Patterns of covariation of DSM-IV personality disorders in a mixed psychiatric sample

Patterns of covariation of DSM-IV personality disorders in a mixed psychiatric sample

Patterns Andrea of Covariation of DSM-IV Mixed Psychiatric Fossati, Cesare Maffei, Monica Maria Bagnato, Marco Fiorilli. Liliana Novella, Person...

926KB Sizes 66 Downloads 65 Views

Patterns Andrea

of Covariation of DSM-IV Mixed Psychiatric

Fossati,

Cesare

Maffei, Monica

Maria Bagnato, Marco Fiorilli. Liliana Novella,

Personality Sample

Disorders

Battaglia, Deborah Donati, and Federico Prolo

Michela

in a Donini,

The covariation patterns of DSM-IV personality disorders (PDs) were studied in 431 consecutively admitted psychiatric patients. The co-occurrence rate was greater than 50% for all DSM-IV PDs. Both bivariate association tests and loglinear models showed distinct significant covariation patterns among PDs which were stable across confounder strata. DSM-IV PD clusters were not replicated, with the exception of cluster A. Principal-component analysis (PCA) showed

the presence of 3 latent dimensions, thus explaining the DSM-IV PD covariation patterns. These results seem to stress the inadequacy of the DSM-IV categorical model of PD assessment. The need for a reduction of axis II categories and the inclusion of a dimensional model in the diagnostic assessment of DSM-IV PDs are discussed. Copyright 0 2000 by W.B. Saunders Company

ONSISTENT DATA from the literature show that at least 50% of the patients diagnosed as having any type of personality disorder (PD) receive 2 or more PD diagnoses.14 Moreover, the use of structured interviews for PD diagnosis increased the rate of PD co-occurrence when compared with clinical interview or chart review.’ This finding is frequently referred to as “comorbidity,” that is. the coexistence of 2 or more independent disorders.“,” However, interpreting the co-occurrence of PDs as the co-presence of independent disorders seems problematic. In fact. previous studies on the co-occurrence of PD diagnoses’,3.5-13 showed significant associations between several PDs. A substantial variation in the size and direction of PD covariation was observed across studies, mainly because of their methodological heterogeneity.“ Moreover, co-occurrence, as well as covariation, depends on the prevalence of the respective PDs. This could be influenced by several confounders such as differences in the diagnostic threshold. method of assessment patient severity,” subject gender,‘” and diagnostic system. Furthermore, it could occur in several ways, such as by including overlapping criteria, emphasizing multiple diagnoses rather than differential diagnosis, or demarcating different categories along a shared spectrum of pathology. I4 The stability of PD covariation pat-

terns across these confounders should be tested before making any conclusion about their generalizability. Unfortunately. only a few studies3-s have tried to assesstheir effect. Despite these methodological problems. the evidence of significant covariation among PDs raised doubts about the validity and clinical usefulness of the DSM-III-R categorical model of PDs. Some author9 suggested that the categorical diagnostic system of PDs could be maintained with several deep modifications. ranging from the elimination of overlapping criteria to the collapsing of present categories into superordinate clusters on the basis of statistically based hierarchies. On the contrary, other authors”-” claimed that the significant covariation observed among PDs could be explained by the presence of common underlying personality dimensions, and suggested that a dimensional model would be more appropriate. A number of studies7.8,“.‘?,‘fi were performed to identify the dimensions underlying the covariation of PDs. Unfortunately. these studies did not provide consistent results, perhaps due to substantial methodological variabi1ity.J No definitive evidence was found for DSM-III-R clusters; rather, some studies showed strong similarities between some of the dimensions underlying DSM-III PDs and those identified by the 5-factor personality model.‘.” With regard to DSM-III-R, DSM-IV’” made a noticeable effort to reduce the overlap between PD criteria and sharpen the boundaries between individual PDs. Diagnostic thresholds were modified for several PDs, sadistic PD was removed, and depressive and passive-aggressive (negativistic) PDs were included as diagnostic categories necessitating further study. All of these modifications. which could poten-

C

From

fhr

Facrtlfy of Milarr: Urhirta.

Clinical Psyhalog?:

aad rhe lastirate Ifa!\:

Address Psicologia

fs.~ho/agy artd P.syho/hrrapy Uaif. “Kra Salaf~” Sari Rafj+&le Uai~vr.G!\:

of Pswholagy~

rqwirrr reyrresrs 10 Cesare Cliaica c Psicotc~rapia.

Uaivrrsify Ma&i. Isfifalo

M.D., Sen*i:io di Scier~lQk~ Sarr

Raffaele-DSNP. via Luigi Prirwrti. 29. 20127 Milano. Capyrighr 0 2000 by W B. Sartader.~ Caatpar~~ 0010-440x/00/4103-0012%10.00/0

206

of Urhirm.

Iraly.

ComprehensivePsychiatry,

Vol.41,

No.3

(MaylJune),2000:

pp 206-215

PAlTERNS

OF COVARIANCE

OF DSM-IV

PDs

207

tially deeply influence the PD base rate and phenomenology, stress the need for a reassessment of PD covariation. Starting from these considerations, the aims of this study were to (I) analyze the patterns of covariation of DSM-IV PDs in a mixed psychiatric sample. (2) evaluate the potential confounding effect of subject gender and severity (i.e., male I’ female, presence r absence of any axis I diagnosis, and inpatient I’ outpatient), and (3) identify the dimensions underlying the covariation patterns of DSM-IV PDs and test their replicability across the confounder levels.

informant, treatment response. etc.) were used in this study. Subjects with axis I diagnoses were administered the SCLD-II at acute

symptom

remission.

according

The sample

consisted

from January Psychotherapy

of 43 I subjects

consecutively

admitted

to October 1997 to the Clinical Unit of the Scientific Institute

Psychology and H. San Raffaele.

Milan. Italy, a specialized unit in the diagnosis PDs. The exclusion criteria were as follows:

and treatment of IQ of 75 or less:

axis I diagnosis of schizophrenia. delusional disorder. or organic mental level less than formed consent

schizoaffective disorder. disorder: and educational

elementary school. All subjects provided into participate in the study after a detailed

description. One hundred sixty-two and 269 (62.4% l were female. The

subjects (37.6%) were male mean age was 29.X I f 8.34

years (mean + SD). Three hundred (75.9%) were unmarried. 72 (16.6%)

twenty-seven were married,

were

widowed.

divorced.

and

thirteen subjects outpatients. Two

2 (OSr/r)

(49.4%) were hundred sixty-two

least I axis I diagnosis; disorders were anxiety disorders substance

ders. tic disorder. age of subjects

inpatients subjects

the most disorders

(n = 70. 16.2%). abuse/dependence

hrief7NOS psychotic jects (4.9%) received

frequency diagnosis

were

Two

hundred

and 21X (50.6%) (60.8%) received at

frequently diagnosed (n = 120. 27.X%).

axis 1 eating

mood disorders (n = 33, 7.7%). disorders (n = 19. 4.4%). and

disorder (n = 10. 2.3%). Twenty-one other axis I diagnoses (e.g., sleep

etc.). The cumulative frequency with specifc axis I diagnoses

and percentage of subjects because of multiple diagnoses.

ders were clinically diagnosed the subjects in treatment. blind base tate of several confounding effect

subjects 30 (7.0%)

suhdisor-

and percentexceeded the

with at least I axis I DSM-IV axis I disor-

by the clinicians who to the axis II diagnosis.

evaluated The low

axis I diagnoses prevented the analysis of the of specifc axis I disorders on DSM-IV PD

covariation patterns. The relatively high base rate of anxiety disorder and eating disorder diagnoses. as well as the low hase

with

the

exception

of

depressive

PD

correlation

coefficient

for dimensional

The

hivariate

DSM-IV greater

association

between

PDs was assessed using than 1.0 shows a positive

ables. whereas an OR less than tion. In the case of independence.

report

screening

questionnaire. available

sources

To increase of information

diagnostic (e.g..

validity. chart data.

categorically

more This

is

diagnosed An OR 2 vari-

I .O indicates negative associathe OR equals I .O. Whenever

table had zero frequency,

a 0.5 constant

was added to avoid OR undefinition.‘-’ The Yates-corrected chi-square (xl) test was used to test the hypothesis that the ORs were

significantly

different

from

I .O.?j

Hierarchical log-linear models were used to identify the specific PD association patterns needed to adequately reproduce the matrix

of PD observed

frequency.

The

independence

was chosen as a baseline model. The following then entered in the model in successive including

(i.e., Bonfetroni-corrected nominal significance = .000X): (2) 2-way interactions with P less than

.Ol: (3) 2-way interactions interactions with P less interaction

model,

based

PD associations

were 2-way

than level,

.000X .05/66

hivariate

model

interactions steps: (I)

interactions.

with

P less

with P less than ,025: (4) 2-way than .05: and (5) 3- and 4-way

on DSM-IV

clusters

(cluster

A, B. and

C). Considering the exploratory nature of these log-linear analyses. the model selection was based on the significance of the likelihood-ratio ing models rion (AK)

x? statistic

(G?) difference

between

goodness-of-fit

of

the

best

model

was

tested

using

statistic. A nonsignificant G? value shows that adequately reproduced the observed frequencies.” OR

homogeneity

across

sample

gender and severity was tested. low power of the homogeneity fluctuation

compet-

and minimization of the Akaike information and Bayesian information criterion (B1Cl.z

of the significance

strata

defined

criteThe the

the

Gz

model

by subject

To find a balance between the xz test?’ and the excessive level

due to the large

number

for each confounder. an extremely liberal

(P = .20/66 homogeneity?’

= ,003). The use of log-linear models to test OR was prevented by the excessive cell sparseness

observed in several Principal-component latent

dimensions

Bonferroni correction nominal significance

of

comparisons applied to

confounder strata. analysis (PCA) explainin,

This

procedure

was justified

between Pearson (i.e.. the correlation

>.20). (median

Wilcoxon

by the almost

PD correlations

with

regard

the PDs.

r) matrix of DSM-IV PDs.

perfect

agreement

r and coefficient phi matrices of dimensionally

correlation assessed

matched-pairs

On the contrary, I’ = -.67) clearly

of DSM-IV

(Pearson of traits)

and categorically scored DSM-IV PDs. median I’ = -.05; correlation between P < ,001.

was level

was used to identify

0 the covariation

observed matrices

all additional

were

the odds ratio (OR). association between

respectively.

SCID-II is a I-lo-item (organized by diagnosis) interview designed to diagnose the 12 DSM-IV was preceded hy the administration of its self-

diagnosis

assessment)

PCA was applied to the correlation dimensionally assessed (i.e.. number

(SCID-II).?“The semistructured PDs. SCID-II

categorical

than .X0 for categorical and dimensional PD diagnoses. consistent with previously published data.”

rate of mood disorders. observed in this sample could he explained by the presence in our hospital of 2 large divisions specializing in the treatment of anxiety and eating disorders. All subjects were administered the Structured Clinical Interview for DSM-IV Axis II Personality Disorders. Version 2.0

of the

(Cohen K = .6X). all other joint-interview interrater reliability coefftcients (Cohen K for categorical diagnosis and intraclass

a cell in the contingency

METHOD

to the judgment

treating clinician. by expert trained raters to avoid confounding effects of axis I disorders on axis II diagnoses.“’ In this study,

median phi = -04, matrices, r = .97,

test = - I. 1 I. P (2-tailed)

tetrdchoric correlation coefficients overestimated the size of DSM-IV to phi (Wilcoxon

test = -4.88,

P

FOSSATI

(2-tailed) (2-tailed)

little theoretical justification in using tetrachoric correlation coefficients because of the existence of a directly measured continuous variable (i.e., number of traits) underlying each categorically scored DSM-IV as a rule for determining

