Schizophrenia and Migration: A Meta-Analysis and Review

Using the key words “migration,” “ethnicity,” “psychosis,” and/or “schizophrenia,” a computerized search of MEDLINE was systematically conducted for possibly relevant publications appearing between January 1977 and April 2003. A database of all schizophrenia incidence studies published during this period was also searched (provided by Dr. John McGrath, Brisbane, Australia). Bibliographies from identified articles were cross-referenced.

To be included in the meta-analysis, studies had to fulfill the following criteria: 1) the study reported schizophrenia incidence rates for one or more migrant groups residing in a circumscribed area or provided numerators and denominators for such calculations, 2) the study included correction for age differences between the immigrant group and the reference group or provided data that made this correction possible, and 3) the study was published in an English-language, peer-reviewed scientific journal.

In most studies, the designation of migrant status was based on the country of birth of the subjects in question or their parents. Denominators for studies conducted in the United Kingdom, however, were based largely on categories derived from national censuses, with some studies using “whites” (11, 13–15) or the “remainder of the general population” (e.g., references 10, 16) as the reference group. Thus, some members of the reference group might actually have recently migrated to the United Kingdom, and some “blacks” or members of ethnic minority populations might possibly have lived in the United Kingdom for more than two generations. This type of classification error would, however, primarily tend toward an attenuation of the relative risks currently obtained for migrants versus nonmigrants.

In selecting studies, we sometimes found that two or more articles reported epidemiological studies within the same region or time frame. When there was a complete overlap of samples, only one study was included in the analysis. The overlap of samples between the study by McGovern and Cope (Birmingham, England, 1980–1983) (17) and that by Cochrane and Bal (England, 1981) (18) was small, and both studies were included. Two Danish register studies (4, 19) overlapped partly in the years studied. The study by Cantor-Graae et al. (4) was selected because the information on ethnicity was more accurate and because that study included outpatients after 1995. The overlap between the studies by van Os et al. (14) and Boydell et al. (20), however, was large, and the former study was selected because it provided a better definition of ethnicity. One study conducted in Manchester, England, was excluded because no age correction could be performed (21).

Studies that combined the results for schizophrenia with those for other psychoses (e.g., schizophrenia and paranoia or schizophrenia and paranoid psychosis) were included. When the same study reported separate results for psychoses and for schizophrenia, the results for schizophrenia were selected.

Some studies presented figures for the first generation only. Other studies reported separate relative risks for first- and second-generation migrants or combined the results for both generations into one figure. Consequently, we prepared three different data sets. The first data set concerned the effect sizes for first-generation migrants only; the second data set concerned those for second-generation migrants only. To make use of a large data set that could be regarded as an aggregate of the general migrant effect and that could yield maximal power for a series of statistical analyses, we also composed a third data set. This data set included 1) studies that did not distinguish between first- and second-generation migrants, 2) studies that made this distinction but from which one could extract figures for the two generations combined, and 3) studies that reported effect sizes for the first or the second generation only. Only one effect size for each migrant group described in the selected study was included in this third data set.

From each study, relative risks were obtained for one or more immigrant groups. Extraction of data and calculation of relative risks were performed by the two authors independently, and consensus was reached in case of discrepancies. The available data for numerators and denominators were used to calculate age- and gender-adjusted relative risks by Poisson regression analysis. When possible, separate relative risks were computed for first- and second-generation migrants and for male and female migrants.

There were large differences between studies in the presentation of details concerning numerators, denominators, and rates. In order to use the same method of variance estimation for all studies, we used the formula V=1/N_n + 1/N_m, where N_n is the number of native cases and N_m is the number of migrant cases (derived from the formula for the variance of odds ratios) (22), as previously done by Aleman et al. (23). In order to prevent studies with very large samples from dominating the analyses, the number of subjects in studies with more than 500 patients was set at 500, as suggested by Shadish and Haddock (24). A homogeneity statistic, Q_W, was calculated to examine whether the various effect sizes that are averaged into a mean value can be assumed to estimate the same population effect size. Significant values of Q_W indicate heterogeneity across studies. The Q_B statistic was used to test whether differences in effect sizes between groups (e.g., risks for male versus female migrants) were statistically significant.

Since the effect size distributions remained heterogeneous even after modeling the between-study differences, analyses were carried out within the mixed-effects model. This model assumes that the effects of between-study variables are systematic but that there is a remaining unmeasured random effect in the effect size distribution in addition to subject-level sampling error. All analyses were performed with the Meta Win 2.0 statistical package (25).

We first analyzed the studies pertaining to first-generation migrants (data set 1, analysis 1) and second-generation migrants (data set 2, analysis 2). For the remainder of the analyses (analyses 3, 4, 4a, 5, 5a, and 6), data set 3 was used, starting with the computation of the mean relative risk for the studies in data set 3 (analysis 3).

