Deriving behavioural phenotypes in an international, multi-centre study of eating disorders

Psychological Medicine, 2001, 31, 635 645. 2001 Cambridge University Press Printed in the United Kingdom Deriving behavioural phenotypes in an international, multi-centre study of eating disorders THE PRICE FOUNDATION COLLABORATIVE GROUP, ABSTRACT Background. An international, multi-site study funded by the Price Foundation has collected 237 affected relative pairs to identify potential genetic factors that may contribute to the pathogenesis of anorexia nervosa (AN). The current report utilized this sample to derive phenotypes from the personality and behavioural traits assessed in a large number of individuals with eating disorders. Methods. Multivariate analytical techniques were used to characterize the relationships among personality (e.g. trait anxiety, perfectionism, harm avoidance, novelty seeking) and behavioural traits (obsessions and compulsions) in individuals with eating disorders (primarily AN; N 348) and assess the effectiveness of these traits in classifying subjects into diagnostic subtypes. Results. Factor analysis revealed that the most influential factor was one of trait anxiety, harm avoidance, perfectionism, obsessive compulsive behaviours, and diminished self-directedness, although the precise nature of this factor differed slightly across sites. Discriminant analysis was used to evaluate the utility of these personality behavioural factors in predicting subdiagnosis. Overall, the misclassification rate was 34%; however, there was an 80% rate of accurate classification of those individuals with a diagnosis of restricting-type AN. Conclusions. Trait anxiety, harm avoidance, perfectionism, obsessive compulsive behaviours and diminished self-directedness may best be conceptualized as parts of the same underlying construct among individuals with eating disorders, particularly AN. These personality and behavioural traits were also found to be of some utility in classifying eating disordered individuals into diagnostic subgroups, particularly those with restricting-type AN. We expect these phenotypic findings to benefit our ongoing search for genetic loci underlying the liability to eating disorders. The investigators in this multi-site study group are: Dr L. R. R. Lilenfeld, Department of Psychology, Georgia State University, Atlanta, GA, USA; Drs B. Devlin, S. Bacanu and W. H. Kaye, Western Psychiatric Institute and Clinic, University of Pittsburgh School of Medicine, Pittsburgh, PA, USA; Dr C. M. Bulik, Virginia Institute for Psychiatric and Behavioural Genetics, Department of Psychiatry, Virginia Commonwealth University, Richmond, VA, USA; Dr M. Strober, Neuropsychiatric Institute and Hospital, School of Medicine, University of California at Los Angeles, Los Angeles, CA, USA; Dr W. H. Berrettini, Center for Neurobiology and Behavior, Department of Psychiatry, University of Pennsylvania School of Medicine, Philadelphia, PA, USA; Dr M. M. Fichter, Klinik Roseneck, Hospital for Behavioural Medicine, Munich, Germany; Dr D. Goldman, National Institute on Alcohol Abuse and Alcoholism, National Institutes of Health, Bethesda, MD, USA; Dr K. A. Halmi, New York Presbyterian Hospital Westchester, Weill Medical College of Cornell University, White Plains, NY, USA; Drs A. Kaplan and D. B. Woodside, Department of Psychiatry, The Toronto Hospital, Toronto, Canada; and Dr J. Treasure, Institute of Psychiatry, Maudsley and Bethlem Royal Hospital, London, UK. Address for correspondence: Dr Walter H. Kaye, Western Psychiatric Institute and Clinic, 3811 O Hara Street E-724, Pittsburgh, PA 15213, USA. INTRODUCTION Psychometric studies have consistently linked anorexia nervosa (AN) to a cluster of moderately heritable (Heath et al. 1994) personality and temperamental traits, specifically, obsessionality, perfectionism and harm avoidance (Strober, 1980; Brewerton et al. 1993; Kleifield et al. 1993; Waller et al. 1993; Sohlberg & Strober, 1994; Vitousek & Manke, 1994; Bulik et al. 1995a; Srinivasagam et al. 1995; O Dwyer et al. 1996; Lilenfeld et al. 1998; Von Ranson et al. 1999; Kaye & Strober, 2000), as have earlier writings (e.g. Palmer & Jones, 1939; DuBois, 1949). In this regard, it has been speculated that phenotype similarities between these traits and the rigidly persevering, obsessional and anxiety- 635

636 The Price Foundation Collaborative Group reducing character of the anorexic s dietary restraint may be based upon shared genetic and environmental factors (Strober, 1991, 1995; Vitousek & Manke, 1994). These findings are remarkably consistent across studies (Lilenfeld & Kaye, 1998) and continue to characterize individuals with AN even after recovery (Casper, 1990; Kaye et al. 1993; Srinivasagam et al. 1995). While such traits may be exaggerated by starvation (Keys et al. 1950; Pollice et al. 1997), their persistence after recovery supports the speculation that such traits may be vulnerability factors for AN, rather than merely consequences of the disorder. Personality traits associated with bulimia nervosa (BN) have been less well-studied and appear more heterogeneous; however, certain consistencies are evident. While impulsivity and emotional instability are more common in BN than AN (Casper et al. 1980; Garfinkel et al. 1980; Strober, 1981), traits such as harm avoidance, compulsivity, perfectionism and obsessionality are prominent (Vitousek & Manke, 1994; Bulik et al. 1995a; Joiner et al. 1997) and often persist after clinical recovery (Kaye et al. 1998). The common occurrence of particular personality traits in AN and BN, especially those associated with anxiety and restraint, converges with evidence of significant co-morbidity between AN and BN (Kendler et al. 1991; Walters & Kendler, 1995), the frequency in both of antecedent anxiety disorders (Deep et al. 1995; Bulik et al. 1996, 1997a) and evidence of serotonergic hyperactivity in both disorders after long-term clinical recovery (Kaye et al. 1991, 1998). Research using twin and adoption study methodology suggests that a broad range of personality traits are heritable, with perhaps 50% of measured personality variation attributed to genetic diversity (Tellegen et al. 1988). A heuristic model proposed by Cloninger (1986) hypothesizes the existence of three core dimensions underlying most personality variation. Importantly, four independent studies (Brewerton et al. 1993; Kleifield et al. 1993; Waller et al. 1993; Bulik et al. 1995b) converge in showing that one particular dimension on the Temperament and Personality Questionnaire (TPQ) (Cloninger, 1986), namely harm avoidance, is significantly elevated in both AN and BN subjects compared to normal controls. This commonality is consistent with aetiological models positing the role of phobic anxiety as a predisposing factor in these conditions (Vitousek & Manke, 1994). The overarching goal of these collaborative studies (Kaye et al. 2000) is to identify genes that may influence susceptibility to AN. Belmaker & Biederman (1994) suggest that it may be more prudent to search for genetic markers of personality traits (e.g. behavioural inhibition) as expressions of genetic vulnerability to psychopathology than to look for markers of specific illness, as defined solely by psychiatric diagnosis. This