PD. Parallel analysis?‘~?” was used the number of components to be

retained. The parallel-analysis values of a correlation matrix

technique requires that the eigenof uncorrelated random variables

are contrasted with those of the real data set, based on the same sample size and number of variables. The components of the matrix

of interest

that have eigenvalues

greater

than those of the

comparison random matrix would be retained. This method is based on the principle that a researcher would not be interested in a principal component that does not account for a variance greater than the corresponding from the distribution of random showed

that

parallel

analysis

principal data.zh-z8

component obtained Monte Carlo studies

is one of the best

methods

for

determining the number of components to retain, particuharly in association with the scree test.?” In this study. several (N = 60) identity correlation matrices were computed from normally distributed observed

random variables.

over the 60 random and those obtained were plotted, the number normal DSM-IV

numbers with the same mean 2 SD as the Values for each latent root were averaged data sets.?“,?’ These averaged eigenvalues from the DSM-IV PD correlation matrix

and the point at which the curves cross indicates of principal components.z6.z7 The marked non-

distribution observed PDs (Shapiro-Wilks

for all dimensionally assessed W statistic, schizoid PD = 24 to

narcissistic PD = .S7. all P < ,001) prevented us from using the Bartlett x? test.z9 The Kaiser rule (i.e.. eigenvalue > 1.0) was not used particular,

in this study because several Monte Carlo

mates consistently tained.z*.30,3t Both

Table

and Pearson r (Wilcoxon test = -5.05. P correlation coefficients. Moreover, there was

c.001) <.OOl)

it lacks statistical validity.z9 In studies showed that it overesti-

the number of components to be reorthogonal (i.e., varimax) and oblique (i.e..

direct oblimin) rotations were used to stabilize the extracted principal components. and their congruence was formally tested. in case of low correlations among the oblique components large congruence with the orthogonal rotation, the latter varimax) was retained as the final solution. The replicability the final principal-component

solution

across

was tested using the congruence coefficient r. The CC ranges from I.0 (when the factor reproduced) to - I .O (when opposite tained), with 0.0 showing completely

confounder

(CC)” structure

and (i.e., of strata

and Pearson is perfectly

factor patterns are obindependent factor solu-

tions.

RESULTS According to the SCID-II, 3 10 subjects (7 1.9%) received at least 1 DSM-IV PD diagnosis; the mean number of PD diagnoses was 1.16 + 1.OO.DSM-IV PD descriptive statistics are shown in Table 1. A high prevalence of cluster B PDs was observed (namely, narcissistic and borderline PDs). The rate of co-occurrence was high (i.e., >50.0%) for all DSM-IV PDs. The 5 DSM-IV PDs with the highest co-occurrence rate were, in increasing order, antisocial, depressive, histrionic, passive-aggressive, and schizoid PDs. In particular, schizoid PD always

1. DSM-IV

PD Descriptive

ET AL

Statistics

Co-Occurring PD Diagnoses No.

%

Mean c SD

%

NO.

No. of Co-Diagnoses (mean z SD)

APD

22

DPD OCPD

13

5.1 3.0

0.46 2 1.27 0.41 -t 1.11

59.1 61.5

13 8

0.68 -t 0.65 0.77 -t 0.72

22 53 14

5.1 12.3

0.47 2 1.22 0.89 2 1.65

54.5 90.6

12 48

0.64 2 0.66 1.49 5 0.97

3.3

0.34 f 1.08

78.6

11

27 20 5

6.3 4.6 1.2

0.44 -c 1.28 0.48 2 1.55 0.14 2 0.66

66.7 60.0 100.0

18 12 5

0.86 2 0.53 1.04 t 1.02

PD

PAPD DEPD PPD SZPD SPD HPD

59 154 97

13.7

1.13 2 1.97

81.4

48

NPD BPD

35.7 22.5

2.72 2 2.80 1.82 r 2.80

59.7 52.6

92 51

ASPD

20

4.6

0.32 2 1.21

75.0

15

NOTE.

The cumulative

frequency

0.85 2 0.81 1.40 ? 0.55 1.34 z 0.92 0.88 2 0.92 0.96 -t 1.09 1.63 ? 1.16

and percentage

of subjects

with specific DSM-IV PD diagnoses exceeded the frequency and percentage of subjects with at least 1 DSM-IV PD diagnosis because of multiple axis II diagnoses. Abbreviations (for all Tables): APD, avoidant dent PD; aggressive schizotypal narcissistic

PD; DPD, depen-

OCPD, obsessive-compulsive PD; PAPD, passivePD; DEPD, depressive PD; PPD, paranoid PD; SZPD, PD; SPD, schizoid PD; BPD, borderline

PD; HPD, histrionic PD; ASPD, antisocial

PD; NPD, PD.

co-occurred with other PDs. The confounder effects on DSM-IV PD base rates are shown in Table 2. Female subjects showed a significantly higher base rate of histrionic PD than male subjects; on the contrary, paranoid, narcissistic, and antisocial PDs were significantly prevalent among male subjects. The borderline PD base rate was significantly higher in inpatients than in outpatients. The overall rate of DSM-IV PD diagnoses was significantly higher among male subjects, inpatients, and subjects with a co-occurring axis I diagnosis. On average, inpatients and male subjects received a significantly higher number of DSM-IV PD diagnoses. A trend (P = .051) for statistical signilicance was observed in the association between the number of PD diagnoses and the presence of an axis I diagnosis. DSM-IV PD co-occurrence descriptive statistics are shown in Table 3. As expected. given the differences in DSM-IV PD base rates, several asymmetries in PD co-occurrence rates were observed. For instance, 70% of the subjects who met criteria for a DSM-IV antisocial PD diagnosis also received a borderline PD codiagnosis, whereas only 14% of subjects with a borderline PD diagnosis received an antisocial PD co-diagnosis. Other striking asymmetries in the overlap rates were observed in the following PD co-occurrence patterns: narcissistic PD/passive-

PAlTERNS

OF COVARIANCE

OF DSM-IV

PDs

209

Table 2. DSM-IV

PD

Male

Female

(n = 162)

(n = 269)

NO.

PD: Confounder

Effects Outpatients In = 218)

lnparients

(n = 213)

No Axis I

%

NO.

%

NO.

%

10

6.2

12

14

6.6

3 11

1.9 6.8

10 11

4.5 3.7

9

4.2

4

25 3

15.4 1.9

28 11

4.1 10.4 4.1

12 28 6

5.6 13.1 2.8

10 25 8

18 11

11.1 6.8

9 9

3.3’

18 14

8.5 6.6

9 6

2

1.2

4

1.9

1

14 73 29

8.6 45.1 17.9

3 45

34 76 57

16.0 35.7 26.8

25 78 40

11.5 35.8 18.3#

ASPD Any PD

17 130

10.5 80.2

8 173

3.8 81.2

12 137

5.5 62.8”

Mean

1.321 (1.001)

APD DPD OCPD PAPD DEPD PPD SZPD SPD HPD NPD BPD

NOTE.

no. of PDs (SD) The

cumulative

frequency

and

81 68 3 180

3.3 1.1 16.7t 30.1* 25.3 1.11

66.911 1.071 (0.996)n percentage

of subjects

percentage of subjects with at least 1 DSM-IV PD diagnosis *Yates-corrected x2 = 9.103, df = 1, P c ,005. tYates-corrected *Yates-corrected

x2 = 4.933, df = 1, P < .03. x2 = 9.201, df = 1, P < ,005.

§Yates-corrected [[Yates-corrected

x2 = 18.034, df= 1, P-C ,001. ~2 = 8.252, df = 1, P c ,005.

nt = 2.52, df= #Yates-corrected

1.324 (0.983)

because

with

specific

of multiple

Any Axis I In = 2621

(n = 169)

No.

%

No.

%

8

3.7 1.8

5 4

3.0 2.4

17 9

6.5 3.4

4.6 11.5

6 17

3.6 10.1

16 36

6.1 13.7

3.7 4.1

6 7

3.6 4.2

8 20

2.8 0.5

5 1

3.0 0.6

15 4

7.6 5.7 1.5

19 63

11.2 37.3

40 91

15.3 34.7

37 11

21.9 6.5

60 9

22.9 3.4

1.009 (1.002)tt DSM-IV

No.

%

111 65.7 1.047 (1.034)

PD diagnoses

exceeded

3.1

199 76.0$$ 1.241 (0.978)§§ the frequency

and

axis II diagnoses.

429, P(2-tailed) < .02. x2 = 3.904, df = 1, P-C .05.

**Yates-corrected x2 = 17.119, ttr = 3.29, df = 429, P (2-tailed)

df= 1, PC ,001. < .005.

**Yates-corrected x2 = 4.873. df = 1, P.03. §§r = -1.96, df = 429, P (2-tailed) < .06.

aggressive PD (28% 1’ 81%), narcissistic PD/ histrionic PD (26% 1’ 68%), and schizotypal PD/ schizoid PD (25% 11 100%). In particular, in this study, schizoid PD always co-occurred with schizotypal PD, even if only a minority of subjects with a schizotypal PD diagnosis showed a schizoid PD co-diagnosis. The ORs are shown in Table 4. Among 66 comparisons, 17 (25.8%) were statistically significant. Ten ORs remained significant even at the Bonferroni-corrected significance level (.05/ 66 = .0008). Among the 17 significant ORs, 7 (41.2%) showed negative PD covariations. Significant negative associations were observed between narcissistic PD and, respectively, avoidant, dependent, depressive, paranoid, and schizotypal PDs. Borderline PD showed negative associations with avoidant and obsessive-compulsive PDs. The most significant positive associations were observed between the following PDs: avoidant/dependent, dependent/depressive, avoidant/depressive, paranoid/ schizotypal. schizotypalkchizoid, passivehistrionic/narcissistic, aggressive/narcissistic,

histrionic/borderline, and borderline/antisocial. No significant OR difference was observed between male and female subjects (homogeneity $(df = 1) = 0.00 to 3.65, all P > .003), inpatients and outpatients (homogeneity x*(d = 1) = 0.00 to 5.01, all P > ,003). and subjects without any axis I diagnosis and subjects with at least 1 axis I diagnosis (homogeneity x”(d’= 1) = 0.00 to 3.39, all P > .003). Log-linear models are shown in Table 5. With the exception of schizoid PD/paranoid PD and passive-aggressive PD/antisocial PD interactions, all other bivariate interactions between DSM-IV PDs significantly improved the model. In fact, the best-fitting model included all bivariate associations with P less than .025 and adequately reproduced the observed frequencies (G* = 168.29, cif = 4,066. P > .90). Entering the 4-way interaction model (paranoid PDkchizotypal PD/schizoid PD; histrionic PD/narcissistic PD/borderline PDI antisocial PD; avoidant PD/dependent PD/obsessive-compulsive PD, passive-aggressive and depressive PDs independent), based on DSM-IV PD

FOSSATI

210

Table 3. Co-Occurrence APD

APD

of DSM-IV

PDs: Descriptive

Statistics

DPD

OCPD

PAPD

DEPD

PPD

SZPD

SPD

HPD

NPD

BPD

4

4

0

5

2

0

0

0

0

0

0

(30.8)

(18.2)

(0.0) 0

(35.7) 5

(7.4) 0

Km 0

(0.0) 0

W) 0

lO.0) 0

Pm 0

(0.0) 0

(4.5)

WO) 0

(35.7) 0

(0.0) 4

(0.0) 0

(0.0) 0

(0.0) 0

(0.0) 4

(0.0) 0

(0.0) 0

(0.0)

(0.0) 0

(14.8) 5

Kw

(0.0) 0

(0.0) 11

(2.6) 43

(0.0) 6

Pw

(18.5) 0

(5.0) 0

(0.0) 0

(18.6) 0

(27.9) 0

(0.0) 14 (14.4)

(0.0)

(0.0) 8

(0.0) 2 (40.0)

(0.0) 0

(0.0) 3

(2.1)

(0.0) 0

(1.9) 1

(3.1)

(5.0) 0

(0.0) 0

(0.6) 0

(1.0) 0

(0.0) 0

VW 40

(0.0) 26

(0.0) 4

(26.0)

(26.8) 37 (38.1)

(20.0) 10

DPD

4

OCPD

(18.2) 4

1

PAPD

(18.2) 0

(7.7) 0

DEPD

VW 5

W) 5

(0.0) 0

0

PPD

(22.7) 2

(38.5) 0

(0.0) 4

wJ) 5

0

(18.2) 0

(9.4) 1

(0.0) 0

0

(40.0)

SZPD

(9.1) 0

uw 0

SPD

W) 0

(0.0) 0

(0.0) 0

(1.9) 0

(0.0) 0

HPD

(0.0) 0

(0.0) 0

Kw 0

(0.0) 11

(0.0) 0

(7.4) 0

(25.0) 0

0

NPD

(0.0) 0

(0.0) 0

(25.6) 43

(0.0) 0

Km) 0

40

(81.1)

0

00) 0

(67.8) 26

37

ASPD

(0.0) 0

(0.0) 0

(0.0) 0

14 (26.3) 6

(0.0) 2 (14.3) 0

(5.0)

BPD

Km 0

(0.0) 3 (11.1)

(0.0) 1

WO) 0

(0.0) 4 (18.2)

(26.0) 10

(0.0)

(0.0)

(0.0)

(11.3)

@.O)

NOTE. Values column PD with BPD, whereas

8 (29.6)

14.4% (14 of 97) of subjects

with

5 (100)

2

5

3 (11.1)

are presented as the number (and percentage). the corresponding row PDs. The table should

(5.0) 0

(0.0) 0

(44.1) 4

(3.7)

(0.0)

(0.0)

(6.8)

APD

9.885

DPD

Associations

of DSM-IV

PAPD

DEPD

PPD

SZPD

4.83t

0.15

13.075

1.54

0.42

1.58

0.25

28.409

0.53

0.72

0.15

0.61 0.23

3.73 1.69

0.98 0.36

0.49

0.67 13.755

SPD

HPD

1.63 2.78 1.63

0.13 0.22 0.13

0.63 2.59

1.80

10.69* 292.05

NPD

BPD ASPD x2 (df = 1) was used to test the hypothesis

§P<

.0008 (Bonferroni-corrected

nominal

(70.0) 14 (14.4) of each also had

significance

level).

that the DB = 1.

BPD

ASPD

0.045 0.06t

0.07t

0.42

0.12

0.72

0.38 10.345

0.07t 1.28

0.42 3.32*

0.56 0.41 0.17

0.67 0.78 0.47

0.31

1.81

3.345 1.14

1.62 1.85 9.222

0.21 0.11

0.06t 0.21 t

0.14 0.56

0.09* 0.16 4.761

.025. .Ol.

(50.0) 14

PDs: OR

SPD HPD NPD

tP< SP<

(0.0)

to adequately reproduce the matrix of DSM-IV PD observed frequencies. However, no direct evidence for DSM-IV PD clusters was found. Dimensionally assessed DSM-IV PD correlation coefficients (Pearson 1.)are listed in Table 6. Parallel-analysis results are shown in Fig I. Even if 5 eigenvalues were greater than 1.0, only the first

PPD SZPD

NOTE. Yates-corrected ‘PC .05.

(30.0) 0

3

Column percentages are the percentage of co-occurrence be read as follows: 70.0% (14 of 20) of subjects with ASPD

OCPD

OCPD PAPD DEPD

6.5)

2

ASPD

BPD also had ASPD.

Table 4. Bivariate DPD

(0.0)

1

clusters, did not significantly improve the G? statistic and produced a worsening of the fit function according to AIC and BIC. In other words, according to log-linear analysis results, DSM-IV PDs could hardly be considered as independent categories (baseline model). In fact, almost all bivariate associations (both positive and negative) are needed

APD

ET AL

PAlTERNS

OF COVARIANCE

OF DSM-IV

PDs

Table

211

5. Covariation

Condition

of DSM-IV

PDs: Log-Linear

G’ldO

404.07

(4,083)

2-way 2-way

interactions* interactionst

220.58 209.27

(4,073) (4,072)

2-way

interactionsSfl

168.29

(4,066)

32.98

2-way 4-wav

interactions5 interactions

164.60 149.35

(4,064) (4,053)

3.69 (2) 15.25 (11)

(cluster

A, C, and B)ll

‘All bivariate independent.

associations

with

tAll bivariate independent.

associations

with

P < .Ol are

*All bivariate independent.

associations

with

P < ,025 are entered

§All bivariate independent.

associations

with

P < .05 are

II(Paranoid PD schizotypal PD. obsessive-compulsive VBest-fitting

P < .0008

are entered

as 2-way

interactions interactions

as 2-way

P c ,001 P < ,001

(6)

interactions

as 2-way

entered

AIC

183.49 (11) 11.31 (1)

as 2-way

entered

Models

G7 Diff. (dr)

interactions

BIG

2.849.19

2.902.05

2.687.70 2.676.98

2.785.29 2.676.39

P < ,001

2.647.41

2.769.39

P> P>

2.647.72 2.654.47

2,777.84 2.829.31

.lO .lO

in the model;

the remaining

PDs are considered PDs

in the

model;

the

in the

model;

the remaining

PDs are considered

in the

model;

the

PDs

PD schizoid PD), (histrionic PD narcissistic PD borderline PD), passive-aggressive PD. depressive PD. (. = interaction).

