Frailty in older-age European migrants: Cross-sectional and longitudinal analyses of the Survey of Health, Aging and Retirement in Europe (SHARE).

Frailty correlates with morbidity and is superior to chronological age in predicting mortality. Frailty of older migrants has important implications for the demands placed on healthcare systems. Examining 95,635 Europeans in the Survey of Health, Aging and Retirement in Europe, we investigated cross-sectional and long- itudinal associations between migration and frailty at ages>50 years. We examined whether associations di ﬀ ered by countries' level of healthcare coverage and access for migrants and tested mediation by home- ownership and citizenship. Cross-sectionally, ﬁ rst-generation migrants>50 years old were, on average, 16.4% (95% con ﬁ dence interval [CI]: 14.6, 18.2%) frailer than non-migrants after confounder-adjustment. This de- creased to 12.1% (95% CI: 10.1, 14.1%) after adjustment for citizenship. The strength of association between migrant status and frailty was greater in migrants from low-or-middle-income countries, compared with mi- grants from high-income countries. Migrants into Northern, Western and Eastern Europe were 37.3% (95% CI: 33.2, 41.5%), 12.2% (95% CI: 10.0, 14.6%) and 5.0% (95% CI: 0.5, 9.6%) frailer than non-migrants, respec- tively, but migrants into Southern Europe were no frailer than non-migrants. The strength of association between migrant status and frailty was greater in countries with lower healthcare coverage and access for migrants. However, citizenship attenuated this di ﬀ erence. Longitudinally, migrants were frailer than non-migrants at 50 years old and trajectories converged over time until migrants and non-migrants were equally frail by 80 – 90 years. Our work ﬁ nds no evidence of the ‘ healthy migrant e ﬀ ect ’ outside of Southern Europe in older migrants and suggests that acculturation is a key determinant of migrant health.


Introduction
By 2010, Europe hosted 72.6 million migrants (8.7% of the European population) (Koser and Laczko, 2010) and the migrant proportion continues to grow (Rechel et al., 2013). Reasons for migration are heterogeneous and may be driven by ill health, educational and occupational possibilities, or by external forces such as trafficking, persecution, conflict or natural disaster (Laczko and Appave, 2013).
Although immigrants, on average, have lower socio-economic status (SES) than native-born comparators (Weitoft et al., 1999), migrant populations are frequently healthier than non-migrant populations on arrival in their host countries. This is known as the 'healthy migrant effect' (Kennedy et al., 2015;Rechel et al., 2013). Their health advantage may decrease or reverse over time because (a) acculturation causes immigrants to behave increasingly like native-born populations in socio-economic terms and health-related behaviours and/or (b) adverse exposures during early life may be latent at the time of migration and emerge in later life (i.e. post-migration) (Solé-Auró and Crimmins, 2008;Williams, 1993). In addition, citizenship rights and access to health care for migrants may have long-term impacts on economic and health outcomes (Malmusi, 2014). Countries with policies that are more inclusive of migrants are often countries with greater wealth, nationalised healthcare systems, a strong commitment to equal rights and opportunities and countries with higher migrant proportions (Huddleston et al., 2015).
Frailty is a multifactorial state of physiological dysregulation, vulnerability to stressors and increased risk of poor health outcomes due to diminished reserve (Malaguarnera et al., 2013) which is superior to chronological age in predicting mortality in European populations Romero-Ortuno and Kenny, 2012). It is influenced by biological factors such as ageing, multi-morbidity and functional status (Gobbens et al., 2010), cognitive and psychological factors https://doi.org/10.1016/j.socscimed.2018.07.033 Received 2 October 2017; Received in revised form 1 April 2018; Accepted 19 July 2018 (Levers et al., 2006) and socio-economic factors including education, wealth (Etman et al., 2012) and social vulnerability (Casale-Martínez et al., 2012).
Previous cross-sectional studies using the Survey of Health, Aging and Retirement in Europe (SHARE) have reported that migrants older than 50 years are significantly more likely to be depressed than nonmigrants (Aichberger et al., 2010;Ladin and Reinhold, 2013) and identified significant regional heterogeneity in frailty Solé-Auró and Crimmins, 2008). Solé-Auró and Crimmins (2008) reported that dependant on country, measures of health and functional ability are either non-differential or significantly worse in European migrants compared with natives. Brothers et al. (2014) find that migrants from low or middle income countries (LMIC) exhibit significantly increased frailty compared to migrants born in high income countries (HIC) and natives in Northern and Western Europe but are no frailer than natives in Southern and Eastern Europe. Observations that migrant versus non-migrant health differences vary by host country migrant integration policies (Giannoni et al., 2016;Malmusi, 2014), by their country of origin economic status and by European host region, and are inversely related to whole-population health Solé-Auró and Crimmins, 2008), highlight the potential role that country-level factors may play in migrant versus non-migrant health inequalities. However, descriptions of the health of migrants in Europe are scant, particularly longitudinally (Rechel et al., 2013).
We hypothesised that differences in frailty, and change in frailty over time, between first generation migrants and non-migrants in Europe would be modified by migrants' economic origins, host region and, for the first time, country-level migrant integration policies (the magnitude of any migrant health inequality being reduced by more inclusive migrant health policies). We aimed to quantify the extent to which post-migration socio-economic status and, for the first time, citizenship may explain these expected differences in frailty and hypothesised that acquisition of citizenship may attenuate effect modification by host country migrant integration policies. This work builds upon existing cross-sectional studies of European migrant health in SHARE (Aichberger et al., 2010;Brothers et al., 2014;Ladin and Reinhold, 2013;Solé-Auró and Crimmins, 2008) by incorporating further waves of data collection. This increases the number of subjects' initial interviews available for cross-sectional analyses and permits longitudinal modelling of frailty trajectories (i.e. frailty over time) at ages over 50 years. It aims to enrich our understanding of the health issues facing an aging and expanding European migrant population to support policymakers in the strategic planning of health and social care delivery.

