The association of migration experiences on the self-rated health status among adult humanitarian refugees to Australia: an analysis of a longitudinal cohort study

We conducted a secondary examination of the first three waves of the BNLA study, collected from October 2013 to March 2016. The BNLA is a five-year national study (2013–2018) conducted by the Australian Government’s Institute of Family Studies to examine how humanitarian refugees settle into a new life in Australia [18]. The Australian Federal Government’s Department of Social Services (DSS) funded the BNLA study. Further information about the BNLA study design can be found in publicly available documents [18].

The BNLA study received ethics approval by the Australian Institute of Family Studies Human Research Ethics Committee. De-identified BNLA data is accessible by authorized researchers who have obtained permission from the Australian Department of Social Services. This permission was obtained by study authors AD, JE and GR. Ethics exemption to use the data was granted by the Monash University Research Ethics Committee (see Additional file 1).

The BNLA cohort comprised of individuals aged 15 years and over who had been granted a permanent humanitarian visa by the Australian Government in the 3–6 months preceding the baseline BNLA study [18]. Eligible participants were identified via the Australian Department of Immigration and Border Protection settlement database from eleven locations around Australia to ensure adequate sample size. Migrating Units (MU) were the primary sampling units for the study and consisted of principal applicants (PA) and secondary applicants (SA). Principal Applicants were the lead participant for the study, and were the initial individuals contacted for participation. Secondary Applicants comprised other members of the Migrating Unit listed on the Principal Applicant’s visa application. The Principal Applicant had to consent to participating before SAs could be invited to do so [18]. A total of 2031 out of 4035 eligible Principal Applicants were contacted and 1509 Principal Applicants and 890 Secondary Applicants completed baseline interviews. The respondents are from 35 different countries and speak 50 different languages and at baseline, aged between 15 and 83 years [18].

The first three waves of ‘Building A New Life in Australia’ data were obtained using either home visits (Waves 1 and 3) or telephone interviews (Wave 2). The BNLA written survey was translated into 14 different languages for Wave 1. A total of 19 languages were covered for the survey with the aid of interpreters [18].

A number of predictor variables were selected from the BNLA survey and related to migration and resettlement, socio-demographics and the presence/absence of chronic conditions.

A conceptual framework used for this study was informed by Cross-Denny & Robinson’s social determinants of health model [19]. The Cross-Denny & Robinson’s model uses five key areas or determinants that are particularly relevant for oppressed and marginalised populations. We added “Political, Socio-Economic” category to encompass the significant associations between structural/political factors (such as spent time in immigration detention centres, migration pathway) and poorer health among refugees. Our adapted model includes six key determinants of refugee health and 25 candidate variables (See Fig. 1). A review of the literature reporting predictors of refugee health outcomes informed the selection of candidate variables from the BNLA dataset to populate the social determinant of health model.

Factors relating to migration and resettlement in host country and health were measured by the ‘Building A New Life In Australia’ self-report survey [18]. Our principal outcome variable was self-rated general health and used the question “Overall, how would you rate your health during the past 4 weeks?” selected from the SF-36 [20]. The SF-36 is a generic measure of health status with reliability and validity among refugee populations [3]. The BNLA survey uses a 6-point Likert response scale (“Excellent” to “Very Poor”). For our study, we use a dichotomous variable comprising of two health response categories of “Excellent-Good” and “Fair-Very Poor”. The decision regarding the creation of the two response categories was based on a review of the literature, which confirmed that our dichotomous response groupings were appropriate [21].

Analysis included descriptive and generalized linear mixed modelling (GLMM). Descriptive statistics describe the overall characteristics of the sample population at baseline, such as socio-demographics, migration experiences using means and standard deviations (SDs) for continuous variables and frequencies and percentages for categorical variables. All analyses used adjusted weights, provided by the BNLA project. Variables with over 10% missing data at baseline were excluded if we were unable to account for the missing data. Only variables asked across all three waves were included in the analysis. Continuous outcome variables, such as age, were transformed into categorical variables. All analyses were done with Stata/SE 15.1 [22].

To examine the relationships longitudinally between outcomes and predictor variables, GLMM were used. As the BNLA data involved multi-level, clustered sampling, GLMMs accounted for inter-individual random effects between different ‘migrating unit’ clusters (ie. the families) and individual participants, as well as intra-individual fixed effects or the repeated measures within individual participants. GLMMs were appropriate for our analysis as ordinary linear models are unable to handle clustered data. Variable selection for the final models occurred in several steps. Firstly, univariate regressions were used to examine associations between baseline candidate variables and outcome variables, with variables retained at p < 0.1. In the second step, collinearity was then examined using Pearson’s correlation coefficient to remove highly correlated variables from the final model. In this study, one candidate variable was selected from two or more correlated variables when r > 0.5. The third step involved examining associations between candidate variables and outcome variables using multivariate GLMM. A variable denoting study time: ‘wave’ was included in these models. Variable at p < 0.1 were retained. In the last step, candidate variables were modelled over 1000 bootstrap samples at 95% resampling for the data set. Candidate variables were retained in final models if they were identified as significant (p < 0.05) in 1000 bootstraps over 50% of the time.

The BNLA is funded by the Australian Government’s Department of Social Services and were not involved in the preparation of this manuscript or the analyses reported.

