Dimensionality of depression in acute schizophrenia: a methodological study using the Bech-Rafaelsen Melancholia Scale (BRMES)

\ PERGAMON

Journal of Psychiatric Research 21 "0887# 258Ð267

Dimensionality of depression in acute schizophrenia] a methodological study using the Bech!Rafaelsen Melancholia Scale "BRMES# Matthias J[ Muller\ Hermann Wetzel Department of Psychiatry\ University of Mainz\ Untere Zahlbacher Stra)e 7\ 44020 Mainz\ Germany Received 00 March 0887^ received in revised form 14 May 0887^ accepted 14 May 0887

Abstract Despite the great clinical importance of depressive symptoms in schizophrenia there is a lack of studies on the assessment and evaluation of depression in acutely psychotic patients[ For the Bech!Rafaelsen Melancholia Scale "BRMES#\ among other advantages\ the concept of unidimensionality was con_rmed in patients with major depression by di}erent methodological approaches including Rasch analysis[ The present evaluation was designed to investigate the scale properties of the BRMES in acutely schizophrenic patients with particular emphasis on the dimensionality of the scale[ Three di}erent statistical approaches were used] principal component analysis in combination with computer simulation\ polytomous Rasch analysis using advanced latent trait and latent class models and con_rmatory factor analysis "CFA# by means of linear structure model approaches[ The statistical methods were applied to BRMES baseline data of 021 acutely schizophrenic patients with predominantly positive symptoms participating in a multi!center pharmacological trial[ The di}erent methodological approaches revealed converging results indicating] "0# a substantial proportion of acutely ill schizophrenic patients showed depressive symptoms^ "1# the hypothesis of unidimensionality of the BRMES had to be rejected for the sample of acutely schizophrenic patients and "2# a three!factorial model of depressive symptoms as measured by the BRMES "{retardation|\ {depressive core symptoms|\ {unspeci_c depressive symptoms|# yielded the best _t of the present data[ Depression in acutely psychotic patients has to be considered rather as a heterogeneous construct than as a well!de_ned syndrome[ The di}erentiation of depressive symptomatology should facilitate treatment evaluation and help to clarify relationships between di}erent symptom classes in further studies[ Þ 0887 Elsevier Science Ltd[ All rights reserved[ Key words] Depression^ Schizophrenia^ Rating scales^ Dimensionality^ Con_rmatory factor analysis

0[ Introduction Psychotic features are often associated with clinically relevant depressive symptoms in acute schizophrenia and schizophrenia spectrum disorders "Siris\ 0884#[ Fur! thermore\ depression and suicidal ideation seem to be related to both positive and negative symptoms in psy! chotic disorders "Addington and Addington\ 0881^ Fen! ton et al[\ 0886#[ Whereas several factorial concepts of schizophrenia based on the Positive and Negative Syn! drome Scale "Kay et al[\ 0876# or the Brief Psychiatric Rating Scale "BPRS^ Overall and Gorham\ 0851# com! prise a depression dimension "Van der Does et al[\ 0882^ Bell et al[\ 0883#\ the standardized assessment of depression is not established in psychotic patients[ More! over\ evaluation of depressive symptoms is complicated

 Corresponding author[ Tel[] ¦38!5020!06!1819^ fax] ¦38!5020!06! 5589^ e!mail] mjm04Ýmainz!online[de 9911!2845:87:,08[99 Þ 0887 Elsevier Science Ltd[ All rights reserved[ PII] S 9 9 1 1 ! 2 8 4 5 " 8 7 # 9 9 9 1 8 ! 5

by the well!known overlap with negative symptoms and extrapyramidal symptoms "EPS#[ Beside the Calgary Depression Rating Scale for schizophrenia "Addington et al[\ 0889#\ which is internationally not yet fully validated\ there are at least three standard rating scales for the assessment of depression severity available for a long time] the Hamilton Depression Rating Scale "HDRS#  sberg Depression "Hamilton\ 0859#^ the Montgomery A  sberg\ 0868# Rating Scale "MADRS# "Montgomery and A and the Bech!Rafaelsen Melancholia Rating Scale "BRMES# "Bech et al[\ 0868#[ Among these scales\ the BRMES was proposed to be superior with respect to homogeneity and internal consistency "Maier and Phil! ipp\ 0874#[ The BRMES seems to ful_l di}erent criteria of unidimensionality "Bech et al[\ 0870^ Bech\ 0873^ Maier and Philipp\ 0874^ Chambon et al[\ 0889# including the restrictive assumptions of latent!trait theory "Rasch\ 0859#\ which was\ however\ only applied to dichotomized item scores derived from major depressive patients[ To make use of global scores of rating scales as an index

