Sex Influences on Shared Risk Factors for Bulimia Nervosa and OtherPsychiatric Disorders

Many studies show that bulimia nervosa (BN) co-occurs with a range ofpsychiatric disorders, including major depressive disorder, anxiety disorders,and psychoactive substance dependence.¹ Additionally,the presence of BN is associated with a wide range of personality styles includingincreased neuroticism,² higher levels of noveltyseeking,^{3 - 4} and lower premorbidself-esteem than psychiatric control subjects.⁵ Avariety of community-based studies suggest that at least some of this observedcomorbidity is not spurious. In a case-control study of women with BN, womenwith psychiatric disorders, and healthy control subjects, rates of exposureto most of the putative risk factors differed in the women with BN and thecontrol subjects.⁵ However, far fewer differenceswere evident between women with BN and women with other psychiatric disorders,consistent with the hypothesis that BN is the result of exposure to nonspecificrisk factors for psychiatric disorder.

Women with BN in the general population are significantly more likelyto be experiencing an affective disorder than women in the general population.⁶ A community study found significantly more women withBN or partial-syndrome BN to have an anxiety disorder,⁷ withsocial phobia and simple phobia more likely to be elevated than other typesof anxiety disorders. Findings from 4 large community-based studies examiningsubstance abuse problems^{8 - 11} suggestthat BN and alcohol-related disorders do co-occur at a rate greater than chance,with the presence of BN being associated with an approximately 3-fold increasein the risk of alcoholism.⁸ Furthermore, individualswith BN have been found to be more likely to have used illicit drugs thaneither non–eating disordered or psychiatric control subjects.¹¹ The correlational nature of these studies does notpermit inference of causality.

The question of whether BN shares a common causal factor with otherpsychiatric disorders and personality styles has therefore been a focus ofresearch in recent years. Family studies provide a design that can addressthis research question. Some family studies have methodological deficiencies,such as the use of unstructured or semistructured diagnostic procedures ofunknown reliability and validity, lack of interviewer blinding, limited statisticalpower, and reliance on proband rather than relative report in family studies,¹² all of which compromise the interpretability of thestudies. However, when the findings of well-designed studies are consideredtogether, some conclusions about shared risk factors between BN and otherpsychopathology can be made.

There is consistent evidence across a range of family studies to suggestthat there are no shared risk factors between alcohol dependence and BN.^{13 - 17} Onetwin study¹³ investigated the interplay ofgenetic and environmental causal factors simultaneously for 6 psychiatricdisorders in women, including broadly defined bulimia, phobia, panic disorder,major depression, generalized anxiety disorder (GAD), and alcoholism and foundthe genetic and environmental risk factors for the psychiatric disorders andalcoholism were quite distinct. This study, however, did find evidence forshared genetic vulnerability between phobia, panic disorder, and bulimia.A further study has supported the presence of shared vulnerability betweenBN and panic disorder.¹⁸ This same study supportedthe hypothesis that both GAD and major depressive disorder share a commoncause with BN,¹⁸ but another study found thatthe mode of familial transmission for GAD was unclear and failed to supporta shared vulnerability with panic disorder.¹⁴ Interestingly,cluster B personality disorders have been found to be elevated among probandswith BN, suggesting that affective instability and impulsivity are importantin the development of BN and substance dependence.¹⁴ ClusterB personality disorders are associated with high novelty seeking and low self-directednessin patients receiving treatment for major depression and BN.¹⁹

Previous family studies suffer from low base rates of eating disordersand while 2 studies investigated the sex of the relative of the proband asa covariate,^{14 ,18} they lackedpower to investigate any sex influences on shared risk factors. Given thelow rates of eating disorders in men, it seems reasonable to hypothesize thatthere may exist a "male equivalent" of BN or that sex influences the presenceof shared risk factors between BN and other psychopathology. For example,psychopathology that is of higher prevalence in men and that is related toimpulsivity, such as substance abuse, might share risk factors with BN inmen but not women.

Given the absence of any data on whether sex influences the presenceof shared risk factors between BN and other forms of psychopathology, thepurpose of the current study was to address this question. In order to achievethis purpose, we investigated the presence of such risk factors in 1192 opposite-sexdizygotic (DZ) pairs (male-female) and 467 same-sex DZ twins (female-female),where each twin was interviewed regarding their own psychopathology status,and used reliable diagnostic procedures and interviewer blinding. This representeda novel and powerful design where we had access to a large sample of opposite-sexDZ pairs, the twins were correlated for age and shared environment, and theprevalence of eating disorders in men in the general population was low, sowe could explore risk for other psychopathology independent of the presenceof an eating disorder. As with other family studies, shared risk factors identifiedin the study of siblings can include either shared familial factors or within-individualfactors.

The data were from a population-based longitudinal study of white twinsdrawn from the Virginia Adult Twin Study of Psychiatric and Substance UseDisorders, formed from a systematic review of all birth records in Virginiaafter 1918. There were 2 samples examined in the current study. The firstwas a subset of a group where at least 1 twin in the pair was male and bornbetween 1940 and 1975, as described previously.²⁰ Thispopulation was involved in 2 waves of interviews, the first consisting ofa telephone interview (1993-1996) where 6814 (72%) of those approached weresuccessfully interviewed. The second interview was face to face (1994-1998),with 5629 interviews being completed, 83% of the sample interviewed in thefirst wave. Of this group of male-female twins, 1183 opposite-sex DZ twinpairs participated in both waves of interviews. At the second interview, wherelifetime BN was assessed in women, psychiatric diagnoses were available for1192 twin pairs. The mean (SD) age of this group at the second interview was36.6 (8.9) years.

The second sample, using female twin pairs born between 1934 and 1971,completed 4 waves of interviews and self-report questionnaires, previouslydescribed and reported extensively in the literature.²¹ Datafor this group were taken from all 4 waves. At the third wave, when lifetimeBN was assessed in the same manner as the opposite-sex twins, 88% of the 2163women approached at wave 1 were interviewed, and this included 467 female-femaleDZ pairs. The mean (SD) age at wave 3 was 36.0 (7.6) years.

Zygosity was determined by standard questions and photographs that havemore than 95% accuracy²² and validated againsthighly polymorphic polymerase chain reaction markers.²³ Writteninformed consent was obtained prior to face-to-face interviews and verbalconsent, prior to telephone interviews.

Interviews were conducted blind to information about co-twins. Informationabout interviewer characteristics was presented earlier.^{8 ,13} Onlywomen were interviewed for lifetime BN, using questions from the Structured Clinical Interview for DSM-IV, ²⁴ withadditional questions assessing the severity of bulimic symptoms. The sameinterview questions were used with the male-female (wave 2) and female-female(wave 3) twin pairs. To maximize statistical power in the study of a low-prevalencedisorder, a broad definition of lifetime BN was adopted. This included 2 modificationsto the narrow definition of BN. First, the DSM-III-R Dcriterion was omitted because there appeared to be few meaningful differencesbetween women who binge and use associated weight-loss methods twice a weekand those who use such behaviors less than twice a week.^{7 ,25} Onaverage, women with broadly defined BN had experienced 9 binges each monthduring a 3-month period. Second, women had to express at least "a little bitmore concern than most women" about their body shape and weight rather than"a lot more" concern. The reliability of this broad definition has been examinedpreviously.²⁶

In addition to the diagnosis of BN, we examined lifetime psychopathologyand continuous measures of body mass index (current and highest) and personality.These measures and the interview waves at which they were collected are outlinedin Table 1.^{27 - 30} Lifetimepsychopathology was assessed as being present if participants met the criteriaat any of the waves of interviewing.

