Examining differential effects of psychological treatment of depressive disorder: An application of trajectory analyses

www.elsevier.com/locate/jad

Journal of Affective Disord

Research report

Examining differential effects of psychological treatment of

depressive disorder: An application of trajectory analyses

Pim Cuijpers a,b,*, Pol A.C. van Lier c,d, Annemieke van Straten a,b, Marianne Donker e

a Department of Clinical Psychology, Vrije Universiteit Amsterdam, The Netherlandsb Trimbos Institute, Utrecht, The Netherlands

c Department of Developmental Psychology, Vrije Universiteit Amsterdam, The Netherlandsd Department of Child and Adolescent Psychiatry, Erasmus MC/Sophia Children’s Hospital, Rotterdam, The Netherlands

e Institute of Health Policy and Management, Erasmus MC, Rotterdam, The Netherlands

Received 23 March 2005; received in revised form 31 August 2005; accepted 5 September 2005

Abstract

Background: Although different psychological treatments of depression seem equally effective, studies in this area have not

taken sufficient account of the heterogeneity among patients. Modern techniques for longitudinal data analysis can be helpful in

examining differential effects of psychological interventions on specific subpopulations of patients.

Methods: Outpatients in mental health care, diagnosed with DSM-IV major depressive disorder, were randomly assigned to

cognitive behavior therapy (N =199) or treatment as usual (N =226). Every 3 months for a total of 1.5 years, depressive

symptomatology was measured using the SCL-90. Growth mixture modeling techniques were used to identify different

trajectory classes of patients. The impact of type of treatment (treatment as usual vs. cognitive behavior therapy) was examined

for each identified trajectory.

Results: On average, patients in both test conditions improved significantly from baseline to posttest, and no significant

difference was found between the conditions. However, four different trajectory classes could be distinguished within the

sample. Most patients were classified into the two classes with the lowest depression scores at baseline (31% and 33% of the

total sample). For these two classes, no significant differences in the course of depressive symptoms were found between the

two conditions. In the two classes with the more severe depression scores (10% and 26% of the sample), however, cognitive

behavior therapy was significantly more effective than treatment as usual.

Conclusions: Although different treatments may seem to be equally effective, this does not have to be true for all classes of

patients. Longitudinal research on the treatment of mental disorders should take heterogeneity among patients into account.

D 2005 Elsevier B.V. All rights reserved.

Keywords: Depressive disorders; Randomized controlled trial; Cognitive behavior therapy; Growth mixture modeling

0165-0327/$ - s

doi:10.1016/j.jad

* Correspondi

The Netherlands

E-mail addre

ers 89 (2005) 137–146

ee front matter D 2005 Elsevier B.V. All rights reserved.

.2005.09.001

ng author. Department of Clinical Psychology, Vrije Universiteit Amsterdam, Van der Boechorststraat 1, 1081 BT Amsterdam,

. Tel.: +31 20 598 8757; fax: +31 20 598 8758.

ss: [email protected] (P. Cuijpers).

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146138

1. Introduction

Dozens of well-designed randomized trials have

shown the efficacy of brief psychological interven-

tions for major depression, such as cognitive behavior

therapy (Churchill et al., 2001; Gaffan et al., 1995;

Scott, 2001), interpersonal psychotherapy (Reynolds

et al., 1999), and problem solving therapy (Barrett et

al., 2001; Dowrick et al., 2000; Williams et al., 2000).

Whether all psychological interventions are equally

effective or some are superior to others, however, is

the subject of continuing debate (Beutler, 2002;

Luborsky et al., 1975, 2002; Shadish and Sweeney,

1991). Currently, most meta-analytic evidence indi-

cates that different types of brief psychological

interventions result in comparable effects on depres-

sion (Luborsky et al., 2002). This lack of differential

effects is usually explained by nonspecific elements

which are common to all interventions, such as an

intensive relationship between the patient and the

therapist, the expectation of the patient to be cured,

the dritualT of the therapy, and the presentation of a

clear drationaleT with which the problems of the

patient can be explained (Duncan, 2002).