PD

antisocial

remaining

remaining

PD), (avoidant

are considered

are considered PD

dependent

model

3 were superior to the eigenvalues obtained from random data. After oblique rotation, only trivial correlations were observed among the 3 extracted principal components (.09 to -.22). An almost perfect congruence was observed between principalcomponent loading matrices obtained, respectively, from oblique and orthogonal rotations (CC = .992. I’ = ,992, P < .OOl). Given these results, the orthogonal-rotated solution was retained. Principalcomponent loadings obtained after varimax rotation are listed in Table 7. Only substantial (i.e., >.30 in absolute value) loadings are displayed. The tirst principal component was bipolar, contrasting PDs characterized by insecurity and clinging behavior (dependent PD), pathological shyness and social anxiety (avoidant PD). and low self-esteem (depressive PD) with PDs characterized by grandiosity and arrogance (narcissistic PD). attentionseeking behavior (histrionic PD), and passive resisTable APD

DPD

DCPD

APD DPD

1 .ooo ,457

1.ooo

OCPD PAPD

,137 -.164

,023 -.178

1.000 -.134

DEPD

.366

,440

PPD SZPD SPD

,038 -.014 ,055

-.083 p.065 - ,037

,001 ,107

HPD NPD

-.201 - ,277

-.122 - ,272

BPD ASPD

-.156 - ,067

- ,097 - ,078

- ,022 ,030 -.139 -.121 -.174 - .084

6. DSM-IV PAPD

1.000 -.151 ,046 -.074 - ,072 ,133 ,434 ,078 ,145

tance to other’s normal requests (passive-aggressive PD). Paranoid, schizotypal, and schizoid PDs showed their highest loadings on the second component, which closely reproduced the DSM-IV cluster A. The third component seemed to contrast PDs characterized by marked instability (in mood, affect, relationships. and identity), impulsiveness, emotional discontrol (borderline PD), antisocial behavior, irritability, failure to plan ahead, irresponsibility, and lack of remorse (antisocial PD) with obsessive-compulsive PD. characterized by rigidity, overconscientiousness, excessive devotion to work, and excessive perfectionism. The factor loading matrix could be safely reproduced across all confounder strata: (1) male versus female subjects (CC = .900, I’ = .902, P < .OOl); (2) inpatients versus outpatients (CC = .952, r = .952, P < .OO1); and (3) presenceversus absenceof at least 1 axis I diagnosis (CC = ,946, r = ,947, P < .OOl).

PDs: Correlation DEPD

Matrix

PPD

SZPD

- .042 - .056

1 .ooo ,482

1.000

- ,028 -.173

.255 -.158

- ,270 -.092

-.178 -.038

(Pearson

r) NPD

BPD

,362 ,202

1.000 ,054

1.000

,023

,117

,265

SPD

HPD

,684 -.129

1.000 -.115

1.000

- ,202 - .058

-.139 - ,086

- ,046

-.046

ASPD

1.000

-.057

,005

1.000

212

FOSSATI

6

8

ET AL

10

Eigenvalues Fig 1.

DSM-IV

PD PCA. Real data (0); random

DISCUSSION

In agreement with previous findings.‘.? the cooccurrence rate was high for all individual PDs and multiple diagnoses were the rule rather than the exception. Both bivariate association tests and log-linear models failed to support the DSM-IV assumption that PDs are discrete, independent diagnostic categories. Interestingly, bivariate associations between DSM-IV PDs were stable across confounder strata, even if the base rate of several PDs was influenced by subject gender and severity. The replicability of these patterns of covariation seems to show that they were not simple artifacts due to PD prevalence and sample composition. Table 7. DSM-IV

PD PCA: Factor

Loadings

PC1

APD

,679

DPD OCPD

,749

PAPD DEPD

-.436 ,719

PC3

,676 .a98 ,789

SPD HPD

-.392

NPD BPD

-.628 ,714

ASPD Eigenvalue Only

Abbreviations: nent.

rotation)

-576

PPD SZPD

NOTE.

(varimax PC2

(% of sz) loadings

2.59 (21.6)

2.02 (16.9)

> .30 (absolute

s2, explained

variance;

value)

.647 1.25 (10.4)

are displayed.

PC, principal

compo-

data (D).

Several significant bivariate associations (for instance. narcissistic PD with histrionic PD. narcissistic PD with passive-aggressive PD, avoidant PD with dependent PD. paranoid PD with schizotypal PD, and borderline PD with histrionic PD) replicated similar findings of previous studies.5 Other previously reported significant associations such as borderline PD with schizotypal PD’ and avoidant PD with schizotypal PDS were not replicated in this study. This result could be due to several factors; among these, we believe that the modifications introduced by DSM-IV in the diagnostic criteria of borderline, avoidant. and schizotypal PDs played a major role in sharpening the boundaries between these specific disorders. For instance. entering the stress-related paranoid ideation and severe dissociative symptoms among borderline PD criteria helped in differentiating these transient (state) reactions frequently observed in subjects with borderline PD from the lifelong patterns (traits) of suspiciousness, ideas of reference, and perceptual distortion characterizing subjects with schizotypal PD. The reduced risk of state/trait confusion with regard to these characteristics could have dramatically decreased the covariation between these disorders. On the other hand, the DSM-IV emphasis on suspiciousness. rather than fear of being criticized. as a source of discomfort in social situations in subjects with schizotypal PD could have played a role in lessening the covariation between this PD and avoidant PD.

PATTERNS

OF COVARIANCE

OF DSM-IV

PDs

In summary, the modifications entered in DSM-IV PD diagnostic criteria seemed to help in differentiating only some PDs, likely by reducing spurious sources of covariation. However, the majority of PD covariation patterns seemed to be unaffected by the modifications entered in DSM-IV criteria. We agree with the previous suggestion9 that the evidence of significant covariations and asymmetries in the co-occurrence rates observed among DSM-IV PDs does not directly imply a lack of validity of the categorical model of PDs. However, it seems to stress the need for a substantial reduction of PD diagnostic categories. According to our results, for instance, there seems to be little sense in retaining schizoid PD and schizotypal PD as separate diagnoses. In agreement with previous findings,j3 schizoid PD was rarely observed in clinical settings and always cooccurred with schizotypal PD. This result seems to show that schizoid PD is better described as a subtype of schizotypal PD. For instance, a diagnosis of schizotypal PD with prominent withdrawal features could be used instead of the co-diagnosis of schizotypal PD. schizoid PD. However, it should be considered that the presence of an overlapping criterion (no close friends or confidants) could spuriously increase the covariation between these 2 PDs. Moreover, the base rate of schizoid PD observed in this study was too small to draw any definitive conclusion and prevent idiosyncratic findings. An extensive significant asymmetric overlap was observed between narcissistic PD and. respectively, passive-aggressive and histrionic PDs. In fact, roughly 80% of subjects with passive-aggressive PD and 68% of subjects with histrionic PD received an additional diagnosis of narcissistic PD, whereas the rate for subjects with narcissistic PD who received a co-diagnosis of passive-aggressive PD and histrionic PD was, respectively, 28% and 26%. As for the schizotypal/schizoid PD covariation pattern, this finding seems to question the DSM-IV assumption that histrionic and passive-aggressive PDs are diagnostic categories distinct from narcissistic PD. Rather, according to our results. narcissistic PD could be better described as a single diagnostic category articulated in 3 main subtypes. Finally, similar findings were observed also for DSM-IV antisocial PD. In this study, antisocial PD was rarely diagnosed without at least 1 co-

