Health service use and health outcomes among international migrant workers compared with non-migrant workers: A systematic review and meta-analysis

Objectives The review aimed to synthesise recent evidence on health service use and health outcomes among international migrant workers, compared with non-migrant workers. Methods A search was carried out in MEDLINE, PubMed, Embase, and CINAHL for studies published between Jan 1, 2010, and Feb 29, 2020. Included outcomes were: occupational health service use, fatal occupational injury, HIV, and depression. Two authors independently screened records, extracted data, assessed risk of bias and judged quality of evidence. We meta-analysed estimates and conducted subgroup analyses by sex, geographical origin, geographical destination, and regularity of migration. Results Twenty-one studies were included comprising >17 million participants in 16 countries. Most studies investigated regular migrant workers in high-income destination countries. Compared with non-migrant workers, migrant workers were less likely to use health services (relative risk 0·55, 95% confidence interval 0·41 to 0·73, 4 studies, 3,804,131 participants, I2 100%, low quality of evidence). They more commonly had occupational injuries (1·27, 95% confidence interval 1·11 to 1·45, 7 studies, 17,100,626 participants, I2 96%, low quality of evidence). Relative risks differed by geographical origin and/or destination. There is uncertainty (very low quality of evidence) about occupational health service use (0 studies), fatal occupational injuries (5 studies, N = 14,210,820), HIV (3 studies, N = 13,775), and depression (2 studies, N = 7,512). Conclusions Migrant workers may be less likely than non-migrant workers to use health services and more likely to have occupational injuries. More research is required on migrant workers from and in low- and middle-income countries, across migration stages, migrating irregularly, and in the informal economy.

strategy for MEDLINE was adapted to search other databases (see S1 Appendix). We hand searched reference lists of previous systematic reviews [5][6][7] and records of studies included in this systematic review. The first 100 hits on Google and GoogleScholar and the webpages of ILO, IOM, UNHCR, and WHO were also searched. Experts were asked to identify eligible published and unpublished studies.
Eligibility criteria. The outcomes of interest were any health service use, any occupational safety and health service use, death from an occupational injury, any non-fatal occupational injury, HIV infection and clinical depression. S1 Table presents the eligibility criteria for populations, comparators, and outcomes. All included outcomes are or align with Sustainable Development Goals indicators [9] with relevance for health and migration (S2 Table).
Study selection. At least two review authors independently screened the titles and abstracts of potentially eligible study records from the search results against the eligibility criteria. Of records that we identified as potentially eligible, review authors independently screened the full texts to determine inclusion in the systematic review. The third review author resolved any disagreements.
Data collection process. Data extraction was conducted independently by two review authors. A standard data extraction sheet was developed and trialled by the data extractors. From each included study, we extracted data on study design (data analytic method, model used, and confounder adjustment), participants (number and type of participants), comparator, outcomes, relative risk (RR) measure, point estimate and 95% CI. The third review author resolved disagreements.
Risk of bias assessment. Following Cochrane's approach, we assessed risk of bias by outcome at the level of the individual study and then at the level of the entire body of evidence. Risk of bias assessment tools are lacking for studies of differences in prevalence. We used the RoB-SPEO tool [10] for studies estimating prevalence in occupational health. The assessed domains were: (1) bias in selection of participants into the study; (2) bias due to a lack of blinding of study personnel; (3) bias due to exposure misclassification; (4) bias due to incomplete exposure data; (5) bias due to conflict of interest; (6) bias due to selective reporting of exposures; (7) bias due to difference in numerator and denominator; and (8) other bias. For each domain, the risk of bias was rated as "high", "low" or "unclear" [11]. Two review authors conducted the assessment independently, and the third author resolved disagreements. Consensus ratings for each domain for each study are presented in a "Summary of risk of bias" table [11].
Evidence synthesis. Two authors independently assessed the clinical heterogeneity of included studies on the same outcome, with the third author resolving differing opinions. We combined studies judged sufficiently homogenous empirically in a meta-analysis. Measures of relative differences other than RRs (eg, odds ratios) were converted into RRs if possible using Cochrane's guidance [12]. Studies were pooled using the inverse variance method with random effects models, because included studies estimated different, yet related effects [13]. Review Manager (RevMan 5.3) computer software was used [14]. We assessed statistical heterogeneity of studies with the I 2 statistic. We expected high levels of heterogeneity, so report pooled effect estimates from meta-analyses also when statistical heterogeneity is high (e.g. I 2 > 95%). For outcomes with evidence rated as "very low quality" (meaning that we were very uncertain about the outcome), we report forest plots of meta-analyses, but do not report pooled estimates. The results of the evidence synthesis are presented in a "Summary of findings" table [15].

