The Medical Research Council Cognitive Function and Ageing Study (MRC CFAS) is a multi-centre population based study of individuals aged 65 years and over living in the community, including care homes. The study began in 1991 and was designed to determine the incidence of dementia .
The study has six centres across England and Wales chosen to represent the national variation of urban–rural mix, socio-economic deprivation and rates of chronic disease . Five of these centres with identical study designs (Oxford, Nottingham, Newcastle, Cambridgeshire and Gwynedd), are used in the present investigation. The sixth centre (Liverpool) used a different design and is not included. Random samples of people in their 65th year and above were obtained from Family Health Service Authority lists from these five centres. The sample was stratified by age (65–74 years and 75 years and over) and equal numbers were randomly selected from these two age groups with the aim of recruiting 2500 to each centre.
Of those 16258 eligible and available to take part in CFAS, 13004 (80%) agreed to participate. All study centres obtained ethical approval from local research committees and from the Eastern Multicentre Research Ethics Committee Ref: 05/MRE05/37. Eligible participants (or their proxies where appropriate) provided informed consent. Trained interviewers undertook baseline interviews in the participants’ homes.
Socio-demographic factors collected included age, sex, marital status, type of accommodation and social class using the Registrar General’s Occupational Classification .
The presence of stroke was determined from self-report through the question: “Have you ever had a stroke that required medical attention?” Time since stroke was determined by subtracting the response to the question: “How old were you when you had the last stroke?” from age at data collection.
General subjective health status or self-rated health (SRH) was determined with the question: “Would you say that for someone of your age, your own health in general is” followed by a list of options from poor to excellent.
Participants were asked about health behaviours including smoking status and alcohol intake. Comorbidities were assessed by the questions: “Have you ever suffered from high blood pressure, angina, heart attack, diabetes or head-injury”?
Functional status was determined by enquiring about activities of daily living (ADL) and instrumental activities of daily living (IADL). ADL disability was defined as requiring help at least several times per week with activities of daily living such as washing, cooking, dressing, or if the respondent was housebound. IADL disability was defined as needing help with heavy housework or shopping and carrying heavy bags.
Cognitive status was determined using the Mini Mental State Examination  and the Verbal Fluency Test .
History of depression was determined by asking the following questions: “Have you ever consulted a doctor about emotional problems, or problems with your nerves?” followed by “What did the doctor say you had?”
Social variables were assessed by questions including: “Does anyone else live here?”
“How often do you see any of your (children or other) relatives to speak to?”, “Do you have friends in this community?”, “How often do you see any of your neighbours to have a chat or do something with?”, “Do you attend meetings of any community or church or social groups, such as over 60’s clubs, evening classes or anything like that?”, and “In general, do you get out and about as much as you would like to?”
All analyses were performed using Stata 12.0. Any participants who had missing data regarding presence of stroke or SRH or had an MMSE that was missing or less than 18 (as their responses could not be considered reliable for the purposes of this analysis), were excluded from the analyses. The distribution at baseline of demographic, physical, psychological, cognitive and social characteristics were described for participants with and without stroke.
For purposes of logistic regression SRH was initially dichotomised into two groups (good = excellent/good vs. poor = fair/poor). The association between each potential predictor and SRH was assessed using multiple logistic regression. Ordinal and linear regression models were considered in order to increase the power of the analysis, however Brant tests following ordinal logistic regression showed that the proportional odds model was strongly violated by this data (suggesting that linear/ordinal models were not appropriate), although findings based on the linear and logistic models were rarely qualitatively different. Further inspection using different possible cut-points for the dichotomisation of SRH showed that where there were discrepancies between the linear and logistic models, effects were more consistent between the excellent/good vs fair/poor cut-off and the excellent/good/fair vs poor cut off. Either of these dichotomisations would be reasonable choices for estimating the determinants of self-rated health. When using the excellent vs good/fair/poor cut-off, however, estimates of the effects of potential predictors were different in these cases, and may have driven the difference in the linear model. Since we are not concerned with identifying predictors of excellent as opposed to good SRH, logistic regression using the initial cut-point was finally selected as the most robust and informative analysis.
Potential predictors of SRH were defined as follows: Psychosocial variables included age group (65–74, 75–84, and 85+), gender, social class (divided into manual (IIIb, IV, V) and non-manual (I, II, IIIa)), disability (in three groups: no impairment, impairment of IADL only, impairment of ADL), cognition (MMSE divided into four groups, less than 18 or missing, 18–21, 22–25 and 26–30), time since stroke (<1 year, 1–2, 3–5 and >5 years), and presence of depression (yes or no). Prevalence estimates were weighted to adjust for oversampling in the study population of those over 75. Differences in characteristics between participants with and without stroke were calculated using logistic regression adjusting for age and sex.
A multivariate logistic regression model adjusting for demographic, physical, psychological, cognitive and social factors was constructed to explore the association of SRH with stroke. The associations of these factors with SRH in those with and without stroke were then calculated using univariate and multivariate logistic regression. The statistical significance of the differences in the associations between the covariates and SRH for those with stroke compared to those without stroke were calculated by estimating a final logistic regression model for SRH using data from the whole sample and including each covariate and the interaction of each with stroke as independent variables.