Racial and Ethnic Differences in Activities of Daily Living Disability Among the Elderly: The Case of Spanish Speakers

Data and Study Sample 

A prospective cohort from a national sample of community-dwelling U.S. adults included 19,597 respondents age 51 or older who self-identified as Hispanic or Latino, African American or black, and Caucasian or white interviewed in 1998 (baseline) and biennially through 2004 as part of the HRS. Analyses are restricted to 15,333 respondents free of baseline ADL disability who were alive at the subsequent 2000 interview. Those who died between 2000 and 2004 were censored after death. Excluded by design were those who reported disability at baseline (n=1773), those who died before 2000 (n=724), and those who had a proxy respond to the interview (n=1125), for a total of 3622 persons. Another 642 persons with missing follow-up disability or insufficient baseline information were excluded.

Disability 

Following the International Classification of Functioning, Disability and Health,7 disability is defined as inability to perform ADLs. Disability in certain ADL tasks (dressing, toileting, bathing, eating, walking across a room, transferring) expected to last 3 months or more was ascertained from self-reported ability to do the task, not do the task, or to use equipment to perform the task (see appendix 1 for the ADL instrument). This assessment captures chronic dependence in basic self-care tasks that could jeopardize a person's ability to live independently. The development of ADL disability is identified by the first report of any ADL task disability at a subsequent 2000, 2002, or 2004 interview.

Instrument and Translation 

The original English questionnaire was translated into Spanish to recruit and interview Spanish speakers. When the HRS surveys began in 1992, they adopted a team translation approach. In this method, a fluent bilingual lead translator translated the source English instrument to the target Spanish language. Then at least 2 different bilingual translators reviewed the translation and noted errors and discrepancies. The translators then met with a member of the English instrument design team to discuss modifications to the translation. Last, when the instrument was in its final form, the national bilingual interviewing staff of the Survey Research Center at the University of Michigan conducted pretests. They found that this procedure, particularly the choice to include HRS-experienced bilingual interviewers living in different parts of the United States, helped in achieving a translation that had comparable comprehension levels for English-speaking and Spanish-speaking respondents and addressed the issue of dialectical variation in the Spanish spoken in the United States (written personal communication with HRS office staff, Sep 25, 2006).

Covariates 

Baseline (1998) demographic factors included race and ethnicity, age, sex, and living arrangement. We used self-reported HRS race and ethnicity information to classify people into 4 mutually exclusive groups: Spanish-speaking Hispanic (Hispanic-Spanish), English-speaking Hispanic (Hispanic-English), non-Hispanic black, and non-Hispanic white. We distinguished Hispanic-English from Hispanic-Spanish adults because previous literature8 suggests persons from different linguistic groups have unique explanatory models for health that may influence conventional disability measures.

Self-reported health factors were assessed at each interview. Chronic physical conditions included arthritis, cancer, diabetes, heart disease, hypertension, pulmonary disease, stroke, and bad vision. Mental health conditions included high levels of depressive symptoms (determined by an abbreviated Center for Epidemiologic Studies Depression score in the upper tenth percentile)9 and cognitive impairment (assessed by a score of ≤8 from a composite score based on the Telephone Interview for Cognitive Status and serial 7's test).10 Physical limitations were assessed by inability or avoidance in upper-extremity (pushing or pulling large objects, lifting or carrying weights >4.5kg) and lower-extremity (walking several blocks, climbing several stairs without rest) tasks.

Health behaviors assessed at each interview consisted of current smoking, current alcohol consumption, weight status, and regular vigorous physical activity. Current smoking was ascertained from a positive response to the question, “Do you smoke cigarettes now?” Alcohol consumption was based on a positive response to the question, “Do you ever drink any alcoholic beverages such as beer, wine, or liquor?” Weight status was determined from weight gain or loss of 4.5kg (10lb) or more, obesity, and underweight. BMI ([weight (kg)]/[height (m2)]), calculated from self-reported height and weight, was used to define obesity (BMI ≥30kg/m2) and underweight (BMI <20kg/m2). Regular vigorous physical activity was ascertained from the report of participation at least 3 times a week over the past 12 months in activities such as sports, heavy housework, or a job that involved physical labor.

Socioeconomic factors assessed at each interview consisted of education, wealth, family income, and health insurance. Education, a measure of human capital, was dichotomized as 12 or more versus fewer completed years of education. For analytic purposes, family income (all sources received by the respondent and spouse/partner during the preceding year) and wealth (the sum of housing and nonhousing assets) were dichotomized using the lowest baseline HRS population-weighted quartiles.11 If only partial income or wealth information was provided during the interview, dichotomized values were based on imputed estimates developed by the University of Michigan. Health insurance was classified into 4 mutually exclusive groups: any private insurance coverage; Medicaid enrollment; Medicare or other government insurance programs such as the Civilian Health and Medical Program of the Uniformed Services, the Civilian Health and Medical Program of the Department of Veterans Administration, and the Veterans Administration, without additional Medicaid or private insurance coverage; and no coverage.

Statistical Analysis 

The HRS is a national probability sample. All analyses were weighted and adjusted for the complex HRS sampling design.12, 13 Analyses adjusted for potential bias because of missing information and/or nonresponse by handling respondents with completed data as another stage of sampling to obtain adjusted sampling weights, using standard sampling methodology.14 Baseline comparisons between the demographic, health, and economic characteristics of minority subgroups versus whites used chi-square tests of categorical outcomes and F tests of continuous outcomes. Statistical testing was conducted at a nominal 5% α significance level.

Survival analysis for discrete data was used to analyze racial and ethnic differences in the development of ADL disability.15 The development of ADL disability was measured in discrete rather than continuous time because ADL disability is monitored only at 2-year intervals (1998–2000, 2000–2002, 2002–2004). A discrete hazard rate modeled the probability of developing disability by the next 2-year interview given a disability-free status and risk profile at the current interview. To be included in the analyses, a person's disability status had to be known at the start and the end of a 2-year period. All completed interview pairs (1998/2000, 2000/2002, 2002/2004) from a person before death contributed to the analyses. This method specifically accounted for repeated measures on the same subject and used time-varying covariates (ie, demographic, health, and medical access factors). We used the SAS GENMOD procedurea with a complementary log-log link to estimate the discrete hazard model. To account for the complex sampling design, variance was estimated using balanced repeated replication, a form of bootstrapping.14 HRs and associated 95% CIs estimated from the discrete hazard model were reported.

We used direct standardization rates to calculate adjusted prevalence rates by estimating probabilities of disability from the discrete hazard-rate model (1) for each member of the white reference sample, and then (2) as (hypothetically) a member of the black subpopulation, (3) as (hypothetically) a member of the Spanish-speaking Hispanic subpopulation, and (4) as (hypothetically) a member of the English-speaking Hispanic subpopulation.14

Study Limitations 

Several limitations common to secondary databases may affect our findings. First, we use self-reported data. However, self-reported information is standard in epidemiologic research, and its reliability has been documented.21, 22 Second, we recognize that dividing Hispanics into 2 groups on the basis of interview language is a gross distinction of cultural differences. However, even this simplistic distinction is sufficient to identify differences in health outcomes consistent with other literature. Finally, this study addresses the issue of Spanish-language interviews only and cannot be generalized to other languages or racial and ethnic groups.