At baseline, 2399 individuals took part in the BNLA study. The mean age was 35.5 years (SD = 13.9) and 54.0% were male. Over half of the participants (61.0%) were married or partnered and nearly half (45%) were born in the Middle East. Almost a third (32.5%) had little to no schooling upon arrival to Australia and 12.1% held university qualifications. Before coming to Australia, 33.6% could speak English ‘very well-well’. A vast majority (88.9%) of participants had exposure to one or more traumatic experiences. Mostly, participants had been in Australia less than six months (82.1%) and were living in metropolitan cities (90.7%). There was little financial stability with high unemployment (90.7%); high dependency on government income support (88.0%) and at least one financial daily hardship (42.3%). Only 9.3% were in paid employment at baseline. Stress caused by not having work, stable housing and language barriers was reported by 33.0, 28.1 and 56.0% of participants, respectively. Having a ‘chronic’ health condition was reported by 23.0% of participants at baseline. See Table 1.

Table 2 provides the results of generalised linear mixed effect models, with general health as the outcome variable and pre- and post-migration factors as the predictor variables across the three years of follow-up. Individual characteristics, such as being female, older in age, originating from the Middle East and having a university qualification were significant positive predictors of poorer general health. For example, our results show that over the three years of follow up, females have 2.02 the odds of reporting ‘Fair-Very Poor’ health compared with males. In terms of post-migration stressors, several economic factors were associated with the reporting of ‘Fair-Very Poor’ health among this refugee population. An increasing number of daily financial stressors was significantly associated with higher odds of reporting poorer health. For example, those with 3–6 daily financial hardships were at increased risk for reporting poorer general health than those without any daily financial stressors. “Other” main income source (ie. spouse/partner/parent’s income) was also positively associated with poorer health (OR: 3.25, 95% CI: 0.013).

Not having a chronic disability/injury or condition was associated with reporting ‘better’ general health (OR: 0.15; 95% CI 0.09–1.04) among this refugee population. That is, those without a chronic health condition have 0.15 the odds of reporting ‘Fair-Very Poor’ health compared with those with a chronic health condition. Another way to interpret the results is that those who do not have a chronic health condition are 85% less likely to report poorer general health. We found no significant association between the health outcomes and duration of stay in Australia, education, accommodation, social support or discrimination.

The mean general health prevalence among humanitarian refugees was calculated in two ways: (1) using BNLA sample data, and (2) using the final model adjusting for model independent variables. Both methods used the provided sample weights. Both methods showed no significant changes in general health over time. See Table 3.

Using BNLA sample data, 35.7% (33.8–37.7%) of participants reported “Fair-Very Poor” self-rated health at baseline (see Table 2). At Wave 2, 691 (36.8%) participants’ self-reported ‘Fair-Very Poor’ general health (95% CI 34.6–39.0%) and 685 (39.7%) had ‘Fair-Very Poor’ general health (95% CI 37.4–42.0%) at Wave 3.

Table 3 shows the prevalence estimates of “Fair-Very Poor” self-rated health calculated using the final GLMM model. In Wave 1, the prevalence for ‘Fair-Very Poor’ general health was 34.6%. In Waves 2 and 3 the prevalence was 34.0 and 33.4%, respectively. This shows that the prevalence of ‘Fair-Very Poor’ general health remained high across the 3 years but did not decrease or increase substantially over time. The significance of change in ‘Fair-Very Poor’ general health over time was also examined using the final GLMM model.

Between Waves 1 and 2, there was no significant decrease in the prevalence of “Fair-Very Poor” self-rated health (p = 0.779). See Table 3. Waves 1 and 2 saw no significant differences in “Fair-Very Poor” self-rated health (p = 0.802). There was also no significant overall decrease between Wave 1 and 3 for “Fair-Very Poor” self-rated health (p = 0.512). Therefore, the results indicate that there was no significant change in “Fair-Very Poor” self-rated health over the three years of follow-up. In summary, on a population level, the point prevalence of “Fair-Very Poor” self-rated health remained high, but no overall significant increase (or decrease) in “Fair-Very Poor” self-rated health was seen over time.

There are several methodological limitations in this study. Retrospective reporting and reliance on self-reporting may run a risk of not remembering or misrepresenting the events and non-accurate measurement of symptoms [40], presenting a risk of recall bias, personal interpretations across languages and cultures. Further, the BNLA study collected information that is more comprehensive from principal visa applicants than from secondary applicants, which constrains the depth and breadth of findings of the study. Alternating face-to-face with telephone interviews may have introduced some inconsistencies across waves in that some respondents may have answered some questions differently depending on the interview style. Despite these limitations, this study has contributed to the literature by providing information about the migration experiences of a large and ethnically diverse cohort of refugees during their first three years of resettlement in Australia and the impact of such experiences on general health.

Our results show that there is persisting high prevalence of poorer general health among adult refugees across the first three years of their resettlement in Australia. This finding suggests unmet health needs which may be compounded by the challenges of resettlement in a new society. This result highlights the need for increased clinical awareness of this sustained health burden to help inform and prepare refugee health care and settlement service providers. It also underscores the importance for health professionals to consider broad health issues among refugee populations and highlights the potential value of the Australian Medical Benefit Schedule subsidy of refugee health assessments for recent refugee arrivals [41]. The association between poorer general health and settlement economic stressors highlight the importance of understanding and eliminating the barriers, which lead to the high rates of unemployment among refugees, as a way of preventing negative health consequences. Qualitative research into the barriers to employment, using the perspectives of refugees themselves; would be a valuable addition to the current knowledge gap. This study also highlighted that attention should be given to assessing the specific needs of groups of refugees based on their demographic characteristics; for example, older, Middle Eastern females. Given the vulnerabilities of these groups, sustained targeted interventions upon entry to Australia would be beneficial to offset any further decline in health.