Given that men were not assessed for BN, more traditional model-fittingtechniques could not be used. Therefore, we used logistic or linear regressions.First, we examined the within-twin association between BN status and our measuresof psychopathology in all DZ women. Continuous measures were standardizedso that the resulting odds ratios (ORs) and 95% confidence intervals (CIs)indicated the change in risk for the dependent measures for every standarddeviation change in the independent variable.

To examine whether risk for BN in a sister influences risk of psychopathologyin her brother, the data of the male and female co-twins were paired withthe brother always appearing as twin 1 and the sister as twin 2. In each case,the psychopathology under investigation was included as the dependent variablewith the corresponding variable for the sister used as a covariate to ensurethat associations between BN and psychopathology were not due to comorbidpsychopathology in the sister. The BN status of the sister was then enteredas the independent variable. Linear regression was used for the continuousvariables (reporting βs, standard errors, and t statistics),and logistic regression was used for lifetime psychopathology (reporting ORsand 95% CIs, along with the χ² statistic). For comparison purposes,the same analyses were carried out for the female-female pairs, with all womenwith BN positioned to appear as twin 2. For greater conceptual clarity, the1 female-female DZ pair concordant for BN was removed from the analyses.

Finally, female-female DZ twins were added to the male-female sample.To provide a comparison, the regressions were run first as described earlierand the statistics were reported. The regressions were then run with the additionalindependent variables of the sex of twin 1 and an interaction between sexand twin 2 BN status. Despite clear directional hypotheses, we used an αlevel of .05 for all the analyses to reduce our chances of a type I error.

Strict criteria for BN were met in 26 (2.18%) women from the male-femalepairs. A further 49 women met the broad definition of lifetime BN. Therefore,a total of 75 (6.3%) women met criteria for broadly defined lifetime BN. Thiscompares with a 5.1% lifetime prevalence of broadly defined BN in women fromthe female-female pairs. There was no significant difference between the prevalenceof broadly defined BN between the male-female pairs and the female-femalepairs (χ² = 1.24; P = .27). Age wasnot significantly associated with BN status for the male-female or female-femaletwins, with an OR of 1.03 (95% CI, 0.99-1.06) and 1.00 (95% CI, 0.96-1.04),respectively.

All associations with BN for women are reported in the second columnof Table 2 (continuous measures)and Table 3 (dichotomous lifetimepsychopathology). Significant within-person associations with BN in the femaletwins existed for only 3 continuous measures: a higher body mass index, neuroticism,and novelty seeking. All forms of lifetime psychopathology were significantlyassociated with BN, with major depression and panic disorder having the strongestassociations.

When the relevant covariate trait for the sister was included in themodel, the neuroticism and novelty seeking in the brother were associatedwith BN in the sister (Table 2,column 3), with a 0.27-fold and 0.28-fold increase in risk for BN in the sisterassociated with every SD = 1.00 increase in neuroticism and novelty seeking,respectively, in the brother. The presence of GAD and use of psychoactivesubstances in the brother predicted BN in the sister (Table 3, column 3). The brother was 8 times more likely to haveGAD if his sister had BN and twice as likely to have used psychoactive substances.When the brother's levels of neuroticism and novelty seeking, presence ofGAD and major depression, and use of psychoactive substances were enteredinto a stepwise logistic regression, only GAD (OR, 7.07; 95% CI, 2.57-19.46)and use of psychoactive substances (OR, 2.77; 95% CI, 1.42-5.43) continuedto predict the presence of BN in the sister.

The predictors of BN status in the female-female pairs are shown inthe fourth column of Table 2 and Table 3. Unlike the male-female pairs,where there are larger numbers of twins, none of the relationships reachedsignificance. With respect to body mass index, extraversion, and novelty seeking,the direction of the association was in the opposite direction than for themale-female pairs. The associations are all in the same direction for lifetimepsychopathology.

