There is a consistent finding, supported by many studies, that lower socioeconomic status is related to higher mortality and morbidity. This has also been extensively documented for Belgium (Gadeyne and Deboosere, 2002a, b; Bossuyt et al., 2004). In this regard, the lower mortality level observed among adults of some Mediterranean migrant communities in Belgium compared with the Belgian native population is especially striking, given their generally lower socioeconomic status. Lower mortality among adult migrants has been widely reported for other countries (Courbage and Khlat, 1995; Razum et al., 1998; Palloni and Morenoff, 2001; Kouris-Blazos 2002). This finding has been conceived by many demographers and epidemiologists as a paradox: the “Hispanic paradox” (Franzini et al. 2001; Palloni and Morenoff, 2001; Palloni and Arias, 2003, 2004), the “Latino paradox” (Abraído-Lanza et al., 1999), the “Mediterranean migrant mortality paradox” (Khlat and Darmon, 2003) or the “Greek paradox” (Kouris-Blazos, 2002).

Low levels of mortality in migrant communities are often accompanied by health problems associated with social deprivation. This observation led to the conclusion that low mortality levels have nothing to do with better health among migrants, but are artefacts due to selection processes or spurious relationships (Ringbäck Weitoft et al., 1999). However, in recent years several studies have refuted the hypothesis that reduced mortality in some migrant communities is a statistical artefact. Abraido-Lanza et al. (1999) rejected the explanation of the “salmon bias” [1] and of the “healthy migrant” selection effect. In a study on life expectancy of migrant groups living in Amsterdam, Uitenbroek and Verhoeff (2002) conclude that, with regard to their data, it seems likely that some migrant groups do have high life expectancy, although the morbidity in these groups can be quite high. They suggest that many health problems associated with detrimental working and living conditions can be seriously disabling without being life-threatening.

Excluding the importance of statistical artefacts as the main explanatory factor of lower mortality, many researchers have been looking for lower risk factors in lifestyle and diet (Wanner et al., 1995; Brussaard et al., 2001; Darmon and Khlat, 2001; Landman and Cruickshank, 2001) or for factors on the psychological level such as the “migrant hope effect” [2] (Anson, 2004). In a few cases, biological predisposition has also been proved to be a factor of influence.

As better life expectancy is the result of lower mortality for some specific causes, several studies have focused on this aspect. Differences have been reported for a number of causes such as cancers (Rosenwaike, 1990; Geddes et al., 1993; Bouchardy et al., 1995), heart diseases (Sundquist and Johansson, 1997; Razum et al., 1998) or suicide (Trovato, 1986; Hjern and Allebeck, 2002). Cause-specific mortality patterns are undoubtedly the most powerful tool to unlock the mystery of the paradox. Abraido-Lanza et al. (1999) conclude that a systematic test examining different causes of death is needed to verify the hypothesis that cultural factors contribute to Latino health.

Belgium is well suited to further explore these major explanations thanks to the presence of large migrant communities making up more than 10% of the population, and the existence of a national population register (Eggerickx et al., 1999). The present analysis looks into new data to give a detailed description of cause-specific mortality patterns for the largest migrant communities: Italian, Spanish, Moroccan and Turkish, compared with the Belgian native population. Other populations are considered, such as migrants from sub-Saharan Africa taken together as one group, and the nationals of the main neighbouring countries (France, Germany and the Netherlands) as a control group. Before making comparisons between communities, we use information obtained by cross-matching population registers and death certificates to explore the plausibility of various selection mechanisms as an explanation for the lower mortality of migrant populations.

Belgium has a long history of migration. As a small country and an active trading centre with a geographically open and central location, the country has been a place of settlement for many migrants. Before the twentieth century, the overwhelming majority of migrants originated from the neighbouring countries. Traditionally, national state borders were not so clear-cut for populations living on either side. There have always been large communities of French, Dutch and Germans residing in Belgium, and Belgians formed the biggest migrant community in France during the nineteenth century. To this day, many Germans, French and Dutch reside in Belgium. Most of them are living in the border regions of their country of origin or in the major cities. In terms of socioeconomic status, the situation of these populations is quite similar to that of the Belgians.

Belgium was mainly an emigrating country in the nineteenth century, when harsh economic conditions and crises in agriculture and the textile industry forced hundreds of thousands of Belgian workers to emigrate. After the First World War, the migration balance became positive. Demand for labour in heavy industry and the coal mines attracted more migrants during the twenties. Immigrants did not only originate from neighbouring countries. Large contingents of Italians and East Europeans started to settle in Belgium to escape the difficult economic and political conditions of their home countries. At the end of the 1920s, the Italian community approached 30,000 persons (Morelli, 2004).

Immediately after the Second World War, labour migration rose sharply, with tens of thousands of mainly Italian workers recruited for the Belgian mines. Italian migration continued over a long period, even after the official migration stop of 1956, and was still appreciable in the 1960s and 1970s. Spanish and Greek workers formed the next migration group. After the labour migration agreement of 1956 between Belgium and Spain, the Spanish community grew quickly, approaching 70,000 individuals in 1970 (Sanchez, 2004). Later on, as the post-war economic boom spread across Europe, a new, low-cost labour force was recruited from Turkey and Morocco (Grimmeau, 1984; Gaudier and Hermans, 1991; Lesthaeghe, 2000). This massive labour migration ended abruptly with the economic crisis of the early 1970s. Migration since then has been restricted to family reunion and the import of brides and grooms by the Moroccan and Turkish populations (Lievens, 2000). Meanwhile, a new cycle of migration started with a substantial influx of political refugees. Finally, a small African community developed over time as a consequence of Belgium’s historical colonial ties with sub-Saharan Africa. It is a predominantly Congolese community (Democratic Republic of Congo) with a large proportion of highly educated persons.

However, as the low mortality of many migrant populations was especially at odds with their lower socioeconomic status, most research has focused on mortality. The contradiction between low socioeconomic status and high life expectancy in adult migrant populations has generated abundant research and a large set of possible explanations. Five groups of hypotheses have been proposed in the literature, sometimes as the sole or main explanatory factor, sometimes in combination with each other.

1. The statistical artefacts theory focuses on data problems and covers a wide range of possible sources of error: numerator and denominator errors or mismatches between deaths and the population at risk, misreporting of age (Rosenwaike, 1990), failure to correctly identify subpopulations (Rosenberg et al., 1999) and registration errors in general (Ringbäck Weitoft et al., 1999). Although the salmon bias hypothesis refers to a (return) migration process, this is sometimes intrinsically linked to a registration problem and the incorrectness of numerators or denominators as suggested by the word “bias” in the denomination (Abraído-Lanza et al., 1999).

2. Selection process hypotheses consider the creation of a selective population as the main source of diverging mortality rates. Both hypotheses (healthy-migrant effect and unhealthy return-migration effect) are looking at the migration process as a basis for their explanation.

