Data and Study Populations
We used data from the 2002 waves of the Health and Retirement Study (HRS) and the English Longitudinal Study of Ageing (ELSA). The HRS and ELSA are biennial, longitudinal, nationally representative surveys of, respectively, US adults aged 51 and older, and English adults aged 50 and older.[8, 9] The two studies were developed collaboratively with significant overlap in survey questions in order to facilitate valid cross-national comparisons of aging-related changes in health, wealth, and well-being. More detail on the studies, including all survey questions, can be found at the HRS[10] and ELSA[11] web sites.
For this study, we included individuals who were aged 65 and older. To maximize the comparability of the US and English samples, we only included respondents of white race since ELSA contains very few (n = 156) non-white individuals. We also excluded individuals represented by a proxy (1,171 [12%] in the HRS and 96 [2%] in the ELSA) since the cognitive tests were not administered to these respondents. The overall response rate among all eligible respondents was 87% for the 2002 HRS and 67% for ELSA. The final study samples included 8,299 individuals from the HRS and 5,276 individuals from the ELSA.
Both the HRS and ELSA studies include population weights that can be used to draw valid inferences for the entire US and English age 65+ populations, respectively. In the HRS, weights are constructed in a two-step process, where the first step develops post-stratified household weights using the initial sampling probabilities for each household, as well as birth year, race/ethnicity, and gender of household members. The second step uses these household weights to then construct post-stratified respondent-level weights which are scaled to yield weight sums corresponding to the number of individuals in the US population as measured by the US Census Bureau's Current Population Survey (CPS) for the month of March in the year of data collection.
ELSA study participants were drawn from the Health Survey for England, which has an equal probability sample design, so weights to account for selection probabilities are not needed. However, weighting is needed to take into account household and individual non-response which was done by analyzing stage and extent of drop-out from the study, and took into account factors including region of residence, age of oldest person, household size, social class, and incidence of longstanding illness. This information was used in logistic regression models to predict non-response probability and the resulting values were inverted for responding households to provide an initial non-response weight. A second round of weighting was used to adjust the initial household non-response weights to ensure the weighted sample matched the English population as assessed by the 2001 National Census. Full details on the development of population weights for the HRS and ELSA studies are available at the study websites[10, 11].
The HRS survey was administered by telephone (71% of the sample) and in-person (29% of the sample). The HRS attempts to interview older respondents (age ≥ 80) in-person whenever possible. All ELSA interviews were performed in-person.
Measurement of Cognitive Function in the HRS and ELSA
Both the HRS and ELSA assessed cognitive function using tests of immediate and delayed recall of 10 common nouns. A list of ten words was presented orally to study participants, who were then asked to recall as many words as possible immediately after the list was read, and then again after an approximately five-minute delay during which they completed other survey questions. The same four randomly assigned lists of 10 nouns were used in both the HRS and ELSA. Orientation to the day, date, month, and year were also assessed in the same way in both the HRS and ELSA. These three tests resulted in a cognitive scale ranging from 0 to 24 possible points (10 points for immediate recall, 10 points for delayed recall, and 4 points for orientation). If a respondent refused to provide an answer for any of the 3 tests, they were assigned a score of "0" for that test. Ninety-five (1%) HRS respondents and 43 (1%) ELSA respondents refused to answer for one or more of the cognitive tests. Detailed information on the cognitive measures used in this analysis, including their derivation, reliability, and validity, is available at the HRS website.[12]
Measures of Health Conditions, Risk Factors, and Treatments
Participants in the both the HRS and ELSA were asked about the presence of common chronic health conditions using the question, "Has a doctor ever told you that you had...?" For this analysis, we included stroke, diabetes, heart disease, hypertension, lung disease, and cancer. Depressive symptoms experienced in the last week were assessed using an 8-item version of the Center for Epidemiologic Studies Depression (CES-D) scale.[13] The 8-item version of the CES-D scale used in the HRS and ELSA has comparable reliability and validity to the widely used and validated 20-item CES-D Scale.[13, 14].
Smoking status (never, former, or current), and alcohol consumption (average number of days/week that alcohol was consumed over the last 3 months [HRS] or 12 months [ELSA]) were assessed similarly in both the HRS and ELSA. Among those who reported hypertension or diabetes, current use of prescription medications to treat these conditions was determined. (Diabetes treatments were categorized as none, oral medications only, and insulin)
The HRS and ELSA determine the presence of limitations in independent function by asking about difficulty with 6 activities of daily living (ADLs; eating, getting in and out of bed, toileting, dressing, bathing, walking across a room) and 5 instrumental activities of daily living (IADLs; preparing meals, grocery shopping, making phone calls, taking medications, managing money).[15]
Sociodemographic Measures
We included age (65 – 74, 75 – 84, ≥ 85), gender, household net worth (tertiles; 2002 US dollars) and level of education as sociodemographic measures in the analyses. British pounds were converted to 2002 US dollars using the average currency exchange rate for 2002.[16] Measures of education differ in the HRS and ELSA. Following Banks and colleagues, [2] we identified low, middle, and high education categories for each country (0 to 12 years of school, 13 to 15 years, and ≥ 16 years, respectively, in the US; qualified to lower than "Ordinary-level" [O-level], O-level to lower than "Advanced-level" [A-level], and A-level or higher, respectively, in England).
Analysis Plan
We calculated scores for the 3 individual cognitive tests (immediate recall, delayed recall, and orientation) as well as for the combined 24-point scale. Mean scores for the full sample, and for age-, gender-, and education-stratified samples for the US and England were compared using t-tests.
We then pooled data from the US and England and estimated an ordinary least squares (OLS) regression model with total cognitive score as the dependent variable. A dichotomous variable indicating the country (0 for England; 1 for the US) was used to determine cross-national differences in cognitive function, after accounting for sociodemographic and health variables. We estimated 8 separate linear regression models with different sets of independent variables (e.g., demographic variables, chronic health conditions, hypertension and diabetes treatments, depressive symptoms) in order to determine which variables were associated with US – English differences in cognitive score.
All analyses used the HRS or ELSA population weights to adjust for survey non-response and for the complex sampling design (stratification and clustering) of each study.
The University of Michigan Medical School Institutional Review Board (IRB) reviewed this study and determined that it was exempt from IRB review since all data used for the study were publicly available.