Data from the first wave (2009/2011) of The Irish Longitudinal Study on Ageing (TILDA) were used. This is a nationally representative study of community-dwelling adults aged 50+ (and their spouses or partners of any age) residing in Ireland. The study is closely harmonised with leading international research, including The English Longitudinal Study of Ageing (ELSA), the Survey of Health, Ageing and Retirement in Europe (SHARE) which is pan-European, and the Health and Retirement Survey (HRS) conducted in the United States.
A total of 8,504 participants were recruited to the study (8,175 aged 50+ and 329 younger partners of eligible individuals). Ethical approval was obtained from the Trinity College Dublin Research Ethics Committee, and all participants provided written informed consent. Participants first completed a computer-assisted personal interview (CAPI) in their own homes. If individuals had known or suspected dementia, they were ineligible for participation in the study. The overall response rate was 62%.
Participants attended a health centre for a comprehensive health assessment. Participants who were unable/unwilling to attend were offered a modified and partial assessment in their own home. All assessments were carried out by qualified and trained research nurses. Among other clinical parameters, cardiovascular measures were assessed and collated. Of the 8,175 participants aged 50+, 5,897 underwent an assessment (85.4% in the health assessment and 15.6% in their own home). Details of the health assessment have been reported elsewhere .
Outcome variables: subjective and objective measures of cardiovascular disorders
TILDA collected data on individual self-reports of specific conditions with the general question: “Has the doctor ever told you have any of the following conditions on this card?” The diagnoses analyzed in this paper are: hypertension and high cholesterol.
TILDA also collected data on objective evidence of these cardiovascular disorders. Focusing first on hypertension, objective measurements of blood pressure were taken by TILDA research nurses in the health or home assessment. Blood pressure was measured using the OMRON™ digital automatic blood pressure monitor with arm cuff (Model M10-IT). Participants had been seated for at least 30 minutes when the measurement was obtained. Three separate readings were taken one minute apart; the first two with the respondent seated and the third immediately after the respondent stood up. The mean value for seated blood pressure from the first and second readings was used in this analysis.
Objective hypertension was defined as systolic blood pressure (SBP) ≥ 140 mm Hg and/or diastolic blood pressure (DBP) ≥ 90 mm Hg and/or on antihypertensive medication [11, 23, 24]. Data on medication was collected in the CAPI. Respondents were asked to show the packaging (bottle, tube, blister pack) of the medications they were taking on a regular basis to the interviewer, who then recorded the names of the medications into a computer-based medication inventory. All medications were then classified according to the WHO Anatomical Therapeutic Chemical (ATC) classification system. This ensured that drugs were classified in a standardized way, according to primary indication. We considered antihypertensive medication to include respondents taking antiadrenergic agents, diuretics, beta blocking agents, calcium channel blockers and ACE inhibitors.
Non-fasting blood samples were collected from TILDA respondents in the health or home assessment. Measurements of total cholesterol, HDL-C (high-density-lipoprotein cholesterol) and LDL-C (low-density-lipoprotein cholesterol) were taken. Respondents were defined as having objective hypercholesterolemia if: total cholesterol ≥5.2 mmol/L and/or on cholesterol-lowering medication. We considered cholesterol-lowering medication to encompasse HMG CoA reductase inhibitors, fibrates, bile acid sequestrants, nicotinic acid derivatives and other lipid modifying agents such as ezetimibe, as well as lipid lowering drugs in combination with other agents. Measurements of HDL-C and LDL-C were also used in our analyses.
Measures of SES
Two key SES measures were used in this study: educational attainment and wealth, each divided into three groups. Education was separated into: ‘no/primary’, ‘secondary’ (junior certificate, leaving certificate or equivalent), and ‘tertiary/higher’ (diploma, first degree or higher).
Data on wealth was collected through a battery of questions on self-valuation of: current residence, properties other than current residence, cars, savings, other financial assets and other assets including business and land. Unfolding brackets were used when respondents refused or said that they “did not know” the value of their house, other financial assets or other assets. The unfolding brackets technique is often used in empirical studies to reduce item non-response on financial measures .
Unfortunately, unfolding brackets were not employed after the question on savings. Hence, savings were imputed for those respondents who, although did not provide an estimate of savings, provided information on all the other assets (N = 742). Multiple imputation was performed using the multiple imputation suite (mi) of commands available is STATA 12 . The mi set of commands was used to generate a regression model to impute missing data based on a range of conditional covariates. The predictive mean matching method was used. This is a partially parametric model that matches the missing value to the observed value with the closest predicted mean [26–28]. A total of five nearest neighbours were included in the set of possible donors . This process was repeated 20 times, creating 20 separated imputed datasets. These 20 datasets were finally combined into one dataset. The wealth gradient was then constructed by dividing respondents in tertiles.
A wide battery of controls was added to the model. These included: i) demographic and socio-economic characteristics: sex, single year of age, marital status (married/cohabiting or not), current area of residence (Dublin, another town/city, rural area), whether respondent grew up in a poor family or not (self-reported); ii) risk factors associated with cardiovascular disease: smoking (current smoker, past smoker, never smoked), drinking (standard alcoholic drinks per week and CAGE (cut-annoyed-guilty-eye opener) questionnaire score) [29, 30], exercise (kilocalories burnt per week doing physical activity) and waist circumference in centimetres; iii) self-report of doctor-diagnosed diabetes; iv) multiple other doctor-diagnosed cardiovascular diseases specifically angina, heart attack, congestive heart failure, stroke, ministroke or transient ischemic attack, abnormal heart rhythm, heart murmur and any other heart trouble; v) health care utilization: number of times visited a physician or a hospital in the year prior to the interview (self-reported); and vi) health insurance coverage: private (yes/no), full public i.e. free health care (yes/no), partial public (yes/no). The regressions on hypercholesterolemia also included a dichotomous variable for whether the respondent reported to have had at least one blood test for cholesterol.
Binary and multinomial logistic and linear regression analyses were used. Analyses were first adjusted only for age and sex. Subsequently, the full battery of covariates was entered simultaneously. All models used appropriate sample weights and included 4,179 observations. Analyses were performed using STATA 12 . A prior level of significance was set at p ≤ 0.05 for all analyses.
First, the associations between SES and self-reported and objectively measured hypertension and hypercholesterolemia were investigated. Binary logistic regressions were used and odd ratios (ORs) and 95% confidence intervals (CIs) were reported. Then, the associations between SES and objectively measured hypertension, LDL-C and HDL-C were investigated more closely. Multinomial logistic regressions were used for objective hypertension and, for ease of interpretation, parameter estimates were converted to estimates of average marginal effects (AMEs). Linear regressions were used for objective measures of LDL-C and HDL-C and coefficients (and 95% CIs) were reported.