Under the first hypothesis, selection occurs in the population of origin. Migrants belong to a subpopulation with above-average health (Abraído-Lanza et al., 1999; Palloni and Morenoff, 2001). The healthy-migrant hypothesis consists of two assumptions. First, migrants are healthier than non-migrants. People who decide to move, especially in the case of international migrations, belong to a healthy subset of the population (Soldo et al., 2002). This part of the hypothesis is generally accepted as plausible (Marmot et al., 1984). The second assumption considers that this selective effect is so strong that the selected group is also healthier than the host population, regardless of the fact that they originate from a population with a generally much lower level of health and belong to the lowest socioeconomic group in the recipient country. As mentioned by Razum et al. (1998, p. 297): “People who migrate, (…) are on the average healthier than the population they originate from, and often also healthier than the population of their host country”.

This selection effect is assumed to be even stronger in the case of labour migration, as potential migrants have to be healthy to be accepted in the work force. In some particular cases they are even screened on their health status before migration (Turks and Moroccans in Belgium, Turks in Germany). Uitenbroek and Verhoeff (2002) refute this hypothesis on the basis that most migrants migrate in their twenties or early thirties, age groups that are usually very healthy and where symptoms of the major causes of death are rarely present. And they conclude that: “It is difficult to imagine how these young people could have been selected – prior to migration – on their future susceptibility to the most important causes of death.” (Uitenbroek et Verhoeff, 2002, p. 1386). Razum et al. (1998) also point out that many migrants simply follow individuals who migrated previously.

In the unhealthy return-migration effect or salmon-bias hypothesis, selection occurs inside the migrant population, leaving behind in the host country a population with above-average health. This hypothesis has been developed with regard to the return of older migrants to their country of origin. Formulated in a broader sense it includes all migrants, regardless of their age, who return to their country of origin because they are less adaptive or not healthy enough to endure harsh working and living conditions, and who are more likely to experience higher mortality (Razum et al., 1998; Uitenbroek and Verhoeff, 2002).

Palloni and Arias make a distinction between “return-migrant effect or bias” (type 1) and the “salmon-bias effect” (type 2). In type 1, return migration creates a downward bias in the estimated mortality rates, irrespective of whether return migrants are less healthy than those who stay, because the denominator in the mortality rate refers to a baseline population that has been reduced by unobserved migration (Palloni and Arias, 2004). We consider this as a statistical artefact. The “salmon-bias effect” hypothesis (type 2) suggests a situation of returning migrants in poor health with higher risks of mortality. But the line between these definitions is thin and in practice we can see that the “return-migrant effect” and the “salmon bias hypothesis” are used indiscriminately, both covering the whole range from return migration of the less healthy to non-registration of return migration.

Palloni and Arias conclude that the bulk of the foreign-born Mexican advantage in the United States is related to return migration of those who are in poor health. But they also stress that they are unable to account for the mortality advantage observed among other foreign-born Hispanics (Palloni and Arias, 2003, 2004).

3. Cultural explanations propose that culture affects mortality risks by influencing health behaviours and specific diets among ethnic or national groups. Again, this category covers a broad range of mechanisms. The Mediterranean diet is often proposed as an explanation for the lower adult mortality of labour-force migrants of Mediterranean countries (Darmon and Khlat, 2001; Landman and Cruickshank, 2001; Kouris-Blazos, 2002). Nutritionists argue that a low-fat diet combined with a high consumption of fruit and vegetables has a positive effect on the expected life expectancy. Conversely, over-consumption of alcohol and tobacco are known to be extremely important factors in adult mortality. Moreover, specific brands and forms of consumption of alcohol and tobacco could moderate or intensify health effects. Differences in food conservation, food preparation and food consumption patterns (in quality, quantity and timing) might be responsible for the development of differences in the general health status of populations. Differences in sexual behaviour can also be a source of different health risks. In general, the concept of health and the attitude towards life, death and one’s own body, are a core part of cultural and religious tradition. They are strongly linked to the country of origin among first-generation migrants who have not undergone the socialization process of the host country’s educational system. The entire process of acculturation is by no means unidirectional and homogeneous. Differences in these core values and attitudes can be very profound and even filter through several generations. The role of cultural factors is stressed by Abraido-Lanza et al. (1999) in their study of the Latino Mortality Paradox as they conclude that “neither the salmon nor the healthy migrant hypothesis explain the pattern of findings. Other factors must be operating to produce the lower mortality”. The French study by Khlat and Courbage (1995) tends to accept a combination of the health selection process and a cultural explanation reformulated as a “positive adaptation”. Immigrants adapt positively by taking the best (health services, living conditions, etc.) and leaving the worst (over-eating, traffic accidents etc.). A hypothesis also defended by Powles (“the best of both worlds”) when he explains the low mortality of Greek migrants in Australia (Powles, 1990).

4. Biological and genetic differences have in some particular cases been proposed as partial explanations. Khlat and Courbage (1995) suggest that low lung cancer mortality among Moroccans could possibly be explained by some protective genetic characteristics. There have been some claims that genes could predispose a person to cancer and it is estimated that inherited forms of cancer account for about 5% of the total of all cancers (Weitzel and McCahill, 2001). Genes are implicated in some cases of breast cancer and could possibly be part of population differences in breast cancer incidence. Sickle cell anaemia is a known example of predisposition to disease in subpopulations due to genetic differences. Some other examples of genetic factors of mortality have been described (Soliani and Lucchetti, 2002). It has been shown that high rates of cardiovascular disease and diabetes among migrants from the Indian subcontinent are related to insulin resistance and the resultant lipid disturbances (McKeigue et al., 1991).

5. Explanations based on sociological or psychological effects specific to the migration process seek to identify individual or group-level mechanisms that might generate a mortality advantage. The theories about “social capital”, “cultural capital” and “social networks” are all well-known. Although some research shows the positive effect of social networks in migrant communities in helping individuals overcome the stress of being a minority group, the measured effect appears to be relatively small compared with individual characteristics (Franzini and Spears, 2003). On the individual level, “the migrant hope hypothesis” has been proposed as a psychological mechanism to better explain health among migrants (Anson, 2004).

Based on our research of mortality differentials in migrant communities in Belgium, we are convinced that the health outcome is the result of a complex interplay of several of these factors. But the relative importance of the different explanatory factors is not without interest for a better understanding of morbidity and mortality processes.

