) and education (0–6, 7–12 and > 12 years of education) group-specific means and standard deviations, excluding participants rated as likely to have dementia by the interviewer at visit 1. Each NP test score was standardized against its normative values, and each cognitive domain Z score was obtained by taking the average of the standardized NP test scores for the available component tests in a given domain.

Adjudicators accessed a Redcap web portal that displayed demographic information, visit eyesight and hearing status, neuropsychological domain and literacy Z-scores, dementia rating, MMSE and CES-D scores, most recent medication list and prior history of antidepressant and other psychiatric medications, psychiatric history, CFQ score, IQCODE score (cutoff for dementia > 3.6), and stroke history between visits (if applicable) [15]. Downloadable case-specific test forms were available for review to aid in assessments. Adjudicators were randomly assigned cases blinded to each other’s ratings based on an algorithm that identified participants with possible cognitive impairment. We derived an algorithm to segregate cases who had neuropsychological testing at Visit 2 into one of two pools: 1) those requiring adjudication, and 2) those rated as cognitively healthy based on an algorithm. The inclusion criteria for consensus adjudication were:

Missing at least one of the 11 NP battery tests; or

Statistical analysis

Participant characteristics were compared across race/ethnicities. Characteristics of participants who only attended visit 1 and those who returned for visit 2 were also compared. We used ordinal proportional odds models to examine race/ethnic differences in prevalence using a three-level ordinal dependent variable of healthy, MCI, and dementia, and difference of association across the three ethnic groups collectively was tested using a likelihood ratio test with 2 degrees of freedom (LRT 2 df). Less than 2% of the sample self-identified as other race (Asian or Pacific Islander, American Indian or Alaskan Native) and were excluded from the analysis. Hispanic and non-Hispanic Black (referred to as Black) were entered into the model (reference = White). Odds of MCI or dementia comparing Hispanic participants to White and to Black participants, and comparing Black to White participants were calculated. Model 1 was adjusted for age, sex, years of educational attainment, medical insurance status (proxy for income), and literacy; model 2 was adjusted for variables in model 1 plus systolic and diastolic blood pressure, antihypertensive medication use, diabetes mellitus, smoking status, moderate alcohol use, physical activity, low and high density lipoprotein levels, and waist-to-hip ratio; model 3 was adjusted for variables in model 1 plus brain parenchymal fraction, log-WMHV, and SBI status. We tested for potential effect modifiers of race and ethnicity by entering multiplicative terms for significant covariates into the model. We performed a sensitivity analysis among participants with healthy cognition at visit 1 (MMSE > 26) to limit confounding by pre-morbid cognitive problems. Analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC).