|Home | About | Journals | Submit | Contact Us | Français|
Twin studies suggest that Conduct Disorder (CD) is under substantial genetic influence, which is stronger for aggressive than for non-aggressive symptoms. Studies of migrating populations offer an alternative strategy for separating environmental and genetic influences on psychopathology.
To examine variation in the prevalence of CD associated with migration from Mexico to the US and whether this variation is similar for aggressive and non-aggressive CD symptoms and symptom profiles.
The prevalence of CD, different types of CD symptoms and CD symptom profiles were compared across three generations of people of Mexican origin with increasing levels of exposure to American culture: families of origin of migrants (in Mexico), children of Mexican migrants raised in the US and Mexican-American children of US-born parents.
General population surveys conducted in Mexico and the US using the same diagnostic interview.
Adults age 18–44 in the household population of Mexico and the household population of people of Mexican descent in the US.
CD criteria assessed using the World Mental Health version of the Composite International Diagnostic Instrument (WMH-CIDI).
Compared with families of origin of migrants, risk of CD is lower in the general population of Mexico (OR=0.54, 95% CI 0.19–1.51), higher in children of Mexican-born immigrants who are raised in the US (OR=4.12, 95% CI 1.47–11.52) and higher still in Mexican-American children of US-born parents (OR=7.64, 95% CI 3.20–18.27). The association with migration is markedly weaker for aggressive than for non-aggressive symptoms.
The prevalence of CD increases dramatically across generations of the Mexican-origin population following migration to the US. This increase is of larger magnitude for non-aggressive than for aggressive symptoms, consistent with the suggestion that non-aggressive symptoms are more strongly influenced by environmental factors than aggressive symptoms.
Conduct disorder (CD) is defined in the DSM-IV by persistent patterns of child or adolescent behavior involving aggression or other violations of age appropriate norms that cause significant clinical impairment1. Children who meet criteria for CD are at high risk for substance use disorders2 and high school dropout 3 as adolescents and unemployment, incarceration and early mortality as adults 4–6. CD is also a strong predictor of a broad range of adult psychopathology, including other impulse control disorders, such as intermittent explosive disorder, substance use disorders, and internalizing disorders, such as major depression and generalized anxiety disorder 7,8. Recent nationally representative studies in the US have estimated the lifetime prevalence of CD to be 9.5%, based on retrospective self-reports of adults age 18–44 9, and the past-year prevalence to be 2.1%, based on parental reports on children age 8–1510.
A large literature in behavioral genetics aims to identify environmental and genetic influences on CD11–13. Comparisons of concordance between monozygotic and dizygotic twins suggest that about 50% of population variability in CD symptoms is attributable to inherited liability, with significant components of risk attributable to both shared and unshared environmental factors14,15. Genetic and environmental influences may differ across CD symptoms. Some studies report stronger heritability and weaker environmental influence for aggressive symptoms (e.g. cruelty to others, fighting with a weapon) than for non-aggressive symptoms (e.g. stealing without confrontation, truancy) 16,17. Although there are significant exceptions to this finding 18, a recent meta-analysis estimated that genetic factors account for 65% of population variation in aggressive symptoms and 48% of population variation in non-aggressive symptoms 19.
Migration provides an alternative strategy for separating genetic and environmental influences on CD20,21. Populations that migrate alter their environmental conditions without altering their genetic composition. Variability in the risk for disorder associated with migration from one country to another indicates an environmental influence22. Epidemiological comparisons of immigrant sending countries in the Caribbean and immigrant receiving countries in Europe have found higher risk for psychotic disorders in the receiving countries, consistent with a hypothesized environmental influence on these disorders23. In the US, large differences associated with US vs. foreign nativity have been found in risk for mood, anxiety and substance use disorders for some, but not all, immigrant groups 24,25. A study of Puerto Ricans found no differences in disruptive behaviors between samples of children in San Juan, Puerto Rico and the South Bronx, New York26.
In this study we examine variation across immigrant generations in CD symptoms in a transnational sample of people of Mexican-origin in Mexico and the US, the largest sustained international labor migration in the world today. Differences in CD symptoms are examined dimensionally, according to symptom counts, and categorically, according to symptom profiles. This study is unique in migration studies in psychiatry because it includes representative samples on both sides of the border, assessments with the same structured interview, and the ability to distinguish sub-samples representative of successive generations with increasing exposure to American culture, from families of origin in Mexico to descendants of US-born Mexican-Americans in the US. We address two questions. First, to what extent does the prevalence of CD vary across immigrant generations in this population? Second, are differences in prevalence across immigrant generations consistent for aggressive and non-aggressive symptoms? We hypothesize that, due to differences in heritability, there would be less variation in prevalence of aggressive than non-aggressive symptoms.
Data come from surveys conducted in Mexico and the US using the same face-to-face interview, the World Mental Health version of the Composite International Diagnostic Instrument (WMH-CIDI) 27. The Mexico survey, the Mexican National Comorbidity Survey28, is a stratified, multistage area probability sample of household residents in Mexico aged 18 to 65, living in communities of at least 2500 people. 5782 respondents were interviewed between September 2001 and May 2002. The response rate was 76.6%. Data on the Mexican-origin population in the US come from two component surveys of the Collaborative Psychiatric Epidemiology Surveys (CPES)29, the National Comorbidity Survey Replication (NCSR)30 and the National Latino and Asian American Survey (NLAAS).31 The NCSR is a stratified multistage area probability sample of the English-speaking household population of the continental United States32. The NLAAS was based on the same sampling frame as the NCSR, with supplements to increase representation of the survey’s target ethnic groups, including monolingual Spanish speakers 31,33. Spanish language interviews in the NLAAS used the same translation of the diagnostic interview as the MNCS. The NCSR was conducted from 2001 through 2003 and had a 70.9% response rate; the NLAAS was conducted from 2002 through 2003 and had a 75.5% response rate for the Latino sample. A total of 1442 respondents in the CPES are of Mexican origin.
In order to reduce the impact of recall bias the sample was restricted to respondents under the age of 45. The resulting sample is comprised of 1736 MNCS respondents and 927 CPES respondents (268 from the NCSR and 659 from the NLAAS).
MNCS respondents were asked whether they had ever migrated to the US and whether they had a member of their immediate family [Miembro de la familia nuclear o inmediata] living in the US. Respondents to the CPES were asked their country of birth, the nativity (US vs. other) of their parents, and, if they were born outside of the US, the age at which they arrived in the US. Using this information, the sample was divided into four groups based on their migration history, the presence of migrants in their immediate family, and the nativity (Mexico vs. US) of their parents. Definitions of the four groups along with the sample size originating from each survey are presented in Table 1.
Comparisons between these four groups, adjusted for age and sex, were specified to test study hypotheses regarding associations of migration with CD and CD symptoms. Group 2, which represents the families of origin of Mexican migrants in the US, was specified as the reference group. Differences between Group 2 and Group 1 indicate household level migrant selection, i.e. differences between individuals in migrant vs. non-migrant households in Mexico34. Differences between Group 2, Group 3, and Group 4 indicate potential influences of migration.
The conduct disorder module of the WMH-CIDI assesses 15 CD symptoms and additional information required to establish a DSM-IV diagnosis, i.e. impairment, clustering of symptoms within a 12-month period and age at symptom onset. A clinical validity study in an adolescent sample found excellent sensitivity (96.8) and specificity (98.7) for the diagnosis of CD compared with a structured clinical interview35. Studies of the NCSR (adult) sample show strong associations of the WMH-CIDI diagnosis of CD with established demographic correlates, including male sex, low educational attainment36, urban residence, and divorce9. Moreover, NCSR respondents who met DSM-IV CD criteria are more likely than other respondents to have a broad range of other psychiatric disorders, including anxiety, mood, substance use and impulse control disorders9. It is important to note that diagnoses of CD in this study refer to a history of CD in childhood.
Information from the CD assessment was used to identify respondents who met DSM-IV criteria for CD and respondents who met the symptom count criteria (>=3 symptoms) regardless of whether they met the additional DSM-IV criteria. Previous research suggests that respondents who retrospectively report >=3 symptoms and do not meet additional DSM-IV criteria for CD are at equally high risk for adverse life events and psychiatric co-morbidity as respondents who meet full DSM-IV criteria37.
The sample design variables for the combined MNCS/CPES employed the primary stratum and primary sampling unit identification codes and the sampling weight variables developed by the sample design team at the Institute for Social Research, including the integrated CPES sampling design and weight variable developed to account for the overlapping coverage of Mexican-Americans by the NCSR and NLAAS29,38,39. Sampling stratification variables were modified to ensure that codes used for the CPES and MNCS components did not overlap40. Sampling weights reflect adjustments for unequal selection and response probabilities as well as poststratification adjustments to enhance the representativeness of weighted inferences with respect to contemporaneous national census estimates of target population sizes. An additional rescaling factor was applied to the sampling weights in the cross-national dataset so that the weighted sample sizes would reflect the relative sizes of the Mexican-origin target populations in Mexico and the US, enhancing the suitability of the weights for use in design-based analyses involving the full population as well as population subgroups39,41.
Associations between migrant generation and CD outcomes, with statistical adjustment for age and sex, were estimated using the appropriate logistic regression model, using the SUDAAN software package to correct standard errors for survey design effects42. Binary outcomes, e.g. presence of DSM-IV CD, were examined using binary logistic regression models, symptom counts were examined using ordered logistic regression models, and polytomous outcomes, e.g. CD profiles, were examined using multinomial logistic regression models. Latent class analysis (LCA) was conducted with the MPLUS software package43.
Table 2 shows the age and sex distribution of the sample. Members of group 3 are slightly younger than the other groups.
The prevalence of CD symptoms in the entire sample ranged from 0.1% (Forced Sex) to 12.8% (Truancy)(Table 3). 11.5% of the sample had 3 or more symptoms and 2% met DSM-IV criteria for CD. There is a strong pattern of higher symptom prevalence in groups 3 and 4, relative to groups 1 and 2, with differences in prevalence reaching statistical significance for 12 of the 15 symptoms. Symptom prevalence is at least twice as common among group 4 relative to group 2 for all but 3 symptoms and is frequently much larger. For instance, respondents in group 4 are more than 13 times as likely to endorse damaging property and more than 9 times as likely to endorse stealing without confrontation as respondents in group 2. The 3 symptoms where this difference in prevalence is less than 2-fold are all aggressive symptoms (cruelty to animals, stealing with confrontation, and forced sex). Forced sex is endorsed extremely rarely in this sample (n=3).
The prevalence of 3 or more symptoms and DSM-IV CD similarly increases across these groups. Respondents in group 4 are more than 5 times as likely to have 3 or more symptoms and more than 7 times as likely to have DSM-IV CD as respondents in group 2.
The pattern of increasing risk for CD is confirmed in logistic regression models for having 3 or more CD symptoms or meeting DSM-IV CD criteria, with statistical adjustment for age and sex (Table 4). In these models group 2 is specified as the reference group because this group represents the families of origin of US migrants and thus is the appropriate point of comparison for subsequent generations. Members of group 2 are more likely to have both outcomes than Mexicans from non-migrant families (group 1), although this difference does not reach statistical significance for DSM-IV CD. Compared with group 2, risk for both outcomes is higher in group 3 (OR=3.3 and 4.1) and higher still in group 4 (OR=9.7 and 7.6).
Ordered multinomial regression models were specified to examine differences in number of symptoms across the migration groups, with statistical adjustment for age and sex (Table 4). When all symptoms are considered together there is a strong relationship between migrant group and CD symptoms, with risk increasing across generations. Note that the confidence intervals for the ORs associated with groups 3 and 4 (relative to group 2) do not overlap.
In order to examine associations of migrant generation with different types of symptoms, separate ordered logistic regression models were specified for counts of aggressive and non-aggressive symptoms. Migration is significantly associated with both types of symptoms, but the association is weaker for aggressive than for non-aggressive symptoms. The ORs relating risk for aggressive symptoms in groups 3 and 4 to that in group 2 are 1.5 and 3.1 for aggressive symptoms and 3.5 and 10.5 for non-aggressive symptoms.