The different sets of explanations fall into two very distinct categories: those that are more exogenous to the mortality process (migration and selection processes, and data artefacts) and those that are an intrinsic part of health conditions and the mortality process (biological, behavioural, sociological and psychological factors). The latter can help us understand health and mortality processes while the former enable us to understand or interpret the data. Statistical artefacts and selection processes are in fact phenomena that make it difficult to assess whether there is a “real” difference in health and mortality between populations and whether this difference is linked to more intrinsic factors specific to the populations in the country of origin. This means that from an epidemiological viewpoint, migrant populations create an exceptional situation. They can be seen as a laboratory experiment in “real life” where populations have a specific set of characteristics and live in the same environmental conditions as a control population. In this context, migration and the effects of migration can be considered as confounding factors, and acculturation must be regarded as bringing the experiment to an end. And we can only conclude with Franzini et al. (2001) that “a rare window of opportunity now exists to learn more about how cultural factors influence one’s health.” Migrant studies have been an important factor in the development of the risk factor approach, and we believe that the new possibilities opened up by computer registration and data linkage should be exploited to investigate this approach in more depth.

Our data are not well suited to proving or disproving the selection hypothesis based on the healthy-migrant effect. We are convinced by logical reasoning and by knowledge of the labour migration process that a healthy-migrant effect does in fact exist. Although it is difficult to assess the magnitude of this selection on mortality, the examination of cause-specific mortality can help us analyse and contextualize the selection effect. Our data are much better suited to evaluating the importance of the salmon-bias effect.

The analysis of cause-specific mortality and the comparison of these results with the host population enable us to draw some conclusions about the possible explanations of the adult migrant mortality advantage. In regard to the aetiology of specific causes of mortality, we suggest plausible explanations of the adult migrant mortality advantage. This pattern of cause-specific mortality seems to be consistent with, or at least does not contradict, the observation of worse self-reported health or higher prevalence of morbidity among migrant populations.

Several analyses have taken migrants’ duration of residence into account. For the Moroccan adult male population, a significant acculturation effect could be measured, the relative advantage diminishing progressively with the duration of residence in Belgium. However, as a more in-depth analysis of the data would be needed to assess the quality of immigration dates, it will not be presented in this paper. Moreover, given the heterogeneity of the migration periods between the migrant communities, and the small size of the groups involved with regard to cause-specific mortality, duration of residence was difficult to integrate in the present analysis.

The analysis of a particular case, the migrant population in Belgium, does not necessarily explain the observations of migrant mortality in other countries. It can only be illustrative of how differences in mortality and morbidity can occur.

The study of the health of a migrant population in a host country must take into account not only the health situation of the migrant population, but also the health situation of the host population. Lung cancers are less frequent among migrant groups than in the Belgian population. Belgian males are known to have a very high lung cancer mortality rate, so migrant populations can easily make a gain in life expectancy for this particular cause of death, even if their own lung cancer mortality is higher than the average world level. We must take into account not only the native population’s health in the host country, but also the general social system (including the health system) both in the host country and in the country of origin. A “salmon-bias effect” is much more likely to operate when migrants in bad health cannot expect better health care in the host country than in their country of origin. This is the case of Mexican-born Hispanics living in the United States who are much more likely to return to their country of origin than Moroccans living in Belgium, as the latter have full access to a high-quality health care system in Belgium.

It does not mean, however, that the register has no flaws or that every person residing in Belgium is actually registered. Illegal immigration is by definition not registered and registration is limited to what people want to be administratively recorded. However, apart from illegal immigrants, errors are marginal. But as we only look at deaths among the population at risk (the baseline population being the registered population in the 1991 census), illegal immigration has no impact on the results of this study. Moreover, apart from the fact that registration is compulsory, people have numerous incentives to register correctly. Pension funds, social security, unemployment benefits are all dependent upon registration in the national register.

The analysis is based on data from the 1991 Belgian census, linked to registration records of all deaths and emigrations between the census date (1 March 1991) and 1 January 1996. The total Belgian population on the census date amounted to 9,978,681 persons. Thanks to Statistics Belgium, a direct individual link was established between census data and register data (Deboosere and Gadeyne, 1999), thereby avoiding the classic problem of the numerator and the denominator bias. The link with the national population register guarantees the accuracy of essential information such as exact date of birth, nationality, nationality at birth, date of arrival in Belgium, date of registration in the population register and eventual dates of emigration or death. Covariates such as educational level, house ownership or level of domestic comfort are based on the census form.

The current study was limited to adults aged 25-54 at the time of the census covering a total of 4,140,559 persons. Total exposure time was 58 months for survivors and truncated for migrants and persons dying during the observation period on the exact day of migration or death. Person-years at risk could be calculated exactly for each subpopulation and for each age group.

Working with register data has two implications:

1. These data reflect administrative reality. The population included in the analysis was registered in the municipality on the date of the 1991 census. As mentioned above, illegal residents are excluded. We know that illegal residents are probably more likely to be in bad health, but our prime interest lies in morbidity and mortality differences between officially registered migrants and the native Belgian population. For our research, the selection hypothesis must be reformulated to take into account the possible selection effect of being included or not in the register.

2. The population analysed is semi-closed: nobody can enter the population. People can leave the population by emigration or death. Emigration is not always registered and some people leave Belgium to return to their country of origin or to another foreign country without reporting their departure. This is rather rare for immigrants who have been part of the labour force. Anyone who has worked legally in Belgium has derived rights (social security, pension entitlements) which are an incentive to inform the authorities of a change of address, though there is no legal obligation to do so. Moreover, when people leave, their departure does not long remain unnoticed at municipal level. It is exceptional for people to leave their home without others coming to replace them. When a new inhabitant registers at an address, the data concerning the previous inhabitants is updated. If necessary, a police officer is sent to the address to certify the departure of the previous residents. Very often, utility companies or the tax authority inform the municipal authorities of the change and the register is updated accordingly. However, it is theoretically possible for someone to remain in the national population register after leaving the country. This can be the case if one member of a household emigrates and the remaining members have no reason or (financial) incentive to declare this departure to the authorities.

A suggested hypothesis is that mortality rates of minorities may be artificially lowered by age underestimation (Elo and Preston, 1994). In the present analysis, rates cannot be affected by this mechanism as the population register contains the exact date of birth. Immigrants are required to produce a birth certificate at the time of registration. In the study populations, the exact date of birth is only missing for a few individuals. In most of those cases, at least the year of birth was attested by a local authority. This problem may affect mortality analysis of the oldest migrants, but certainly not the selected adult population.

The study population includes non-Belgian nationals and foreign-born persons who have acquired Belgian nationality. Based on the nationality of origin, the population is divided into nine groups and a rest category. Nationality, former nationality, place of birth and mother’s nationality were used to maximize population groups of Italian, Spanish, Moroccan, Turkish and sub-Saharan African origin. The population originating from neighbouring countries (France, Germany+Luxembourg and the Netherlands) is included in the analysis as a control group. The population of Belgian origin and a rest group of migrants from all other countries form the two remaining groups. Table 1 gives a summary description of the study population data.

Table 1
Description of the study population (adults aged 25-54 in 1991), Belgium, 1991-1995

To avoid this source of uncertainty, death certificates for the period 1991-1995 were linked to the population register. The linkage was based on anonymous data using a large set of variables. For more than 85% of the death certificates, it was possible to find a unique match with the death registration in the population register. For the remaining 15%, a probabilistic linkage was performed through an algorithm using the date of birth and death, the sex and the municipality of residence. For more than 98% of people registered at census time and dying during the observation period a match was established. For less than 2% no match could be found, sometimes due to minor errors in the death certificates [3].