269

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267

of severity\ the assumption of unidimensionality of the underlying construct must hold[ Unidimensionality implies that the item parameters are independent of glo! bal severity and independent of characteristics of a given patient sample[ Otherwise\ patients could be rated as equally {severe| while presenting with qualitatively rather di}erent syndromes "Gibbons et al[\ 0882#[ In the case of depression in psychotic patients\ dimen! sionality is of both theoretical and practical importance] dimensional analysis of a well!known standard rating scale for depression could give insight into the organ! ization of depressive symptoms in psychotic disorders[ On the other hand\ a dimensional structure of depressive symptomatology in psychotic patients could be useful to unravel di}erent associations and overlaps of depression with negative and positive symptoms[ The present study investigated the dimensionality of the BRMES in a large sample of acutely psychotic pati! ents[ To avoid methodological shortcomings of tra! ditionally employed methods\ e[g[ Cronbach|s coe.cient alpha "Cronbach\ 0840#\ exploratory factor analysis and principal component analysis "Hattie\ 0874#\ three di}erent advanced methods comprising con_rmatory fac! tor analysis "Long\ 0872#\ comparison of factor analyses on empirical versus random data "Horn\ 0854^ Ham! bleton et al[\ 0880# and mixed Rasch analysis applicable to rating scale formats "Rost\ 0877# with improved item response model tests "Molenaar\ 0872#\ were applied to baseline data of acute schizophrenics in order to draw conclusions about the dimensionality of the BRMES on a sound basis of converging results from di}erent methods[ The core methodology\ however\ to test the unidi! mensionality of BRMES items in psychotic disorders was con_rmatory factor analysis "CFA# using linear structure models "EQS\ Bentler\ 0875\ 0884^ LISREL\ Joreskog and Sorbom\ 0875\ 0877#[

Psychiatric Rating Scale "BPRS^ item score 0Ð6# at base! line[ Exclusion criteria comprised other DSM!IIIR diag! noses like schizoa}ective disorder\ dementia or organic brain disorder\ prevailing negative schizophrenic sympto! matology as assessed by a SANS score "Andreasen\ 0873# above 44 points\ history of alcohol or substance depen! dence or abuse\ serious suicidality\ relevant medical or neurological diseases\ pregnancy or lactation and depot neuroleptic medication within a time period of 2 months prior to study inclusion "Wetzel et al[\ 0887#[ The study was carried out in accordance with the latest version of the Declaration of Helsinki^ all patients had given fully informed written consent prior to the study and the study was approved by the local ethics commit! tees[ Among other scales\ the Bech!Rafaelsen Melancholia Scale "BRMES# "Bech et al[\ 0868# was employed by experienced clinicians who were extensively trained before the start of the pharmacological trial[ Interrater reliability scores were not assessed[ The BRMES consists of 00 items "see Table 0\ for item contents# each to be rated on a 4 point scale "9\ absent^ 3\ extremely#[ Data on outcome evaluation are published elsewhere "Wetzel et al[\ 0887#[

1[ Materials and methods

1[1[0[ Exploratory and con_rmatory factor analysis "CFA# Exploratory principal component analyses "PCA# with varimax rotation were carried out and one!\ two!\ three! and four!factor models with a simple structure for load! ings as obtained by PCA were subsequently used to cal! culate CFA[ Latent dimensions "factors# were allowed to correlate for theoretical reasons "Long\ 0872#[ For computation of CFA\ LISREL 6[19 "Joreskog and Sor! bom\ 0877# and EQS 2[9 "Bentler\ 0884# software was used[ To attain results not liable to deviations of item distributions from multivariate normality\ both maximum likelihood "ML# estimations "LISREL# and alternative estimations "robust ML^ ERLS\ iteratively reweighted generalized least squares parameter estimates following elliptical\ non!normal\ distribution theory^ EQS# were calculated[ Beside loadings "l# and factor correlations "f#\ goodness!of!_t indices as suggested in

1[0[ Patients characteristics and assessment In the present evaluation\ baseline data of a multi! center study at 00 German centers on the treatment of acute schizophrenia were analyzed[ All patients were o} neuroleptic medication for 0Ð2 days before they were enrolled in a controlled pharmacological trial "Wetzel et al[\ 0887#[ Anticholinergics were not withdrawn during baseline assessments[ The study comprised 021 acutely schizophrenic patients "age 22[828[7 years\ 33) female# according to DSM!III!R with predominantly positive symptoms[ Acutely admitted in!patients of either sex aged 07 to 54 years with a primary diagnosis of schizo! phrenia\ paranoid type "184[2# or undi}erentiated type "184[8# according to DSM!III!R\ were included[ Patients were required a score of at least 25 points on the Brief

1[1[ Statistics After initial descriptive analysis of all BRMES items including frequency of ratings\ means\ univariate and bivariate distribution analysis\ item analyses according to classical test theory were performed using Cronbach|s coe.cient alpha "Cronbach\ 0840# as an index of internal consistency as well as corrected item!score!interrelations "rit# and squared multiple regressions of individual item scores on all other items "Rit1#[ These preliminary analyses showed no apparent irregularity precluding further analyses[

260

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267 Table 0 BRMES] descriptive statistics "n021# BRMES] items

Mean S[D[ Md Range

Kurt

Skew

rit

Rit1

"0# Motor retardation "1# Verbal retardation "2# Intellectual retardation "3# Psychic anxiety "4# Suicidality "5# Depressed mood "6# Self depreciation "7# Emotional retardation "8# Sleep disturbances "09# Tiredness and pains "00# Work and interests

