‘The Healthy Migrant Effect’ for Mental Health in England: Propensity-score Matched Analysis Using the EMPIRIC Survey

This study utilised data collated for the EMPIRIC national population survey in 2000 (n = 4281). The EMPIRIC study is a cross-sectional survey carried out to assess the prevalence of common mental disorders in different ethnic groups including White, Black-Caribbean, Bangladeshi, Indian and Pakistani groups in England. Participants from the ethnic minority groups were obtained from the 1999 Health Survey for England [15]. White participants were taken from the 1998 Health Survey for England [16]; the general population questionnaire and measures being very similar to the ethnic minority 1999 Survey. The EMPIRIC study excluded the 8% of participants who did not want to be re-contacted and selected their sample from participants aged 16–74. Data from the EMPIRIC study were accessed through the UK data archive and this study adhered to the conditions to the agreements of use. Ethical approval for EMPIRIC was obtained from the North Thames Multi-centre Research Ethics Committee (MREC) and ratified by all Local Research Ethics Committees (LRECs) in England.

The EMPIRIC study utilised questions from existing validated instruments such as SF-12 physical and mental health summary scales [17], and the revised clinical interview schedule (CIS-R) [18]. Social support was measured using tools such as Whitehall II study of British civil servant [19], and social functioning questionnaire [20]. Full details of the EMPIRIC survey including the measures recorded are described in greater detail [21, 22].

An assessment for the presence of a common mental disorder was undertaken using the CIS-R, a widely used tool in the UK [23, 24]. The CIS-R is considered by many researchers to be the most valid and reliable tool for assessing common mental disorders, particularly anxiety and depressive disorders, according to ICD-10 and DSM-IV diagnostic criteria [25, 26]. The CIS-R enquires about the presence and severity of 14 non-psychotic symptoms in the week before the interview. These cover somatic complaints including fatigue, problems with memory and/or concentration and sleep disturbance, and emotional symptoms such as depressed mood and thoughts, anxiety and obsessional thoughts. A CIS-R score of 12 or more has been shown to be a suitable case threshold in determining the presence of a common mental disorder, and was used in the present study [21]. The EMPIRIC survey documented whether participants were immigrants allowing the present study to define the risk of common mental disorders by immigrant status.

The present study controlled for several factors that have been shown to affect the risk of developing common mental disorders [27]. The participants gender and age were recorded, the latter being categorised into three bands; 16–34, 35–54 and 55–74 years old. Social class was based on the occupation of the head of the household and classified according to the Registrar General’s classification using the six-point scale; class I (professional), II (managerial/technical), IIIa (skilled, non-manual), IIIb (skilled, manual), IV (semi-skilled), V (unskilled) [28]. Participants’ marital status was recorded into five categories; single, married, separated, divorced and widowed. Educational achievement was classified into NVQ4/5/degree equivalent, higher education below degree, NVQ3/A-Level equivalent, NVQ2/O-Level equivalent, NVQ1/CSE equivalent, foreign/other educational achievement or no qualifications. The religious or spiritual beliefs of participants were recorded into religious, spiritual or none. Ethnic origin was self-defined using the census classificatory system [29]; White, Black-Caribbean, Indian, Pakistani and Bangladeshi. The short form 12-item survey (SF-12), derived from the longer SF-36, measures the participants’ physical health and has been demonstrated to be reliable in clinical and population-based applications worldwide [17, 30]. The participant’s answers are computed to give a score from 0 to 100, where 0 indicates the lowest level of health and 100 indicates the highest level of health [17]. In order to get a notion of the social interaction and potential adversity faced by participants this study utilised information on whether there was regular contact with friends (yes/no), number of people participants were close to (0–60) and whether they had been physically attacked in the last 12 months (yes/no). Chronic social strain was measured by questions from the Whitehall II study [19]; a five-point Likert scale was used for domains considering problems with relatives, financial problems, housing problems, and difficulties in the local neighbourhood. These were combined to give an overall score (on a scale from 0 to 5), where a higher score indicated more frequent or greater social strain [20].

Baseline characteristics of the sample were calculated using proportions and medians as appropriate and were compared between the immigrant and non-immigrant groups using χ² test for categorical variables and Mann–Whitney U test for continuous variables. We determined a propensity score for common mental disorders according to immigration status using a logistic regression model [31, 32].

Propensity-score matching is a statistical principle that attempts to overcome a key limitation of making causal inferences from observational study designs, where no systematic methods have been used to maintain control groups. Propensity-score matching involves pairing exposure (or treatment) and comparison units in terms of their observable characteristics. This produces two groups that are matched according to a set of measured covariates. Propensity-score matching methods are being increasingly used in medical research to provide unbiased estimates from observational data where assignment to treatment or exposures is typically not random [33, 34].

The covariates entered into the propensity score are described in detail earlier and include socio-demographic factors (gender, age, social class, education attainment, ethnic background and religious or spiritual beliefs), physical health status (SF-12 general health score) and social measures (chronic social strain score, number of people close to, regular contact with friends and whether physically attacked in the previous 12 months). Immigrants were matched to non-immigrants on the logit of the propensity score by using calipers of width equal to 0.2 of the standard deviation of the logit of the propensity score [35]. A matching ratio of 1:1 was used [36]. The matched characteristics of the cohort represent reweighted figures on the propensity score [37].

We estimated standardised differences for all covariates before and after matching, with a standardised difference of 20% or more considered indicative of imbalance [34, 37]. The presence of common mental disorders between immigrant and non-immigrant participants were compared after matching using methods appropriate for the analysis of matched data to estimate the effect of immigration status and its significance. We determined the odds ratio of common mental disorders using a logistic regression model. We determined three models of common mental disorder risk development according to immigrant status; unadjusted model (model 1), adjusted model for covariates (model 2) and for covariates entered into the propensity score for matching both immigrants and non-immigrants (model 3).

To assess the sensitivity of the results we tested the influence of unmeasured confounders, which if neglected may influence the participation probabilities, i.e. two individuals with the same observed covariates may have differing chances of allocation to treatment/exposure group. In a study free of hidden bias, participation probability will be based solely on their observed measurements. However, with hidden bias an unmeasured factor may be driving allocation to treatment and thus moderating any perceived treatment effect. Using the Mantel–Haentzel statistic using the STATA programme mhbounds [38], we calculated the test statistic for the risk of screening positive for a common mental disorder for differing degrees of assignment due to unobserved factors. All analyses were performed in STATA version 13 for Windows [39].