may be particularly applicable to restricting-type AN, which may have the most homogeneous personality presentation of any psychiatric illness (Vitousek & Manke, 1994; Strober, 1995). Thus, our collaborative studies (Kaye et al. 2000) will correlate personality and behavioural traits, as well as psychiatric diagnosis, with genotypes. The goal of the current report was to utilize a multivariate analytical approach to improve the characterization of behavioural phenotypes of our subjects. This was accomplished by examining the inter-relationships among personality and behavioural traits and evaluating the utility of these traits in classifying subjects in our sample. The concise characterization of these personality and behavioural traits will aid in the search for liability loci. METHOD Study design and collaborative arrangements In an effort to map genetic susceptibility loci involved in AN, and international, multi-site study, funded by the Price Foundation, was designed (Kaye et al. 2000). This study utilized affected relative pairs to identify genetic factors that might contribute to the pathogenesis of AN. In the Kaye et al. (2000) study, all probands met modified DSM-IV (American Psychiatric Association, 1994) criteria for AN, whereas all affected first-, second- and third-degree relatives met modified DSM-IV criteria for AN, DSM-IV BN, or eating disorder not otherwise specified (NOS) (refer to Participants section). The investigators, using a collaborative consensus, selected a battery to assess psychological

Deriving behavioural phenotypes of eating disorders 637 and personality features that have been shown to be associated with, and may underlie vulnerability to, eating disorders. In general, we attempted to evaluate persistent traits rather than personality features confined to the acute phase of the illness (Brewerton et al. 1993, Kleifield et al. 1993, Bulik et al. 1995 a, Srinivasagam et al. 1995; O Dwyer et al. 1996; Kaye et al. 1998; Von Ranson et al. 1999). These traits included harm avoidance, perfectionism, and anxiety. Six sites in North America and Europe were identified that had well-established programmes for the treatment of AN. The sites included University of Pittsburgh (W. H. K.), Cornell University (K. A. H.), University of California at Los Angeles (M. S.), University of Toronto (A. K. and D. B. W.), University of Munich (M. M. F.) and University of London (J. T.). An additional site, Thomas Jefferson University in Philadelphia (W. H. B.), was included for its expertise in molecular genetic research on psychiatric illness. See Kaye et al. (2000) for a complete description of the overall study design. Participants The subjects reported in this manuscript (N 348) are a subset of the sample from the multisite project (Kaye et al. 2000). This manuscript reports on all AN probands (N 196) and their siblings who had a lifetime diagnosis of modified DSM-IV AN (N 116) or DSM-IV BN (N 36). All other relatives from the original sample (e.g. half-siblings, cousins, aunts, uncles), as well as siblings with a diagnosis of NOS, were excluded in the current report. Our reasoning for this sampling approach is as follows. Both genetic and environmental effects on the correlation of personality behavioural traits should be consistent within full siblings and greater than these effects on more distant relative pairs, such as aunt niece pairs. Thus, in statistical terms, observations of sibling-pairs are exchangeable across pairs (i.e. the identity of the pairs is irrelevant), but a sibling-pair and aunt niece pair are not exchangeable. By analysing only full siblings we maintain the majority of the sample while also meeting this important statistical criterion of exchangeability. Probands were required to meet the following criteria for acceptance into the study: (a) an unequivocal lifetime core diagnosis of AN by DSM-IV criteria, waiving the single criterion of amenorrhoea for 3 consecutive months, since some subjects were menstruating due to treatment with exogenous hormone replacement and some were male; (b) age between 13 and 65; and (c) fulfilment of AN diagnostic criteria for not less than 3 years prior to ascertainment. Exclusion criteria included a lifetime history of any of the following: organic brain syndrome; IQ 70; dementia, schizophrenia; bipolar illness; obesity; medical illness that could affect appetite, eating behaviour, or body weight (e.g. diabetes); DSM-IV binge eating disorder, and any regular binge eating. The latter was defined as bingeing at least once weekly for 3 or more consecutive months. All siblings had similar exclusion criteria, with the exception of binge eating, as some of these subjects had bingeing symptomatology in the context of an AN diagnosis and some had a BN diagnosis. BN diagnoses were made according to DSM-IV criteria. Assessment instruments The assessment battery was selected, based on existing literature, to include those personality and behavioural features most consistently associated with eating disorders and not confined to the acute phase of the illness (Brewerton et al. 1993; Kleifield et al. 1993; Waller et al. 1993; Bulik et al. 1995a; Srinivasagam et al. 1995; O Dwyer et al. 1996; Kaye et al. 1998; Von Ranson et al. 1999). However, we did ask subjects to report some behaviours (e.g. obsessions and compulsions) at the time when their eating disorder was at its worst in order to access the maximum lifetime severity of the behaviour. Methods for interviewer training and reliability are described elsewhere (Kaye et al. 2000). Eating disorder diagnostic assessment Structured Interview of Anorexia Nervosa and Bulimic Syndromes (SIAB) This instrument was used to assess lifetime history of eating disorders among probands and affected relatives. The SIAB (Fichter et al. 1998) is a detailed structured interview schedule that derives eating disorder diagnostic information, as well as additional psychopathological factors

638 The Price Foundation Collaborative Group related to eating disorders. The internal consistency of the six SIAB subscales on the lifetime version of the instrument is moderate to high, with Cronbach s alpha ranging from 0 64 to 0 89 (Fichter et al. 1998), and their inter-rater reliability for these same subscales is excellent, ranging between 0 80 and 0 90 (Fichter et al. 1998). No psychometric information has yet been published on the English translation of the SIAB. Subjects were asked to report worst lifetime symptoms. Behavioural assessments Yale Brown Obsessive Compulsive Scale (Y- BOCS) The Y-BOCS (Goodman et al. 1989) is a semistructured interview designed to rate the presence and severity of obsessive thoughts and compulsive behaviours typically found among individuals with obsessive compulsive disorder (OCD). It has excellent inter-rater reliability (Goodman et al. 1989) and is considered to be the gold standard for measuring obsessive compulsive symptom severity (Pato et al. 1994). There is a published German translation of this instrument as well (Goodman et al. 1991). Subjects were asked to report worst lifetime symptoms. Yale Brown Cornell Eating Disorder Scale (YBC-EDS) The YBC-EDS (Sunday et al. 1995) is similar to the Y-BOCS; however, it assesses core obsessions and compulsions specific to eating disorders (e.g. those related to food, eating, weight and exercise). Excellent inter-rater reliability, internal consistency and convergent validity have been demonstrated for the YBC- EDS (Mazure et al. 1994). No psychometric information has yet been published on the German translation of the YBC-EDS. The YBC- EDS was modified with the authors of the instrument to assess worst lifetime symptoms, as well as current symptoms. Self-report psychological symptom and personality assessments State-Trait Anxiety Inventory (STAI) The STAI (Spielberger et al. 1970) is a widely used instrument for the assessment of anxiety. The state anxiety assessment asks subjects to report how they feel at this moment and the trait anxiety assessment asks subjects to report how they generally feel. There is also a published German translation of this instrument (Laux et al. 1981). Multidimensional Perfectionism Scale (MPS) The MPS (Frost et al. 1990) is a 35-item, factoranalytically developed self-rating instrument that consists of an overall assessment of perfectionism, as well as six specific dimensions of perfectionism. The coefficients of internal consistency for the factor scales range from 0 77 to 0 93 and the reliability of the overall scale is 0 90 (Frost et al. 1990). The MPS successfully discriminates between subjects with and without eating disorders (Srinivasagam et al. 1995). No psychometric information has yet been published on the German translation of the MPS. To assess worst lifetime symptom expression, subjects were instructed to respond to the questions according to how they felt at the time when their concerns about eating and weight were strongest. Temperament and Character Inventory (TCI) The TCI (Cloninger et al. 1993) is a 226-item factor-analytically developed self-rating instrument that measures seven dimensions of personality. The TCI is an extension of the Tridimensional Personality Questionnaire (TPQ) (Cloninger et al. 1991), which assesses the temperament dimensions of novelty seeking, harm avoidance and reward dependence (Cloninger et al. 1991). The authors original model has been extended to measure the additional temperament factor of persistence and the three character dimensions of self-directedness, cooperativeness, and selftranscendence. The internal consistency of all seven scales is high, ranging from 0 76 0 89 (Cloninger et al. 1993). There is a published German translation of the TCI as well (Cloninger et al. 2000). Procedure Subjects were recruited through several sources, including: (1) identification of potential subjects through clinic data bases; (2) referrals from clinicians with knowledge of the study; (3) advertisement in a variety of different media at local and national levels. Potential subjects were first screened to determine study suitability. If a

Deriving behavioural phenotypes of eating disorders 639 likely proband denied a family history of eating disorders, or refused permission to contact possibly ill relatives, the screening was terminated. Otherwise, a preliminary verification of the diagnosis of AN was undertaken and eating disorder histories on possibly affected relatives were obtained. If probands met all inclusion and no exclusion criteria, and gave a history suggestive of eating disorder in a nonparent, non-child, and a non-mz twin blood relative, the proband was asked to discuss the study with the affected relative and obtain permission for study personnel to contact the relative for informed consent. Study personnel then contacted the relative who was screened for eligibility. If the relative fulfilled entrance criteria, both the proband and affected relative were scheduled for the complete battery of evaluative procedures. At the time that the interviews were scheduled, the proband and affected relative were told that they would be mailed a packet of self-rating assessments. They were asked to complete the assessments and bring the packet to the interview. For those probands and relatives who were not able to come to a satellite centre for an in-person interview, the interview was conducted by telephone and the self-rating assessments were returned by mail. Only AN probands and siblings with AN and BN diagnoses are reported in the current study due to reasons described in the Participants section. Data analysis Data analysis was conducted in several stages. An initial principal components factor analysis applying a varimax rotation was performed on individual TCI scales to evaluate whether the factor structure would be replicated with our eating disorder sample. Next, to characterize the personality and behavioural attributes of individuals with eating disorders, we performed an exploratory factor analysis with varimax rotation using the responses of our study participants to the following personality and behavioural assessments: the seven TCI scales, two measures of obsessive compulsive symptoms (Y-BOCS and YBC-EDS), a multidimensional measure of perfectionism (MPS), and a measure of trait anxiety (STAI). The initial factor analysis grouped data for all sites (or populations). To determine whether there was heterogeneity among populations, a delete-one jackknife procedure was done. This evaluated the results of the factor analysis after deleting one population at a time. Any population that was an outlier, as determined by the jackknife analysis, was deleted from the sample and the initial factor analysis was repeated. To explore further potential differences among populations, a factor analysis was performed on data from the three populations with the largest sample sizes (excluding any outliers ). To evaluate whether the personality behavioural constructs extracted from the factor analysis were useful for predicting eating disorder subdiagnoses, a discriminant analysis was performed using the factors and the four diagnostic subclassifications: anorexia nervosarestricting type (AN-R), anorexia nervosa-purging type (AN-P), anorexia nervosa-bingeing or bingeing and purging type (AN-B), and BN. Discriminant analysis is a statistical method that utilizes trait attributes to discriminate members of two or more groups from one another. That is, this analytical approach allows one to predict the group membership status of subjects based upon their scores on quantitative trait attributes. Our methods follow classic discriminant analysis procedures as implemented in S-Plus (Venables & Ripley, 1994). First, using a random selection of 80% of the data (the training set), a classification tree was built based on factors 1 4 from the factor analysis. The tree itself is based on bifurcating factor splits that yield optimal separation of populations into diagnostic categories. Using the remaining 20% of the data as the test set, the tree built with the training set is pruned to provide a more accurate classification tree. The accuracy is improved by avoiding overfitting the training data. All data analyses were performed using the SAS statistical package (version 6.12) RESULTS Factor analysis of the TCI with an eating disordered sample Factor loadings for each individual scale were similar and positive, confirming the TCI factor structure for our eating disorder sample. In addition, the fraction of variance explained by the first factor was substantial for all scales.

640 The Price Foundation Collaborative Group Table 1. Raw scores on the personality and behavioural measures across all sites Variables Munich (N 78) New York (N 65) Pittsburgh (N 59) Toronto (N 52) Los Angeles (N 48) London (N 37) Philadelphia (N 9) YBC-EDS 21 55 25 23 25 07 26 51 23 54 25 46 25 78 (6 53) (5 59) (4 66) (7 53) (5 52) (5 74) (5 64) Y-BOCS 8 57 15 88 17 88 18 78 17 69 12 38 17 25 (10 70) (11 84) (10 89) (13 50) (10 46) (12 31) (11 02) MPS 84 71 95 86 101 63 104 94 102 74 101 86 106 27 (23 97) (21 06) (21 68) (23 61) (20 44) (18 50) (27 15) STAI-T 50 32 52 63 50 50 55 61 53 95 53 59 49 75 (12 35) (14 10) (14 24) (15 92) (13 25) (13 01) (16 60) NS 18 88 19 53 15 55 16 37 18 04 14 53 20 22 (7 20) (7 00) (7 07) (7 74) (7 57) (7 80) (7 10) HA 20 66 21 56 20 47 23 40 22 11 22 97 19 79 (8 52) (7 79) (7 91) (8 09) (7 57) (7 28) (9 09) RD 15 72 16 90 17 33 16 37 15 89 16 69 18 21 (3 86) (3 84) (3 97) (3 96) (4 50) (3 50) (3 71) PS 4 76 5 80 6 76 6 11 5 72 6 38 6 22 (2 11) (2 28) (1 71) (1 96) (2 22) (1 99) (1 32) SD 22 13 24 50 28 48 25 74 25 81 24 23 24 36 (8 96) (9 40) (9 75) (9 59) (9 54) (7 60) (9 44) C 29 45 32 85 35 24 34 08 31 66 33 89 30 04 (5 57) (6 51) (6 30) (7 24) (6 93) (5 98) (9 46) ST 13 62 13 74 16 01 15 87 14 04 13 09 18 00 (6 41) (6 52) (7 00) (7 61) (6 41) (6 46) (9 27) Values are means (standard deviations). YBC-EDS, Lifetime worst total score from the Yale Brown Cornell-Eating Disorder Scale. Y-BOCS, Lifetime worst total score from the Yale Brown-Obsessive Compulsive Scale. MPS, Lifetime worst total score from the Multidimensional Perfectionism Scale. STAI-T, Trait Anxiety from the State-Trait Anxiety Inventory. Temperament and Character Inventory (TCI) scales: NS, Novelty Seeking; HA, Harm Avoidance; RD, Reward Dependence; PS, Persistence; SD, Self-Directedness; C, Cooperativeness; ST, Self-Transcendence. Thus, we had confidence in using the seven temperament and character scales for the remaining analyses. Factor analysis of personality and behavioural measures The jackknife analysis revealed one highlyinfluential population, Munich, which differed from other sites with regard to the personality and behavioural traits of their subjects. In viewing Table 1, it is clear that the primary discrepancy between Munich and the other sites is substantially lower Y-BOCS scores, as well as lower YBC-EDS and MPS scores compared to all other sites. Due to the undetermined reason for this significant site difference, we excluded Munich from the remaining factor analyses. The results of the factor analysis are presented in Table 2. The first four factors explained 51% of the variance in the personality and behavioural traits, with factors 1 4 accounting for 14, 13, 13 and 11% of the variance respectively. The weightings for factor 1 reflect a classic profile of eating disordered individuals, which includes moderate to strong positive loadings on trait anxiety (STAI-T), harm avoidance (HA), perfectionism (MPS), general and eating-specific obsessive compulsive symptoms (YBOCS; YBC-EDS). An additional strong negative loading on this factor was found for self-directedness (SD). Factor 2 reveals ties between reward dependence (RD) and cooperativeness (C), with weaker links to SD and lack of anxiety (STAI-T). Factor 3 reveals a connection between perfectionism (MPS), persistence (PS), and self-transcendence (ST), with some weaker associations with general (Y- BOCS) and eating-specific obsessive compulsive behaviours (YBC-EDS). Factor 4 s weightings are dominated by novelty-seeking (NS). To explore differences among populations, a factor analysis was performed on data from the three North American sites with the largest sample sizes. Because a 5:1 ratio of subjects to variables is recommended for factor analysis (Bryant & Yarnold, 1998), data from the Los Angeles, London and Philadelphia sites were not factor analysed individually due to their

Deriving behavioural phenotypes of eating disorders 641 relatively small individual sample sizes. Data from the Munich site were not factor analysed individually due to the undetermined reason for substantial differences on several assessment measures compared to all other sites. Only the weightings for the first factor are reported (Table 3), although four factors were extracted from the data. The first factor explains 23, 26 and 29% of the variance for the data from the New York (N 65), Pittsburgh (N 59), and Toronto (N 52) populations respectively. For individuals from New York, the construct is specifically characterized by high loadings on STAI-T and HA. For individuals from Pittsburgh and Toronto, the construct encompasses a wider range of characteristics, including HA, STAI-T, MPS and obsessive compulsive symptoms (general and eating-related) (Y-BOCS and YBC-EDS). Diminished SD is also strongly tied to this anxious obsessional construct for each of these three sites. Discriminant analysis Discriminant analysis was used to evaluate the extent to which we can accurately classify Table 2. Factor analysis of personality and behavioural measures for individuals diagnosed with eating disorders (N 270) (excluding the sample from Munich) Variable Factor 1 2 3 4 YBC-EDS 0 454 * 0 294 * Y-BOCS 0 486 * 0 274 * MPS 0 531 * 0 508 * STAI-T 0 806 0 197 * * NS 0 187 * 0 201 0 722 HA 0 780 * 0 187 0 383 RD * 0 692 * 0 152 PS * * 0 594 0 204 SD 0 773 0 325 0 212 * C 0 210 0 697 0 186 0 149 ST * 0 155 0 434 0 356 YBC-EDS, Lifetime worst total score from the Yale Brown Cornell-Eating Disorder Scale. Y-BOCS, Lifetime worst total score from the Yale Brown- Obsessive Compulsive Scale. MPS, Lifetime worst total score from the Multidimensional Perfectionism Scale. STAI-T, Trait Anxiety from the State-Trait