9[84 9[72 0[30 1[90 9[30 9[87 9[74 0[16 0[64 9[67 1[18

9[86 0[92 0[09 0[02 9[57 9[83 0[98 0[98 0[18 9[76 0[21

0 9 0 1 9 0 9 0 1 9 2

9Ð3 9Ð3 9Ð3 9Ð3 9Ð2 9Ð2 9Ð3 9Ð3 9Ð3 9Ð3 9Ð3

−9[97 9[93 −9[71 −9[63 0[41 −9[62 0[99 −9[67 −0[21 9[76 −9[88

9[79 0[93 9[27 −9[03 0[41 9[42 0[29 9[25 −9[01 0[06 −9[25

9[38 9[38 9[54 9[38 9[18 9[42 9[41 9[44 9[23 9[26 9[43

9[58 9[58 9[41 9[20 9[11 9[20 9[34 9[26 9[17 9[14 9[24

BRMES total score

02[42

5[84

02

9Ð29

−9[48

9[12

a9[70

BRMES\ Bech!Rafaelsen Melancholia Scale^ S[D[\ standard deviation^ Md\ sample median^ Kurt\ kurtosis of distribution^ Skew\ skewness of distribution^ rit\ item!corrected item!total correlation^ Rit1\ squared multiple correlation "all other items predicting item i#^ a\ Cronbach|s coe.cient alpha[

the literature were computed] a x1!value for model _t which is strictly only usable for estimations based on multivariate normally distributed covariances\ the good! ness!of!_t index "GFI# and its degrees!of!freedom! adjusted counterpart\ the adjusted goodness!of!_t index "AGFI# "Joreskog\ 0870^ Joreskog and Sorbom\ 0877# as well as the corresponding x1 values and comparative _t index "CFI# for EQS analyses "Bentler\ 0884#[ The AGFI value is of great importance as more complex models can trivially account for more variance and covariance than simple models[ Values for CFI×9[89\ GFI×9[89 and AGFI×9[79 are suggested for acceptable models\ the GFI and AGFI indices can be interpreted similarly to common values of R1 and shrunken R1 "Darlington\ 0889^ Muller\ 0885# in multiple regression analysis[ The CFI coe.cient is used to compare a speci_ed CFA model with a null!model assuming independent variables "no factor#[ Accordingly\ relative improvement of nested models can be calculated by x1 tests "Bentler and Bonett\ 0879^ Bentler and Weeks\ 0879#[ Thus\ acceptable _t for a single model and signi_cant improvement compared to a sim! pler model should yield the best _tting and most par! simonious factorial solution[ 1[1[1[ Comparison of empirical and random data "computer simulation# Additionally\ to test for the contribution of chance to _t data in factorial structures\ computer simulations were carried out on the data[ Therefore\ samples of n021 were randomly drawn from normally distributed values with the following restrictions] mean and standard devi! ation as well as the range of data "9Ð3 or 9Ð2# had to be equal to the empirically found distribution "c[f[ Table 0#[ A total of ten random samples was drawn and\ sub! sequently\ PCA were carried out to compare the eigen!

values of empirically and randomly derived principal components "0Ð4 factors#[ This approach can demon! strate exploratively whether randomly distributed item scores could be accounted for by similar factorial models as empirically obtained {real| values from {real| patients[ 1[1[2[ Polytomous mixed Rasch model The third approach comprised developments of latent trait and latent class analysis[ Two improvements of tra! ditional Rasch analysis "Rasch\ 0859^ Fischer and Molenaar\ 0884# were used in the present context] mixed latent trait analysis and latent class models[ Traditionally\ ordinal scale data were dichotomized to calculate item response curves for dichotomous Rasch model analysis\ for example\ in the studies of Chambon et al[ "0889# and Maier and Philipp "0874#[ Although in most cases there should not be a substantial decrease in information\ interpretations of item homogeneity and dimensionality are based on these dichotomized item scores but were commonly generalized to the polytomous item format in use[ Polytomous mixed Rasch models as proposed recently "Rost\ 0880^ Von Davier and Rost\ 0884# and available as software program "WINMIRA^ Von Davier\ 0885# can be used to analyze item response formats of rating scales with di}erent parameter restrictions for cal! culations[ In the present study the {rating scale format| option supposing equally distributed and equidistant di.culty thresholds for all items was used "Von Davier and Rost\ 0884#[ The second improvement regards the application of latent class analysis for testing the _t of latent trait models[ Within the mixed Rasch model approach\ two or more subgroups of the total sample are forced to be detected with a maximum of heterogeneity between and a maximum of homogeneity within the groups "Rost\ 0874\ 0889#[ Thus\ traditional tests "And!

261

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267

Fig[ 0[ Cumulative frequency distribution of BRMES total scores in schizophrenic patients[ N021 "099)#^ cut!o} values according to Bech et al[ "0875#[

ersen\ 0862# can be performed between these groups pro! viding a {worst case scenario| which is highly rec! ommendable[ Moreover\ shortcomings by dividing the group according to their sum score "Molenaar\ 0872# can be avoided by this approach[ Additionally\ Molenaar "0872# suggested a graphical test and proposed to divide the sample according to single item scores "{splitter items|# for evaluation of the _t of data to a speci_ed item response model[ In the present approach\ we used both proposals for illustrative pur! poses and calculated Andersen tests "Andersen\ 0862# between groups split according to the score of a subgroup of items^ additionally\ results of the Andersen test for two latent classes found in our sample will be shown[

2[ Results 2[0[ Descriptive analysis The results from descriptive analyses of BRMES item scores are shown in Table 0[ Additionally\ results from classical test theory are provided[ The BRMES total score of 02[426[9 points indicates substantial depressive symp! tomatology in the study sample of acutely schizophrenic patients "Bech et al[\ 0875#[ The distribution of BRMES total scores is given in Fig[ 0 and shows that only 04)

of acute schizophrenic patients can be regarded as {not depressed| in terms of BRMES scores whereas 31) of acutely schizophrenic patients showed BMRES total scores above 03 points indicating {major depression| or {moderate:severe depression| "Bech et al[\ 0875#\ respec! tively[ A Cronbach|s coe.cient a9[70 indicated a su.cient internal consistency according to its de_nition[ 2[1[ Exploratory factor analysis and results of simulation "PCA# Four factorial models as exploratively suggested by principal component analysis "PCA# with subsequent varimax rotation were tested[ Model 0 comprised all BRMES items on a single factor\ model 1 "1 factors# contained items 0\ 1\ 2 and 7 on the _rst factor and the remaining items on the second[ The proposed models are shown in Table 1[ The eigenvalues and the proportion of explained vari! ance "in parentheses# for the _rst four principal com! ponents of PCA of depressive symptoms in psychotic patients were 3[9 "25)#\ 0[4 "03)#\ 0[1 "09)# and 9[7 "6)#[ A comparison of the eigenvalues and proportions of explained variance of PCA results as derived from either the empirical data or from computer simulation are pre! sented in Fig[ 1[

262

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267 Table 1 BRMES] model speci_cations BRMES] items

"0# Motor retardation "1# Verbal retardation "2# Intellectual retardation "3# Psychic anxiety "4# Suicidality "5# Depressed mood "6# Self depreciation "7# Emotional retardation "8# Sleep disturbances "09# Tiredness and pains "00# Work and interests

M0

M1

M2

M3

F00

F10

F11

F20

F21

F22

F30

F31

F32

F33

¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦ ¦

¦ ¦ ¦ 9 9 9 9 ¦ 9 9 9

9 9 9 ¦ ¦ ¦ ¦ 9 ¦ ¦ ¦

¦ ¦ ¦ 9 9 9 9 ¦ 9 9 9

9 9 9 ¦ ¦ ¦ ¦ 9 9 9 9

9 9 9 9 9 9 9 9 ¦ ¦ ¦

¦ ¦ ¦ 9 9 9 9 ¦ 9 9 9

9 9 9 ¦ ¦ 9 9 9 9 9 9

9 9 9 9 9 ¦ ¦ 9 9 9 9

9 9 9 9 9 9 9 9 ¦ ¦ ¦

M0 to M3 indicate models with 0 to 3 factors "F00ÐF33#\ e[g[ F32 indicates the 2rd factor of a 3!factor model "M3#[ ¦\ hypothetical positive factor loading "l# to be freely estimated^ 9\ loadings _xed to zero "l9#[ Factors were allowed to correlate "f9#[

Fig[ 1[ Results of principal component analyses of BRMES items[ Eigenvalues of principal components "0Ð5# and the cumulative proportion of explained variance "insets# are shown[ On the left side] results of ten computer simulations\ on the right side] results of the present sample data[

Apparently\ three orthogonal PCA components account for more than 59) of BRMES item variance in the empirical data whereas 59) of explained variance is reached only after extraction of the sixth principal component in simulated data[ Correspondingly\ the scree plot of simulation data is rather ~at when compared to the empirical data with its prominent _rst component[ 2[2[ Con_rmatory factor analysis "CFA# The four competing models of Table 1 show a simple structure "{Einfachstruktur|# with the restriction of item loadings to only one of several factors[ The interrelation

of factors in models with more than one factor should be allowed in CFA\ as previous analyses and theoretical considerations support the preference of non!zero relationships between latent traits which can empirically be tested as compared to the assumption of orthogonal factors which favor diluted and inconsistent item loading patterns[ Results of CFA using LISREL and EQS esti! mations are shown for the four competing models in Table 2^ a comparison of model _t is provided in Table 3[ Figure 2 presents the best!_tting CFA solution "three factors# with respect to goodness!of!_t and parsimony of estimated parameters[

263

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267

Table 2 BRMES] Comparison of CFA models

LISREL\ ML df x1 p RMSR GFI AGFI

M9

M0

M1

M2

M3

44 380[3 ³9[990 9[18 9[49 9[39

33 072[1 ³9[990 9[09 9[68 9[58

32 060[7 ³9[990 9[00 9[70 9[60

30 027[2 ³9[990 9[00 9[76 9[65

27 020[2 ³9[990 9[00 9[74 9[63

060[5 ³9[990 9[68

057[4 ³9[990 9[68

011[9 ³9[990 9[76

002[4 ³9[990 9[75

EQS\ ERLS x1 p CFI

M0 to M3\ models with 0 to 3 factors^ ML\ maximum likelihood solution "normal distribution theory#^ RMSR\ root mean squared residuals^ GFI\ goodness!of!_t index^ AGFI\ adjusted goodness!of!_t index^ ERLS\ iteratively reweighted generalized least squares solution "elliptical distribution theory#^ CFI\ comparative _t index[

Table 3 BRMES] Comparison of model _t

D"M0 versus M9# D"M1 versus M0# D"M2 versus M1# D"M3 versus M2#

Ddf

Dx1

p

Improvement

00 0 1 2

297[1 00[3 22[4 6[9

³9[99990 9[9996 ³9[99990 9[961

¦ ¦ ¦ −

Models of Table 2 were compared^ x1 of ML solutions were used to compare hierarchically nested models[ ¦\ signi_cant improvement in _t of k!dimensional model Mk versus model Mk−0^ −\ no signi_cant improvement[

The proposed three!factorial model shows statistically signi_cant superiority when compared to the respective one! or two!factor models[ An additional factor "3!factor model# did not add any further information regarding the structure of the BRMES items in acute schizophrenia