Additional analyses
Subgroup analyses were conducted by sex; WHO region of destination, WHO region of origin, and regularity of the migration. Such analyses were only conducted of data presented in main meta-analyses with pooled estimates.

Quality of evidence assessment
For each outcome, we assessed the quality of its entire body of evidence, using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [16]. Our assessment considered risk of bias, inconsistency, indirectness, imprecision, and size of the RR estimates. We applied the ratings "high", "moderate", "low" and "very low". Our assessment started at "high". For each domain, we downgraded by one level for serious concerns and by two levels for very serious concerns. Evidence was upgraded by one level and two levels if the estimated RRs were high (� 2.5) and very high (� 5.0), respectively. The "Summary of findings" table includes quality of evidence ratings and their justification [15].

Study selection
Of 1607 study records identified by our search, 21 studies with 22 study records fulfilled the inclusion criteria and were included in this review (Fig 1)  .

Study characteristics
The included studies comprised >17 million participants in 16 countries within five WHO regions (Africa, Americas, Eastern Mediterranean, Europe, and Western Pacific) ( Table 1). Thirteen included studies were cohort studies, and eight were cross-sectional. Most studies investigated regular migrant workers from high-income countries working in the formal economy of high-income destination countries.

Risk of bias
For each outcome, we judged the risk of bias of the body of evidence to be high (see S1 and S2 Figs). Over half of all included studies (13 of 21 studies) were judged to carry a high risk of selection bias because they analysed study samples that were non-representative of national populations of either or both of migrant workers and non-migrant workers. Of note, most studies captured neither migrant workers from irregular migration nor workers in the informal economy. The risk of performance and detection bias was often judged to be overall low because most studies relied on administrative data collected for purposes other than comparing the health of migrant workers with that of non-migrant workers. Most studies carried a high risk of bias from selective reporting because migrant workers may have underreported adverse outcomes at disproportionately higher rates due to relatively higher fear of losing their work if reporting adverse outcomes. Risks of confounding and of bias from conflict of interest were generally low. Several studies carried a risk of bias from differences between the numerator and the denominator.