213

occurring PD, and showed a very strong association with borderline PD. These data seem to confirm previous findings showing significant associations between antisocial traits and impulsive personality profiles.3J On the other hand, this result seems to emphasize that DSM-IV antisocial PD could be better described as a subtype of a broader diagnostic category of impulsive/aggressive personality, which should also include borderline PD. With regard to previous studies,5 a larger number of substantial significant negative covariations were observed among DSM-IV PDs. The DSM-IV attempt to sharpen the diagnostic boundaries between PDs did not, in these cases, increase independence. Rather, the negative covariations seem to show that negatively associated PDs are at the opposite extremes of some latent personality dimension. PCA showed the existence of 3 independent, nonrandom latent dimensions that are stable across different sample subgroups and rotation methods, which explained the covariations observed among DSM-IV PDs. On one hand. this result seems to suggest that the individual DSM-IV PDs could be included as subtypes in hierarchical higher-order diagnostic categories, corresponding to principal-component polarities. Interestingly, and contrary to the possible bias related to the SCID-II format, PCA (in agreement with log-linear model results) did not support the DSM-IV clustering of PDs. In fact, only DSM-IV cluster A was clearly reproduced by the second principal component. DSM-IV cluster B PDs seemed to belong to different latent dimensions. Narcissistic and histrionic PDs (together with passive-aggressive PD) formed the negative pole of the first principal component, whereas borderline and antisocial PDs were at the positive pole of the third principal component. The same result was observed for DSM-IV cluster C PDs. Avoidant and dependent PDs (together with depressive PD) were at the positive pole of the first principal component, whereas obsessive-compulsive PD formed the negative extreme of the third principal component. On the other hand, the PCA results seem to stress the need for the introduction of a dimensional model of personality description, complementary to the DSM-IV categorical PD diagnoses. In particular, the first principal component identified a personality dimension positively related to fearful, shy,

214

FOSSATI

dependent, pessimistic PD features (avoidant, dependent, and depressive PDs) and negatively related to grandiose, attention-seeking, negativistic PD traits (narcissistic, histrionic, and passiveaggressive PDs). The second principal component clustered PDs mainly characterized by aloofness, social withdrawal, and suspiciousness (paranoid, schizotypal, and schizoid PDs). Finally, the third principal component seemed to contrast impulsiveness, irritability, failure to plan ahead, and instability in several areas of psychosocial functioning (e.g.. self-image, relationships, long-term goals, etc.; borderline and antisocial PDs) with excessive meticulousness, moralistic attitudes, and rigidity (obsessive-compulsive PD). The DSM-IV model of PDs is not adequate for explaining and describing these latent dimensions. Rather, the introduction of a complementary dimensional model could result in a better understanding of PD covariation patterns. For instance, the first principal component could be described as an extreme variant of extraversion, according to the 5-factor model, or harm avoidance and cooperativeness, according to the Cloninger model of personality.36 The second principal component could be identified as low agreeablenes@ or low reward dependence and cooperativeness.36 Finally, the third principal component could be better described as a variant of neuroticism35 or novelty-seeking and self-directedness.36 In other words, at least 2 major dimensional models of personality could be useful in better describing and understanding the personality dimensions underlying the covariation of DSM-IV PDs. These data seem to suggest at least the need to introduce in DSM-IV a dimensional

ET AL

assessment of personality, complementary to the categorical diagnosis. Finally, it should be considered that this study has several limitations that limit the generalizability of the results. The differences in sampling strategies, axis I disorder base rates and assessment, PD assessment. and diagnostic criteria make it difficult to compare the results of this study with findings obtained in previous studies. In fact, all of these factors were shown to modify both PD base rates and covariation pattems.3.s Moreover, a possible “halo effect” could have occurred, given the SCID-II format.5.‘i However, the large number of interviewers involved in this study, the jointinterview design, and the use of all available sources of diagnostic information should have lessened the risk of bias due to the halo effect. The use of additional diagnostic interviews in this study was prevented by the lack of convergence between different diagnostic instruments for PD assessment”.“’ and the absence of definite guidelines for handling the diagnostic discrepancies.?’In other words, we believed that entering an additional interview to diagnose DSM-IV PDs could lead to confusing results without controlling the diagnostic bias. Finally, it should be considered that even if the sample in this study is large and composed of consecutively admitted subjects, it is not a random sample in statistical terms and shows an excess of anxiety disorder and eating disorder diagnoses. This may have led to sampling bias, which limits the generalizability of the results. However, it should be noted that all previous studies on PD covariation (as well as the wide majority of clinical studies in psychiatry) were performed on clinical or, at best, consecutively admitted samples.

REFERENCES 1. Pfohl

B, Coryell

W, Zimmerman

M, Stangl

DA.

DSM-III

L, Davies

personality disorders: diagnostic overlap and internal consistency of individual DSM-III criteria. Compr Psychiatry 1986:27: 2 l-34. 2. Loranger

AW,

Personality Disorder Disord 1997;1:1-13. 3. Morey

LC.

Susan VL,

Oldham

Examination:

Personality

M,

4. Widiger Manning

Oldham

JM

TA, Frances AJ, D. Comorbidity (ed):

Diagnostic Validity. Press, 1991:163-194. 5. Oldham

Personality Washington,

JM, Skodol

Russakoff

a preliminary

disorders

III-R: convergence, coverage. Psychiatry 1988;145:573-577.

JM.

in DSM-III

and internal

Harris M, Jacobsberg LB. among axis II disorders. Disorders: DC:

AE. Kellman

New American HD,

Hyler

Reliability comparison

J Pers

Am

J

Fyer In:

Perspective

on

Psychiatric SE. Rosnick

of

patterns

personality

of comorbidity.

