| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
ORIGINAL ARTICLES |
From the Division of Epidemiology and Clinical Applications (G.W., S.S.K., M.W.H.), National Heart, Lung, and Blood Institute, Bethesda, MD; Division of Biostatistics (T.R., M.A.P., D.C.R.) and Departments of Psychiatry and Genetics (D.C.R.), Washington University School of Medicine, St. Louis, MO; and Boston University School of Medicine (R.C.E.), Boston, MA.
Address reprint requests to: Gerdi Weidner, PhD, Department of Psychology, State University of New York at Stony Brook, Stony Brook, NY 11794-2500. Email: gweidner{at}sunysb.edu
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
METHODS: Analyses were based on 680 European-American families (2525 individuals) from the NHLBI Family Heart Study (FHS), a population-based study of genetic and nongenetic determinants of CHD, atherosclerosis, and cardiovascular risk factors. The influence of family relationships, age, and education on the variation in each of the four hostility scores were estimated.
RESULTS: Significant familial resemblance in all hostility scores was found, accounting for 42% of the variance in total hostility, 30% in cynicism, 38% in aggressive responding, and 18% in hostile affect. Very little of this resemblance could be explained by similarities in education. Familial resemblance for cynicism was solely due to significant parent-offspring and sibling correlations (ie, no spouse resemblance), suggesting the possibility of genetic influences. Gender and generation differences were also evident in the familial correlations.
CONCLUSIONS: Hostility aggregates in families. Both family environmental and genetic sources of resemblance are suggested for hostility.
Key Words: hostility • education • gender • familial aggregation
Abbreviations: CHD = coronary heart disease; FHS = family heart study; NHLBI = National Heart, Lung, and Blood Institute.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
In contrast, little is currently known about these influences on coronary-prone personality attributes, such as anger and hostility. Recent research has paid considerable attention to hostility and its link to CHD (see Ref. 9 for review). In several prospective studies, hostility as measured by the Cook-Medley Hostility Scale (Ho) was linked to CHD mortality (10–12), severity of atherosclerosis (13), and potentially pathogenic physiological reactions to stress (14–17).
Several studies of adult twins suggest modest heritability for some measures of hostility (primarily cynical mistrust/paranoia) for both male and female twin pairs (18–21). The most powerful twin design, comparing identical and fraternal male and female twins raised together and apart, supports the notion that both genetic and shared rearing environments may be important determinants of twin similarity in hostility, explaining 21% and 20% of the variance, respectively (19). In these studies, no significant gender differences were noted.
The study of familial aggregation of hostility may shed additional light on the origins of this variable. That is, the pattern of significant familial correlations suggests whether a trait is caused in part by familial factors. For example, genetic influences may be inferred when parent-offspring and sibling correlations are significant but spouse correlations are not because spouses (generally) do not share common genes, whereas siblings and parent-offspring pairs have half of their genes in common. Both genetic and familial environmental factors are inferred if the sibling, parent-offspring, and spouse correlations are all significant. Generational influences may be hypothesized if the correlations among siblings are greater than those between parents and offspring (cf. Ref. 22).
Only one family study examined hostility (ie, cynicism/paranoia) using nontwins (23). No familial aggregation was evident. However, spouses resembled each other in hostility. Similarities in hostility between spouses could be due to either assortive mating (eg, marrying those with similar attitudes) and/or shared environmental influences (eg, marital conflict). There was no resemblance in hostility among siblings, but a slight resemblance between fathers’ and daughters’ cynicism scores and mothers’ and daughters’ paranoia scores. The absence of significant family resemblance in hostility may have been due to the rather young age of the children in that study (range = 6–18 years). It is conceivable that familial aggregation of hostility is evident only in families with older offspring who are more stable in their personalities (24, 25). Additionally, the same personality scales can be administered to families with older offspring, minimizing error variation associated with the use of different (ie, developmentally appropriate) personality scales (22). The purpose of the present study was to examine familial aggregation of hostility among adult family members of the NHLBI Family Heart Study. The NHLBI Family Heart Study was designed to identify and evaluate genetic and nongenetic determinants of CHD in individuals and families recruited from existing population-based studies (26). In addition to detailed assessments of biomedical variables on a random sample of families as well as a high-risk sample of families chosen because of increased rates of CHD within the family, psychosocial measures of hostility (11, 27, 28) and social support (28, 29), work stress (30), and socioeconomic status (years of education) were also assessed. The focus of this paper is on hostility in the random sample because this sample is more representative of the population than the high risk sample. To the extent that genetic factors and/or shared household environments play a role in the expression of hostility, it was expected that hostility scores among related family members would be more similar than those of nonrelatives. Because of the comparatively large sample size, it was also possible to examine gender- and generation-specific associations.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Participants in the present study were selected from the random group of European-Americans, which consisted of 1441 randomly selected probands from parent studies and their relatives (a total of 9221 individuals). A complete and detailed description of participant selection procedures has been published previously (26). The following section briefly summarizes this information. In phase I of the FHS, the "family history component," probands were given standardized questionnaires to ascertain family histories of CHD, other types of heart disease, and related conditions. All participants (probands and relatives) were subsequently asked to provide medical histories for themselves. Reported CHD events were validated by reviewing medical records and death certificates. Individuals identified as high risk-probands (based on computed risk scores using risk equations from the Framingham Heart Study) were further examined along with their family members (eg, spouses, siblings, children, and parents) during phase II.
In phase II of the study, the "clinical examination and follow-up component," an equal number of families was selected randomly from the parent cohorts as a comparison group. This random sample of families (representing the entire range of family risk of coronary heart disease) was also offered clinical examinations (including psychosocial questionnaires) and laboratory tests. This study is based on data from the random sample collected in this phase. There were 680 European-American families (2525 individuals), consisting of the probands, their parents, siblings, spouses, and children. Table 1 shows the sample characteristics by pedigree structure (gender, age, education, and hostility levels).
|
Hostility.
All psychosocial questionnaires, including the 27-item version of the Cook-Medley Hostility Scale were self-administered. Hostility was assessed by three subsets of items from the Cook-Medley Hostility Scale (11, 27, 28), labeled cynicism (13 items), hostile affect (5 items), and aggressive responding (9 items). The three subsets are not independent; rather, they reflect three components of the same construct: cognitions or beliefs about the world (cynicism), emotions associated with social relationships (hostile affect), and behavioral tendency to use anger and aggression as a response (aggressive responding). Construct validity of this 27-item version scale has been provided by Barefoot et al. (11) and Brummett et al. (31), who demonstrated associations of self-reported hostility to behavioral indicators of hostile emotions (facial expressions during social interaction (31)). The 27-item scale was chosen rather than the 50-item scale because of its documented validity in predicting mortality in a male sample (11) and myocardial ischemia in women (but not men) (17) and because of the need to reduce participant burden in such a large epidemiologic study. Representative items from the cynicism subset include: "I think most people would lie to get ahead"; "No one cares much what happens to you"; and "I have often met people who were supposed to be experts who were no better than I." Sample items of the hostile affect subset are "It makes me impatient to have people ask my advice or otherwise interrupt me when I am working on something important"; "Some of my family have habits that bother and annoy me very much"; and "I am easily angered." Aggressive responding is measured by items such as "When someone does me a wrong I feel I should pay him back if I can, just for the principle of the thing"; "I do not try to cover up my poor opinion or pity of a person so that he won’t know how I feel"; and "I strongly defend my own opinion as a rule." Each of the 27 items is scored 1 (true) or 0 (false). The average for each of the three subsets was computed and analyzed separately (see Ref. 28 for scoring details). In addition, the sum of the three subsets was used as an indicator of overall hostility level. The rationale for investigating all three subsets as well as the overall hostility score in the NHLBI Family Heart Study is that their relationship to coronary risk may be different in women than in men (11, 27).
Data Adjustments
Two analysis variables were constructed for each of the four hostility measures: one adjusted for age only and the other adjusted for both age and education. These adjustments were carried out separately by gender and involved corrections in both the mean and variance. The general procedure for each of the variables utilized a stepwise multiple regression procedure. Only terms that were significant at the 5% level were retained. In summary, a given measure was regressed on a polynomial in age (age, age2, and age4) and education. The residual from this mean regression was retained. The squared residual was then regressed on another polynomial in age, and the predicted score from this second regression (heteroscedasticity) was retained. The final phenotype used in the familial analysis was computed as the residual from the mean regression divided by the square root of the predicted score from the second (variance) regression. A final standardization step was taken to ensure a mean of zero and a standard deviation of one.
Familial Correlation Model
A gender-specific familial correlation model was used to investigate the familial resemblance for three subsets of Cook-Medley hostility items and the sum of these subsets (cynicism, aggressive responding, hostile affect, and total hostility).
Nuclear families consist of fathers (F), mothers (M), sons (S), and daughters (D), which leads to eight possible gender-specific correlations among family members (one spouse (FM), four parent-offspring (FS, FD, MS, and MD), and three sibling (SS, DD, and SD)). Eight hypotheses were tested concerning the significance of the correlation coefficients and whether there were gender and/or generation differences in the correlations. These hypotheses are outlined in Table 2.
|
In hypotheses 6 through 8 the significance of the correlations was tested (ie, whether they could be fixed at zero). In hypothesis 6, no spouse resemblance was allowed (ie, FM = 0; df = 1). In hypothesis 7 there was no parent-offspring resemblance (ie, FS = FD = MS = MD = 0; df = 4), and in hypothesis 8, no resemblance among siblings was allowed (ie, SS = SD = SD = 0; df = 3). The most parsimonious model was derived in hypothesis 9 by combining nonrejected hypotheses into a single test. The p values resulting from the tests of each of these null hypotheses are given in Table 2 (age-education adjusted phenotypes).
A maximum likelihood method of analysis provided efficient estimates of the correlations and the means and variances that best describe the covariance structure by fitting the correlations directly to the observed family data. Means and variances for both offspring groups (sons and daughters) were estimated separately, and means and variances for the parents were fixed at their sample values.
The general purpose computer program SEGPATH (32) was used, and the model specifications corresponding to the gender-specific familial correlation model used here are presented elsewhere (33). Generalized heritability estimates (ie, the percentage of variance due to familial factors) were derived for each of the hostility measures. This estimate includes both polygenic and shared environmental effects. Also, because education is known to affect hostility levels (34), analyses were done twice: before and after adjusting for education.
The goal of model-fitting analyses was to reduce the number of estimated parameters to only those few that explained the variability in the data without a significant loss of fit (the most parsimonious model). Thus, we formulated null hypotheses (eg, whether the spouse correlation is nonsignificant, ie, FM = 0) and compared the log-likelihood of this reduced model to that obtained from the general model where all the correlations, including FM, were estimated. The test statistic was the likelihood ratio criterion, which was (minus twice) the difference between the values of the log-likelihood obtained under the general model and a reduced model. The difference in the log-likelihoods was asymptotically distributed as a 
2 with the degrees of freedom being the difference in the number of parameters estimated in the two nested models. A significant 2 (p < .05) suggested that the model reduction was not acceptable, whereas a nonsignificant test suggested that the reduction was acceptable.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Intraindividual Correlations: Hostility Scores, Age, and Education
Intraindividual correlations for the raw measures, separately by family structure and gender, were computed. These correlations were derived using the nuclearized data. Results (not shown) indicated that the four hostility scores seemed to be measuring similar aspects of hostility (eg, they shared much common variance). All correlation coefficients were statistically significant, ranging between 0.35 and 0.78. Correlation coefficients between hostility scores and age were either nonsignificant or of a small magnitude (ranging from -0.04 to 0.20). Years of education were related significantly (negatively) and consistently to hostility scores (ranging from -0.13 to -0.32).
Effects of Age and Education Adjustment on Hostility Scores
In general, adjustment for age had a small effect on hostility scores, explaining less than 1.4% of the variance in hostility scores. Specifically, for the total hostility score, no age effects were significant in men, but in women the age2 and age4 terms accounted for 1.3% of the variance. For the cynicism subscale, age and age4 terms accounted for 1.2% in the men, and age4 accounted for 1.4% in the women. No age effects were noted in either gender for the aggressive responding or the hostile affect subscales. Education accounted for more variance in the traits than did age, especially among men. Among men, education terms accounted for 6.6% of the variance in total hostility, 9.4% in cynicism, 3.8% in aggressive responding, and 1.3% in hostile affect. The corresponding values for women were 3.0%, 5.1%, 1.2%, and 0.7%. In every case, the signs of the education regression coefficients were negative, indicating that higher education was associated with reduced hostility scores.
Familial Correlations
The results for testing each of the null hypotheses are outlined in Table 2. For total hostility, because there were significant gender differences in the offspring (model 1; p = .007), the following two gender and generation tests (models 2 and 3) were not reported because they depended on the sibling test being nonsignificant. The test for gender-specific resemblance with no generation differences (ie, model 4; male-male, female-female, and opposite gender correlations) was significant (p = .006). The test for same- vs. opposite-gender resemblance with no generation differences (ie, male-male = female-female, opposite gender correlations in model 5) was not performed because the prior one was significant. For the significance tests, each of the spouse (model 6), parent-offspring (model 7), and sibling (model 8) correlations were significantly different from zero (p < 0.006).