Critics of this view, however, have pointed at the

complexity of determining the presence of specific

effects. They have argued that significant differences

between psychotherapies do exist but that the number

of potential effect predictors is very large and that

most studies do not have sufficient statistical power to

detect them (Beutler, 2002; Kazdin and Bass, 1989).

Furthermore, the mean effects of various psycholog-

ical treatments do not necessarily imply that there are

no identifiable differences in effects on individuals or

subpopulations within a sample. The research on

mediators and moderators of outcome increasingly

shows that some subpopulations do benefit less than

others do from psychological interventions (Shadish

and Sweeney, 1991). For example, there is growing

evidence that comorbid anxiety disorder (Albus and

Scheibe, 1993; Brown et al., 1996) and comorbid

personality disorder (Reich, 2003) both reduce the

effects of psychological treatment of depression.

A fundamental problem inherent in the research

studies on moderators of outcome is that these studies

are looking for subpopulations who benefit more or

benefit less from an intervention, but actually examine

only characteristics which may be indicative of these

subpopulations. Recently, methods have become

available to identify distinct groups of individuals,

differing in the initial level and course of a specific

behavior, through the empirical identification of

developmental trajectories (Muthen and Shedden,

1999; Nagin, 1999). The advantage of these methods

is that the identification of subpopulations within a

sample is based on the target behavior itself (in our

case depressive symptoms over time). These techni-

ques also make it possible to examine whether the

effects of interventions differ for different categories

of patients, to ascertain which characteristics predict

membership of one of these categories, and to

establish whether outcomes are different for each

category (Muthen, 2001; Muthen et al., 2002).

In the current study, data from a large pragmatic

randomized trial comparing cognitive behavior ther-

apy and treatment as usual were used to demonstrate

the possibilities of these analytical techniques in

examining the relative effects of interventions on

specific subpopulations of patients. We focus on the

following questions: (1) what is the impact of

cognitive behavior therapy and treatment as usual on

the development of depressive symptoms; (2) how

many trajectory classes of depressive symptoms can

be distinguished; (3) do treatment as usual and

cognitive behavior therapy have a different effect on

patients in different trajectory classes; and (4) how can

patients who follow each of the trajectory classes be

identified?

2. Method

2.1. Procedure and respondents

We used the data of a pragmatic randomized trial

designed to examine whether the effects of psycho-

logical treatments for patients with either mood or

anxiety disorders, which have been typically studied

in research settings, can also be generalized to routine

practice settings. The methods of this study have been

described in detail elsewhere (Van Straten et al.,

submitted for publication). In brief, subjects were

recruited at 7 outpatient mental health centers

(MHCs) at 12 different locations (both rural and

urban regions) in the Netherlands. Patients were

enrolled in two steps (Fig. 1). First, all patients

Fig. 1. Patients included in the analyses.

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146 139

between 18 and 65 years old who were in need of

mental health care were screened during the inclusion

period (February 2000 – October 2001) on exclusion

criteria: psychotic or manic symptoms, cognitive

impairments such as dementia or mental retardation,

illegal hard drug dependence (patients with alcohol

were not excluded), high suicide risk, or poor

command of the Dutch language. The remaining

patients were then screened for the presence of mood

and/or anxiety disorders. Second, all remaining

patients were interviewed at home (baseline assess-

ment) by a trained research assistant to determine the

presence of mood and/or anxiety disorders with the

Composite International Diagnostics Interview

(CIDI). The CIDI, a structured interview developed

by the World Health Organization (World Health

Organization, 1994; Ter Smitten et al., 1998), enables

trained lay interviewers to assess psychiatric diagno-

ses according to the DSM-IV. Patients with the

following DSM-IV diagnosis were included: major

depressive disorder (single episode or recurrent),

dysthymic disorder, panic disorder (with or without

agoraphobia), social phobia, or generalised anxiety

disorder, including comorbid diagnoses. All eligible

patients were asked to participate in the study. After

full explanation of the study, 702 patients gave written

informed consent (Fig. 1). These 702 patients were

randomized. We used block randomization, stratified

by MHC setting. The randomization scheme was

derived by computer and managed centrally at the

research centre. When a patient was included, the

researchers opened an envelope and the randomization

outcome was reported to the MHC. Patients and

therapists were informed about the randomization

outcomes, but the research assistants who performed

the interviews were kept blind during the whole study.

Because of the complexity of the analyses and the

illustrative purposes of this study, only the subjects

randomized to CBT and TAU, who had a diagnosis of

major depressive (MDD) are included in this study

(TAU, n =226, CBT, n =199, Fig. 2).

Patients were interviewed at baseline and then

every 3 months until 7 measurements were taken

(1.5 years after baseline). The presence of DSM-IV

mental disorders was established using the CIDI

during face-to-face interviews at baseline and 1.5 to

2 years after baseline. Other interviews were

conducted by telephone.

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146140

Baseline characteristics of the included subjects

have been described in detail elsewhere (Van Straten

et al., submitted for publication). In brief, 61% of the

patients were female, their mean age was 36 years

(standard deviation (S.D.)=10), 66% had a paid job,

and 35% used anti-depressive medication. The per-

centage of subjects who used anti-depressive medica-

tion at baseline was similar in the TAU and CBT

condition (v2 (1, N =393)= 0.865, p N0.05). At

baseline, 40% (TAU) and 41% (CBT) had a severe

MDD and 46% (TAU) and 47% (CBT) had an anxiety

disorder.

2.2. Interventions

Cognitive Behavior Therapy is characterized by a

structured approach and is primarily aimed at restruc-

turing irrational cognitive assumptions about oneself

which are responsible for negative interpretations of

situations, negative emotions, and finally to self-

undermining actions. CBT has proven its efficacy for

depressive and several types of anxiety disorders

(Churchill et al., 2001; Gaffan et al., 1995; Scott,

2001; Van Balkom et al., 1997), although its effective-

ness in real life conditions has not beenwell established.

Treatment as usual is the treatment as delivered in

current daily practice in the MHCs in the Netherlands.

In TAU, a multidisciplinary team chooses a therapeu-

tic approach which is tailored to the individual’s

needs, taking into account the specific constellation of

problems and characteristics of each different patient.

As a result, any kind of therapy may be chosen from a

wide variety of approaches.

Each therapist, in any of the participating centers,

was expected to apply only one of the treatments.

Therapists had to be experienced in the treatment

provided in this trial. A limited protocol was

developed for CBT by experts in cooperation with

the participating therapists. The therapists received a

2-day training in using the protocol. Since there are

long waiting lists in the Netherlands, we allowed a

waiting time before the start of the treatment with a

maximum of 4 months for both TAU and CBT.

Furthermore, severely ill patients were allowed to

receive antidepressant medication in addition to the

treatment to which they were randomized. Prescrip-

tion was protocolized in accordance with current

guidelines.

The mean number of days after randomization at

which the treatment started was 75 days (S.D.=61).

The mean number of sessions received was 10

(S.D.=11). The mean treatment duration was 224

days (S.D.=202), with CBT (223 days; S.D.=180)

being significantly shorter than TAU (265 days;

S.D.=232). A considerable proportion of the subjects

(16%) did not receive the treatment they were

randomized to, but this was not significantly different

in CBT and TAU.

2.3. Instruments

Depressive symptomatology was assessed with

the Depression Scale of the Symptom Check List-

90 (SCL-90; Ettema and Arrindell, 2003). This

scale measures the severity of symptoms and

consists of 16 psychological symptoms which must

be rated on a 5-point scale, ranging from 1 (not

distressed by the symptom) to 5 (extremely

distressed by the symptom). A total score can be

obtained by adding the item scores, ranging from

16 (no depressive symptomatology) to 80 (high

level of depressive symptoms). For this study, we

recalculated the SCL-D scores so that 0 was the

lowest score.

Mood and anxiety disorders were assessed at

baseline and 18 to 24 months after baseline, using

the CIDI-Auto (World Health Organization, 1994)

Dutch version (Ter Smitten et al., 1998). The CIDI is a

standardized diagnostic interview for the assessment

of mental disorders, developed by the World Health

Organization. It was designed for use by trained

interviewers who are not clinicians. Its reliability has

been demonstrated to be good to excellent, and the

validity has been demonstrated to be adequate

(Andrews and Peters, 1998; Wittchen, 1994). With

the CIDI, the severity of major depressive disorder

can be assessed (mild, moderate, severe), and a

distinction is made between first or recurrent major

depressive disorder.

2.4. Attrition

The response of the 702 patients initially included

in the study was 65% (n =459) after 1 year and 69%

(n =484) for the last diagnostic interview (at 1.5 to 2

years after baseline).

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146 141

Over the 7 repeated assessments, the majority of

subjects (75%) had missing values on one or more

SCL depression reports but 76% of the sample

completed at least four assessments. One hundred

and thirty-seven subjects (32%) did not participate in

the final clinical interview. Missing one or more self-

reported SCL assessments was not related (at p b0.05

level) to presence of MDD, severe MDD, dysthymic

disorder or anxiety disorder, nor to SCL depression

scores at pre-treatment. Likewise, missing at the final

clinical interview was not related ( p b0.05) to the

presence of MDD, severe MDD, dysthymic disorder

or anxiety disorder, nor to SCL depression scores at

baseline.

2.5. Statistical analyses

We started by analyzing the course of depressive

symptomatology according to the SCL-depression

scale over the follow-up period by means of trajectory

analyses. Trajectory analyses describe the course of

depressive symptomatology through a regression

function, using continuous latent growth factors

(intercept, slope, and quadratic slope). The intercept

represents the level of depressive symptoms at pre-

treatment. The change in depressive symptoms over

time is accounted for by one or more growth factors

i.e. linear slope or quadratic slope.

In the second stage, growth mixture modeling

(GMM) was used to determine the number of

developmental trajectories of depressive symptom-

atology for subjects who received TAU and CBT

(Muthen and Shedden, 1999; Nagin, 1999). The

objective of GMM was to find the smallest number

of classes of subjects with similar courses of

depressive symptoms. GMM estimates mean growth

curves, based on the initial status (intercept) and

change (linear slope/quadratic slope), for each class of

subjects and captures individual variation around

these growth curves by the estimation of factor

variances for each class. Separate GMM models were

analyzed for TAU and CBT.

Different considerations may be used in deciding

on the optimal number of classes (Muthen and

Muthen, 1998–2004). The first is the Bayesian

information criterion (BIC; Kass and Raftery, 1993;

Schwartz, 1978) in which lower BIC values indicate

improvement of the model when compared to the

model with one class less. Another consideration is

the usefulness of the classes, which is the subjective

comparison of the developmental trajectories, the

number of subjects in each class, and the differences

in outcomes between classes. In this study, both

criteria were used.

In the third and final stage, general growth mixture

modeling (GGMM; Muthen and Muthen, 2000) was

used to compare the developmental trajectories of

subjects who received TAU with subjects who

received CBT. In GGMM, the effect of type of

treatment (0 =TAU, 1=CBT) on the course of

depressive symptoms can be estimated by regressing

the continuous latent variables on treatment status in

each of the classes (Muthen et al., 2002). Factors

predicting class membership (presence of a severe

depression, and a comorbid dysthymic disorder or

anxiety disorder) were added to the model. The

estimated parameters of this GGMM were: (1) latent

class membership probabilities, which gives the

probability for each individual to belong to each of

the classes; (2) the means and variances of the

continuous latent growth factors of the depressive

symptoms for each of the classes; (3) estimates of the

regression coefficient of type of treatment on the

continuous latent growth factors for each of the

classes; and (4) the regression coefficients (odds

ratios), predicting class membership for each patient,

by each of the predictor variables.

Trajectory analyses were conducted with Mplus

3.0 (Muthen and Muthen, 1998–2004). Since missing

data on the SCL-depression scores were random, the

missing data module was used to optimally use the

available data.