The addition of female DZ pairs to the data set showed that there wereno cross-twin associations between BN status (twin 2) and any continuous measures(Table 2, column 5) but that significantassociations existed with the lifetime psychopathology of major depression,panic disorder, GAD, and psychoactive substance abuse (twin 1) when controllingfor the corresponding psychopathology in twin 2 (Table 3, column 5). In the final column of Table 2 and Table 3,interactions between sex and twin 2 BN status are examined. Two interactionswere significant, where vulnerability to BN in the sister is expressed asan increased risk of GAD and novelty seeking in male siblings only.

The results of the present study provide important replication of previousfindings pertaining to shared risk factors between BN and other forms of psychopathology.With respect to the within-twin associations, we replicated findings of previousstudies showing that BN is associated with novelty seeking,^{3 - 4} neuroticism,² and an array of lifetime psychopathology.¹ When we examined the cross-twin associations thatcould inform us of the presence of shared risk factors, we found a slightlydifferent pattern of results depending on whether we used a male-female groupor a combined male-female and female-female group. With the male-female group,novelty seeking, neuroticism, GAD, and psychoactive substance use in the brotherwere associated with increased risk for BN in the sister. In the female-femalegroup, major depression, panic disorder, GAD, and psychoactive substance usein one sister were associated with increased risk of BN in the other sister.With respect to psychopathology, the reasons for the different pattern ofresults is due to higher levels of depression and panic disorder in the womencompared with the men, giving more power to the combined analyses. With respectto the continuous measures, it seems that neuroticism and novelty seekingin women have somewhat different relationships with BN in women than men.Neuroticism has a negligible relationship with BN in women but a significantone in men, whereas the directions of association with novelty seeking aredifferent across men and women.

Unlike many family studies, we found that there was strong evidenceto suggest that BN does share familial risk factors with GAD. Consistent withthis evidence was our finding that neuroticism and BN share familial riskfactors. In the present study, there exists a significant association betweenGAD and neuroticism (OR,1.29; 95% CI, 1.20-1.34). The results of one previousstudy have been consistent with this result,¹⁸ whileanother study found the association to be unclear.¹⁴ Yetanother study¹³ found that BN only shared asmall degree of genetic risk factors with GAD in women. One notable limitationof this latter study was the absence of data from men, leading the authorsto warn that sex differences in prevalence rates and risk factors for commonpsychiatric disorders may not permit extrapolation of the results to men.Our finding that risk for BN in one sibling is translated to risk for GADin the other sibling only if the other sibling is male supports the existenceof sex-specific manifestation of familial liability.

We also found that risk for BN in one sibling is translated to riskfor higher levels of novelty seeking in the other sibling only if the othersibling is male. This temperament style may have a behavioral marker in theuse of psychoactive substances. In support of this argument, use of psychoactivesubstances was significantly associated with novelty seeking in the male-femalesample, with an OR of 1.03 (95% CI, 1.00-1.06). Novelty seeking, a tendencytoward frequent exploratory activity and excitement in response to novel stimuli,⁴ is associated with cluster B personality disorders¹⁹ and therefore represents an agreement with the resultsof Lilenfeld et al.¹⁴ Psychoactive substanceabuse and GAD were the only variables retained in our multivariable analysisand may represent the most salient markers of the temperament traits thatput people at risk of developing BN. With a positive but nonsignificant associationbetween GAD and psychoactive substance use, one wonders whether in some peoplethere is an aversive juxtaposition of a general anxiety and novelty seeking,where binge eating may serve to temporarily alleviate the tension betweenthe 2 opposing tendencies. However, in some individuals, this may represent2 different subtypes of risk factors.