The same pattern of results was found for the cynicism scores, except that the spouse correlation (model 6) was not significant (p = .165). For the aggressive responding scores, although there were gender differences (model 1, p = .034), these differences were attributable to same- vs. opposite-gender patterns (models 4 and 5). The spouse, parent-offspring, and sibling correlations were significant (p < .001). Finally, for hostile affect, there were no gender or generation differences of any sort, and the spouse, parent-offspring, and sibling correlations were all significant.
The most parsimonious model for the total hostility score was the general one because all of the gender and significance tests were significant. For the cynicism scale, the most parsimonious model was for no spouse correlations (model 6). For the aggressive responding scale, the most parsimonious model was for same- vs. opposite-gender correlations (model 5). For the hostile affect scale, the most parsimonious model was for no gender or generation differences (model 3). The corresponding model-fitting results for the age adjusted phenotypes (not shown) revealed a similar pattern of results.
The maximum likelihood estimates under the most parsimonious model is given in Table 3. Familial resemblance in hostility and cynicism was greatest for sons and their parents, as well as among sisters.
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Before discussing our findings on familial resemblance, it should be noted that three of the hostility scores (cynicism, hostile affect, and aggressive responding) are subsets of items from the total hostility scale used in the present study. They are highly correlated with the total hostility score and interrelate highly. Thus, cynicism, aggressive responding, and hostile affect should be considered descriptive labels rather than distinct subscales (11).
Resemblance Between Spouses
Spousal resemblance in hostility was evident for the total hostility, the hostile affect, and the aggressive responding score. Because husband and wife are not genetically related, their similarity in these aspects of hostility may be due to selective mating or the effects of cohabitation. One of the factors involved in selective mating, for instance, could be similarity in educational level, which is negatively related to hostility (34). When we adjusted for years of education, spousal resemblance in these three subscale scores remained significant, but was reduced. The nonsignificant resemblance for cynicism among spouses suggests that there is not much evidence for the effects of shared family environment on this trait (cf. Ref. 22).
Resemblance Among Related Family Members
Maximal heritability estimates (ie, percentage of variance due to familial factors adjusted for spouse resemblance) were similar for the total hostility scale and the subsets of items measuring aggressive responding (42% for total hostility and 38% for aggressive responding). The familial effects for cynicism and hostile affect were somewhat lower (30% and 18%, respectively).
These findings indicate that a considerable amount of variance in hostility scores among related family members can be accounted for by shared family effects. Because spouses also resembled each other in three of the four aspects of hostility (ie, total hostility, aggressive responding, and hostile affect), we can conclude that at least some of the familial resemblance may be due to familial environmental causes, such as cohabitation. Very little of this environmental effect is explained by education, as evidenced by the pattern of similar heritability estimates after adjustment for education.
Additional evidence for the influence of other shared environmental factors not measured by us can be seen in the pattern of results obtained for aggressive responding: correlation coefficients for aggressive responding scores between spouses and between other family members were of similar magnitude, which suggests that familial aggregation of aggressive responding may be largely influenced by shared environmental factors that are unrelated to educational level.
In contrast, the pattern of familial resemblance in cynicism (ie, parent-offspring and sibling correlations, but no spouse correlation) suggests the possibility of primarily genetic influences. It is interesting to note that adjusting the cynicism scale for education tended to reduce familiality (although not statistically significantly), suggesting that there may also be some kind of pleiotropy (ie, the same genes and/or family environments influence both education and cynicism), and removal of education also removed some of the familial effects on cynicism. However, the fact that the familial effects were not eliminated after adjusting for education suggests that there are additional familial factors influencing cynicism levels that are not associated with education.
In summary, our findings suggest that cynicism holds the strongest promise for a genetic component. This interpretation is consistent with results from twin studies that have suggested heritable variance in cynicism (18–21). For example, the heritability estimate in the NHLBI Twin Study (male participants) was 28% for both the full Ho scale and the cynicism subscale (18). Smith et al. (21) reported heritability estimates ranging from 56% to 64% (unadjusted for education) for subsets of items measuring cynical hostility in a sample of male twins. Rose’s (20) findings suggest heritability for both male (38%) and female twins (33%) for cynical hostility. Corresponding estimates in the Pederson et al. (19) study were 21%.