Findings
Use of any health services. Four studies with a total of 3,804,131 participants from four countries in three WHO regions (Africa, Europe, and South-East Asia) reported an estimate of the use of any health service among migrant workers, as compared with non-migrant workers [22,28,31,36]. The populations and definitions and measurements for this outcome were somewhat heterogeneous (Table 2). Nevertheless, we judged them to be sufficiently comparable to combine them in one meta-analysis. Compared with non-migrant workers, migrant workers were an estimated 45% less likely to use any health services (RR 0�55, 95% CI 0�41 to 0�73, I 2 100%, Fig 2).
We downgraded the quality of this body of evidence by two grades from "high" to "low" quality of evidence. We downgraded by one grade each for serious risk of bias and serious indirectness. We did not downgrade for inconsistency; we had expected heterogeneity to be very high, even due to the heterogeneous study population alone, and we also found very high heterogeneity in the analysis. In conclusion, migrant workers may be less likely to use any health service than non-migrant workers. Further research is very likely to have an important impact on our confidence in the conclusion and is likely to change it.
In subgroup analyses by WHO region (Table 2), the likelihood among migrant workers was reduced in all WHO regions with data included in the systematic review, but it may be particularly lower in South-East Asia. There may be considerable differences in this outcome between regions (Test for subgroup differences: p = 0�02).
Use of any occupational safety and health services. Our systematic review identified no eligible study on the relative likelihood of using any occupational safety and health service among migrant workers, compared with non-migrant workers.
Fatal occupational injury. Five studies with a total of 14,210,820 participants reported an estimate of having a fatal occupational injury among migrant workers, compared with among non-migrant workers [17,20,21,29,35]. We judged the studies to potentially be sufficiently homogenous and combined them in a meta-analysis (Fig 3). We downgraded the quality of this body of evidence by three grades from "high" to "very low" quality of evidence. We downgraded by two grades for very serious risk of bias and by one grade for serious indirectness. Because of the very low quality of evidence, we do not present a pooled estimate in the forest plot. We are very uncertain about this outcome among migrant workers, compared with nonmigrant workers.
Any occupational injury. Eight studies with a total of 17,100,626 participants from five countries in three WHO regions (Eastern Mediterranean, Europe, and Western Pacific) reported an estimate of having had any occupational injury among migrant workers, compared with non-migrant workers [17,18,21,23,26,30,32,33]. We judged seven studies [17,18,21,23,30,32,33] with 21 individual estimates from 13,063,936 participants to potentially be sufficiently homogenous despite some heterogeneity in the population and in the outcome definition and measurement. Migrant workers were 27% more likely than non-migrant workers to have an occupational injury (RR 1�27, 95% CI 1�11 to 1�44, I 2 96%) (Fig 4). The eighth study [26] reported an OR of 1�07 without an estimate of statistical variance for the prioritised measure (occupational bone fracture) and was therefore not be included in the meta-analysis; We downgraded this body of evidence by two grades to "low quality evidence" for serious risk of bias and serious indirectness. In conclusion, migrant workers may perhaps have a higher likelihood of having any occupational injury than non-migrant workers. Further research is very likely to have an important impact on our confidence in the conclusion and is likely to change it.
Subgroup analyses by WHO region found that the risk among migrant workers was elevated in all WHO regions with data included in this systematic review, but it may be particularly elevated in the Eastern Mediterranean and Western Pacific ( Table 2). There may be considerable differences in this outcome between regions (Test for subgroup differences: p < 0�000). HIV. Four studies with a total of 13,775 participants from three countries in three WHO regions (Americas, Europe, and Western Pacific) reported an estimate of having HIV among migrant workers, compared with among non-migrant workers [24,27,37,38]. All studies investigated differences among sex workers. We judged three studies [24,37,38] to potentially be sufficiently homogenous and combined them in a meta-analysis (see S4 Fig). The fourth study [27] reported that a ratio of the odds of migrant workers divided by the odds of nonmigrant workers could not be calculated because no non-migrant workers had HIV. We downgraded the quality of this body of evidence by three grades for serious risk of bias, indirectness and imprecision from "high" to "very low" quality of evidence. Because of the very low quality of evidence, we do not present a pooled estimate in the forest plot. We are very uncertain about this outcome among migrant workers, compared with non-migrant workers.
Depression. Two studies with a total of 7,512 participants from two countries in Europe reported an estimate of being clinically depressed among migrant workers, compared with non-migrant workers [25,34]. We judged both studies to potentially be sufficiently homogenous and combined them in a meta-analysis (see S5 Fig). We downgraded the quality of this body of evidence by three grades for serious risk of bias, indirectness and imprecision to "very low" quality of evidence and do not present a pooled estimate in the forest plot. We are very uncertain about this outcome.