However, the main reason for non-matching is that a number of death certificates are simply missing. When the place of death is outside Belgium, no Belgian death certificate is established and the register of death certificates holds no trace of this death. The population register, on the other hand, is notified by the family or by an administrative body that the person is deceased. For instance, out of a total of 102,489 deaths in the National Population Register for the year 1992, it was possible to link 100,697 records with death certificates. For 1,792 persons, no match was found. Among persons who died between 1991 and 1995 in the adult study population (aged 25-54 in 1991), 1,566 persons were registered as deceased in the population register but could not be linked to a corresponding death certificate (Table 2). Immigrants have a significantly higher probability of dying abroad than Belgian nationals. Generally, migrants travel abroad more often and for longer periods, either for holidays or family visits. Therefore, relying exclusively on death certificates introduces the risk of underestimating the mortality of migrant subpopulations. This problem can be avoided however, since the National Population Register informs us about the exact date of death; only information on the cause of death is lacking.

Table 2
Number of deaths in Population Register without matching death certificates in the study population (adults aged 25-54 in 1991) Belgium, 1991-1995

We can conclude that the population register is fairly accurate in registering death, but that the non-establishment of death certificates for deaths occurring outside Belgium results in a large proportion of unknown causes of death, especially for certain migrant communities.

The death certificates in the period under study used the 9^th International Classification of Diseases (ICD-9). Our classification of causes of death is based on the “underlying cause of death” concept as defined by the WHO. It represents the cause of death considered the most relevant, as reported on the death certificate. As our analysis is limited to the 25-54 age group, we assume that the information on causes of death is relatively accurate, even for the migrant population.

The aggregation used follows the general lines of the classification adopted by the 9^th ICD. Major single causes of death such as suicide, diabetes mellitus or breast cancer are treated separately. The other causes are grouped according to the chapters or subchapters used by the ICD-9. Alcohol-related diseases are the one major exception. All deaths caused by alcohol abuse, chronic or acute, are grouped together independently of the chapters of ICD-9 (Table 3).

Table 3
Causes of death among the study population (adults aged 25-54 in 1991, all nationalities), Belgium, 1991-1995

The aim is to organize all causes of death in groups that are large enough, have a similar aetiology and do not exceed an acceptable level of uncertainty in terms of diagnosis or classification.

Let SMR be the standardized mortality rate, P_x the population of Belgian origin, m_x the age-specific mortality rates, s a given subpopulation and c a given cause of death. We obtain:

Evidently, the sum of standardized cause-specific mortality rates equals the overall standardized mortality rate.

To estimate the impact of socioeconomic status on differences in total mortality, we used a Cox regression analysis. The Cox model estimates a non-parametric baseline risk of death at time t for an individual who has survived up to this date t. The register records all deaths and emigrations, thus permitting an exact definition of the period of exposure to death or to censoring. Note in this regard that the Cox regression model is able to treat censored observations. The duration is expressed in days.

For each covariate, the model estimates the multiplicative effect on the risk for each unit increase in the covariate. For categorical covariates, the result indicates the multiplicative effect of each modality on the baseline risk associated with the reference modality.

If h(t) is the mortality risk at time (t), then the Cox regression models it as:

with: h_i(t) the mortality risk or hazard at time (t) for individual i

and h_o(t) the baseline risk.

Thus, the mortality risk at time t is the product of the baseline risk function and the exponential of the sum βx, where x is the vector of individual covariates (Kleinbaum, 1996). The exponentials of β express the relative mortality risk during the observation period associated with the different modalities of a variable with respect to the reference modality.

In the initial model, the covariates are nationality at birth and age; age is introduced as a continuous variable since mortality increases proportionally with age in the study group (age 25-54). The Cox regression is applied separately by gender. In subsequent models, educational level and “housing” are introduced as control variables. For educational level, higher education (associated with the lowest mortality) is taken as reference modality. Other educational levels are thus compared to the baseline of Belgian men aged 25 with the highest educational level. The “housing” variable is a composite variable which takes account of ownership status and level of domestic comfort. The analysis of the Belgian population census data demonstrates that the housing variable is a good proxy for income and wealth. Here again, the group with the lowest mortality – house owners living in a house with highest comfort – are chosen as the reference modality.

The study of cause-specific mortality is useful for identifying factors which characterize healthier populations. By using Cox regression we can estimate the position of each of the different immigrant groups relative to the Belgian population for a specific cause of death. The regressions are applied separately by gender. The initial model takes account of age and nationality at birth; here age is introduced as a categorical variable due to its non-linear relationship with certain causes of death. The regression exhibits a nationality effect on the risk of dying from a given cause during the observation period. When modelling the cause-specific mortality risk, we consider dying from all other causes of death as censored cases in the same way as people are censored when they emigrate. Adding covariates to the initial model we can assess their impact on the cause-specific mortality risk by nationality of origin.

As we focus on the factors beneficial to health, the importance and the role of specific causes of death in the resulting general mortality must be estimated. We therefore compare mortality patterns of migrant communities with those of the native Belgian population using a simple decomposition technique based on the table of directly standardized cause-specific mortality rates for each nationality. The method (Elo and Drevenstedt, 2002) estimates the contribution of each cause of death to the difference in total mortality between each subpopulation and the population of Belgian origin. The results do not permit a comparison between nationalities, because similar relative contributions can hide enormous differences in absolute mortality figures. But the graphical representation creates footprints that allow a first easy reading of the contribution of different causes of death in the total outcome. Relative contributions sum to one.

We remedied the specific problem of unidentified causes of death by redistributing the unlinked causes over all other causes of death on a proportional basis. Although it is a very rough method, it limits the underestimation of specific causes. As we can assume that people who are seriously ill are less inclined to travel, there is probably a slight overestimation of deaths from chronic diseases (cancers, chronic obstructive pulmonary disease) and a possible underestimation of causes of sudden death such as traffic accidents and ischaemic heart diseases. Insofar as we lack other information, proportional redistribution seems most acceptable.

The salmon bias hypothesis actually operates in two possible ways (or is introduced in two different ways in the literature):

1. The first line of reasoning states that people on the verge of dying (like the returning salmon) go back to their country of origin. The postulate is that even if their emigration is registered, mortality figures are seriously biased because by leaving the country just before dying they make the numerator artificially low.

2. The second elaboration of the salmon bias hypothesis argues that people leave the country without being registered as emigrants. This affects both numerator and denominator. As they are still assumed to be in the country, they are counted in the population at risk. But their death outside the country is never registered and they become statistically immortal.

When the population at risk is small, a few deaths can be very significant, especially if mortality is measured at old ages.