Respondents were then classified into 4 mutually exclusive CD profiles based on the presence of aggressive vs. non-aggressive symptoms: 1) No-CD, zero CD symptoms, 2) Aggressive, one or more aggressive symptoms and no non-aggressive symptoms, 3) Non-aggressive, one or more non-aggressive symptoms and no aggressive symptoms, and 4) Combined, one or more of both aggressive and non-aggressive symptoms. Based on this categorical scheme, the contrast between aggressive and other CD classes is even more pronounced (bottom rows of Table 4). There are strong associations between migrant group and both the non-aggressive and combined CD profiles, with increasing risk in groups 3 and 4 relative to group 2. However, there is no difference in risk for the aggressive profile in groups 3 or 4 relative to group 2. Note that here too, group 2 is at higher risk for all three CD classes relative to group 1, significantly so for the aggressive and combined classes.
In order to identify sub-types of CD based on empirically derived symptom configurations we estimated latent class models. The best fitting model according to model fit indices included 4 latent classes. Model fit statistics for latent class models with 2 through 5 classes are presented in supplemental table 1. The entropy of the 4 class model was 0.898, above the threshold of 0.8 recommended for modeling associations between latent class assignments and auxiliary covariates44.
Table 5 shows the prevalence of each CD symptom in each of the 4 classes identified by LCA. The first class is characterized by low prevalence of all 15 symptoms and is labeled ‘no CD’. The second class is characterized by high prevalence of rule-breaking and theft/deceit symptoms and low prevalence of most aggressive symptoms. This class has been labeled rule-breaking/deceit. The third class is characterized by high prevalence of aggressive and low prevalence on all other symptoms and is labeled ‘aggressive’. The fourth class is characterized by high prevalence on all symptoms and is labeled ‘severe’. Respondents in the severe class are more severe not only in having higher prevalence of symptoms characteristic of the rule breaking/deceit and aggressive classes, but also in having high prevalence of destructive symptoms (fire setting, damage to property) and breaking and entering which are uncommon in the other classes.
No respondents in the No CD class had >=3 symptoms. In the two intermediate classes, rule breaking/deceit and aggression, the prevalence of >=3 symptoms was high, 53.3% and 40.8% respectively, while the prevalence of DSM-IV CD was 7.9% and 5.5% respectively. All of the respondents in the severe class endorsed >=3 symptoms and 40.4% met criteria for DSM-IV CD.
The prevalence of the No-CD class decreases sharply across the migrant groups, from 86.7% in group 2 to 38.6% in group 4 (Bottom rows of Table 5). The increase is not evenly distributed across the three remaining classes. The prevalence of the rule-breaking/deceit and of severe classes increases dramatically across groups, while the prevalence of the aggressive class increases only slightly, from 2.4% in group 2 to 3.3% in group 4.
This pattern of association between migrant group and latent class membership is sustained with adjustment for age and sex (Table 6). The strength of this association is modest for the aggressive class: compared to group 2 (families of origin of migrants), risk for being in the aggressive class is not significantly elevated in group 3, but is elevated in group 4 (OR=3.55, 95% CI 1.28–9.88). Associations between migration and the other two classes are much stronger. Compared with group 2, the ORs associated with being in the rule breaking/deceit or severe classes are 3.55 and 6.08 in group 3 and 10.7 and 29.4 in group 4. Differences between group 3 and group 4 are statistically significant.
This study has several unique strengths. Data were collected in both sending and host countries, using the same survey instrument, and information was available to identify sub-samples representing families of migrants in Mexico and successive generations of Mexican-Americans in the US. The ability to account for migrant selection bias34,45 by identifying families of migrants in Mexico (group 2) turns out to be particularly important because respondents from these families were more likely to report having had CD symptoms than respondents in families without migrants (group 1). Had we not made this distinction and simply compared Mexican-Americans in the US with the general Mexican population, the increase in risk for CD associated with migration to the US would have appeared larger. Selection effects are minimized though not eliminated by including migrants who migrated at age 16 or higher and return migrants in the Mexico group. In addition, differences between children of migrants raised in the US (group 3) and children of US-born parents (group 4) suggest that risk continues to increase within the US population, independent of possible migrant selection effects.