3. Results

3.1. Course of depressive symptoms

Fig. 2 presents the 1.5-year course of SCL

depression of subjects who received TAU, or CBT.

Regardless of type of treatment, subjects had on

average a similar course of depressive symptoms, with

a significant improvement in SCL depression scores

over time. The percentage of subjects with a DSM-IV

diagnosis of MDD (mild, moderate, or severe),

subjects with a severe MDD, and subjects with an

0

5

10

15

20

25

30

35

0 3 6 9 12 15 18

SC

L D

epre

ssio

n

months

Cognitive Behavior Therapy

Treatment as Usual

Fig. 2. Developmental course of SCL-depression for subjects receiving treatment as usual or cognitive behavior therapy.

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146142

anxiety disorder at post-test (1.5–2 years after

baseline) are presented in Table 1. The large decreases

in the percentages of subjects with any of these

psychiatric conditions at post-test substantiated the

decreases in SCL depression scores. At post-test,

approximately 24% (28–20%) of the subjects had an

MDD. 11% of the subjects who received TAU had a

severe MDD at post-test compared to 5% of the

subjects who received CBT. This difference was

marginally significant (v2 (1, N = 425) = 3.4,

p b0.06). 28% of subjects who received TAU and

19% of subjects who received CBT had an anxiety

disorder at post-test. This was a trend towards

significance (v2 (1, N =425)=2.9, p b0.10), suggest-

ing a (small) positive effect of CBT over TAU.

Table 1

Post-test comparison of prevalence (%) of severe major depressive disord

received treatment as usual or cognitive behavior therapy for the total sam

DSM-IV condition at the end of the study

All patients with MDD Patien

TAU CBT Test TAU

Total sample 28 20 ns 11

Trajectory classes

Class 1 100 64 * 69

Class 2 49 14 * 14

Class 3 22 23 ns 6

Class 4 0 7 ns 0

MDD=major depressive disorder, TAU=treatment as usual, CBT=cogniti

significant.

3.2. Trajectory classes of depression

The number of trajectory classes was identified for

TAU and CBT separately. Two (BIC: TAU=8491,

CBT=7370), three (TAU=8409, CBT=7292), four

(TAU=8371, CBT=7261) and five (TAU=8353,

CBT=7265) trajectory classes were fitted. Fitting

more trajectory classes resulted in non-converging

solutions. Allowing the indicator variance and inter-

cept variance for the last trajectory class (low

depressive symptoms) to be different from the other

trajectory classes improved the fit of the four-class

model (BIC: TAU=8233, CBT=7125). Allowing for

variances to be different between trajectory classes did

not result in a better fit in the five-class model,

er, major depressive disorder, and anxiety disorder for subjects who

ple and for trajectory classes 1 to 4

ts with a severe MDD Patients with an anxiety disorder

(with or without MDD)

CBT Test TAU CBT Test

5 ** 28 19 **

36 ** 100 50 *

0 * 46 31 ns

2 ns 20 13 ns

2 ns 4 9 ns

ve behavior therapy. v2 tests (df =1); *p b0.05; **p b0.10; ns=not

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146 143

compared to the four-class model. It was therefore

decided that the four-class model best fitted the data

for both TAU and CBT. The percentage of subjects

who had missing SCL depression scores (v2 (3,

N =425)=0.9, p N0.05) or who did not participate in

the final clinical interview (v2 (3, N =425)=4.9,

p N0.05) was similar for each of the four trajectory

classes.

In the final model, all subjects were analyzed

simultaneously in the GGMM analyses. Predictor

variables were included if they predicted class-

membership at p b0.05. Using this criterion, three of

the four examined predictors (presence of a severe

MDD, a comorbid dysthymic disorder, and a comor-

bid anxiety disorder) were included in the model,

whereas the fourth examined predictor (first depres-

sion vs. recurrent depression) was not. To account for

the effect of type of treatment on the course of

depressive symptoms in each of the trajectory classes,

the continuous latent variables were regressed on

treatment status. No effect of type of treatment on the

intercepts was found for subjects who were classified

to each of the four classes. This indicates that given a

trajectory class, no differences in SCL depression

scores at baseline were found between TAU and CBT.