In analyses 4 and 4a, the risks for migrants from developing countries were compared to those for migrants from developed countries. Countries were classified by using the United Nations Conference on Trade and Development (UNCTAD) classification of countries for the year 1995 (26). UNCTAD classifies countries into five categories: 1) countries with developed market economies, 2) Eastern European countries, 3) developing countries with relatively high income (e.g., Trinidad), 4) developing countries with middle income (e.g., Jamaica), and 5) developing countries with low income (e.g., India). In analysis 4, migrants from UNCTAD groups 1 and 2 were compared to migrants from developing countries as an aggregate category (UNCTAD groups 3–5). To compare extremes in developmental levels, we also examined a tripartite division: UNCTAD groups 1 and 2 versus groups 3 and 4 versus group 5 (analysis 4a). Migrants from regions (e.g., the Middle East) were classified according to the index of those countries from which most migrants originated. Effect sizes for groups with regionally mixed birthplaces, i.e., the groups studied by Zolkowska et al. (27), the groups designated “other” in the study by Selten et al. (3), blacks in the study by Goater et al. (11), and second-generation migrants in the study by Cantor-Graae et al. (4) were excluded from this analysis.

The effect of ethnicity as defined on the basis of skin color was examined in analysis 5 by comparing migrants from areas where the majority of the population is classified as white (Europe, North America, North Africa, Turkey, the Middle East, Australia; skin color group 1) to migrants from areas where the majority of the population is classified as black (the Caribbean, sub-Saharan Africa; skin color group 2) and to migrants from areas where the majority of the population cannot be classified as white or black (India, Pakistan, Asia, South America, Greenland; skin color group 3). In addition to excluding the three effect sizes for groups of mixed birthplace (3, 4, 27), we excluded three effect sizes for migrants who constituted a mix of groups 2 and 3; i.e., the Asian and Caribbean migrants in the study by Hitch and Clegg (28) and the Surinamese migrants in the study by Selten et al. (3, 29). Migrants from Turkey, Morocco, and the Middle East were assigned to skin color group 1, i.e., white (analysis 5), and also, in a separate analysis, to skin color group 3 (analysis 5a). Finally, possible gender effects were examined in analysis 6.

The most comprehensive of the meta-analyses, which included both first- and second-generation migrants, yielded a mean weighted effect size of 2.9 (95% CI=2.5–3.4) for the risk of developing schizophrenia among migrants. It should be observed that significant heterogeneity was found across studies, indicating that the different migrant groups cannot be regarded as coming from one population having a common effect size. It is important to note that few studies examined all migrant groups in a given country, and most studies were directed at specific groups with high risks, such as the Surinamese in the Netherlands. However, an overall effect size of this magnitude indicates that there can be little doubt about the existence of an association between migration and schizophrenia.

The largest and most well-documented risk factor related to the development of schizophrenia is a family history of the disease (35). The effect size associated with migration is, however, greater than the effect size associated with most other risk factors putatively implicated in the etiology of schizophrenia. A recent meta-analytic review showed that the odds ratios for the risk for developing schizophrenia associated with obstetric complications were generally less than 2.0 (36). Winter birth is associated with a 5%–8% increase in the risk for schizophrenia (37), corresponding to a relative risk of 1.05–1.08. The only risk factor with an effect size approaching that of migration would be urbanization, for which relative risks of 2.4 (38) and 2.75 (39) have been reported.

A growing number of empirical findings suggest that Ödegaard’s selection hypothesis (5) can be rejected as the sole explanation for the findings on migrants. Findings of fivefold higher incidence rates among Surinamese migrants to the Netherlands than in the native-born population constitute one of the strongest arguments against the selection hypothesis, insofar as more than one-third of the Surinamese population emigrated (6, 29). Moreover, selective migration cannot explain the increased risk for schizophrenia found among second-generation migrants (Table 2) or among migrants residing in Denmark before their 15th birthday (4). Similarly, the median ages of migration for Surinamese and Moroccan-born patients in the study by Selten and colleagues (3) were 9 and 15 years, respectively. Although a role for selective migration cannot be excluded in these migrants’ parents, the risk of developing schizophrenia among second-generation migrants in Denmark was lower for those with a parental history of psychiatric disorder than for those without such a parental history (4).