Data sources and study design
SHARE is a multinational, multidisciplinary representative observational survey of health, economic factors, social and family networks in community-dwelling persons aged ≥50 years and their partners or spouses, regardless of age. Participating households contained at least one non-institutionalised member aged ≥50 years old who spoke the official language of the country and was not living abroad at the time of survey. Interviews were conducted by trained lay interviewers from professional survey agencies in respondents' homes using structured computerised questionnaires. Sampling methods varied by participant country as described in the SHARE documentation (see below). Participating countries at wave 1 (2004)(2005)   . We retained a final study sample as described in Fig. 1 which included 95,635 unique individuals from 66,600 households who participated in 193,747 interviews.

Migrant status, host and origin exposures
First-generation migrants were defined as individuals who reported being born outside their current country of domicile in SHARE. Using similar methods to Brothers et al. (2014), we grouped migrants into having originated from LMIC or HIC according to their country of birth using the World Bank Analytical Classification and taking the modal group over 1987(World Bank, 2014. Age at migration and time since migration were derived from self-reported year of migration. We classified their destinations into Northern, Western, Eastern and Southern Europe as per United Nations definitions (United Nations Statistics Division, 2015) but re-coded Slovenia from Southern to Eastern Europe for the purposes of studying migration.  2014), we mapped deficits onto a 0 (absent) to 1 (present) scale (a) with as many deficits as possible coded as ordinal (graded severity of deficit) in the 'ordinal' coding scheme and (b) only dichotomous values in the 'dichotomous' coding scheme. FI scores were derived by summating an individual's mapped deficits and dividing by the number of deficits recorded. Frailty indices were coded as missing at wave 3 (23,846 interviews) where health deficits were not measured and in 2758 other interviews where > 20% of deficits were missing. FI scores were log-transformed and entered as a continuous variable into regression and multilevel models so that residuals were approximately normally distributed.