Anxiety Inventory. Temperament and Character Inventory (TCI) scales: NS, Novelty Seeking; HA, Harm Avoidance; RD, Reward Dependence; PS, Persistence; SD, Self-Directedness; C, Cooperativeness; ST, Self- Transcendence. * Factor loading was 0. Table 3. Factor analyses for the three North American recruitment sites with the largest sample sizes. Only the first factor loadings are presented % Variance explained New York (N 65) 23% Sites Pittsburgh (N 59) 26% Toronto (N 52) 29% YBC-EDS 0 277 0 558 0 541 Y-BOCS 0 395 0 471 0 645 MPS 0 368 0 659 0 660 STAI-T 0 830 0 728 0 830 NS 0 146 0 236 * HA 0 827 0 758 0 655 RD * * 0 119 PS 0 213 * 0 259 SD 0 771 0 844 0 807 C 0 227 * 0 326 ST 0 168 0 141 0 163 YBC-EDS, Lifetime worst total score from the Yale Brown Cornell-Eating Disorder Scale. Y-BOCS, Lifetime worst total score from the Yale Brown- Obsessive Compulsive Scale. MPS, Lifetime worst total score from the Multidimensional Perfectionism Scale. STAI-T, Trait Anxiety from the State-Trait Anxiety Inventory. Temperament and Character Inventory (TCI) scales: NS, Novelty Seeking; HA, Harm Avoidance; RD, Reward Dependence; PS, Persistence; SD, Self-Directedness; C, Cooperativeness; ST, Self- Transcendence. * Factor loading was 0. subjects into diagnostic subgroups using only the extracted personality behavioural factors. The degree to which these factors can accurately sort our subjects has direct implications for using these traits to define a useful behavioural phenotype. To evaluate whether the factors were indeed useful for predicting subdiagnosis, discriminant analysis was performed using the factors and the four diagnostic subclassifications: AN-R, AN-P, AN-B and BN. Discrimination into the four diagnostic subcategories based upon the personality behavioural factors yielded an overall misclassification rate of 33 8%. The misclassification varied by subdiagnosis (Table 4A), with a strong trend toward increasingly accurate classification of subjects with AN-R. That is, most AN-R subjects (80%), but only slightly more than half of all other subjects, were classified into the correct diagnostic subtype based upon the personality behavioural factors. However, this presentation does not give the full picture. Another way to view the data is to ask: Given

642 The Price Foundation Collaborative Group A Table 4. Discriminant analysis into four diagnostic subcategories using the personality behavioural trait factors from Table 2 Percentage of subjects with a given subdiagnosis who were correctly labelled as having that subdiagnosis Actual subdiagnosis Correctly classified (true positive) % Incorrectly classified (false negative) % Restricting anorexia nervosa 80 0 20 0* Purging anorexia nervosa 52 0 48 0 Bingeing anorexia nervosa 49 0 51 0 Bulimia nervosa 57 6 42 8 * These subjects were incorrectly labelled with some subdiagnosis other than their true subdiagnosis of restricting anorexia nervosa when classified using the personality behavioural factors. B Percentage of subjects labelled as having a given subdiagnosis who actually have that subdiagnosis Subdiagnosis based upon personality behavioural factors Correctly classified (true positive) % Incorrectly classified (false positive) % Restricting anorexia nervosa 68 9 31 1 Purging anorexia nervosa 64 6 35 4 Bingeing anorexia nervosa 60 0 40 0 Bulimia nervosa 47 5 52 5 The actual subdiagnosis of these subjects is something other than restricting anorexia nervosa. Bingeing anorexia nervosa includes those anorexic subjects with bingeing only and those with bingeing and purging symptomatology; Purging anorexia nervosa includes those anorexic subjects with purging, but no history of bingeing. the assigned subdiagnosis based upon the personality behavioural factors, what fraction of individuals actually have that subdiagnosis?. By this view, the discriminant analysis was successful approximately two-thirds of the time for all AN subdiagnoses, but slightly less than half the time for a BN diagnosis (Table 4B). DISCUSSION Phenotypic characterization Factor analysis revealed that the strongest factor underlying the cluster of personality and behavioural traits in this sample is a combination of trait anxiety, harm avoidance, perfectionism, obsessive compulsive behaviour, and diminished self-directedness. Importantly, the current study raises the possibility that these five traits may be best conceptualized as parts of the same underlying construct among individuals with AN and related eating disorders. This primary factor supports past research (Casper, 1990; Brewerton et al. 1993; Kleifield et al. 1993; Kaye et al. 1993; Waller et al. 1993; Bulik et al. 1995a, b; 1997a; Srinivasagam et al. 1995; Joiner et al. 1997; Fairbum et al. 1997; Pollice et al. 1997; Kaye et al. 1998; Lilenfeld et al. 1998, 2000) demonstrating anxiety, harm avoidance, perfectionism and obsessive compulsive behaviour to be potential vulnerability factors for the development of eating disorders, AN in particular. Of all five traits, self-directedness has been the least studied. The central elements of selfdirectedness are the acceptance of responsibility for one s own choices, identification of individually valued goals and purposes, development of skills and confidence in solving problems, and self-acceptance (Cloninger et al. 1993). Therefore, low levels of self-directedness may be conceptualized as immaturity, ineffectiveness, and lacking an integrated view of oneself. Low self-directedness has recently been demonstrated to be a reasonably good predictor of personality disorder (Bulik et al. 1995 b) and high selfdirectedness has been found to predict rapid and sustained treatment response (Bulik et al. 1999) in a bulimic sample.

Deriving behavioural phenotypes of eating disorders 643 Several differences emerged across subjects from different parts of the world when sites were compared. For those individuals from New York, the primary personality factor was best characterized by trait anxiety, harm avoidance and diminished self-directedness. In addition to these three traits, for those from Pittsburgh and Toronto, the factor also encompassed perfectionism and obsessive compulsive behaviour. Notably, the Munich sample differed from the other sites. There are at least two possible explanations for this difference. There could be cultural differences in the personality behavioural constructs that characterize individuals with eating disorders from Munich compared with those from other parts of the world. Such differences may suggest the importance of cultural influences in the development of eating disorders and related personality characteristics. Alternatively, the differences could be due to translation of the assessment instruments, as Munich was the only non-english speaking site. Overall however, there were more similarities than differences across sites. Classification Our first question was: How often are subjects assigned the correct subdiagnostic label, based solely upon the personality behavioural factors?. Results of the discriminant analysis revealed that on average, subjects were accurately classified into the proper diagnostic subtype category based solely upon the extracted personality behavioural factors approximately two-thirds of the time. However, we were much more accurate (i.e. 80% accuracy) in classifying subjects with restricting-type anorexia nervosa into the appropriate category than any other eating disorder subtype. Our second question was: Of all those subjects labelled as belonging to a particular diagnostic subgroup based solely upon the personality behavioural factors, how many actually have that diagnosis?. Comparable diagnostic accuracy was demonstrated across all anorexic subgroups, whereas there was substantially less accuracy in classifying subjects as having bulimia nervosa. That is, more than half of subjects who were classified as bulimic based upon the personality behavioural factors, were actually anorexic. One possible reason that there was more accuracy in classifying subjects with restrictingtype AN is because they have a more homogeneous phenotype. Our findings support the frequent observation that these individuals have a remarkably consistent personality and behavioural presentation (Sohlberg & Strober, 1994). If the restricting anorexic subtype indeed has the clearest phenotypic characterization, this subset of individuals may be the most promising focus for future genetic work. Subtype misclassification may be partly explained by the fact that a significant minority of individuals with AN later develop BN (e.g. Eckert et al. 1995; Bulik et al. 1997b; Strober et al. 1997). However, if this conversion is to take place, it most often occurs within 3 to 5 years after the onset of AN. This is precisely why we required restricting-type AN probands to have a pure diagnosis; that is, the onset of their illness must have been at least 3 years prior to ascertainment, with no prior or subsequent history of bulimic behaviour. However, a similar purity criterion was not applied to affected relatives; that is, they simply needed to meet DSM-IV criteria in order to receive a diagnosis of restricting-type AN. Therefore, some subtype misclassification may be explained by impure diagnostic categories, consisting partially of relatives classified as AN-R who may later go on to develop BN. Caveats Inference for this study is restricted to a population of individuals with eating disorders. This population is commonly characterized by extreme values on personality and behavioural measures. Therefore, the relationships among these measures cannot be taken as representative of the general population. In addition, the nature of the current sample is very selective (i.e. anorexic probands and their eating disordered siblings). Therefore, we also may not be able to generalize our findings to all individuals with eating disorders. Next, the current study lacks a comparison group, such as a random sample from the population. A comparison sample would undoubtedly magnify the similarities among individuals with eating disorders, who tend to score more extremely on these personality and behavioural measures than would randomly-chosen individuals from the popu-

644 The Price Foundation Collaborative Group lation. Finally, our findings should be replicated on an independent sample. Conclusions The current findings suggest that the phenotypic characterization of eating disorders may be enriched for genetic studies by the inclusion of empirically derived personality and behavioural constructs. An important implication of the current findings is that trait anxiety, harm avoidance, perfectionism, obsessive compulsive behaviour, and diminished self-directedness may constitute a single underlying construct among individuals with AN and related eating disorders. These data are consistent with a substantial literature showing that anorexic individuals have extreme levels of behavioural rigidity, emotional restraint and perfectionism. The phenotypic variables assessed in this cohort may prove useful in elucidating the genetic susceptibility to disease in several ways. For example, analysis of genotypic data on pairs of affected relatives by allele-sharing methods can include quantitative covariates, such that pairs of relatives characterized by high trait anxiety, harm avoidance, perfectionism, obsessive compulsive behaviour, and diminished self-directedness scores are given greater weight than pairs without this profile. This may allow for the definition of more genetically homogeneous subgroups among the cohort, facilitating the detection of loci whose action to increase risk for AN is more pronounced on the background of these personality behavioural traits. We are hopeful that this effort to identify quantitative traits will offer enhanced power to search for accompanying genotypes that contribute to the pathophysiology of eating disorders. This work was financially supported by the Price Foundation of Geneva, Switzerland. REFERENCES American Psychiatric Association. (1994). Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition). American Psychiatric Press: Washington, DC. Belmaker, R. H. & Biederman, J. (1994). Genetic markers, temperament and psychopathology. Biological Psychiatry 36, 71 72. Brewerton, T. D., Hand, L. D. & Bishop, E. R. (1993). The Tridimensional Personality Questionnaire in eating disorder patients. International Journal of Eating Disorders 14, 213 218. Bryant, F. B. & Yarnold, P. R. (1998). Principal-components analysis and exploratory and confirmatory factor analysis. In Reading and Understanding Multivariate Statistics (ed. L. G. Grimm and P. R. Yarnold). American Psychological Association: Washington, DC. Bulik, C. M., Sullivan, P. F., Weltzin, T. E. & Kaye, W. H. (1995a). Temperament in eating disorders. International Journal of Eating Disorders 17, 251 261. Bulik, C. M., Sullivan, P. F., Joyce, P. R. & Carter, F. A. (1995b). Temperament, character, and personality disorders in bulimia nervosa. Journal of Nervous and Mental Diseases 183, 593 598. Bulik, C. M., Sullivan, P. F., Carter, F. A. & Joyce, P. R. (1996). Lifetime anxiety disorders in women with bulimia nervosa. Comprehensive Psychiatry 37, 368 374. Bulik, C. M., Sullivan, P. F., Fear, J. L. & Joyce, P. R. (1997a). Eating disorders and antecedent anxiety disorders: a controlled study. Acta Psychiatrica Scandinavia 96, 101 107. Bulik, C. M., Sullivan, P. F., Fear, J. L. & Pickering, A. (1997b). Predictors of the development of bulimia nervosa in women with anorexia nervosa. Journal of Nervous and Mental Diseases 185, 704 707. Bulik, C. M., Sullivan, P. F., Carter, F. A., McIntosh, V. V. & Joyce, P. R. (1999). Predictors of rapid and sustained response to cognitive behavioural therapy for bulimia nervosa. International Journal of Eating Disorders 26, 137 144. Casper, R. C. (1990). Personality features of women with good outcome from restricting anorexia nervosa. Psychosomatic Medicine 52, 156 170. Casper, R. C., Eckert, E. D., Halmi, K. A., Goldberg, S. C. & Davis, J. M. (1980). Bulimia: its incidence and clinical importance in patients with anorexia nervosa. Archives of General Psychiatry 37, 1030 1035. Cloninger, C. R. (1986). A unified biosocial theory of personality and its role in the development of anxiety states. Psychiatric Developments 3, 167 226. Cloninger, C. R., Przybeck, T. R. & Svrakic, D. M. (1991). The Tridimensional Personality Questionnaire: US normative data. Psychological Reports 69, 1047 1057. Cloninger, C. R., Svrakic, D. M. & Przybeck, T. R. (1993). A psychobiological model of temperament and character. Archives of General Psychiatry 50, 975 990. Cloninger, C. R., Przybeck, T. R., Svrakic, D. M. & Wetzel, R. D. (2000). Swets Test Services GmbH. Schaumainkai 69, D-60596. Deep, A. L., Nagy, L. M., Weltzin, T. E., Rao, R. & Kaye, W. H. (1995). Premorbid onset of psychopathology in long-term recovered anorexia nervosa. International Journal of Eating Disorders 17, 291 297. DuBois, F. (1949). Compulsion neurosis with cachexia (anorexia nervosa). American Journal of General Psychiatry 106, 107 115. Eckert, E., Halmi, K., Marchi, P., Grove, W. & Crosby, R. (1995). Ten-year follow-up of anorexia nervosa: clinical course and outcome. Psychological Medicine 25, 143 156. Fairburn, C. G., Welch, S. L., Doll, H. A., Davies, B. A. & O Connor, M. E. (1997). Risk factors for bulimia nervosa: a communitybased case-control study. Archives of General Psychiatry 54, 509 517. Fichter, M. M., Herpertz, S., Quadflieg, N. & Herpertz-Dahlmann, B. (1998). Structured interview for anorexic and bulimic disorders for DSM-IV and ICD-10: updated (third) revision. International Journal of Eating Disorders 24, 227 257. Frost, R. O., Marten, P., Lahart, C. & Rosenblate, R. (1990). The dimensions of perfectionism. Cognitive Therapy and Research 14, 449 468. Garfinkel, P. E., Moidofsky, H. & Garner, D. M. (1980). The heterogeneity of anorexia nervosa: Bulimia as a distinct subgroup. Archives of General Psychiatry 37, 1036 1040. Goodman, W. K., Price, L. H., Rasmussen, S. A., Mazure, C., Fleischmann, R. L., Hill, C. L., Heninger, G. R. & Chamey, D. S. (1989). The Yale Brown Obsessive Compulsive Scale (Y-BOCS): I. development, use and reliability. Archives of General Psychiatry 46, 1006 1011.

Deriving behavioural phenotypes of eating disorders 645 Goodman, W. K., Rasmussen, S. A., Price, L. H., Mazure, C., Heninger, G. R. & Charney, D. S. (1991). The Yale Brown Obsessive Compulsive Scale. Verhaltenstherapie 1, 226 233. Heath, A. C., Cloninger, C. R. & Martin, N. G. (1994). Testing a model for the genetic structure of personality: a comparison of the personality systems of Cloninger and Eysenck. Journal of Personality and Social Psychology 66, 762 775. Joiner, T. E. Jr., Heatherton, T. F., Rudd, M. & Schmidt, N. B. (1997). Perfectionism, perceived weight status, and bulimic symptoms: two studies testing a diathesis-stress model. Journal of Abnormal Psychology 106, 145 153. Kaye, W. H. & Strober, M. (2000). Neurobiology of eating disorders. In Neurobiological Foundations of Mental illness (ed. D. E. Charney, E. J. Nestler and B. S. Bunney). Oxford University Press: New York. Kaye, W. H., Gwirtsman, H. E., George, D. T. & Ebert, M. H. (1991). Altered serotonin activity in anorexia nervosa after longterm weight restoration: does elevated cerebrospinal fluid 5- hydroxy indoleacetic acid level correlate with rigid and obsessive behaviour? Archives of General Psychiatry 48, 556 562. Kaye, W. H., Weltzin, T. E. & Hsu, L. K. G. (1993). Relationship between anorexia nervosa and obsessive