It may be pertinent to note that the premorbid and prodromal characteristics of schizophrenia patients, which include negative symptoms and frontal dysfunction, do not seem compatible with the effort required to emigrate. Considering that even more effort may be required for persons to migrate from developing countries to Europe, such individuals should have lower risks for schizophrenia, contrary to the current findings. Also relevant in this regard, Rosenthal et al. (48) found that emigration rates of adopted children born to a biological parent with schizophrenia spectrum disorder were lower than for children born to parents who did not have a schizophrenia spectrum disorder. It may be further noted that incidence studies conducted in the Caribbean (e.g., reference 49) and Surinam (50) show normal incidence rates for schizophrenia. Thus, genes alone cannot explain why individuals from these countries develop a higher risk for schizophrenia when they migrate.

There is still some controversy about whether environmental factors contribute to the etiology of schizophrenia. Nevertheless, meta-analytical evidence from twin studies indicates that schizophrenia has complex multifactorial determinants that would include environmental factors (51). The current findings similarly suggest the operation of environmental factors, albeit their exact nature remains uncertain. Moreover, viable hypotheses concerning migrants must meet the challenge posed by increased rates of schizophrenia found in first-generation and in second-generation migrants. Thus, for example, an impaired immunity to a neurotropic virus found in Europe would not explain the findings for second-generation migrants, and impaired intrauterine immunity would not explain the higher rates in first-generation migrants. Studies analyzing the effects of Borna disease and influenza viruses have yielded negative results in migrants (52, 53), but other viral mechanisms are possible. Delayed exposure to a common infectious agent such as Epstein-Barr virus has been suggested as a model for multiple sclerosis, a disorder that also shows migration effects (54). Nevertheless, the intergenerational risk pattern of multiple sclerosis in migrants differs from that of schizophrenia. Migrants from countries with a warm climate generally retain the low risk for multiple sclerosis associated with their country of birth, whereas the risk in their offspring approaches that of the host country population. Moreover, whereas climate and latitude are strongly correlated with multiple sclerosis rates, climate factors seem increasingly unlikely, considering the increased risks for schizophrenia in migrants to Denmark from both Australia and Greenland (4). Prenatal exposure to a vitamin D deficiency associated with dark-skinned migrants living in cold climates might increase the risk for schizophrenia among second-generation migrants but not among first-generation migrants (55). Childhood exposure to animal-borne infectious agents less frequently found in the country of origin, such as Toxoplasmosis gondii (e.g., reference 56), might merit further study, since such exposure would affect persons migrating early in life as well as second-generation migrants.

Some data suggest that higher rates of schizophrenia are found in so-called modern countries (57). The mechanism involved in modernization would require further specification, since numerous factors could be involved. Concepts akin to modernization are urban birth and urban upbringing, factors strongly related to higher risk for schizophrenia (38, 39), yet whose specific components are still unknown. The notion that the migrant phenomenon is a manifestation of the urban effect would, however, require that migrants be more intensively exposed to the urban factor or be more sensitive to it. Use of illicit drugs, cannabis in particular, might tentatively fulfill these requirements, in that cannabis use may contribute to the development of schizophrenia (e.g., reference 58). However, it should be noted that the twofold risk associated with cannabis (59) is smaller than the effect size indicated by migration. Moreover, cannabis use would predict a greater risk for schizophrenia among male migrants (60), but a gender difference was not found in the current study. Migrant studies examining illicit drug use in first-onset psychosis patients have failed to find higher rates in African Caribbeans in the United Kingdom (60) or in migrants to the Netherlands (61). Also, a study of schizophrenia risk in Swedish conscripts found that the urban effect remained after adjustment for cannabis use (62).

Aspects of reproduction may be affected by migration. There is some preliminary evidence that greater paternal age contributes to the etiology of schizophrenia (63), and this factor could be further explored in migrants. More speculatively, it has been proposed that cephalopelvic disproportion arising from improved nutritional standards could lead to an elevated incidence of schizophrenia among second-generation migrants (64). Two studies have investigated history of obstetric complications in African Caribbean versus white British-born patients (65, 66). Both studies found fewer obstetric complications in African Caribbean patients, and no differences in first- versus second-generation African Caribbean patients (65). Also, although a decline in breast-feeding among first-generation migrants might lead to a reduction of any putative protective effects in second-generation migrants, data from two national birth cohorts failed to find any protective effect of breast-feeding with regard to risk of schizophrenia (67).

In general, the broad spectrum of migrant groups implicated would seem to refute the notion that any single biological or genetic factor could provide an adequate explanation. Between-population variation of this type does rather strongly support a causal role for aspects of the social environment (68). Explanations that incorporate socioenvironmental elements need not negate the genetic or biological nature of schizophrenia. On the contrary, an approach that seeks to integrate psychology and biology may provide a constructive step forward.