Gender and Generation Effects
Both gender and generation differences were evident in the familial correlations for total hostility and cynicism. A pattern of same- vs. opposite-gender effects suggests a gender-limited trait. Additional generational differences suggest that there may be developmental factors involved, which may be staged differently in males and females. Generational differences alone suggest that the trait is primarily influenced by cohort effects among individuals who are more similar in age. For example, cohort effects are most likely to account for attitudinal similarity among adolescent peers.
In general, familial resemblance in hostility and cynicism was greatest for sons and their parents, and among sisters. The greater resemblance among parent-son pairs than among parent-daughter pairs is consistent with the observation that sons’ coronary-prone personality attributes may be more susceptible to parental influences (eg, modeling) than daughters’ (see Ref. 35 for father-son resemblance in Type A behavior).
Among siblings, greater similarity in cynicism was found among sisters than among brothers or opposite-gender siblings. This finding supports the results from a twin study that included twins of both genders and reported a correlation of 0.287 between dizygotic twin brothers, and a correlation of 0.383 for dizygotic twin sisters (20). One reason for the greater similarity among sisters may be that they engage in more frequent social interactions with each other than brothers (especially in adulthood), which may increase their similarity in personality. Some support for this interpretation comes from a study by Rose et al. (36), in which greater frequency of social contact was observed among adult female twins than adult male twins (both for monozygotic (MZ) and dizygotic (DZ) twins). This study also found that increased social interaction accounted for some of the variability in personality traits: significant variance in intrapair resemblance for neuroticism scores was contributed by social contact frequency among both male and female twins. That is, there was greater similarity among female MZ twins with daily contact than female MZ twins with rare contact; the same pattern was evident for their MZ male counterparts; similarly, female DZ twins with daily contact were more similar than female DZ twins with rare contact. Unfortunately, we do not have information on the frequency of social contact between family members in our study. The gender-specific associations of hostility scores in the FHS, however, suggest that the effects of both shared genes and common experience on hostility are modulated by gender.
In conclusion, these simple analyses using familial correlations have been useful in several respects. First and foremost, we demonstrated that there are familial factors underlying the hostility scales. Second, the variability in some of these scales may be influenced by genetic factors. Third, gender may impact on the familial aggregation of these traits. Together, these findings suggest that follow-up analyses using more complex models are warranted to characterize more precisely the source of the phenotypic variation in this trait. These models should be designed to explore whether there are a) separate factors (ie, major gene, polygenic, and familial environmental), and b) whether these factors are related to those that lead to increased risk for cardiovascular disease. Finally, it would be of interest to see the results of multivariate model fitting examining the overlap of familial influences on hostility and those on actual CHD or some biological marker of CHD risk.
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
Participating institutions and principal staff are as follows: Gerardo Heiss, Jeannette Bensen, Gregory Burke, Beth Newman, Catherine Paton, Delilah Posey, and H.A. Tyroler, Field Center, Forsyth County, University of North Carolina, Bowman Gray School of Medicine; Michael Sprafka, Aaron Folsom, Robert Jeffery, and Gail Murton, Field Center, Minneapolis, MN, University of Minnesota; R. Curtis Ellison, Richard H. Myers, Andrew G. Bostom, and Greta Lee Splansky, Field Center, Framingham, MA, Boston University; Roger Williams, Paul Hopkins, and Jan Skuppin, Field Center, Salt Lake City, University of Utah; D. C. Rao, Michael Province, Ingrid Borecki, Jeanne Jory, Philip Miller, Derek Morgan, Treva Rice, Kenneth Schlechtman, and Avril Adelman, Coordinating Center, Washington University; Millicent Higgins, Jacob Keller, Sarah Knox, Phyllis Sholinsky, and Lorraine Silsbee, National Heart, Lung, and Blood Institute, Project Office; Lisa O’Neill, National Heart, Lung, and Blood Institute, Contract Office. We thank John Barefoot for his helpful comments to this manuscript.
Received for publication April 5, 1999.
Revision received September 14, 1999.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONACKNOWLEDGMENTSREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  K. Rikkonen, A.-K. Pesonen, K. Heinonen, J. Lahti, E. Kajantie, T. Forsen, C. Osmond, D. J.P. Barker, and J. G. Eriksson Infant Growth and Hostility in Adult Life Psychosom Med, April 1, 2008; 70(3): 306 - 313. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 L. Pulkki, M. Kivimaki, M. Elovainio, J. Viikari, and L. Keltikangas-Jarvinen Contribution of Socioeconomic Status to the Association between Hostility and Cardiovascular Risk Behaviors: A Prospective Cohort Study Am. J. Epidemiol., October 15, 2003; 158(8): 736 - 742. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