The results suggest that there is a large difference in risk for CD between Mexicans in Mexico and people of Mexican descent in the US. While only 2% of people in families of migrants (group 2) met DSM-IV criteria for CD, 11.5% of US-born Mexican-Americans with at least one US-born parent (group 4) met these criteria. The prevalence of CD in group 4 is close to that in the Non-Mexican-American, US-born sample of the CPES of 10.6%. Adjusting for age and sex, the odds of CD are 7.64 times as high in group 4 as in group 2. This is a dramatic increase in risk for one of the most serious childhood psychiatric disorders. The fact that this increase in risk occurs across generations within a migrating population strongly suggests the influence of early childhood environmental conditions in the US. The finding highlights the importance of identifying environmental factors that are potentially modifiable and may offer opportunities to reduce the prevalence of CD in the US.
The increase in risk for CD in the US is particularly concerning because of the strong relationship of CD in childhood with a wide range of adverse life events in adolescence and adulthood as well as most forms of adult psychopathology 9,46. In fact, CD may be an indicator of influences that affect not only childhood psychiatric disorders, but may also contribute to differences in risk for later mood, anxiety and substance use disorders between immigrants and US-born that have been observed in studies of Mexican-Americans and some other immigrant groups in the US 24,47–50.
Convergent evidence from analysis of symptom counts and symptom profiles suggests that differences associated with migration are weaker for aggressive symptoms than for non-aggressive symptoms. In the analysis of symptom counts, aggressive and non-aggressive symptoms are both significantly higher in both groups 3 and 4 relative to group 2, but the association is much stronger for non-aggressive symptoms than for aggressive symptoms. A limitation of the analysis of symptom counts is that no distinction is made between ‘pure’ and ‘mixed’ symptom profiles. When classes of CD were defined based on the a priori distinction between aggressive and non-aggressive symptoms, there was a strong association of migration with non-aggressive and combined types of CD profile and no association of migration with the aggression only profile. Similarly, analysis of the empirically derived latent classes found much stronger associations with migration for the rule-breaking/deceit and severe latent classes than for the aggressive latent class.
These findings are consistent with hypothesized differences in the extent of genetic influence on different types of CD symptoms. If, as twin studies suggest19, aggressive symptoms are more strongly influenced by genetic factors than non-aggressive symptoms, then we would expect to find the pattern of results observed in this study, that is, a much smaller intergenerational change in aggressive than in non-aggressive symptoms. These data do not allow us to quantify the heritability of either type of symptom, and the results should not be interpreted as evidence that aggressivity is predominantly genetic in origin. Rather, this study provides confirmation of the findings from twin studies regarding the relative contribution of environmental and genetic factors to different types of symptoms using a different methodology. In contrast with twin studies, inferences based on this study do not depend on the equal environments assumption. In addition, the range of variation in environment is greater in this study than in twin studies which are generally limited to a single country or region.
The difference in association with migration between aggressive and non-aggressive symptoms was greater in the analysis of CD profiles than in the analysis of symptom counts. One potential explanation for this pattern is that aggressive behaviors that occur along with other CD symptoms are more likely to be situational, i.e. contextually normative or premeditated51,52, and therefore more influenced by environmental factors than aggressive behaviors that occur in isolation from other CD symptoms. For instance, adolescents who are more likely to break rules may, through that rule-breaking, get into situations or relationships in which aggressive behaviors are more common, such as drug trafficking or gang activity. There is some empirical support for a distinction between aggressive and non-aggressive CD53,54, with the former uniquely associated with emotional dysregulation, that may help explain these differences.
It is important to note that evidence of environmental influences on risk for CD does not rule out the possibility that there are inherited liabilities that are triggered or suppressed by environmental conditions 55–57. Twin studies of substance use58, smoking59 and externalizing behaviors60 have found that in settings where the prevalence of these outcomes is higher, their heritability is greater. For instance, there might be an inherited liability to CD common to Mexico and the US that is more frequently expressed in the US due to environmental influences. This pattern would suggest that CD among Mexican-Americans in the US would be concentrated among offspring of migrants with liability, expressed or not, to CD. Alternately, there could be distinct inherited liabilities to CD in the two environments. This pattern would suggest that CD would be heritable within both populations but not heritable between migrants and their offspring. Family studies with follow-up of migrants on both sides of the border are needed to examine these alternative models.
