Design
This study is based on a secondary analysis of data obtained from a population-based cross-sectional study with a full-census approach in a convenience sample of three administrative districts in Baden-Württemberg, one of the largest Federal States with 44 administrative districts in the South of Germany. The analysis is explorative in the sense that no sample size calculation was performed in advance due to a lack of studies with similar scope among asylum-seeking populations. This means that effect estimates are judged upon regarding their direction and magnitude rather than upon their statistical significance (or non-significance).
Participants
Asylum-seekers were considered individuals who have applied for recognition as refugee in Germany and are awaiting a decision on their application (AsylVfG section 55), are “tolerated” (AufenthaltG section 60) or hold a permit on humanitarian grounds (AufenthaltG section 25). Asylum-seekers aged 18 or above were eligible to participate in the study.
Data collection and recruitment
The study protocol was approved by the ethical committee of the Medical Faculty of the University of Heidelberg prior to onset of the study (Ethical approval nr: S-261/2014). Data were collected between October 2014 and February 2015 on the occasion of monthly payments of welfare benefits to asylum-seekers either in the accommodation centre or in the Welfare Agency, where all registered asylum-seekers are obliged to be personally on-site. In a full census approach, all asylum-seekers meeting the inclusion criteria were invited to participate. They were informed by the research team in written and oral form about the voluntary nature of participation, anonymity and confidentiality of data, emphasizing that participation would neither influence the health care situation nor the asylum procedure or residence status. Data were collected on health care access, morbidity, and SES using a questionnaire with mainly standardized instruments in seven languages (German, English, French, Arabic, Persian, Serbian, Russian) tailored to the languages most frequently spoken among registered asylum-seekers. Details on the translation process are provided in Additional file 1. Demographic data on non-responders and reasons for declining participation were documented to perform a non-responder analysis.
Measurement of access (dependent variables)
Access to health care as an outcome is a multi-layered concept to be best approximated by measuring health service utilization and unmet need [16–18]. To quantify utilisation participants were asked whether they had used any of the following medical services during the past 12 months: (1) Any physician (in-or outpatient); (2) General practitioner (GP); (3) Psychotherapist; (4) Hospital inpatient. Unmet medical need was recorded if participants indicated that at least once in the past 12 months they had felt a need for medical care but did not receive it. Each item was used as a binary dependent variable in different models.
Measurement of SES (independent variables)
Capturing and comparing SES among asylum-seekers in the traditional “objective” sense is challenging as it requires measurement of SES indicators (education, income, profession) in a fashion that is applicable to both a broad array of countries (e.g., .standardised measurement of education and professions) and to the legal particularities entailed by the asylum-seeking process. At the time of data collection for this study, asylum-seekers did not receive working permits during the first 9 months of their stay in the country. Further regulations limiting “real” access to the labour market during the first 15 months after receipt of a working permit rendered questions for income inapplicable and difficult to operationalise [19].
Since the impact of social standing in society on health goes beyond the effects of objective SES and includes the effects of subjective positioning into the social hierarchy that societies create [20, 21], we decided to capture SES using a two-pronged strategy:
Firstly, measurement of the highest educational attainment as “objective” SES indicator by the question “What is the highest education you received?” with the response options “None at all; Primary school; Secondary school; Religious school; Tertiary/University”. The option “religious school” was collapsed with “Secondary school”, so that the highest educational attainment was used as an ordinal variable with four categories (none/primary/secondary/tertiary education).
Secondly, we captured the participants’ subjective social status (SSS) in Germany as comprehensive measure of SES by asking participants to put themselves on a 10-rung social ladder modelled after the MacArthur Scale [20, 22]. The SSS index was grouped into three groups: Low (1–4), Medium (5–6), High (7–10) [23].
Both measures of SES were used as independent variables in separate models. In models that contained both measures jointly, we used SSS as independent and educational attainment as control variable.
Measurement of need (independent variables)
Following common approaches [14], we used self-rated general health status, quantified on a five-point Likert scale and dichotomized into ‚Good’ (very good/good) and ‚Bad’ (fair/bad/very bad), to approximate morbidity and need for health care among asylum-seekers. Further need variables were age (continuous) and sex, which we included hypothesising that higher age and female sex indicate higher health care needs.
Control variables
The place of residence (district 1, 2 or 3) was included as a covariable to assess potential differences in access that are explained through geographic characteristics. Furthermore, a language variable with three categories (‘German’, ‘English’ and ‘Other language’) was generated and included in the multiple regression analysis if the language proved to be significantly associated with health care access in the bivariate models.
Statistical analysis
We calculated sex-stratified absolute and relative frequencies for categorical variables, and means, medians and standard deviations (SD) as well as the interquartile range (IQR) for continuous variables. To assess the relationship between dependent, independent and control variables, we designed causal diagrams to visualise the potential relationship between the variables and guide the development of statistical models (Additional file 1). The number of variables was restricted in favour of a higher model power.
We calculated unadjusted Odds Ratios (OR) with 95 % confidence intervals [CI] in simple logistic regression models for each pair of dependent and independent variables. Multiple logistic regression analysis was performed to investigate the adjusted association between dependent and independent variables as well as the joint explanatory effect of the predictor variables on access to health care. Variation inflation factors (VIFs) were calculated to assess multicollinearity among covariables and all VIFs of included variables ranged between 1 and 2.
We built three models for each combination of SES indicators: model 1 contained educational attainment as main independent variable, model 2 SSS as main independent variable, and model 3 both educational attainment and SSS (whereas the first was used as control variable). Separate models were built for each of the five dependent variables: (A) Physician, (B) GP, (C) Psychotherapist, (D) Hospital, (E) Unmet medical need, so that in total 15 models were analysed (Model 1A–E, Model 2A–E, Model 3A–E). A stepwise inclusion of the need and control variables was performed as illustrated in Additional file 2.
For the analysis of horizontal equity, the final models describe the association between SES and access to care under control of sex, age and morbidity. The analysis of vertical equity includes the SES adjusted association of need variables on access to care. Goodness of fit was assessed using the Bayesian information criterion (BIC). All statistical tests were calculated on a significance level of alpha = 0.05 in an exploratory manner and analyses were carried out with STATA 12.1.
Missing data and non-responder
Missing data were treated as missing at random and a complete case analysis was performed. We performed a non-responder analysis to assess potential differences in sex and language between responding and non-responding AS.