Potential confounders
The following were considered to be potential confounders: age group (≥50 to < 60, ≥60 to < 70, ≥70 to < 80 and ≥ 80 years), gender (female versus male), height (proxy for adverse early life premigration exposures; < 160, ≥160 to < 170 and ≥ 170 centimetres), education (proxy for pre-migration SES) and the existing migrant population based on the United Nations host country migrant stock data (< 5, ≥5% of host country population) (United Nations, 2013). Peak educational attainment was categorised into three tiers based on International Standard Classification of Education (ISCED) 1997 codes: none, pre-primary or primary education (ISCED 0, 1), lower or upper Fig. 2. Association between migration and frailty in models adjusting for confounders and potential intermediaries stratified on high and low or middle income country of origin. a Low or middle income country. b High income country. c Percentage difference in frailty in migrants versus non-migrants from linear models with 95% confidence intervals. secondary education (ISCED 2, 3) or post-secondary education i.e. tertiary or post-secondary non-tertiary education (ISCED 4-6).

Potential mediators
Home ownership and quintiles of household income (measures of adult/post-migration SES) and citizenship (both a marker of migrant acculturation and of legal status (Riosmena et al., 2015)) were considered to be potential mediators of the association between migrant status and frailty.

Subgroup analyses
We postulated a priori that the effect of migration on frailty may be modified by country of origin, both in terms of geography and economic development, and host country policy to make health care accessible to migrants. We therefore assessed whether associations between migrant status and frailty differed in the following subgroups: (i) LMIC-and HIC-born migrants, (ii) Residents of Northern, Southern, Eastern and Western European countries, (iii) Tertiles (1 being most inclusive; 3 being least inclusive) of the Migrant Integration Policy Index (MIPEX) IV (2014) (Huddleston et al., 2015) health stranda proxy of healthcare coverage and access which is described in Online Resource 2. Interactions were formally tested using likelihood ratio tests.
2.7. Statistical analyses 2.7.1. Cross-sectional associations of migration with frailty at baseline interviews Cross-sectional modelling of the association between migration status and frailty used data from subjects' initial interviews. Multiple linear regression models were fitted to estimate differences in frailty between migrants and non-migrants accounting for clustering at the household level. We assessed the following regression models: (i) unadjusted; (ii) adjusted for age and each potential confounder separately (listed above); (iii) adjusting for all potential confounders; (iv) additionally adjusting for each potential mediator separately (listed above); (v) fully adjusted for all potential confounders and mediators. Confounders and potential mediators were assumed to have the same effect on outcome irrespective of migrant status, since exploratory analyses provided no evidence for such interactions. Regression coefficients were exponentiated and are interpreted as the percentage difference in FI score in migrants compared to non-migrants. We derived crude estimates of excess hazard of death associated with differences on the FI score, based on the work of Romero-Ortuno and Kenny (2012) to highlight the public health relevance of these variations.

Associations of migration with trajectories of frailty at ages over 50
Individual trajectories of log FI from age 50 years were estimated using random effects multilevel models, fitted in MLwiN v2.28 using the Stata (StataCorp, College Station, Texas) command 'runmlwin' (Leckie and Charlton, 2013;Rasbash et al., 2013). Random intercepts and random slopes were permitted and log frailty trajectories were modelled using an age and age 2 term. They were estimated for all participants with at least one measure of log frailty and all measures of the included covariates under a missing at random assumption. Full model details are provided in Online Resource 3. An interaction term was included to estimate the difference in log FI trajectories between migrants and non-migrants. The difference in log FI between migrants and non-migrants at ages 50, 60, 70, 80, 90 and 100 years was estimated and Z-tests were used to assess the statistical evidence for these differences. Mean differences in log FI were exponentiated to give a ratio of geometric means, thus, results are interpreted as a percent difference in FI score in migrants compared to non-migrants.

Sensitivity analyses
We compared results from our main analysis in which the 'ordinal' FI coding was used, to those in which the alternative 'dichotomous' FI coding scheme was used (Online Resource 2) (Peña et al., 2014). We also examined the effect of the following alternative MIPEX ( (Huddleston et al., 2015) health strand tertile rank ('top rank' being most inclusive; 'bottom rank' being least inclusive). b Percentage difference in frailty in migrants versus non-migrants from linear models with 95% confidence intervals. c Confounders: age, gender, height, education, proportion of international migrants in the host country population. d Socio-economic status: home ownership, household income.
Resource 2: (i) MIPEX labour market mobility strand (a proxy of workers' rights and access to training, education and employment); (ii) 'five-strand average' MIPEX (a broad proxy of migrant integration in multiple dimensions). To explore possible selection bias by differential loss to follow-up, we examined associations in subjects who contributed one and two-or-more observations. We also examined whether loss to follow-up (defined as only single-wave participation throughout SHARE waves 1-4) was predicted by migrant status and frailty at baseline interview.