While the dramatic increase in CD symptoms across generations within this migrant population implicates environmental influences, specification of the particular environmental factors remains elusive. The environmental changes experienced by migrating populations are comprehensive, involving virtually all aspects of social institutions, identity, family relationships, language, and diet. Generations of researchers have focused on the distinctive social circumstances of immigrants. In the 1920s, Faris and Dunham observed a configuration of social conditions among immigrant groups living in urban areas of Chicago that they suggested lead to increasing risk for juvenile delinquency as well as substance use and mental disorder.61(p.8–9). In their view, immigrant parents were marginalized from mainstream culture, while their children were caught between their culture of origin, to which they often felt little connection, and their host culture, which was likely to reject them. Similar explanations in terms of differential acculturation62 and discrimination 63,64 have been proposed for observations of increased risk for mental disorders in children of immigrants relative to their parents in recent studies.
Explanation of the increase in risk for psychiatric disorders among descendants of migrants in terms of experiences distinctive of migrant communities would suggest that risk is particularly high in these communities relative to more established populations that do not share these circumstances. However, epidemiological evidence does not support this prediction. Differences among ethnic groups in the US are much smaller in magnitude than differences between immigrants and the US-born, and members of ethnic minorities tend to have lower rather than higher risk for most psychiatric disorders65,66. An alternative explanation for the observed changes across immigrant generations is that immigrants in the US share the same cultural and social conditions that account for high risk for psychiatric disorders in the US relative to immigrants’ countries of origin67. This explanation implies that differences observed across migrant generations are due to influences on the US population as a whole that contribute to large cross-national differences in disorder prevalence.
The association between migration and CD symptoms in this sample should not be generalized across migrant groups. Bird and colleagues did not find differences in disruptive behaviors between samples of Puerto Rican children on the island and in the South Bronx26. However, in that study parental acculturation was related to disruptive behavior in both samples68. Other studies have found that the association between US nativity and psychopathology differs between the Mexican and Puerto Rican populations24,47.
Results of this study should be interpreted in light of its reliance on retrospective reporting of CD symptoms and other DSM-IV criteria by adults, a method known to result in underestimates of prevalence due to failures of recall 69,70. In order to minimize the impact of recall errors, the sample was limited to respondents under the age of 45, but this restriction of the sample is unlikely to fully correct for recall errors. Recall errors bias estimates of the prevalence of CD downwards, but would also affect estimates of the relative risk of CD if recall accuracy is associated with migration. Tests for statistical interactions between age and migrant group were not significant in any of the models.
A second potential limitation arises from the reliance on contemporaneous samples of generational groups rather than the samples drawn from actual multigenerational families. This is a common method in migration studies, but the assumption that these groups represent biologically connected successive generations should be noted. The validity of this assumption is supported by the consistency in the finding that Mexican-born immigrants to the US have lower risk for psychiatric disorders than the US-born Mexican-American population from the early 1980s71 through the early 2000’s72.
International migration across societies presents a valuable opportunity for biobehavioral research to advance understanding of the interplay between environments and gene expression in the etiology of CD and other complex psychiatric disorders. Though migration studies lack the mathematical formalism that underlies the quantitative estimation of trait heritability in twin studies, they provide valuable insight into the role of environmental factors in the etiology of disease73. In this study, comparisons across representative samples of successive generations within a migrating population suggest that the environmental influence on CD is quite large, but restricted to certain sub-types of the disorder. Future studies may be able to identify the specific genetic and environmental factors involved in this complex epidemiological shift in psychiatric morbidity, particularly if they include samples of migrant families with members in both the sending and receiving communities.
This research was supported by grants from the National Institute of Mental Health ((R01 MH082023 (PI: Breslau), K24-MH072756 (PI: Kravitz)), the UC Davis Clinical and Translational Science Center (NIH, UL1 RR024146), and the University of California Migration and Health Research Center. The lead author had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Conflict of Interest
Funding agencies had no role in the design and conduct of the study, collection, management, analysis, or interpretation of the data, or in the preparation, review, or approval of the manuscript. The authors have no conflicts of interest to disclose. The corresponding author (Breslau) takes responsibility for the integrity of the data and the accuracy of the data analysis, and that all authors had full access to all the data in the study.