"Table 3#[ The three factors can be labeled as {retardation| "factor 0#\ {depressive core symptoms| "factor 1#\ and {non!speci_c| or {accessory depressive symptoms| "factor 2#[ Nevertheless\ the obtained _t indices of the three! factor model indicate still deviations from perfect _t[ All loadings l0Ð00 of BRMES items on the latent factors were statistically signi_cantly di}erent from zero "p³9[90#^ the mutual relationship of latent factors was only moderate\ only factors 1 and 2 of the proposed model showed a signi_cant overlap "f12¦9[46^ p³9[90#[ Factor 0 "{retardation|# showed only a shared variance of about 4) with the other factors "f01¦9[11^ f02¦9[13#[ 2[3[ Mixed Rasch analysis Mixed Rasch analyses were computed to test the hypothesis of unidimensionality of the BRMES items in the present sample[ The model test revealed a conditional log!likelihood of −0793[8 "df04# for the total sample[ The two!class solution "latent class analysis# with class I comprising 35) and class II 43) of patients\ yielded a product of conditional log!likelihoods of −0651[0 "df20#[ The respective Andersen test was highly sig! ni_cant "x174[6\ df05\ p³9[999990# indicating a lack of transferability of the BRMES items[ The order of item di.culties di}ered substantially in both empirically derived classes "Spearman rank correlation coe.cient\ rs9[37\ 1!tailed p×9[09#[ Dichotomizing of item responses "9\ 09^ 1\ 2\ 30# as used in traditional Rasch analysis yielded also a signi_cant deviation from trans! ferability "Andersen test] x144[0\ df02\ p³9[999990#[ Similar results "Figure 3# were obtained when the group was divided according to the median of the sum score of BRMES items 0\ 1\ 2 and 7 "factor 0#[ When checking transferability of the proposed three! factorial structure using the highly restrictive latent!class approach\ the following results were found] factor 0 "3 items\ x115[6\ df8\ p9[991#\ factor 1 "3 items\ x106[4\ df8\ p9[94# and factor 2 "2 items\ x15[5\ df7\ p9[47#[ These results illustrate a clear advantage

Fig[ 2[ Three!factorial model of BRMES items in schizophrenia "CFA solution#[ Numbers in parentheses are item numbers of BRMES "0Ð00#[ Numbers represent either loadings of items on latent factors "straight arrows\ l# or correlations of latent factors "curvilinear arrows\ f#[ All loadings l are statistically signi_cantly di}erent from zero "p³9[90#[

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267

264

Fig[ 3[ Graphical _t test of homogeneity of BRMES items[ The sample was divided according to the median of the sum score "S# of BRMES items 0\ 1\ 2 and 7 "median3#\ and item parameters "s# were calculated in both subsamples by means of mixed Rasch model[ Item parameters for BRMES items were graphically represented as proposed by Molenaar "0872# with item parameters s0 "subsample 0\ S below the sample median# on the abscissa and item parameters s1 "subsample 1\ S above the sample median# on the ordinate[ Similar item parameters in both subsamples would imply that all data points are close to the dotted line "s0s1#[ The Andersen test "Andersen\ 0862# compares _t statistics of subsamples and indicates a signi_cant deviation from the hypothesis of transferability of item parameters for the present subsamples[

of the three!factorial structure also with respect to the restrictive criterion of transferability[

3[ Discussion The present study used di}erent methodological approaches to investigate the scale characteristics of the Bech!Rafaelsen Melancholia Scale "BRMES# in acute schizophrenia[ The analyses revealed converging results indicating that "0# a substantial proportion of acutely ill schizophrenic patients showed depressive symptoms^ "1# the hypothesis of unidimensionality of the BRMES had to be rejected for the sample of acutely schizophrenic patients and "2# a three!factorial model of depressive symptoms as measured by the BRMES "{retardation|\ {depressive core symptoms|\ {unspeci_c depressive symp! toms|# yielded the best _t of the present data[ The limi! tations of the BRMES sum score to measure depression in acute schizophrenia is a main outcome of the present study[ The results suggest that depression in acutely psy! chotic patients has to be recognized rather as a het! erogeneous construct than as a well!de_ned homogenous syndrome[ The di}erentiation of depression in schizo! phrenia allows further insight into the functional organ!

ization of symptomatology and should facilitate treat! ment evaluation in further studies by clarifying relationships between di}erent symptom classes[ Traditionally\ homogeneity and unidimensionality of a scale was investigated by classical item analysis including calculation of Cronbach|s coe.cient alpha "Cronbach\ 0840# and by PCA applying several heuristic criteria as the scree plot "Cattell\ 0855#\ Kaiser!Guttman criterion of eigenvalues ×0[9 for factor extraction "Kaiser and Ca}rey\ 0854# and indices for the dominance of a _rst component "19Ð39) of explained variance] Carmines and Zeller\ 0868^ Reckase\ 0868#[ According to these criteria\ the present study could be regarded as evidence for unidimensionality of the BRMES in schizophrenic patients[ Cronbach|s coe.cient a exceeded the generally accepted value of 9[64 for homogeneity "Lienert\ 0878#\ a pronounced _rst principal component explained about 25) of variance and the scree plot could be interpreted also in favor of unidimensionality[ However\ there is general agreement that the afore! mentioned criteria are to some degree necessary but not su.cient to prove unidimensionality of a scale "Green et al[\ 0866^ Hattie\ 0874^ Clark and Watson\ 0884#[ The most powerful and straightforward approach to test dimensional models at present seems to comprise con!