While we did find that psychoactive substance use shared risk factorswith BN in this group, we found no evidence to suggest that substance abuseor dependence did. These results are consistent with many risk factor studies^{13 - 18} andsuggest that despite rigorous evidence pointing to a co-occurrence of BN withalcohol and drug-related disorders beyond chance,^{10 - 11} thisco-occurrence is not due to shared familial risk factors. Given that maleand female adolescents who binge eat are more likely to use all types of substancesthan nonbingers and both behaviors are associated with lower self-esteem ingirls and higher levels of depression in boys,³¹ plausiblealternative explanations for the co-occurrence are that the experience ofBN leads to drug or alcohol problems (or vice versa) or that a mediating thirdvariable (eg, low self-esteem) accounts for the co-occurrence.

In summary, results from this study suggest the presence of either familialor nonfamilial shared risk factors between BN and GAD, neuroticism, psychoactivesubstance use, novelty seeking, major depression, and panic disorder. Theseshared risk factors between BN and GAD and BN and novelty seeking are onlypresent in men, thereby supporting the existence of a sex-specific manifestationof familial liability for BN. This suggests that sex of family members mustbe incorporated into future analyses so that an effect of sex on shared riskfactors can be detected.

Our findings must be interpreted in the context of 7 limitations. First,we were unable to comment on the nature of these shared risk factors. Theymay be due to genetic factors, shared environmental factors, or some complexcombination of both. Although there is some evidence to suggest a small influenceof the shared environment on the development of BN¹³ anddisordered eating,^{32 - 33} the sharedenvironment only contributes between 1% and 4% of the variance of the etiologyof GAD, panic disorder, or major depression.¹³ Therefore,the common risk factors are unlikely to be from the shared environment. Second,we did not assess BN in the male twin and therefore may have obtained largereffect sizes than would be the case if we had controlled for the BN statusof the brother. As the prevalence of BN in men in the general population islow—around 10% of that expected in women—the effect of this covariateis unlikely to be substantial. A third limitation relates to our clinicaldefinition of BN. For the purposes of power considerations, we used a broadlydefined BN. Given the consistency of our overall results with many other studiesand the findings that the so-called subthreshold eating disorders are associatedwith the same degree of psychopathology as the full-syndrome eating disorders,⁷ the validity of our diagnosis is likely to be sound.Fourth, although our results apply to the white paired-twin sample from Virginia,they may not extrapolate to different populations. Fifth, results may be somewhatdifferent if all family members were incorporated or other forms of psychopathologyexamined. Sixth, we only used one occasion of BN measurement for the male-femalepairs, thereby introducing some unreliability of measurement or measurementerror. Finally, the interactions should be interpreted qualitatively as wellas quantitatively (ie, absence of an interaction does not mean that a trendtoward interaction is not present). Conversely, multiple testing could leadto a higher likelihood of type I errors.

Corresponding author and reprints: Tracey Wade, PhD, School of Psychology,Flinders University, PO Box 2100, Adelaide, South Australia 5001, Australia(e-mail: tracey .wade@flinders.edu.au).

Submitted for publication May 5, 2003; final revision received September23, 2003; accepted September 23, 2003.

This study was supported by grants MH-40828, MH/AA/DA-49492, DA-11287,and AA-09095 from the National Institutes of Health, Bethesda, Md.

An earlier version of this study was presented at the Eating DisordersResearch Society annual meeting; November 20,1999; San Diego, Calif.

We thank the Virginia Twin Registry, now part of the Mid-Atlantic TwinRegistry (MATR), Richmond, Va, for ascertainment of subjects for this study.The MATR, directed by Linda Corey, PhD, and Lindon Eaves, PhD, has receivedsupport from the National Institutes of Health, Bethesda; the Carman Trust,Richmond; the WM Keck Foundation, Los Angeles, Calif; the John Templeton Foundation,Radnor, Pa; and The Robert Wood Johnson Foundation, Princeton, NJ.

We also thank Patsy Waring; Frank Butera, MA; Sarah Woltz, BA; BarbaraBrooke, MSW; and Lisa Halberstadt, MS, for supervising data collection andIndrani Ray, BA, BSc, and Steven Aggen, PhD, for database management.