Several claims have been made in the literature that the salmon bias hypothesis can be refuted on the basis of logical reasoning. Why should people who are seriously ill return to their home country where health care is less developed than in the migration country? Why should they leave their family when dying [4]? The entire logic of pension funds and health infrastructure does not support this hypothesis; neither does the fact that family ties and children hold them in their new country. Older emigrants, once retired and no longer bound by work and small children, often stay for longer periods in their country of origin. But this hardly applies to the selected age group.

If we assume that all emigration is registered, we can calculate the hypothetical mortality rate emigrants would need to have if a salmon bias was at work. Applying mortality rates of the native Belgian population to the migrant group gives the theoretical number of deaths that would be observed if mortality of the migrant group equalled that of the Belgian population. Subtracting the number of observed deaths yields the difference that has to be generated by emigrants of this subpopulation.

Table 4 shows that mortality rates among return emigrants would need to be extremely high to offset the low immigrant mortality observed in Belgium (column 7). We can conclude that based on officially registered emigration, the salmon bias hypothesis can be dismissed for the adult population aged 25-54.

Table 4
Estimation of the mortality rate of return migrants and of the total number of return migrations required to equal Belgian mortality rate (men aged 25-54 in 1991, mortality in 1991-1995)

The second working hypothesis of the salmon bias effect is based on unreported emigration. In a system of computerized and centralized population registers, this is most likely to occur when one member of a household leaves the country while the others stay behind (adolescents leaving the parental home, older persons leaving their children, etc.). This seems most unlikely for the adult population of working age however. We can simulate the unobserved number of people who would have to leave the country to influence mortality figures so as to obtain the same mortality rate as the Belgian population. The calculation is simplified by applying standardized mortality rates and a theoretical maximum exposure time of 58 months. This rough calculation suffices to demonstrate the huge scale of unreported return migration that would be necessary for some nationalities in order to use “statistical immortality” as an explanation for low mortality (Table 4, column 10).

Another approach can be used to estimate the extent of unregistered deaths among people who were enumerated in the census (and hence included in the population register). As mentioned earlier, unregistered death occurs when people die outside the country. In the case of Belgian nationals, the consular services will be informed and a time lag in registration is the only potential problem. Although the correct date of death is registered, some information may be missing at the time of analysis.

The situation is different for the death of a person without Belgian nationality. If the deceased did not work, and without an incentive for relatives to declare the death outside Belgium, some cases may remain unreported. To estimate the scale of the problem, a distinction was made within each group between those who had acquired Belgian nationality and the others. Survival analysis using Cox regression only revealed a potential problem for the Turkish population. The relative mortality risk was found to be significantly higher for those who had acquired Belgian nationality, indicating that mortality among the other Turkish immigrants may be underestimated.

Given that we merely tried to equal the mortality rate of the Belgian population, while most migrant populations have a marked socioeconomic disadvantage, we can conclude that the salmon bias hypothesis cannot explain the “migrant paradox”, at least not in the Belgian case.

The data at hand are less helpful for evaluating the selection hypothesis. Under the “healthy migrant” hypothesis, lower mortality of adult migrants compared with the host population is explained by a selection process in the population of origin. It is obvious that general health among migrants measured shortly after migration is probably better than the health of the population they originate from. However, the effect of health selection can only influence mortality in the period immediately after migration or lower the incidence in the migrant population for some specific chronic health problems such as mental illness. It is difficult to imagine the influence of a selection process in the country of origin that would be responsible for lower cancer mortality several years later. Specific mortality patterns can thus be useful for estimating the possible influence of selection processes.

Table 5
Influence of the country of origin on mortality risk according to the different independent variables used in the Cox regression model (adults aged 25-54 in 1991), Belgium, 1991-1995

Among women, the relative risk of sub-Saharan Africans is significantly higher than for Belgians, whereas German and French women do not differ significantly. Spanish women have the lowest relative risk. After including education in the model, the relative risk of several migrant populations is even lower, especially for the Moroccan and Turkish communities. This shows that these groups have a lower mortality than could be expected on the basis of their educational level. Controlling for the housing variable operates in the same direction, meaning that most migrants have a better mortality profile than could be expected on the basis of their socioeconomic status. The reduction of the relative risk for French men shows that their lower socioeconomic status is partially responsible for their high mortality compared with Belgians.

Obviously, the major causes of death are the prime suspects for explaining the large differences in total mortality between population groups. In Belgium, in absolute terms, ischaemic heart disease is the single main medical cause of death for both sexes taken together, followed by lung cancer. However, in the study age group, external causes of death also take a heavy toll, suicide being the most important, responsible for one tenth of all deaths.

Looking at each sex separately, the picture is very different for men and women. Apart from the fact that male mortality is about twice as high as female mortality, we find a different proportional distribution of cause-specific mortality. Suicide, ischaemic heart disease and lung cancer are the three main causes of death among men, all in a similar order of importance. Among women in this age group, breast cancer is the dominant cause of death, followed by suicide. The mortality rates from these causes vary substantially between migrant groups (Appendix tables A and B) and there are some very large differences with respect to the general pattern of mortality in Belgium.

Table 6 shows the result of a Cox regression for the main causes of death: suicide, lung and breast cancer and ischaemic heart disease. Age has been entered as a categorical variable because of the non-linear relationship between age and mortality for some specific causes of death. Taking into account possible artificial differences caused by diagnostic problems, we also grouped all heart diseases, except cerebrovascular diseases and “other diseases of the circulatory system”. Because of their particular contribution to the cause-specific differences in migrant mortality with respect to the rest of the population, we also added infectious and parasitic diseases, alcohol-related mortality and cancers of the digestive system.

Table 6
Influence of country of origin on mortality for the main causes of death among adults aged 25-54 in 1991, Belgium, 1991-1995 (relative risks and 95% confidence intervals)

Turkish and Moroccan men have lower relative risks for virtually all causes. Traffic accident mortality of Turkish men is an exception. For most causes the lower relative risk is statistically significant, except for lung cancer and ischaemic heart disease among Turkish men. Taking all heart diseases together, the main results stay identical, although the relative positions of Moroccan and Turkish men become more similar. The relative risk of Italian men also becomes significantly lower than that of Belgian men.

The clearest contrast is observed for mortality from suicide and alcohol consumption, which is much lower in the Islamic migrant communities. Suicide is also very low in the Italian community, as is mortality from alcohol consumption in the Spanish one. Lung cancer mortality is low for Moroccans, as is breast cancer mortality for Moroccan, Turkish, Spanish and German women.

The introduction of the socioeconomic status (not presented) does not change much in this general pattern, except for reducing the relative risks of the migrant population, as could be expected from earlier results for total mortality. Suicide is an exception; after controlling for education, we observe an increased risk for French and Dutch women.