To test wheter the results of the final model were

consistent among the samples, two random sub-

samples were drawn and the model was reran. First,

parameters were freely estimated for each of these

sub-samples. We then fixed the parameters at the

0

10

20

30

40

50

0 3 6 9

SC

L D

epre

ssio

n

mon

Fig. 3. Developmental trajectories of SCL-depression for subjects

estimates found in the model on the total sample. For

each sub-sample, it was found that the parameter

estimates were similar to those found in the total

sample (sub-sample 1: v2 (15, N =221)=7.1, p b0.05;

sub-sample 2: v2 (15, N =204)=8.9, p b0.05) indi-

cating that the found results were consistent within

this sample.

3.3. Trajectory classes of depression and type of

treatment

Fig. 3 presents the four resulting trajectory classes.

Ten percent of all subjects were allocated to class 1.

These subjects had the highest SCL depression scores

at baseline. The level of depressive symptoms

remained consistently high in subjects who received

TAU. In contrast, the significant estimate of the type

of treatment on the development of SCL depression

(est.=�1.5, S.E.=0.4, p b0.01) indicated that sub-

jects who received CBT had a significant improve-

ment in their depressive symptoms over the 1.5-year

follow-up period. To assess the clinical relevance of

the mean difference at post-test (after 1.5 years),

Cohen’s d was calculated, indicating the standardized

difference between TAU and CBT at post-test. The

standardized difference was found to be 0.75,

indicating a large effect (Cohen, 1988).

The percentage of subjects who had (severe) MDD

or comorbid anxiety disorder at post-test for each of

the classes is presented in Table 1. The reduced risk of

12 15 18

Class 3

Class 4

ES = .75

ES = .86

Class 1

Class 2

Class 3

Class 4

ths

Treatment as Usual

Cognitive Behavior Therapy

receiving treatment as usual or cognitive behavior therapy.

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146144

having these psychiatric conditions for class 1

subjects who had been treated with CBT, compared

to class 1 subjects who had been treated with TAU,

substantiates the positive effect of CBT over TAU,

which was found in the GGMM analyses on the SCL-

scale. For instance, all class 1 subjects (100%) who

received TAU had an MDD and an anxiety disorder at

post-test, compared to 64% (MDD) and 50% (anxiety)

for class 1 subjects who received CBT.

Twenty-six percent of the subjects were allocated

to class 2. These subjects had, on average, the second

highest SCL depression scores of all subjects at

baseline. All of these subjects improved on their

depressive symptoms. The significant estimate of type

of treatment on the slope of class 2 depressive

symptoms indicated, however, that subjects who

received CBT had significantly larger decreases in

depressive symptomatology than subjects who re-

ceived TAU (est.=�1.3, SE=0.4, p b0.01). The size

of the standardized mean difference of SCL depres-

sive symptoms at post-test is 0.86, which is a large

effect (Cohen, 1988). 14% of the subjects receiving

CBT, compared to 49% of subjects who received TAU

still had an MDD at post-test (Table 1). Likewise,

none of the subjects who received CBT still had a

severe MDD, compared to 14% of those who received

TAU.

The remaining subjects were allocated to class 3

(31%) and class 4 (33%). The differences between

these two trajectory classes are the SCL depression

scores at baseline (S.D. class 3=9.3, S.D. class

4=11.2). The depressive symptoms improved rapidly

to low level (class 3) or almost absence (class 4) of

SCL depressive symptoms after approximately 9–12

months, regardless of type of treatment (Fig. 3). The

percentage of subjects who had a (severe) MDD and/

or anxiety disorder at post-test (Table 1) is similar for

subjects in both treatments, which indicates that CBT

Table 2

Association between trajectory class-allocation and presence of severe de

Comobrid Class 1 vs. Class 4

OR 95% CI

Severe depression 4.6 (1.8–11.8)*

Dysthymic disorder 2.9 (1.1–8.0)*

Anxiety disorder 5.9 (2.3–15.1)*

All participants met criteria for DSM-IV major depressive disorder. All asso

* p b0.05.

and TAU are equally effective for adults in class 3 and

class 4.

