Data Source
Data were obtained from Health Survey for England (HSE) 2014, the 24th cross-sectional survey representative of the general population of all ages living in private households in England. The series of annual surveys have been designed and carried out since 1994 by the Joint Health Surveys Unit of Nicen Social Research and the Research Department of Epidemiology and Public Health at University College London and commissioned by the Health and Social Care Information Centre. The HSE has a core set of questions used every year, plus additional questions to focus on specific topics, which change each year. In 2014 mental illness was one such specific topic, therefore only one wave of the survey could be used. The HSE 2014 dataset was accessed via the UK Data Service.
For survey data collection an interview was conducted, followed by a visit from a specially trained nurse for all those who agreed. The interview was conducted by fully trained personnel using computer-assisted interviewing. Data collected included socio-demographic characteristics, attitudes to mental illness, and use of mental health services. The 12-item General Health Questionnaire (GHQ-12) was also administered as a self-complete measure. A total of 8077 adults aged 16 and over participated in the interview stage of the survey and 5491 of them expressed their consent for a nurse visit. During the nurse visit, participants were asked questions about mental illness experience; familiarity with someone else with a mental illness; and prescribed medicines.
Missing values’ occur for several reasons, including refusal or inability to answer a particular question; refusal to co-operate in an entire section of the survey (such as the nurse visit or a self-completion questionnaire); and cases where the question is not applicable to the participant. Missing values were coded originally in the survey protocol as ‘-9’ and have been omitted from all tables and analyses. Cases of incomplete answers or refusals were excluded from the analysis (Fig. 1).
Figure 1 here
Outcome Measures
Help-seeking from a professional for a mental health problem was measured using three binary variables derived from a variety of survey questions. Firstly, contact in the last two weeks with a doctor for a mental health problem was generated from two questions: participants were asked if they have talked to a doctor during the last two weeks, and those who answered yes were then asked a second question about the reason. Possible answers were ‘physical health problem’, ‘mental health problem’ and ‘both of these’. A binary variable was derived by combining ‘mental health problem’ and ‘both of these’. Responses of ‘no’ to the first question or ‘physical health problem’ to the second were combined and used as the reference category. Contact with a doctor included contact with a general practitioner, as well as a specialist or any kind of doctor. Any type of contact i.e. telephone, health centre, home visit, was recorded. The second outcome was use of any type of healthcare service in the past twelve months. Participants were asked about attending a hospital as: an inpatient; outpatient or day patient; through the Emergency department; or any other way, along with the reason for attendance. In addition, a list of mental health community care services was presented and participants were asked if they had used any. Responses to ‘Why were you in hospital?’ of ‘mental health problem’ or ‘both of these’ (i.e. mental and physical) were combined with any use of mental health community care services. Those who did not attend a hospital or did not use any community care service were combined, together with physical health reasons, and used as the reference. The third outcome variable was having antidepressants currently prescribed, which we interpret as being for common mental disorder, as rates of use for other conditions are much lower [33]. This was derived from a single question and was defined as any versus none.
Covariates
Attitudes to mental illness were assessed using the 12-item scale derived from the original Community Attitudes toward the Mentally Ill (CAMI) scale developed by Taylor and Dear [34]. The CAMI-12 scale is a subset of the original statements, selected to show levels of mental health-related stigma and tolerance, and first used in the survey evaluating the Time to Change social marketing campaign [35]. This scale was administered in the self-completion questionnaire during the interview visit.
For the HSE 2014, factor analysis was performed on the 12 attitudes statements and generated two internally reliable factors of ‘Tolerance and Support for community care’ and ‘Prejudice and Exclusion’ [36]. The identification of the two factors was consistent with previous research, which ran a factor analysis on the 27 item version of the CAMI questionnaire used in the National Attitudes to Mental Illness survey [37]. A more detailed description can be found elsewhere [38, 39]. The 12 CAMI statements were phrased in both positive and negative directions. Positive views were expressed by agreement with ‘Tolerance and Support’ items, for example ‘The best therapy for many people with mental illness is to be part of a normal community’; and disagreement with ‘Prejudice and Exclusion’ items, such as ‘I would not want to live next door to someone who has been mentally ill’. The degree of a respondent’s agreement or disagreement was rated on a 5-point Likert Scale which was scored as follows: 0 for ‘disagree strongly’, 25 for ‘disagree slightly’, 50 for ‘neither agree nor disagree’, 75 for ‘agree slightly’ to 100 for ‘agree strongly’. Negative statements were scored in reverse so that in each case, a higher score represented a more positive attitude. There was also a sixth option of ‘Don’t know. Those choosing this option were excluded from the calculation of the mean score.
A single measure was derived for each of the two factors and takes the average of the mean scores from the individual statements relating to that factor. These variables were derived from valid answers to 6 CAMI items and constitute the two subscales: prejudice/exclusion and tolerance/support [39].
Symptoms of common mental disorder were measured using the General Health Questionnaire 12 item version (GHQ-12) score. The GHQ has been widely used and validated in different settings and different cultures as an effective instrument to screen for common mental disorders [40, 41]. The questionnaire consists of twelve items measuring general levels of happiness, depression and anxiety, sleep disturbance and ability to cope over the last few weeks. Answers of ‘not at all’ or ‘no more than usual’ were scored 0 and responses indicating the symptom is present ‘rather more than usual’ or ‘much more than usual’ were scored 1. Consistent with previous HSE surveys, a GHQ-12 score of 4 or more is referred to as a ‘high GHQ-12 score’, indicating probable psychological disturbance or mental ill health [38]. Scores below 4 suggest the absence of mental health problems. Incomplete answers were excluded from the calculation of the GHQ-12 score (Fig. 1).
GHQ-12 score as well as CAMI score were entered as continuous variables to maximise power.
Consistent with previous research [38, 42, 43] the analysis included other factors reported to be associated with attitudes towards mental illness or to help-seeking or receipt. Demographic factors and familiarity with mental illness were included in the regression models as follows: gender, age, educational qualification, income, employment status, ethnicity and familiarity with a mental illness. Age was included as a categorical variable because it has previously been found to show a nonlinear relationship with attitudes to mental illness [44]: ’16–24’; ’25–44’; ‘45–64’ and ‘65+’. Highest educational qualification was recorded in four categories using the public examinations held in England taken at age 16 and 18, National Vocational and university qualifications: ‘NVQ4/NVQ5/Degree or equivalent’; ‘Higher education below degree (inclusive A level)’; ‘NVQ2/GCE O Level /NVQ1/CSE other grade equivalent’ and ‘None’. The measure of equivalised household income was used, which takes into account the number of adults and dependent children in the household as well as overall household income. Based on this measure, households are divided into quintiles (fifths) and used as categories. Employment status was recorded as a binary variable with ‘in work’/’not in work’ categories.
Information about familiarity with a mental illness was drawn from the question ‘Who is the person closest to you who has or has had some kind of mental illness’. The response categories included the participant himself/herself, others (immediate family, partners, friends and acquaintances), and no one if the participant did not know anyone with mental illness. Endorsing the view that one has a mental illness is strongly associated with more positive attitudes [45] and with help seeking and receipt [46], while knowing someone else is associated with more positive attitudes but not as strongly as is personal experience [45].
Statistical analysis
The analysis followed the HSE design of multistage stratified probability sampling using postcode sectors as the primary sampling units (PSUs) and Postcode Address File as the sampling frame for households. Data were weighted for dwelling unit and household selection and for population profile including age, sex, ethnicity and regional structure of England’s population. A non-response adjustment was added for interview visit as well as for the sample that received the nurse visit. The full sampling design is described elsewhere [36].
Fifteen percent of participants did not provide details of their income and less than 1% had missing information on ethnicity, educational qualification and employment status.
We calculated basic descriptive statistics for participant characteristics in addition to help-seeking variables, CAMI-12 and GHQ-12 scores.
For each of the three outcomes, we ran a logistic regression model with and without the interaction between total GHQ-12 and CAMI-12 scores. Socio-demographic characteristics and familiarity with mental illness were controlled for. Statistical significance of the interaction was tested using the Wald test. Statistical analysis was conducted using Stata V.14.2.