Figure 1 shows the contribution of specific causes of death to the lower total mortality for some selected nationalities. For each group, differences are analysed relative to the Belgian population and cannot really be compared with each other. The charts show the contributions of specific causes to the difference in the total mortality rate compared with the native Belgian population. The causes that contribute to higher mortality are shown on the right-hand side. For Dutch men, whose total mortality rate is much lower than that of Belgian men, almost all bars point to the left. For sub-Saharan African women, with a higher mortality risk than Belgian women, the graph shows the contribution of infectious diseases to the higher mortality rate.

In the WHO Mortality Data Base, which is a compilation of official data reported by the member states, the difference in life expectancy at birth for the period 1990-1995 is about 15 years for males and females between the extremes of France and Morocco. Compared with Belgium, where the mean life expectancy of men and women is 73.3 and 80 years respectively, four countries have a better life expectancy for the same period: France (73.3 and 81.4 years), Spain (73.8 and 81 years), Italy (74 and 80.5 years) and the Netherlands (74.3 and 80.2 years), followed closely by Germany (72.6 and 79.1 years).

A much lower life expectancy at birth is found in Morocco (62.8 years for men and 66.2 years for women) and in Turkey (65 and 69.7 years respectively). Most of the countries of sub-Saharan Africa have even shorter life expectancies. Although child mortality is responsible for the biggest share of this difference between the low and high mortality countries, the information at hand still suggests large differences for all age groups.

When we look at the risk ratio of mortality computed for the Belgian migrant groups in our study (Table 5), the amplitude and direction of the differences appear to bear no correlation with the countries of origin. The low mortality rate for Mediterranean migrants is consistent with the findings in other European countries (Brahimi, 1980; Razum et al., 1998; Uitenbroek and Verhoeff, 2002). The high risk ratio for African women is not unexpected. A remarkable result is the high mortality among French men, especially when compared to Germans and Dutch.

Is this broad spectrum between nationality groups belonging to countries with comparable life expectancy plausible? Why is the French risk ratio so much higher? If we control for education and housing (Table 5), the French comparative disadvantage can largely be explained by a lower socioeconomic status compared with the Belgian population. The French, Dutch and German migrations are largely border migrations and in fact the risk ratios obtained are remarkably in line with the mortality of the local Belgian population (Van Oyen et al., 2002). In an analysis of regional differences in mortality (Deboosere and Gadeyne, 2002) risk ratios between districts go from 0.7 for the district of Maaseik, in the northern part of Belgium on the border with the Netherlands, to 1.38 for the district of Arlon, in the southernmost tip of Belgium at the French border. French migrants reside overwhelmingly in higher mortality districts, while the Dutch are in the lower end of the distribution. Moreover, the distribution of mortality in France is also geographically unequal and the northern part of France, bordering Belgium, is clearly a region with the lowest life expectancy in France. Male mortality exceeds the national average by 25% in the Nord Pas-de-Calais and by 12% in Picardie and for women by 14% and 11%. This higher mortality appears to extend to almost all causes of death, but is particularly marked for alcohol-related mortality (Vallin and Meslé, 1988; Salem et al., 1999). The high mortality rates and the mortality pattern for residents of French origin in the southern part of Belgium are thus closely matched with the patterns on the other side of the border.

The conclusion that can be drawn from this observation is that large differences in risk ratios are possible inside the same country for migrant groups originating from countries with comparable levels of development. So we should not be surprised when our results show mortality rates that are twice as high for one subpopulation as for another. Even small differences in absolute numbers can create substantial differences in relative risks, and this does not necessarily reflect poor quality of data. The fact that our regional data coincide with the results for Dutch and French residents can be considered as an additional external test for the consistency of our dataset.

The lower risk ratios for Italian and Spanish migrants are in line with the mortality figures in their home countries and with international findings. These results are probably largely related to the Mediterranean diet. Both countries have traditionally supplied workers to Belgium. But, as these are “old wave” migrant groups, they have gradually evolved towards the socioeconomic composition of the Belgian population.

Turks and Moroccans were part of the latest labour migration in the seventies. Most of these “guest workers” belong to the population with the lowest socioeconomic status in Belgium. Nevertheless, they show low mortality and it falls even lower (as expected) after controlling for socioeconomic status. These results tally with the observations in other European countries. Low mortality has been observed for Turks in Sweden (Ringbäck Weitoft et al., 1999) and Germany (Razum et al., 1998), for Turks and Moroccans combined in the Netherlands (Uitenbroek and Verhoeff, 2002) and for Moroccans in France (Khlat and Courbage, 1995). All these studies document different explanations for this observation. Khlat and Courbage propose the healthy migrant effect. Razum and Uitenbroek and Verhoeff refute the healthy migrant effect and argue for the return-migration effect. Ringbäck Weitoft et al. highlight the importance of non-registered return, whereas Uitenbroek and Verhoeff do not see registration problems as a major factor. But all of them observe an unexplained residual difference that does not tally with the social status of these migrant groups.

In general, given their overall lower mortality rate, migrant communities can be expected to have a lower relative risk than the native population for most causes. However, for three one-sided graphs (Moroccan men, Italian men and women) total mortality may be underestimated.

It is possible to discern the specific causes which contribute most to the low total mortality rate. Just a few sets of causes are responsible for the bulk of the difference: a mere six causes account for half of the adult mortality. Suicide alone is responsible for 10% of all deaths in the studied age group. The other important causes are ischaemic heart disease (8.5%), lung cancer (8%), traffic accidents (7%), cancers of the digestive system (7%) and breast cancer (16% of female deaths). Large differences in total mortality are generated by some of these leading causes of death. Extremely low rates in less important causes can also contribute to the overall outcome however. This is the case, for instance, for alcohol-related mortality where all alcohol-related causes combined account for only 3.5% of adult mortality, but where some migrant groups tend to have virtually no mortality at all for this cause.

Suicide is the leading cause of death in this age group in Belgium and appears to be the main explanatory factor for the lower mortality among Moroccan, Turkish and Italian men. The low suicide rate among adult Moroccans was also found in France (Khlat and Courbage, 1995) [5]. This demonstrates the importance of the cultural or religious factor, suicide being a strong taboo in Islamic countries. Southern European countries (Italy, Spain) are also known to have very low suicide figures. The low suicide rate for Dutch men and women compared with the Belgian rates correlates entirely with the differences observed between the countries concerned (Diekstra and Gulbinat, 1993). The French rates for men are exceptionally high. This reflects the fact that France is among the countries with high suicide rates, particularly in the north-western part of the country (Salem et al.1999). Some migrants may experience such overwhelming levels of anxiety and depression as to resort to suicide (Trovato and Clogg, 1992), but the background culture of the country of origin seems to be the strongest predictor for suicide (Trovato, 2003). Sub-Saharan African men also have a slight excess mortality for suicide. It is well known that suicide can be provoked by a large set of factors ranging from mental health problems to drug abuse and familial or financial difficulties. Unemployment has frequently been related to suicide. Although unemployment is extremely high among Moroccan men, this is not reflected in the suicide rate. However, unemployed Moroccan men have a much higher risk than employed Moroccan men (results not shown). Khlat and Courbage (1995) observe high suicide rates for young Moroccan women in France, showing the possibility of large variations in suicide rate inside the migrant communities related to particular conditions.