3.4. Predicting trajectory classes

Table 2 presents the increase in risk, expressed as

odds ratios, of being allocated to class 1, class 2 or

class 3, compared to class 4 (subjects who completely

recovered) as a function of the presence of a severe

MDD, dysthymic disorder and/or a comorbid anxiety

disorder at baseline. Being classified as class 1 was

predicted by the presence of a severe MDD, the

presence of dysthymic disorder and the presence of an

anxiety disorder at pre-treatment. Class 2 subjects

were predicted by the presence of a severe MDD and

a comorbid anxiety disorder, compared to class 4

subjects. The difference between class 3 and class 4 is

the presence of a severe MDD at baseline.

4. Discussion

We wanted to illustrate how trajectory analyses

techniques can help in identifying subpopulations

with depressive disorders, based on their initial level

of depressive symptomatology and the course of

symptoms over time. We included predictor variables

to identify the subjects in each of the trajectory

classes. We additionally wanted to study if TAU and

CBT had a different impact on the courses of

depressive symptomatology in the different trajectory

classes. We showed that the overall mean course of

depressive symptomatology for subjects treated with

CBT was comparable to the course of symptomatol-

ogy for subjects who received TAU. However, for the

empirically identified subpopulations, this was not

true. CBT proved to be superior to TAU for subjects

with higher initial levels of depressive symptoms.

pression, dysthymic disorder and anxiety disorder at baseline

Class 2 vs. Class 4 Class 3 vs. Class 4

OR 95% CI OR 95% CI

2.7 (1.4–5.4)* 2.6 (1.2–5.5)*

1.6 (0.8–3.2) 1.1 (0.5–2.6)

3.2 (1.5–6.9)* 1.4 (0.6–3.4)

ciations are given as multiple odds ratios (95% confidence interval).

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146 145

These findings are important for several reasons.

First, we demonstrated that several subpopulations

within a population of subjects, diagnosed with MDD,

can indeed be distinguished. This means that not all

patients with MDD follow an identical course of

depressive symptomatology. Further research is nec-

essary to examine whether the four trajectory classes

we found are typical for outpatients or are also found

in other patient groups, for example in primary care.

Comparable trajectories may possibly be found for

patients who do not receive treatment. It is also

important to examine whether these trajectory classes

can be distinguished in subjects with only major

depression, without comorbid anxiety or mood

disorders. If these categories can indeed be validated,

this will allow the development of specific treatments

for each of these categories.

The development of specific treatments for differ-

ent subpopulations diagnosed with MDD seems to be

necessary, especially for the patients with the highest

level of initial depressive symptoms. These patients

do benefit more from CBT compared to treatment as

usual, but still have the highest level of depressive

symptoms at post-test, compared to the other classes.

A more intensive treatment seems to be necessary for

these patients. Possibly, the patients in the categories

with less severe symptoms might improve even with a

minimal intervention.

Second, despite the many indications from primary

research and meta-analyses that different psycho-

therapies are equally effective (Luborsky et al., 2002),

this is not necessarily true when different trajectory

classes are distinguished. This earlier research did not

account sufficiently for the heterogeneity among

patients. In our study, a clinically significant superi-

ority of CBT over TAU was demonstrated for those

most in need of treatment. This suggests that to

thoroughly compare the effectiveness of different

psychotherapies, research should give due consider-

ation to the heterogeneity within samples of patients

with MDD.

A third reason why our finding of different

trajectories is important is that this can strengthen

the research on moderators of effects of interven-

tions. We no longer have to limit this research to

examining specific moderators, which can be

compared to searching for a needle in the haystack,

but we can concentrate on empirically defined

trajectory classes. The identification of predictors

of class allocation will allow us to predict much

better who will benefit most from which type of

intervention. On the basis of the current study,

several moderators were identified. The presence of

a severe MDD, a dysthymic disorder or an anxiety

disorder was used to predict which subjects were

allocated to each of the four trajectory classes.

