The Medical Research Council Cognitive Function and Ageing Study (MRC CFAS), is a multi-centre longitudinal study that is fully described elsewhere . Between 1990 and 1991 a random sample of individuals aged 65 years and older living in five centres representative of rural and urban areas in England and Wales (Cambridgeshire, Gwynedd, Newcastle, Nottingham and Oxford) were contacted, with a response rate of 82%. At baseline, 13,004 participants completed the screening interview. A subgroup of those screened including all participants with severe cognitive impairments and a stratified subsample of the remainder (n = 2640) also completed a more detailed assessment. This included selected items from the Geriatric Mental State Automated Geriatric Examination for Computer Assisted Taxonomy (GMS-AGECAT), Mini Mental State Examination (MMSE)  and the multi-domain Cambridge Cognitive Examination (CAMCOG) .
Persons diagnosed with dementia (defined as an AGECAT organicity rating case level of O3 or above, n = 587 ), an unknown dementia status (n = 3), a MMSE score that is missing or below 18 (n = 243), or Parkinson's disease (n = 26) were excluded. Therefore of the 2640 who undertook the assessment interview 1781 persons were included in this study. The aim of this study was to analyse the psychiatric correlates of cognitive impairment, and so those with underlying psychiatric conditions were not excluded from the study.
Informed Consent and Confidentiality
MRC CFAS has Multi-centre Research Ethics Committee's approval and ethical approval from the relevant local research ethics committees. All participants gave informed consent and patient confidentiality was not breached.
Assessment of behavioural and psychological symptoms
Symptoms assessed during the screen, assessment and informant interviews include depression, apathy, anxiety, feelings of persecution, hallucination, agitated behaviour, elation, irritability, sleep problems, wandering, confabulation and misidentification. Symptoms were assessed using interviews conducted with both study participants and their informants, combined with observations made by the interviewer during the course of the interviews. Depression was measured with the Cambridge Mental Disorders of the Elderly Examination (CAMDEX) . Details of the questions used to assess each symptom have been described previously .
Assessment of cognitive function and definition for 'mild cognitive impairment' criteria
Cognitive function was assessed using the MMSE  and by classification based on definitions of mild cognitive impairment (MCI) and its subtypes . Using the MMSE, cognitive function was classified as low (18-21), intermediate (22-25) or high (26-30). Classifications of MCI aim to identify individuals with cognitive impairments likely to indicate incipient dementia. Three subtypes are considered, amnestic (A-MCI) including deficits in memory, non-amnestic (N-MCI) in which cognitive impairment is restricted to non-memory domains and multiple (M-MCI) in which both memory and non-memory domains of cognition are impaired .
A clinical diagnosis of MCI requires a subjective complaint of memory impairment by either the respondent themselves or an informant. Performance on cognitive tests must fall below a specific threshold, typically below 1 or 1.5 standard deviations of the mean age-specific performance, or if test scores are highly skewed below the 16th or 7th percentiles. To exclude those with possible dementia, general cognitive function must not be severely impaired. Exclusion criteria can include any health or psychiatric problems that might be the alternative cause of cognitive impairment, or a severe impairment in activities of daily living. The operationalisation of these criteria, including the level of subjective memory complaint, the specific cognitive tests and exact exclusion criteria, are not specified.
For the present study, 'subjective memory complaint' was defined as a report of memory problems by either the respondent or an informant at either the screening or the assessment interview. 'Objective memory impairment' was defined using as a score below the 16th age-specific percentile on one or more of the remote memory, recent memory or learning memory domains of the CAMCOG. 'Non-memory impairment' was defined using as a score of below the 16th age-specific percentile on any of the other cognitive domains measured by CAMCOG which includes orientation, language, attention/calculation, praxis, abstract thinking or perception. CAMCOG cut-off scores were age adjusted using five 5-year age groups including 65-69, 70-74, 75-79, 80-84 and 85+ years. General cognitive function was deemed to be severely impaired when participants scored less than 22 on MMSE. No health-related or functional exclusions were applied, since the relationships between these and behavioural and psychological symptoms were of interest in the present study.
MCI subtypes were assigned as follows: All MCI subtypes required subjective memory complaint and no severe impairment of general cognitive function (ie MMSE > = 22). Those with memory impairment but no non-memory impairment were classified as A-MCI. Those with non-memory impairment but no memory impairment were classified as N-MCI. Those with impairments in memory and non-memory domains were classified as M-MCI. Participants were not excluded on the basis of medical or psychiatric problems. Full details of the mapping of MCI in MRC CFAS have been published elsewhere .
A 'not impaired' group was defined using the following criteria: normal general cognitive function, no severe functional impairment, normal memory and non-memory test performance. Participants who could not be classified as "not impaired" or MCI using any definition were classified as "other cognitive impairment, no dementia" (OCIND). This heterogeneous group includes both respondents with an objective memory impairment but no subjective memory complaint, and those with impaired general cognitive function but not diagnosed with dementia.
As well as those factors that contribute to the definition of MCI, other factors considered in our analysis included sex, age, institutionalization, vascular co-morbidity, education, and social class. Age was reported continuously as well as being dichotomized into 65-74 years versus 75 years and older. Vascular co-morbidity was defined using self or informant reported history of a heart attack, stroke or diabetes. Institutionalisation was divided into two groups including, independent (living alone or in a warden controlled flat) or dependent (living in a residential home, nursing home or hospital). Level of education was measured in number of years of education and was divided into two groups including less than 10 years and 10 or more years of education. Functional impairment was defined as requiring help at least several times per week with activities of daily living (ADLs) such as washing, cooking for themselves, dressing, or if the respondent was housebound. Data were missing in fewer than 2% of individuals for each covariate and was assumed to be missing at random.
All analyses were performed using Stata 10.0. The prevalence of each behavioural and psychiatric symptom was estimated in cognitive groups. Participants were grouped both by MMSE scores, and by MCI classification. Three MMSE categories were defined including low (18-21), mild (22-25) and high (26-30). MCI classifications were 'not cognitively impaired', 'MCI' and 'OCIND'. Prevalences were also compared across MCI subgroups A-MCI, N-MCI and M-MCI as previously described. Differences in prevalence across groups were tested using likelihood ratio tests. Post-hoc pairwise tests comparing subgroups were not conducted.
Participants were back-weighted in prevalence estimation to adjust for over-sampling in the study population of individuals aged 75 years or older and the subsequent stratified sampling for the assessment interview. The relation between symptoms, subjective memory complaint, objective memory impairment, vascular disease, disability and socio-demographic factors was investigated using both univariate and multivariate logistic regression.
Co-occurrence between symptoms was measured using odds ratios estimated using logistic regression. Univariate analyses and multivariate analyses adjusting for the presence of all other symptoms, MMSE, age and sex were conducted. Factor analysis was used to determine the structure of the co-occurrence of symptoms. The Spearman's correlation matrix relating the symptoms was estimated and this was subjected to the principle factor analysis algorithm. The resulting factor solution was rotated for ease of interpretation using the Varimax rotation. This factor analysis was repeated within each of the MMSE and MCI subgroups described above.