Ischaemic heart disease (IHD) is the single most important cause of death worldwide and the most lethal kind of circulatory disease. There are several well-known risk factors for IHD including hypertension, high cholesterol levels, smoking, obesity and diabetes. Prolonged high levels of low density lipoproteins in the blood vessels can cause an accumulation of plaque and obstruction of the circulatory system. This condition is probably linked to genetic predisposition, diet, tobacco consumption or physical inactivity. IHD reflects various aetiologies and IHD mortality is the result of counterbalancing or reinforcing factors. Thus an interpretation in terms of risk behaviour appears difficult. There are large between-country differences both in the levels and trends in mortality from cardiovascular diseases, especially from ischaemic heart disease. The between-country differences for European countries are not completely reflected in our data. Inhabitants of Dutch and French origin show differences with respect to native Belgians which are exactly opposite to those observed between Belgium and the respective countries. While French women are known to be among those with the lowest cardio-vascular mortality in Europe, French women living in Belgium have only a slight advantage over Belgian women. French men living in Belgium perform even worse than Belgian men. There has been some suspicion that differences in coding may affect the differences between countries i.e. between Belgium, France and the Netherlands (Sans et al., 1997). But again, there are large differences between regions of a single country, with the highest IHD mortality in the northern regions of France. Sans et al. emphasize that spatial variability in mortality patterns in Europe can be shown to cross borders. There is a clear north-south gradient inside Belgium for IHD and this is also reflected in the Dutch and French migrant communities, with the former living mainly in the north and the latter mainly in the south of the country.

While Moroccan men show an advantage over Belgian men, the same does not apply to Moroccan women. These findings are similar to the situation in France. Khlat and Courbage (1995) assume that in general both men and women are protected by the typical Mediterranean low-fat diet, but that Moroccan women are less well protected due to unbalanced eating habits. This is also reflected in diabetes mortality. Excess mortality from diabetes mellitus is also found among Moroccan women living in Belgium. The primary risk factor for non-insulin-dependent diabetes seems to be obesity, also a primary risk factor for IHD. The Belgian national health surveys of 1997 and 2001 show a significantly higher body mass index [6] for women of Turkish and Moroccan origin than for all other women (Anson, 2000; Observatorium voor Gezondheid en Welzijn, 2004). At the same time, for men and women alike, health surveys report a higher and more frequent consumption of fruit and vegetables for all Mediterranean migrant populations.

Lung cancer became a leading cause of cancer death for Belgian men in the late 1950s. The primary risk factor is cigarette smoking, an association repeatedly demonstrated in epidemiological studies (Lopez, 1995). Occupational exposure has also been shown to increase lung cancer risk. Belgium is among the European countries with the highest lung cancer mortality (European Commission, 2004). Only French males have higher lung cancer mortality.

As Islamic faith does not prohibit tobacco consumption, no protective role regarding this aspect of health behaviour is observed. Lung cancer mortality among Turkish men is among the highest, at the same level as Belgian, French and Italian men. Smoking is also significantly more prevalent among Turkish men than among the other nationalities. The population of male workers of Mediterranean origin is known to have high exposure to occupational risks. This situation accentuates the low lung cancer mortality of Moroccan men also found in France (Khlat and Courbage, 1995). Moroccan adult men are known to be moderate smokers and consequently have significantly lower lung cancer rates. The 1997 health survey shows that over half of the Moroccan respondents aged 45-64 never smoked, compared with 36% of the Belgian population. The highest proportion of heavy smokers in this age group was observed among southern European respondents (22.8%), the lowest among Moroccans (5.8%) (Anson, 2000). Smoking behaviour is also likely to be partially responsible for the low ischaemic heart mortality among Moroccan men. The cultural attitude towards smoking for women still results in extremely low lung cancer deaths in the Turkish and Moroccan communities. According to health surveys, the prevalence of smoking among young men and women aged 15-24 is very similar; only Moroccan and Turkish women are less likely to smoke. A remarkable exception in the relationship with the country of origin is the high (although not significant) relative risk of Spanish women.

Breast cancer is the single most important cause of female death in the study age group, accounting for one in six deaths. Family history, and late age at first birth or nulliparity are well-established risk factors. All migrant groups have a lower relative risk than the population of Belgian origin. The relative risk of women of French origin is slightly (non-significantly) lower. Moroccan and Turkish women have significantly lower breast cancer mortality. Low breast cancer mortality has also been found among Moroccan women in France and among Turkish women in Germany. Lower incidence of breast cancer among Turkish and Moroccan women is an important factor in their lower general mortality. The single factor of low breast cancer mortality largely explains the low mortality of adult Turkish women. This finding is consistent with the theory of early childbearing as a key risk reduction factor. It is possible that low consumption of animal fat and low exposure to tobacco smoke play an additional role. In Germany an increase in breast cancer mortality among Turkish women (Zeeb et al., 2002) indicates that Turkish adult women may progressively lose their relative mortality advantage.

In the early 1990s, traffic accidents were the fourth most important cause of death in the 25-54 age group. Turkish men and immigrants of French origin are the only groups with (non-significant) higher mortality from this cause than the Belgian population. This does not correspond with observations in other countries. In France, a country with equally high traffic accident mortality, immigrants have excess mortality for this cause (Brahimi, 1980) and the relative risk for adult Moroccans is 50% higher than for the French (Khlat and Courbage, 1995). The low mortality for the Moroccan community in Belgium is probably the result of their residential pattern. More than 50% of all adults of Moroccan origin live in the Brussels capital region and are less exposed to fatal motor vehicle accidents. This is the result of widespread use of public transport and of high traffic densities in the Brussels conurbation. Lower alcohol consumption probably also contributes to the lower traffic accident mortality.

Religion plays a crucial role in alcohol-related mortality. The prohibition of alcohol in Islam results in extremely low figures of alcohol-related mortality among Moroccan and Turkish men. In the epidemiological literature, moderate alcohol consumption is stated to have some positive effects, but alcohol abuse results in high alcohol-related mortality figures. Men and women of French origin have a significantly higher relative risk of alcohol-related mortality. Alcohol abuse can also have an indirect negative effect on other causes of mortality. A strong relationship has been demonstrated with suicide, traffic accidents, liver and gall bladder diseases and cancers of the oesophagus. These results indicate that there is probably an indirect positive effect on all these causes of death in the Islamic communities while the opposite is true for the migrants of French origin.