Post hoc analyses
Further analyses were conducted to assess whether associations were similar in (i) males who migrated after 1949, to examine a more homogenous group of labour migrants (Ladin and Reinhold, 2013) and (ii) those who lived in the host country for ≤10 years to investigate the effect of migrant acculturation. As will be shown below in the results section, Southern Europe was unique in that the direction of association between migration and frailty was opposed to those of other geographical areas. To test whether this inconsistency was more attributable to the Southern European migrant population or the comparator host population, we compared frailty in Southern European migrants versus Northern European migrants. Finally, we explored whether duration of citizenship might explain associations between migration status and frailty.

Cross-sectional analyses of baseline interviews
European migrants were on average 16.4% (95% CI: 14.6, 18.2%) frailer than non-migrants, even after adjustment for age and all potential confounders (Table 2 and Online Resource 5). Chronological age strongly predicted frailty. Women were frailer than men, but after adjustment for age and other confounders this was markedly attenuated. Taller height and higher levels of education were associated with reduced frailty. FI scores were lower in populations comprising ≥5% migrants. Home ownership and higher household income (proxies of post-migration SES) and host country citizenship (a proxy of migrant acculturation) were protective.
Results of intermediary-adjusted models are presented in Table 3 (Europe-wide), Fig. 2 (LMIC-and HIC-born migrants), Fig. 3 (regional models), Fig. 4 (within MIPEX categories) and Online Resources 5-11 and 13-15. SES did not explain migrant versus non-migrant differences in frailty in any analysis. Adjustment for citizenship reduced the migrant frailty burden from 16.4% (95% CI: 14.6, 18.2%) to 12.1% (95% CI: 10.1, 14.1%) at European level, with marked attenuation for LMICborn migrants, HIC-born migrants and within Northern Europe. Citizenship also attenuated the migrant effect in countries of bottom MIPEX healthcare ranking such that the migrant effect no longer varied by MIPEX category after adjustment for citizenship (interaction p = 0.20).

Multilevel longitudinal models of frailty trajectories
We present Europe-wide frailty trajectories in Fig. 5 and percentage differences in frailty between migrants and non-migrants in Table 4. After confounder adjustment, migrants were frailer than non-migrants by 15.4% (95% CI: 11.8, 19.1%) at age 50 years (p < 0.001). We observed that the Europe-wide migrant sample accumulated frailty at a slower rate than non-migrant comparators such that frailty trajectories converged, becoming indistinct between ages 80 and 90 years. Adjustment for potential intermediaries reduced the frailty inequality and accelerated the convergence of migrant and non-migrant frailty trajectories.
Frailty trajectories in migrants and non-migrants were assessed within subgroups of country of birth income classification, European region of domicile and host country MIPEX healthcare ranking (Online Resources 19 and 20). Results were consistent with the cross-sectional analyses. With the exception of Southern Europe, migrants were consistently frailer than non-migrants. Frailty differences between migrants and non-migrants were especially large for LMIC-born and Northern European migrants.