J. Gilmore

M.

disorders

by

Am .I PsychiaS. Brown

S.

criteria for borderline personality disorder: of DSM-III and the Diagnostic Interview

Hurt

a for

Borderline Patients. Am J Psychiatry 1984:141: lO80- 1084. 7. Kass F, Skodol A, Charles E. Spitter RL. Williams JBW.

and DSM-

consistency.

of DSM-III-R

interviews:

try 1992; 149:2 13-220. 6. Frances A, Clarkin

M. The

report.

M. Diagnosis

two structured

Scaled

ratings

of DSM-III

try 1985;142:627-630. 8. Morey LC. Waugh

personality M.

Blashlield

disorders.

Am J Psychia-

R. MMPI

DSM-III personality disorders: their derivation Pers Assess 1985:49:245-25 1.

scales

for

and correlates.

9. Dahl A. Some aspects of the DSM-III personality disorders illustrated by a consecutive sample of hospitalized patients. Acta

Psychiatr

Stand

1986;73:62-66.

J

PAl7ERNS

OF COVARIANCE

OF DSM-IV

PDs

IO. Livesley WJ. Jackson D. The internal consistency and factorial structure of behaviors judged to be associated with DSM-III personality disorders. Am J Psychiatry 1986; 143: 14731474, 1 I. Widiger TA. Trull T, Hurt S. Clarkin J. Frances A. A multidimensional scaling of the DSM-III-R personality disorders. Arch Gen Psychiatry 19X7:44:557-563. 12. Hyler SE. Lyons M. Factor analysis of the DSM-III personality disorder clusters: a replication. Compr Psychiatry 1988:29:304-30X. 13. Zimmemtan M, Coryell W. DSM-III personality disorders in a nonpatient sample: demographic correlates and comorbidity. Arch Gen Psychiatry 1989;46:682-689. II. Frances A. Widiger T. Fyer MR. The influence of classification methods on comorbidity. In: Maser J. Cloninger CR teds): Comorbidity of Mood and Anxiety Disorders. Washington. DC: American Psychiatric Press. 199O:J I-59. 15. Cloninger CR. A systematic method of description and classifcation of personality variants: a proposal. Arch Gen Psychiatry 1987:44:573-588. 16. Svrakic DM. Whitehead C, Przibeck TR, Cloninger CR. Differential diagnosis of personality disorders by the sevenfactor model of temperament and character. Arch Gen Psychiatry I993:50:99 I-999. 17. Widiger TA. Frances A. Toward a dimensional model for the personality disorders. In: Costa PT. Widiger TA teds): Personality Disorders and the Five-Factor Model of Personality. Washington. DC: American Psychological Association, 1994: 1940. 18. Blashlield R. Sprock J. Pinkston K, Hodgin J. Exemplar prototypes of personality disorder diagnosis. Compr Psychiatry 1985;26:ll-21. 19. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. Ed. 4. Washington. DC: American Psychiatric Press. 1994. 20. First MB. Spitzer RL. Gibbon M, Janet BW. Benjamin L. Structured Clinical Interview for DSM-IV Axis II Personality Disorders (SCID-II). Version 2.0. New York. NY: Biometrics Research Department, New York State Psychiatric Institute, 1994. 21. Zimmerman M. Diagnosing personality disorders. A review of issues and research methods. Arch Gen Psychiatry 1994:5 I X225-245. 22. Maffei C. Fossati A. Agostoni 1. Barraco A. Bagnato M. Donati D. et al. Interrater reliability and internal consistency of the Structured Clinical Interview for DSM-IV Axis II Personal-

215

ity Disorders (SCID-II). Version 2.0. J Pers Disord 1997;11:279284. 23. Fleiss JL. Statistical Methods for Rates and Proportions. Ed. 2. New York, NY: Wiley. 198 I. 24. Agresti A. Categorical Data Analysis. New York, NY: Wiley. 1990. 25. Cattell RB. Tbe scree test for the number of factors. Mult Behav Res 1966; I :245-276. 26. Horn JL. A rationale and test for the number of factors in factor analysis. Psychometrika I965:30: 179- 185. 27. Montanelli RG, Humphreys LG. Latent roots of random data correlation matrices with squared multiple correlations on the diagonal: a Monte Carlo study. Psychometrika 1976:41:341348. 28. Zwick WR, Velicer WF. Comparison of five rules for determining the number of components to retain. Psycho1 Bull 1986:99:432-442. 29. Morrison DE Multivariate Statistical Methods. New York. NY: McGraw-Hill, 1990. 30. Yeomans KA. Golder PA. The Guttman-Kaiser criterion as a predictor of the number of common factors. Statistician 1982;31:221-229. 3 I. Zwick WR. Velicer WE Factor influencing four rules for determining the number of components to retain. Mult Behav Res 1982; 17:253-269. 32. Mardia KV. Kent JT, Bibby JM. Multivariate Analysis. London, England: Academic, 1979. 33. Kalus 0. Bernstein DP, Siever LJ. Schizoid personality disorder: a review of current status and implications for DSM-IV. J Pers Disord 1993:7:43-52. 34. Virkunnen M. Kallio E. Rawlings R, Tokola R, Poland RE, Guidotti A. et al. Personality profiles and state aggressiveness in Finnish alcoholic, violent offenders, fire setters, and healthy volunteers. Arch Gen Psychiatry l994:5 I :28-33. 35. Trull TJ. McRae RR. A five-factor perspective on personality disorder research. In: Costa PT. Widiger TA (eds): Personality Disorders and the Five-Factor Model of Personality. Washington, DC: American Psychological Association, 1994:5971. 36. Cloninger CR. Svrakic DM, Przibeck TR. A psychobiological model of temperament and character. Arch Gen Psychiatry 1993;50:975-990. 37. Kavoussi RJ, Coccaro EF. Klar HM, Bernstein D. Siever LJ. Structured interviews for borderline personality disorders. Am J Psychiatry 19903147:1522-1525.