Infectious and parasitic diseases constitute the most notable exception, with higher mortality for almost all migrant groups. However, this cause of death is relatively minor in absolute terms and has little effect on total mortality. The higher mortality from infectious and parasitic diseases among migrant communities has also been observed in other countries. It is a category of diseases typically linked to absolute deprivation (poor housing, malnutrition and limited access to health care). Migrant communities living in the most deprived parts of our cities and at the bottom of the socioeconomic ladder are more vulnerable. The extremely high relative risks for the sub-Saharan African community, both among men and women, reflect the prevalence of HIV/AIDS among this migrant group in the early 1990s.

The fact that most migrant communities (except for French and sub-Saharan African women) have a lower total mortality is indicative of the potential improvement in public health that can be achieved in a rich and highly developed society such as Belgium. The comparison of Belgian health expectancy figures with other countries also shows that there is a considerable margin for improvement. However, the advantage of the present research is that we can exactly pinpoint major causes where progress can be made. One of the most evident observations is the dominant role of man-made causes of death in generating mortality differentials between the Belgian and migrant male population. For adult Belgian men aged 25-54, alcohol abuse, tobacco consumption, traffic accidents and suicide are the main causes of death. Three of these four causes contribute immensely to the migrant mortality advantage. In traffic accidents, the advantage is less clear cut, with Turkish men even having higher mortality rates.

While a breakdown by cause is the best approach for analysing subpopulation differences, the standardized mortality rate is the best measure for making global comparisons over the entire spectrum of nationalities. In that respect, it is important to underline how strongly cause-specific mortality can differ between populations. Bringing gender differences into the picture, it is even clearer that many deaths in the 25-54 age group are rightly classified as “preventable mortality”, giving more support to the thesis that differences in total mortality are not an artefact.

It also gives credit to the assumption that a significant part of the gender gap in life expectancy is due to unhealthy lifestyles and adverse working conditions for men. In recent years there has been a continuous improvement in life expectancy, closely linked to more healthy lifestyles. Between 1954 and 1994 there was a fourfold increase in lung cancer mortality among men in Belgium. The diminishing proportion of cigarette smokers among men is now finally reversing this trend, with an improvement of male life expectancy and a reduction in the life expectancy gap between men and women. However, the differences in standardized lung cancer mortality between Moroccan women and Belgian women show how much more improvement is possible. The highest female lung cancer rate (13 per 100,000 for German women) is still below the lowest rate for men (Moroccan with 18 per 100,000), but the differences within each gender group are much larger.

External causes of mortality, and especially suicide, are consistently much higher for men than for women, over all nationalities. Only the Turkish and, to a lesser extent, the Italian populations have a sex ratio for suicide below two.

We must keep in mind that adult mortality is only one aspect of the global health situation of migrant communities. Looking at the broader picture of total life expectancy we have a much bleaker image. Perinatal mortality in the Turkish and Moroccan population is more than the double the Belgian rate and has been so for many years (Peters and VanderVeen, 1990; Aelvoet et al., 1998). Likewise, mortality among young children has long been reported as higher in migrant communities in Belgium (Maffenini, 1980).

Especially alarming is the mortality rate among young Moroccan men under age 25. A similar kind of “anomaly” has been signalled by Hayes-Bautista et al. (2002) concerning Latino male adolescents. Further analysis is needed to conclude whether this adolescent mortality peak in the Maghreb community heralds a surge in mortality among Moroccan adults.

Finally, all the information about morbidity in adult migrant populations reveals a sharp contrast with the more favourable mortality rates. Belgian health surveys (1997 and 2001) report that subjective health is worse in the Islamic migrant communities than in the Belgian population. The census of 2001, introducing some new questions about subjective health, gives additional support to these negative findings. Taking account of their socioeconomic status and compared with the Belgian population, adult migrants of some Mediterranean countries have lower mortality, but do not necessarily have better health. The Turkish migrant population, with a lower total mortality than the general population, emerges with the most disadvantaged health indicators. Its morbidity is more consistent with the lower socioeconomic position than its mortality. Lower mortality appears to be the result of a more protective diet, less alcohol consumption, very few suicides and early childbearing for women compared with the native Belgian population. The contrast between mortality and morbidity is frequently reported when analysing gender differences, pointing in the same direction: mortality is merely an indicator of the prevalence of fatal diseases, but not necessarily of general health. For many chronic diseases, Razum et al. (1998) observed that among Turkish residents in Germany, morbidity and disability are disproportionately higher than mortality. Ringbäck Weitoft et al. (1999) observe lower migrant mortality among disability pensioners and conclude that this relatively low mortality rate might indicate that they suffer from fewer fatal diseases than Swedish-born disability pensioners. Uitenbroek and Verhoeff (2002) report that data both on self-reported illness and on the use of health services show a high level of chronic though usually non-life-threatening illness. In particular, musculoskeletal problems, such as back complaints, and socio-psychological problems seem to be widespread in migrant populations. And they conclude that although these conditions are not life-threatening, they can be seriously disabling and can have a very negative effect on quality of life.

The salmon bias cannot explain differential mortality and cause-specific mortality patterns. A small effect of statistical immortality is possible and most probable for Turkish women. A healthy migrant selection effect cannot be excluded on the basis of our data, but is unlikely to be the main explanatory factor. Selection plays a role, not with regard to the host country but to the country of departure. The healthy migrant selection effect primarily means that people who emigrate have to be fit enough to emigrate. However this does not necessarily explain why their mortality is lower than mortality within the host population. The reasons for the paradox appear to be multifactorial and at least a combination of lifestyle, dietary intake variations and the effect of the health infrastructure in the host country. Our results show mortality patterns that are more easily explained by beneficial cultural endowments than by selection processes. We cannot affirm that selection processes and data problems do not influence the final result. But we can conclude firstly that they do not suffice to explain the complex mortality patterns and secondly that beneficial cultural and behavioural factors are the most obvious explanations for these patterns.

Cause-specific mortality patterns are consistent with our knowledge about health behaviour, lifestyle and diet of migrant populations. Lower adult mortality in some migrant communities is related to a small number of specific causes mostly depending on health behaviour. Low suicide prevalence is of major importance for Islamic and Mediterranean populations. This contrast is particularly striking given the very high suicide figures for the native Belgian population. Alcohol consumption and smoking behaviour constitute a second group. Diet could be a third explanatory factor.

The relative advantage, especially for the Moroccan community, is at risk of disappearing in the near future. Young Moroccan men are starting to develop very high mortality figures. As the protective health behaviour disappears, these younger generations will bear the full negative impact of a lower socioeconomic status.

These results yield opportunities for a better orientation of health services towards specific subpopulations. The much better outcome for some causes of death in populations with a low socioeconomic status shows the potential for improving the health status of the general population.

Cause-specific standardized mortality rates⁽¹⁾ for 100,000 men aged 25-54 in 1991, by country of origin, Belgium 1991-1995

Cause-specific standardized mortality rates⁽¹⁾ for 100,000 women aged 25-54 in 1991, by country of origin, Belgium 1991-1995