Sensitivity and post hoc analyses
Sensitivity analyses demonstrated that our conclusions may be more conservative than if 'dichotomous' FI coding had been utilised (Online Resource 17). Associations between migration and frailty were altered when subgroups based on the MIPEX five-strand average (as an overall proxy for the inclusivity of host country policies towards migrants) and the MIPEX labour market mobility ranking (as a proxy of migrant workers' rights and access to training, education and employment) were employed (Online Resource 18). For both descriptors, in comparison to MIPEX healthcare categories, the migrant frailty burden was smaller in bottom-ranking countries and higher in top-ranking countries.
Migrants exhibited a different pattern of SHARE wave participation to non-migrants (Table 1) but our findings were unchanged by restriction to single-wave or multi-wave participants (Online Resource 17). Frailty predicted loss to follow-up (p < 0.001). Migrants were 6.2% less likely to be lost to follow-up than natives (p = 0.03). However, migration did not interact with frailty to predict loss to follow-up (p = 0.34).
Restricting the migrant sample to the 3156 men who migrated after 1949 (likely to be a more homogeneous group of labour migrants) did not alter our results (Online Resource 16). We found that the 399 migrants who enrolled in SHARE ≤10 years after migration (median 5.8 [IQR: 3.8, 7.7] years) were no more or less frail than natives after confounder-adjustment (−1.9% [95% CI: −9.0, 5.8%]), p = 0.62 (Online Resource 16). This may suggest that the migrant frailty burden develops with acculturation but is conditioned by citizenship and healthcare access. To investigate the contributions of migrant and host frailty to the 'healthy migrant effect' detected in Southern Europe, we found that Southern European migrants were 18.8% (95% CI: 14.2, 23.1%) less frail than Northern European migrants after confounder adjustment (p < 0.001), supporting the theory that Southern European migrants' reduced frailty is a function of migrants to Southern Europe rather than the specific host population.

Discussion
This is one of the largest studies to examine the impact of migration on older age frailty. Overall, we found that migrants were frailer than non-migrant counterparts and these differences were not fully accounted for by confounding factors. We observed marked heterogeneity both by the geographical region to which they migrated and the level of economic development of the countries of origin. LMIC-born migrants were far frailer than HIC-born migrants, consistent with the findings of Brothers et al. (2014). This is likely to reflect the worse early life environment that such migrants experienced pre-migration and would be consistent with theories of ageing such as 'antagonistic pleiotropy', which postulate that biological investment in early growth and development comes at the expense of later repair and regeneration. Genetic factors or an interaction between genes and environment may also play a role. LMIC-born migrants may also suffer greater barriers to socioeconomic mobility, naturalisation or healthcare access than HIC-born migrants in Europe's host countries (Rechel et al., 2013).
Frailty trajectories converged over the life course so that by 80-90 years old there were no migrant versus non-migrant differences. This could reflect the more deleterious adult life style exposures experienced by non-migrants and/or a stronger healthy survivor effect seen in migrants. Another possibility is the 'unhealthy re-migration effect' (49 ,   Table 4 Percentage differences in frailty between migrants and non-migrants throughout older age adjusting for confounders and potential intermediaries. 50), a survivorship bias in which migrants return to their country of origin to die, leaving an apparently healthier migrant population in the host country. Mortality data ascertained from official sources and by re-contacting households between SHARE waves 1 and 2 demonstrate that death rates that are no different between migrants and non-migrants . Convergence of frailty trajectories in the present study is therefore unlikely to be explained by drop-out bias i.e. migrants being more likely to die than non-migrants at equal levels of frailty, leaving the most healthy migrants alive by age 80-90 years.
Whilst we observed that Northern, Western and Eastern European migrants were frailer than non-migrants, Southern European migrants were no more or less frail than non-migrants. This could have been because South European non-migrants, as the comparator, were themselves frailer but in our sensitivity analysis we found the same pattern of results when we compared the South European migrants to a North European migrants, indicating that this reflects a greater health selection effect for those individuals who choose to migrate to Southern Europein the past more frequently fleeing conflict, war and economic crises (Massey, 1990;Salt, 2005). This is consistent with studies which have considered functional outcomes in Greece, Italy and Spain (Solé-Auró and Crimmins, 2008); chronic disease burden in Greece and Italy (Solé-Auró and Crimmins, 2008) and depression in Southern Europe (Aichberger et al., 2010). Our larger sample size allowed us to extend the observation of Brothers et al. (2014), who reported that migrants were no frailer than non-migrants in a pooled Eastern and Southern European sample, by demonstrating that this was only true in Southern Europe. Not all analyses have found this; for example more chronic disease was reported burden in Spanish migrants (Solé-Auró and Crimmins, 2008) and greater rates if depression has been seen in migrants to Greece, Italy and Spain (Ladin and Reinhold, 2013). Though 9 of the 60 items in the frailty index related to psychiatric symptoms, the overall FI may not adequately capture the burden of psychological morbidity. The possibility of under-reporting of health outcomes and the 'unhealthy re-migration effect' could also explain our findings in Southern Europe.
Migrants are frequently found to be healthier and exhibit lower mortality rates than non-migrants (Angel et al., 2010;Hummer et al., 1999;Rechel et al., 2013). Our findings may therefore appear paradoxical but contribute to a growing literature which finds little evidence of the 'healthy migrant' in later life in Europe (Aichberger et al., 2010;Brothers et al., 2014;Ladin and Reinhold, 2013;Solé-Auró and Crimmins, 2008). It remains possible that any 'healthy migrant' signal in early post-migration life (median age of migration was 21 [IQR: 13,30] year old) was eroded before enrolment in SHARE due to mortality or acculturation (Wakabayashi, 2010). As well as individual-level factors, we also observed potential structural factors that appeared to modify migrant risk. Countries with policies that enhanced migrant health access, as measured by MIPEX health, showed less frailty for migrants compared to the non-migrants, suggesting that this may be mediated by potential barriers to healthcare access. Unfamiliarity with host country healthcare systems and reduced ability to navigate alternative routes of healthcare access may explain this. Adjustment for individual citizenship not only attenuated the frailty differences but also reduced the structural effect modification of health care access, implying that this effect can be reduced by obtaining equal legal rights to non-migrants. Post hoc analyses suggested that increasing duration of citizenship may be additionally protective.
In agreement with other studies Santos-Eggimann et al., 2009;Theou et al., 2013), whole-population FI scores were greater in Southern and Eastern Europe than Northern and Western Europe. This likely reflects the relative economic growth, political stability and success of health policy and healthcare delivery in improving life expectancies in the latter regions over the lifetime of the study participants .