By far the most striking finding in the current meta-analysis is the greater effect size associated with black skin color (relative risk=4.8, 95% CI=3.7–6.2). These findings suggest that despite heterogeneity in birthplace, migrants whose skin color is considerably darker than the background population may share a common risk exposure. Individuals with darker skin are more often discriminated against in Western societies. Experiences of discrimination may foster a paranoid attributional style that facilitates the development of psychotic symptoms (69). A preliminary report from a prospective study conducted in the Netherlands suggests that perceived discrimination predicts the development of psychotic symptoms in healthy persons (70). However, such a hypothesis would also predict a higher rate of schizophrenia in African Americans, yet evidence in support of this association is generally lacking (e.g., reference 71). Indeed, findings from clinical samples suggest that schizophrenia may be overdiagnosed among African Americans, rather than underdiagnosed (72, 73) (but see reference 74). Nevertheless, prevalence rates based on treated samples may not yield valid estimates, because of differences in service use and diagnostic disparities (e.g., reference 75). Population-based incidence studies utilizing well-defined denominators would be required to resolve this potential anomaly.

It may be argued that aspects other than discrimination could contribute to the observed skin color effect. Persons with dark skin color (and migrants in general) may be more subject to the effects of poverty and low social class. It is, however, unlikely that the higher risks for migrants are explained by their low socioeconomic status. The association found between low socioeconomic status and schizophrenia appears largely attributable to social selection (downward mobility of genetically predisposed persons) rather than to social causation (76). Moreover, no clearcut relationship between parental social class and schizophrenia has been indicated across studies (77), and associations with both higher (e.g., reference 78) and lower parental class (79) have been found. It is important to note that a higher relative risk for schizophrenia has not been found among Turkish migrants to the Netherlands versus native-born persons, although Turkish migrants have lower income, educational, and employment levels than Surinamese migrants, whose relative risk for schizophrenia is high (3, 29). Although none of the studies in the meta-analysis controlled for socioeconomic class, Selten et al. (3) found no effect of neighborhood levels of socioeconomic status.

Ethnic disadvantage, more broadly defined, might make it more difficult for migrants to create a life plan (80), insofar as role models and community institutions may be lacking. Thus, the stress of making decisions concerning adult life, perhaps in combination with weaknesses in executive functioning, might contribute to the development of schizophrenia. Although discrimination and/or difficulties in formulating a life plan would explain the even higher rates of schizophrenia generally found in second-generation migrants, both hypotheses would still need to integrate those aspects of the brain that may be etiologically relevant for schizophrenia. A unifying hypothesis concerning migrants should also be able to explain the risk for schizophrenia associated with internal migration to areas of increased urbanization (39) and the protective effect of ethnic density, i.e., the relatively lower incidence of schizophrenia found in nonwhite ethnic minorities that represent larger proportions of the local population (20).

Such a common mechanism might be the long-term experience of social defeat, i.e., the chronic stressful experience of outsider status (77). Social defeat could arise whenever an individual is forced into a subordinate position in relation to a dominant group and could also arise through internal migration to areas of increased urban density, due to the highly competitive atmosphere in urban areas. Animal experiments using the defeated intruder paradigm show that social dominance has an effect on synaptic dopamine levels. In these experiments, the intruder rat is put into the cage of another rat that attacks him and forces him to show submissive behavior. Male rodents subjected to social defeat stress show elevated levels of dopamine in the nucleus accumbens and the prefrontal cortex (81). Repeated exposure to social defeat stress in rats causes increased self-administration of cocaine and amphetamine and increased sensitivity to these substances (82). Moreover, studies of primates have shown a strong relationship between dopaminergic activity and social rank (83, 84). Thus, disturbed brain dopaminergic function resulting from long-term experiences of social defeat could provide a common pathogenetic mechanism for the increased risk for schizophrenia in urban residents and migrants. Such a notion is purely speculative and based thus far on animal models. Moreover, social defeat is unlikely to be a sufficient cause of schizophrenia, but it could well play a contributory role within a multifactorial model. Although the effect of social defeat on the human brain would require further exploration with neuroimaging techniques, accumulating evidence suggests that stress can be responsible for anatomical changes in the human brain, including hippocampal (85) and gray matter (86) volume reductions. Brains with developmental anomalies may make other, possibly unique adaptations to stress (87).

Specific strategies are needed in order to test the factors putatively involved in the migrant effect. For example, little is known about ethnic groups across settings, e.g., the risk of schizophrenia in African Caribbeans residing in the United States versus those residing in the United Kingdom. A study of dark-skinned migrants who move from developing countries with warm climates to developed countries with warm climates (e.g., Australia) could help differentiate between climate and socioenvironmental factors. The effect of urban versus rural upbringing could be further explored in second-generation migrants. Sibling-pair designs have been successful in other contexts and could be used in studies of migrants to explore the relationship between environmental and genetic factors.