265

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267

_rmatory factor analysis in combination with computer simulations and probabilistic model analyses[ Alter! natively\ factor analyses based on item response theory "Bock et al[\ 0877^ Gibbons et al[\ 0882# and the analysis of residuals have been proposed "Hambleton et al[\ 0880#[ The requirements regarding sample size\ the number of estimated parameters\ the distribution of scores as well as the prevalence and variance of items\ for carrying out  berla\ 0857^ Long\ 0872^ Cole\ dimensionality analyses "U 0876^ Muller\ 0885# were ful_lled in the present evalu! ation[ In case of item based statistics such as Rasch analy! sis and latent class analysis\ the present approach used the original 4!point rating scale format in addition to the dichotomization of item responses "Fischer and Molenaar\ 0884#[ Both strategies yielded comparable results\ however\ using the full range of item responses no information is lost[ Another aspect of advanced approaches of item response theory was utilized by appli! cation of latent class analysis for the evaluation of hom! ogeneity of the BRMES scale[ By this method\ two classes of respondents were assembled by maximizing between group variance[ Thus\ a {worst case scenario| was used for inferential and graphical tests "Andersen\ 0862^ Molenaar\ 0872#[ In the present study\ the results of CFA converged with _ndings from PCA\ the comparison of PCA on empirical and random data and with the out! come of probabilistic models[ Accordingly\ a three factor model showed signi_cant superiority to one!\ two! and four!factor solutions[ The three!factor model showed a su.cient model _t although GFI and AGFI did not fully reach the proposed level of 9[89 and 9[79\ respectively "Joreskog and Sorbom\ 0877^ Bollen and Long\ 0882#[ The high prevalence of a {phenocopy of depression| "Siris\ 0884# in acute states of schizophrenic disorders found in the present study is in line with previous results proposing that 4 to 79) of schizophrenic patients su}er from substantial depression without ful_lling criteria of a schizoa}ective disorder or major depression[ The high variability of these results points to the heterogeneity of both patients samples and assessment methodology of depression in schizophrenia[ With respect to the latter problem\ the present study could show that an observer rating scale with su.cient homogeneity in patients with major depression "Maier and Philipp\ 0874# yielded rather heterogeneous results in acute schizophrenics[ Our results also imply that the BRMES sum score seems not to be an accurate index of depression severity in acute schizophrenia and should be interpreted only cautiously[ This _ndings are in line with the results of a study by Goldman et al[ "0881# which demonstrated that the HDRS total score was non!speci_c in measuring depression in schizophrenic inpatients whereas an indi! vidual sub!factor provided a more speci_c index of depression in these patients[ Whether or not a depression rating scale originally developed for severity assessment of patients with major depression is suitable to assess

depression in schizophrenia remains a matter of debate[ However\ there seems no alternative and the recently proposed approach to develop a speci_c scale for the assessment of depression in schizophrenia "Calgary Depression Rating Scale ðCDSSŁ^ Addington et al[\ 0889# also used items of the HDRS "Hamilton\ 0859#[ On the other hand\ di}erent methodological studies using exploratory factor analysis "Baumann\ 0865^ Maier et al[\ 0874#\ Rasch analysis "Maier et al[\ 0874^ Chambon et al[\ 0889# and item response theory "Gibbons et al[\ 0882# have shown that the hypothesis of unidimensionality must be rejected for the HDRS also in patients with major depression[ The model of three relating but distinct factors of the BRMES in acute schizophrenia appears to be of high face validity and is similar to _ndings by Maier et al[ "0884#^ the authors\ however\ proposed a modi_ed two! factor model of the BRMES in schizophrenic patients in the context of a model for depressive and negative symptoms[ This points to the problem of sometimes arbi! trary decisions in the _eld of factor analysis[ We tried to exclude such bias by using rather restricted models with simple structure and without any post!hoc changes of parameter relations[ However\ some data!driven and sub! jective decisions remain\ for example\ in the case of labe! ling and interpreting the derived factors "{nominalistic fallacy|^ Cli}\ 0872#[ Additionally\ it has to be kept in mind that factor analytic methods are signi_cantly depen! dent on the items included in the scale used to measure psychopathology[ However\ in the present study\ a widely accepted scale with high face validity for the assessment of depressive symptoms was used[ A further hint towards external validity of the present results is that three of four items of the latent BRMES factor of {depressive core symptoms| are represented in the CDSS "depressive mood\ suicidal ideation\ self depreciation#[ Thus\ the pre! sent results also partly corroborate the speci_city hypoth! esis of the CDSS[ However\ the CDSS taps further aspects of depression including hopelessness and feelings of worthlessness not covered in the HDRS or the BRMES[ Nevertheless\ the use of the BRMES in schizo! phrenic patients provides a reasonable alternative^ _rstly\ the CDSS was initially designed for a timeframe of two weeks whereas most of the traditional rating scales allow for assessments within one week or less[ However\ the CDSS can also be used for shorter time intervals "D[ Addington\ personal communication#[ Secondly\ the BRMES like the HDRS and the MADRS are widely used and convenient observer rating scales for depressive syndromes and allow therefore for comparisons across diagnostic categories[ Thirdly\ the application of a tradi! tional depression rating scale allows for a broader assess! ment of behavioral depressive symptoms and thus for insight of the structural composition of {depression| in acute schizophrenia[ However\ whether or not the present _ndings on the structure of depressive symptoms in acute

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267

schizophrenic patients are also applicable to patients with chronic schizophrenia or those with predominant nega! tive symptoms\ remains open to further studies[ Fur! thermore\ the stability of the BRMES factors under treat! ment and the relationship between di}erent symptom classes "including EPS# will be addressed in a future paper "Muller et al[\ in preparation#[ Within a functional approach of psychiatric disorders "Benkert\ 0889# the comparability of symptoms across the borders of traditional nosology is of particular interest^ common functional de_cits and treatment approaches of a}ective and psychotic disorders could be delineated by the present approach[ Distinct factors of depressive symptoms should show di}erential relationships to posi! tive\ negative and extrapyramidal symptoms[ Thus\ a more accurate subdivision of depression in acute schizo! phrenia as proposed by the present study could help to understand the apparent overlap of depression with other symptom classes[ Studies replicating and validating the present results are required and under way[

