Commentary: Of salmon and time travellers—musing on the mystery of migrant mortality

Paradoxes abound

Socioeconomic status is known to be strongly and inversely associated with mortality: those who are poor, unemployed, or have a low educational attainment experience higher mortality than the rich, employed, and well-educated. Immigrants tend to have, on average, a lower socioeconomic status than the majority population of the destination country. And yet, their mortality, overall as well as for certain specific causes, is often lower in comparison—a paradox.¹ In this issue of the journal, Singh and Hiatt² report similar findings from the US. Foreign-born persons of all four major racial/ethnic groups—Asians, blacks, Hispanics, and non-Hispanic whites—have a mortality advantage relative to the US-born. Levels of socioeconomic achievement among many immigrant groups, however, are comparatively lower. The reasons for this puzzling finding again remain elusive. What implications for future migrant research in epidemiology should this have?

Effects of study design?

To begin with, the question has to be resolved whether the observed mortality advantage of immigrants is real or due to bias. Singh and Hiatt attempt to arrest the usual suspects, first and foremost the ‘salmon bias’. Its underlying claim is that gravely ill immigrants tend to return to their countries of origin. This leads to a numerator–denominator mismatch and thus to an underestimation of mortality. Singh and Hiatt's study design, a repeated cross-sectional analysis, is prone to this type of bias. For example, in a similar type of study a considerably lower all-cause mortality was observed among male Turkish migrants in Germany than among German men.³ In a longitudinal design, however, their peers in The Netherlands had a higher mortality than Dutch men.⁴ Still, this observation cannot be generalized: Other cohort studies, as well as studies that could rule out return migration, showed real mortality advantages of immigrants, some larger than those observed by Singh and Hiatt.^5,6 To minimize the effect of study design on the size and direction of observed mortality differentials, future migrant studies should be based on individual follow-up.

The datasets, which Singh and Hiatt as well as others⁴ had at their disposal, have an additional limitation: the association between socioeconomic indicators such as unemployment on the one hand, and health outcomes on the other, is merely ecological. As a consequence, it remains unclear whether immigrants with a low socioeconomic status have a higher or lower mortality than those of high socioeconomic status, and how social mobility of immigrants affects their mortality. Only measurements at individual level can help to further clarify the determinants of trends in disparity. They should be complemented by measurements of community attributes such as the extent of discrimination. Clearly, (as Singh and Hiatt also point out), present studies examine only selected aspects of the association between migration and health.⁷

More metaphors

Various explanations for an actually lower mortality of immigrants, relative to the populations of origin and of the host country, have been proposed. One is a selection process, the so-called ‘Healthy Migrant’ effect. Here, the underlying assumption is that mainly healthy and active individuals migrate. Immigrants would thus have a mortality advantage—at least relative to the population from which they originate. To better understand this effect, future migrant studies have to include the populations of origin of immigrants as well.

In another metaphor, migrants from low-income countries have been pictured as time travellers. Relative to the situation in industrialized countries, they are coming from a ‘past’ stage of the health transition.¹ They have been exposed to fewer, or lower doses of, risk factors for cardiovascular and other non-communicable diseases. These migrants will experience a mortality advantage for such diseases even years after migration and the adaptation of a ‘Western’ lifestyle because of the long lag-times between exposure and death. At the same time, they will immediately benefit from better access to medical care for infectious diseases and emergencies, further increasing their mortality advantage. To investigate the dynamics of these effects, duration of stay in the destination country needs to be measured. In addition, a detailed breakdown of ethnic/racial and geographic origin is required because some disease risks are brought from the countries of origin. A category such as ‘Asian’ does not differentiate at all in terms of ethnic background and local epidemiological situation.

Dreams of an ‘ideal’ migrant cohort

Where to go from here? In the field of surveillance and monitoring of migrant health, a definition of ‘migrant’ is needed that puts minimal additional requirements on routine data collection, allows to assessment duration of stay in the host country, includes second-generation immigrants, and is uniform so it can be used by neighbouring countries in a region. Efforts to develop a definition matching these criteria are currently ongoing in Europe, hopefully leading to improved monitoring of migrant health. However, only limited conclusions regarding causes of mortality differentials and their change over time could be drawn from data thus generated. To better understand such differentials, three additional requirements would have to be fulfilled. The first one is to base future studies on an explicit model of migrant health and its determinants,⁸ taking into consideration the complexity of the association between migration and health.⁷ Second, these determinants would have to be measured longitudinally and at an individual level. Such a migrant cohort would, of course, also be useful to investigate the respective roles of nature and of nurture in disease aetiology.

A migrant cohort measuring behavioural and socioeconomic variables at individual level would require considerable resources, which many a migrant researcher has failed to obtain so far. Epidemiological migrant studies appear to have come to a stop for another reason, however (and this is the third requirement): An ‘ideal’ prospective migrant study, which could help to understand the mortality paradox, would have to enrol individuals before they migrate (these individuals would be the very ones that are usually categorized as ‘lost to follow-up’ in cohort studies because of moving abroad). Follow-up would need to continue in the country of destination and, in case of return migration, again in the country of origin. Achieving sufficiently large sample sizes is not entirely impossible. There are particular migration movements in which numerous individuals from one country or region move to one, or a small number of, destination countries—an example being ethnic Germans migrating from former Soviet Union states to Germany. Technically, trans-national follow-up is becoming feasible: Many emerging-economic countries where migrants originate from have sufficiently refined population registration systems. Thus, the idea of a record linkage study spanning countries of origin as well as host countries of migrants is no longer implausible. This kind of design would create a nightmare in terms of data protection, however, and considerable effort would be needed to make it acceptable to society. Until then, the puzzle of migrant mortality will continue to stimulate much musing.

I thank Jacob Spallek for helpful comments.

References