and compulsive behaviours. Psychiatric Annals 23, 365 373. Kaye, W. H., Greeno, C. G., Moss, H., Fernstrom, J., Fernstrom, M., Lilenfeld, L. R., Weltzin, T. E. & Mann, J. J. (1998). Alterations in serotonin activity and psychiatric symptoms after recovery from bulimia nervosa. Archives of General Psychiatry 55, 927 935. Kaye, W. H., Lilenfeld, L. R., Berrettini, W. H., Strober, M., Devlin, B., Klump, K. L., Goldman, D., Bulik, C. M., Halmi, K. A., Fichter, M. M., Kaplan, A., Woodside, D. B., Treasure, J., Plotnicov, K. H., Pollice, C., Rao, R. & McConaha, C. W. (2000). A search for susceptibility loci for anorexia nervosa: methods and sample description. Biological Psychiatry 47, 794 803. Kendler, K. S., MacLean, C., Neale, M., Kessler, R., Heath, A. & Eaves, L. (1991). The genetic epidemiology of bulimia nervosa. American Jorunal of Psychiatry 148, 1627 1637. Keys, A., Brozek, J., Henschel, A., Mickelsen, O. & Taylor, H. L. (1950). The Biology of Human Starvation. University of Minnesota Press: Minneapolis. Kleifield, E. I., Sunday, S., Hurt, S. & Halmi, K. A. (1993). The Tridimensional Personality Questionnaire: an exploration of personality traits in eating disorders. Journal of Psychiatric Research 28, 413 423. Laux, L., Glanzmann, P., Schaffner, P. & Spielberger, C. D. (1981). Das State-Trait-Angstinventar. Testzentrale 25, D-37079. Lilenfeld, L. R. & Kaye, W. H. (1998). Genetic studies of anorexia and bulimia nervosa. In Neurobiology in the Treatment of Eating Disorders (ed. H. W. Hoek, J. L. Treasure and M. A. Katzman), pp. 169 194. John Wiley and Sons: New York. Lilenfeld, L. R., Kaye, W. H., Greeno, C. G., Merikangas, K. R., Plotnicov, K., Pollice, C., Rao, R., Strober, M., Bulik, C. M. & Nagy, L. (1998). A controlled family study of anorexia nervosa and bulimia nervosa. Archives of General Psychiatry 55, 603 610. Lilenfeld, L. R. R., Stein, D., Bulik, C. M., Strober, M., Plotnicov, K., Pollice, C., Rao, R., Merikangas, K. R., Nagy, L. & Kaye, W. H. (2000). Personality traits among currently eating disordered, recovered, and never-ill first-degree female relatives of bulimic and control women. Psychological Medicine 30, 1399 1410. Mazure, C. M., Halmi, K. A., Sunday, S. R., Romano, S. J. & Einhorn, A. M. (1994). The Yale Brown Cornell Eating Disorder Scale: development, use, reliability and validity. Journal of Psychiatric Research 28, 425 445. O Dwyer, A. M., Lucey, J. V. & Russell, G. F. M. (1996). Serotonin activity in anorexia nervosa after long-term weight restoration: response to D-fenfluramine challenge. Psychological Medicine 26, 353 359. Palmer, H. & Jones, M. (1939). Anorexia nervosa, the manifestation of compulsive neurosis: a study of psychogenic factors. Archives of Neurology and Psychiatry 41, 856 860. Pato, M. T., Eisen, J. & Pato, C. N. (1994). Rating scales for obsessive compulsive behaviour. In Current Insights in Obsessive Compulsive Disorder (ed. E. Hollander, J. Zohar, D. Marassati and B. Olivier), pp. 77 91. John Wiley and Sons: New York. Pollice, C., Kaye, W. H., Greeno, C. G. & Weltzin, T. E. (1997). Relationship of depression, anxiety and obsessionality to states of illness in anorexia nervosa. International Journal of Eating Disorders 21, 367 376. Sohlberg, S. & Strober, M. (1994). Personality in anorexia nervosa: an update and theoretical integration. Acta Psychiatrica Scandinavia 89, 1 15. Spielberger, C. D., Gorsuch, R. L. & Lushene, R. E. (1970). STAI Manual for the State Trait Anxiety Inventory. Consulting Psychologists Press: Palo Alto. Srinivasagam, N. M., Plotnicov, K. H., Greeno, C., Weltzin, T. E., Rao, R. & Kaye, W. H. (1995). Persistent perfectionism, symmetry, and exactness in anorexia nervosa after long-term recovery. American Journal of Psychiatry 152, 1630 1634. Strober, M. (1980). Personality and symptomatological features in young, nonchronic anorexia nervosa patients. Journal of Psychosomatic Research 24, 353 359. Strober, M. (1981). The significance of bulimia in juvenile anorexia nervosa: an exploration of possible etiological factors. International Journal of Eating Disorders 1, 28 43. Strober, M. (1991). Family-genetic studies of eating disorders. Journal of Clinical Psychiatry 52, 9 12. Strober, M. (1995). Family genetic perspectives on anorexia nervosa and bulimia nervosa. In Eating Disorders and Obesity: A Comprehensive Handbook. (ed. K. E. Brownell and C. G. Fairburn), pp. 212 218. The Guilford Press: New York. Strober, M., Freeman, R. & Morrell, W. (1997). The long-term course of severe anorexia nervosa in adolescents: survival analysis of recovery, relapse, and outcome predictors over 10 15 years in a prospective study. International Journal of Eating Disorders 22, 339 360. Sunday, S. R., Halmi, K. A. & Einhorn, A. (1995). The Yale Brown Cornell Eating Disorder Scale: a new scale to assess eating disorder symptomatology. International Journal of Eating Disorders 18, 237 245. Tellegen, A., Lykken, D. T., Bouchard, T. J., Wilcox, K. S., Segal, N. L. & Rich, S. (1988). Personality similarity in twins reared apart and together. Journal of Personality and Social Psychology 54, 1031 1039. Venables, W. N. & Ripley, B. D. (1994). Modern Applied Statistics with S-Plus. Springer-Verlag: New York. Vitousek, K. & Manke, F. (1994). Personality variables and disorders in anorexia nervosa and bulimia nervosa. Journal of Abnormal Psychology 1, 137 147. Von Ranson, K. M., Kaye, W. H., Weltzin, T. E., Rao, R. & Matsunaga, H. (1999). Obsessive compulsive disorder symptoms before and after recovery from bulimia nervosa. American Journal of Psychiatry 156, 1703 1708. Waller, D. A., Gullion, C. M., Petty, F., Hardy, B. W., Murdock, M. V. & Rush, A. J. (1993). Temperament and Personality Questionnaire and serotonin in bulimia nervosa. Psychiatry Research 48, 9 15. Walters, E. E. & Kendler, K. S. (1995). Anorexia nervosa and anorexic-like syndromes in a population-based twin sample. American Journal of Psychiatry 152, 64 71.