Acknowledgements The study is part of a multi!center trial sponsored by Synthelabo Recherche\ Paris\ France[ The present analysis was carried out fully independently from sponsorship[

References Addington D\ Addington J\ Schissel B[ A depression rating scale for schizophrenics[ Schizophr Res 0889^2]136Ð140[ Addington DE\ Addington JM[ Attempted suicide and depression in schizophrenia[ Acta Psychiatrica Scand 0881^74]177Ð180[ Andersen EB[ A goodness of _t test for the Rasch model[ Psychometrika 0862^27]012Ð039[ Andreasen NC[ Scale for the Assessment of Negative Symptoms "SANS#[ Iowa City\ Department of Psychiatry\ University of Iowa College of Medicine\ 0873[ Baumann U[ Methodische Untersuchungen zur Hamilton!Depression! Skala[ Arch Psychiatrie Nervenkrankheiten 0865^111]248Ð264[ Bech P\ Allerup P\ Gram LF\ Reisby N\ Rosenberg N\ Jacobsen O\ Nagy A[ The Hamilton Depression Scale] evaluation of objectivity using logistic models[ Acta Psychiatrica Scand 0870^52]189Ð188[ Bech P\ Bowlby TG\ Kramp P\ Rafaelsen OJ[ The Bech!Rafaelsen Melancholia Scale and the Hamilton Depression Scale[ Acta Psy! chiatrica Scand 0868^48]319Ð329[ Bech P\ Kastrup M\ Rafaelsen OJ[ Mini!compendium of rating scales[ Acta Psychiatrica Scand 0875^Suppl[ 215]0Ð26[ Bech P[ The instrumental use of rating scales for depression[ Phar! macopsychiatry 0873^06]11Ð18[ Bell MD\ Lysaker PH\ Beam!Goulet JL\ Milstein RM\ Lindenmayer JP[ Five!component model of schizophrenia] assessing the factorial invariance of the positive and negative syndrome scale[ Psychiatry Res 0883^41]184Ð292[ Benkert O[ Functional classi_cation and response to psychotropic drugs[ In] Benkert O\ Maier W\ Rickels K\ editors[ Methodology of the evaluation of psychotropic drugs[ Berlin] Springer\ 0889[ pp[ 044Ð052[

266

Bentler PM\ Bonett DG[ Signi_cance tests and goodness of _t in the analysis of covariance structures[ Psychol Bull 0879^77]477Ð595[ Bentler PM\ Weeks DG[ Linear structural equations with latent vari! ables[ Psychometrika 0879^34]178Ð297[ Bentler PM[ EQS] Ein Ansatz zur Analyse von Struktur! gleichungsmodellen fur normal! und nichtnormalverteilte quan! titative Variablen[ In] Mobus C\ Schneider W\ editors[ Struktur! modelle fur Langsschnittdaten und Zeitreihen] LISREL\ Pfad!\ und Varianzanalysen[ Bern] Verlag Hans Huber\ 0875[ p[ 16Ð45[ Bentler PM[ EQS structural equations program manual[ Encino\ CA] Multivariate Software\ 0884[ Bock RD\ Gibbons R\ Muraki E[ Full information item factor analysis[ Appl Psychol Measurement 0877^01]150Ð179[ Bollen KA\ Long JS\ editors[ Testing structural equation models[ New! bury Park] Sage\ 0882[ Carmines EG\ Zeller\ RA[ Reliability and validity assessment[ Beverly Hills] Sage\ 0868[ Cattell RB[ The scree test for the number of factors[ Multivariate Behav Res 0855^0]001Ð014[ Chambon O\ Cialdella P\ Kiss L\ Poncet F\ Chevance M\ Milani! Bachman D[ Study of the unidimensionality of the Bech!Rafaelsen Melancholia Scale using Rasch analysis in a French sample of major depressive disorders[ Pharmacopsychiatry 0889^12]132Ð134[ Clark LA\ Watson D[ Constructing validity] basic issues in objective scale development[ Psychol Assessment 0884^6]298Ð208[ Cli} N[ Some cautions concerning the application of causal modeling methods[ Multivariate Behav Res 0872^07]004Ð015[ Cole DA[ Utility of con_rmatory factor analysis in test validation research[ J Consult Clin Psychol 0876^44]473Ð483[ Cronbach LJ[ Coe.cient alpha and the internal structure of tests[ Psychometrika 0840^05]186Ð223[ Darlington\ RB[ Regression and linear models[ New York] McGraw! Hill\ 0889[ Fenton WS\ McGlashan TH\ Victor BJ\ Blyler CR[ Symptoms\ subtype and suicidality in patients with schizophrenia spectrum disorders[ Am J Psychiatry 0886^043]088Ð193[ Fischer GH\ Molenaar I\ editors[ Rasch models[ Foundations\ recent developments and applications[ New York] Springer\ 0884[ Gibbons RD\ Clark DC\ Kupfer DJ[ What exactly does the Hamilton Depression Rating Scale measure<[ J Psychiatric Res 0882^16]148Ð 162[ Goldman RS\ Tandon R\ Liberzon I\ Greden JF[ Measurement of depression and negative symptoms in schizophrenia[ Psy! chopathology 0881^14]38Ð45[ Green SB\ Lissitz RW\ Mulaik SA[ Limitations of coe.cient alpha as an index of test unidimensionality[ Educat Psychol Measurement 0866^26]716Ð727[ Hambleton RK\ Swaminathan H\ Rogers HJ[ Fundamentals of item response theory[ Newbury Park\ CA\ Sage\ 0880[ Hamilton M[ A rating scale for depression[ J Neurol Neurosurgery Psychiatry 0859^12]45Ð51[ Hattie JA[ Methodological review] assessing unidimensionality of tests and items[ Appl Psychol Measurement 0874^8]028Ð053[ Horn JL[ A rationale and test for the number of factors in factor analysis[ Psychometrika 0854^29]068Ð074[ Joreskog KG\ Sorbom D[ LISREL VI] analysis of linear structural relationships by maximum likelihood\ instrumental variables and least squares methods[ Mooresville] Scienti_c Software\ 0875[ Joreskog KG\ Sorbom D[ LISREL 6\ a guide to the program and applications[ Chicago] SPSS\ 0877[ Joreskog KG[ Analysis of covariance structures[ Scand J Stat 0870^7]78Ð81[ Kaiser HF\ Ca}rey J[ Alpha factor analysis[ Psychometrika 0854^29]0Ð 03[ Kay SR\ Fiszbein A\ Opler LA[ The Positive and Negative Syndrome Scale "PANSS# for schizophrenia[ Schizophrenia Bull 0876^02]150Ð 165[