Discussion
This systematic review and meta-analysis included 21 studies of >17 million participants in 16 countries in five regions . Most studies investigated regular migrant workers in highincome destination countries. Compared with non-migrant workers, migrant workers may be less likely to use any health service (Table 3 summarises the findings). They may be more likely than non-migrant workers to have an occupational injury. In subgroup analyses, RRs of these outcomes differed considerably between geographical origin and/or destination regions, but not by sex. We are very uncertain about the likelihood of migrant workers, compared with non-migrant workers, to have used any occupational health services; died from an occupational injury; HIV; and depression.
The body of evidence synthesised in this systematic review and meta-analyses has several limitations. Clinical heterogeneity was high across populations and some outcomes. Migrant workers as a population are diverse in terms of country of origin, stage of migration, formality of migration (regular versus irregular), country of destination, occupation, industrial sector, work in the formal versus informal economy, lengths of residence in the country of destination, gender composition, and level of education, amongst other variables. Differences in health service use and health outcomes are also influenced by up-stream, structural interventions, such as whether migrant workers are covered with employment injury compensation schemes. These upstream interventions will differ between countries. The high clinical heterogeneity observed in this systematic review may reflect these differences in migrant populations and work-related policies and programmes in the studied countries. Some outcomes were also relatively heterogenous in one or both of definition and measurement. Statistical heterogeneity was also high in the meta-analyses for most outcomes (I 2 � 95%). We caution the interpretation of the pooled estimates and suggest these are viewed as indicative, not conclusive.
We judged the current body of evidence to be seriously indirect for most outcomes included in this review. The existing evidence covers a subpopulation of migrant workers. These are primarily migrant workers at the country of destination, residing in high-income destination countries, who have migrated regularly and work in the formal economy. Some of the migrants covered in the included studies are North-North migrants. There is scarce scientific evidence is available on migrant workers coming from low income country and migrating to low-middle income countries. Therefore, this current body of evidence does not capture migrant workers in more vulnerable situations, including those in transit (e.g., in detention) and return (e.g., those who returned after acquiring an occupational or work-related injury or High quality: further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: we are very uncertain about the estimate.
� The basis for the assumed risk is the median control group risk across studies. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI disease in the destination country); migrant workers residing in low-and lower middleincome countries; those who have migrated irregularly (and therefore do not enter civil registration and in turn may not qualify for public services and benefits); and migrants working in informal economies. The evidence reviewed can, therefore, only partially answer the systematic review's research questions. High risk of bias presents a serious concern in this systematic review for several outcomes. Most included studies investigate non-representative samples. Even probability studies or complete censuses of general populations (e.g. all occupational injuries registered in the national injury register or employment injury compensation scheme) have generally either actively or de facto excluded irregular migrant workers because their lack of civil registration and entitlement for compensation means that they do not appear in official registries and compensation records, respectively. There is a high risk of bias from disproportionate underreporting of cases of occupational and work-related diseases or injuries due to fear of losing one's job among migrant workers. For this reason, different approaches to data collection may be necessary in future research: approaches need to be able to capture information on the health outcomes of all migrant workers, including those in the informal economy or with irregular status.
The body of evidence is limited to comparisons of migrant workers in destination countries with non-migrant workers in the same country. If migrant workers who experience changes in health services or health outcomes return to their country of origin as a result (e.g., because they lose their right of residency as soon as they are no longer fit for work or have lost their employment), they are lost from official registries and national cohort studies. The adverse health effects they experienced will also not be captured in these national data and information systems. This can lead to an underestimation of the RR of migrant workers compared with non-migrant workers. Studies that compare migrant workers pre-departure in their country of origin with the same migrant workers after return to their country of origin are needed. These studied were eligible for inclusion in this systematic review, but we did not identify any.
This review was limited to migrant workers who had entered their country of destination through regular migration (with few exceptions). Fitness to work established through medical examination is often a prerequisite for regular migration for employment. This would lead to selection based on health status, where applicants for migration who are in less than ideal health are excluded from regular migration, and only relatively healthy persons can enter a country of destination as regular migrant workers. Additionally, regular migration often also depends on the applicant fulfilling minimum education and wealth requirements that generally lie above the average level achieved in non-migrant workers in the country. In summary, this health selection may explain any health advantage observed in regular migrant workers, compared with a general sample of non-migrant workers in the same country.
This review highlights that more high-quality research is needed on these differences in workers' health services use and health outcomes by migrant status. More research is required on broader and more diverse populations of migrant workers, particularly on the forms of migration and the stages of migration, in which more careful identification of migrant workers and measurement of their health status is made. Longitudinal studies, such as cohort studies, may be particularly informative; however, large scale population-based studies, using data from population censuses for example, are also needed. This could include research on migrant workers from and in low-income countries, in migration stages other than destination, from irregular migration, and migrants working in the informal economy. Harmonized internationally standardized measures will aid comparability between studies. Further research will strengthen the evidence base for designing, planning, costing, implementing and evaluating laws, policies, and interventions that promote the safety, health and wellbeing of workers.