Strengths and limitations
This is a large study based on 95,635 older age Europeans providing statistical power to look for heterogeneity by geographical regions, migrant economic origins and levels of healthcare policy inclusivity. The use of both cross-sectional and longitudinal analyses allows us to have greater insight into the natural history of frailty between the migrant and non-migrant groups.
We had little data on pre-migration variables, other than height and education, or the reasons for migration to differentiate economic migrants, health tourists, refugees and asylum seekers. The effect of residual confounding by pre-migration and selection factors which are likely to favour a healthy, upwardly mobile migrant would be to attenuate the observed associations, suggesting the true associations may be even stronger. In addition, migrants may actively select host countries on the basis of liberal health care access policies, though this bias could operate in either direction so that healthier or less healthy migrants choose the most accessible countries.
It is possible that migrants may have been more likely to report morbidity. However, self-reported diagnoses often closely reflect medical records regardless of advancing age, increasing co-morbidity, poor cognition or education (Simpson et al., 2004). Furthermore, it has been shown that frailty indices are well aligned with self-perceived health (Solé-Auró and Crimmins, 2008) and predict mortality better than chronological age in different populations (Arnold et al., 2010;Romero-Ortuno and Kenny, 2012).
There could have also been differential loss to follow-up between migrants and non-migrants either due to death, non-response or 'unhealthy re-migration' (49, 50). Brothers et al. (2014) reported that short-term death rates were no different between non-migrants and different migrant groups. Our post hoc analyses found that restriction to single-versus multi-wave participants did not alter our results. Furthermore, we found no evidence that migration interacted with frailty to predict loss to follow-up.
The generalisability of the SHARE migrant population is also an issue. Underrepresentation of undocumented migrants, transient workers, trafficked persons, refugees and asylum seekers may have biased our results towards the null as these groups have the least access to citizenship and healthcare and the greatest vulnerability to negative health outcomes, a particular issue in Southern Europe, which is reported to host the largest populations of undocumented migrants (Carta et al., 2005).

Conclusion
This study has highlighted the potential increased burden of frailty in the older migrant population. These differences are modifiable to some degree by the post-migration experience including adequate access to health care. Ideally one should establish migrant specific cohorts with a focus on collecting pre-migration variables as well as better capturing the migration experience and post-migration acculturation with both self-reported and objective outcome measures. Our findings provide evidence in supporting policies that promote migrant-inclusive healthcare policies and citizenship rights as potential solutions to the growing healthcare needs of this population in Europe.

Declaration of interest
Conflicts of interest: None.