267

M[J[ Muller\ H[ Wetzel:Journal of Psychiatric Research 21 "0887# 258Ð267

Lienert GA[ Testaufbau und Testanalyse[ Munich\ Weinheim] PVU\ 0878[ Long JS[ Con_rmatory factor analysis[ Sage University Series on Quan! titative Application in the Social Sciences\ 922[ Beverly Hills and London] Sage Publication\ 0872[ Maier W\ Philipp M[ Comparative analysis of observer depression scales[ Acta Psychiatrica Scand 0874^61]128Ð134[ Maier W\ Philipp M\ Gerken A[ Dimensionen der Hamilton! Depressionsskala "HAMD#] Faktorenanalystische Unter! suchungen[ Eur Arch Psychiatry Neurol Sci 0874^123]306Ð311[ Maier W\ Schlegel S\ Klingler T\ Hillert A\ Wetzel H[ Die Nega! tivsymptomatik im Verhaltnis zur Positivsymptomatik und zur depressiven Symptomatik der Schizophrenie] Eine psychometrische Untersuchung[ In Moller HJ\ Pelzer E\ editors[ Neue Ansatze zur Diagnostik und Therapie schizophrener Minussymptomatik[ Berlin\ Heidelberg\ New York] Springer\ 0884[ p[ 58Ð67[ Molenaar IW[ Some improved diagnostics for failure of the Rasch model[ Psychometrika 0872^37]38Ð61[  sberg M[ A new depression scale designed to be Montgomery P\ A sensitive to change[ Br J Psychiatry 0868^023]271Ð278[ Muller RO[ Basic principles of structural equation models[ New York] Springer\ 0885[ Overall JE\ Gorham DR[ The Brief Psychiatric Rating Scale[ Psychol Rep 0851^09]688Ð701[ Rasch G[ Probabilistic models for some intelligence and attainment tests[ Copenhagen] Danish Institute of Education Research\ 0859[ Reckase MD[ Unifactor latent trait models applied to multi!factor tests] results and implications[ J Educat Stat 0868^3]196Ð129[

Rost J[ A latent class model for rating data[ Psychometrika 0874^49]26Ð 38[ Rost J[ A logistic mixture distribution model for polychotomous item responses[ Br J Math 0880^33]64Ð81[ Rost J[ Rasch models in latent classes] an integration of two approaches to item analysis[ Appl Psychol Measurement 0889^03]160Ð171[ Rost J[ Rating scale analysis with latent class models[ Psychometrika 0877^42]216Ð237[ Siris SG[ Depression and schizophrenia[ In] Hirsch SR\ Weinberger DR\ editors[ Schizophrenia[ Oxford] Blackwell Science\ 0884[ p[ 017Ð34[  berla K[ Faktorenanalyse[ Berlin] Springer\ 0857[ U Van der Does AJ\ Linszen DH\ Dingemans PM\ Nugter MA\ Scholte WF[ A dimensional and categorical approach to the symp! tomatology of recent!onset schizophrenia[ J Nervous Mental Dis! eases 0882^070]633Ð638[ Von Davier M[ WINMIRA] A Windows 2[x program for analyses with the Rasch model\ with the latent class analysis\ and with the mixed Rasch model[ Kiel] Institute for Science Education "IPN#\ 0885[ Von Davier M\ Rost J[ Polytomous mixed Rasch models[ In] Fischer GH\ Molenaar I\ editors[ Rasch models[ Foundations\ recent devel! opments\ and applications[ New York] Springer\ 0884[ p[ 260Ð72[ Wetzel H\ Grunder G\ Hillert A\ Philipp M\ Gattaz WF\ Sauer H\ Adler G\ Schroder J\ Rein W\ Benkert O[ Amisulpride versus Flu! pentixol in schizophrenia with predominantly positive symp! tomatology] a double!blind controlled study comparing a selective D1!like antagonist to a mixed D0!:D1!like antagonist[ Psycho! pharmacology 0887^026]112Ð121[