These pre-existing psychiatric conditions uniquely

predicted who would be classified to each of the

trajectory classes. A severe MDD, a dysthymic

disorder, and a comorbid anxiety disorder predicted

which persons were allocated to class 1. The

presence of a severe MDD and an anxiety disorder

predicted who was to be in class 2. These findings

substantiated the finding that heterogeneity is

present within a sample of clinically referred adults.

The finding that comorbid conditions exacerbate the

severity of depressive symptoms is in accordance

with prior findings (Albus and Scheibe, 1993;

Brown et al., 1996). Future studies should examine

whether comparable or additional moderators can be

identified which can help to identify patients in

need of specific treatments.

There are limitations to this study. First, the

number of subjects in the two classes of subjects

with the most severe depressive symptomatology was

relatively small, resulting in a low statistical power.

The fact that we did find statistically significant

differences between the two treatments, however,

indicates that these effects are considerable. Second,

the interventions were conducted in routine practice

and although intervention fidelity was monitored, we

cannot be sure whether the protocols were always

used as planned. Because of these limitations, we

have to be cautious in the interpretation of these

findings. We also have to stress that the methods of

analysing the data we have used in this study are quite

complicated, which may limit the possibilities of

using it in future research.

Despite these limitations, we found clear indica-

tions that there are not only important differences

between patients with depressive disorders but also

that the effects of psychotherapies are different for

different classes of patients. This results can be used

for future classifications of patients which indicate

which treatment is most effective for which type of

patient.

P. Cuijpers et al. / Journal of Affective Disorders 89 (2005) 137–146146

References

Albus, M., Scheibe, G., 1993. Outcome of panic disorder with or

without concomitant depression: a 2-year prospective follow-up

study. Am. J. Psychiatry 150, 1878–1880.

Andrews, G., Peters, L., 1998. The psychomentric properties of the

composite international diagnostic interview. Soc. Psychiatry

Psychiatr. Epidemiol. 33, 80–88.

Barrett, J.E., Williams Jr, J.W., Oxman, T.E., Frank, E., Katon, W.,

Sullivan, M., 2001. Treatment of dysthymia and minor

depression in primary care: a randomized trial in patients aged

18 to 59 years. J. Fam. Pract. 50, 405–412.

Beutler, L.E., 2002. The dodo bird is extinct. Clin. Psychol. Sci.

Pract. 9, 30–34.

Brown, C., Schulberg, H.C., Madonia, M.J., Shear, M.K., Houck,

P.R., 1996. Treatment outcomes for primary care patients with

major depression and lifetime anxiety disorder. Am. J.

Psychiatry 153, 1293–1300.

Churchill, R., Hunot, V., Corney, R., Knapp, M., McGuire, H.,

Tylee, A., 2001. A systematic review of controlled trials of the

effectiveness and cost-effectiveness of brief psychological

treatments for depression. Health Technol. Assess. 5, 1–173.

Cohen, J., 1988. Statistical Power Analysis for the Behavioral

Sciences, 2nd ed. Erlbaum, Hillsdale, NJ.

Dowrick, C., Dunn, G., Ayuso-Mateos, J.L., Dalgard, O.S., Page,

H., Lehtinen, V., 2000. Problem solving treatment and group

psychoeducation for depression: multicentre randomised con-

trolled trial. Outcomes of Depression International Network

(ODIN) Group. Br. Med. J. 321, 1450–1454.

Duncan, B.L., 2002. The legacy of Saul Rosenweig: the profundity

of the dodo bird. J. Psychother. Integr. 12, 32–57.

Ettema, J.H.M., Arrindell, W.A., 2003. SCL-90. Handleiding bij een

multidimentionele psychopathologie-indicator. Swets Zeitlinger,

Lisse. SCL-90, manual of the Dutch version.

Gaffan, E.A., Tsaousis, I., Kemp-Wheeler, S.M., 1995. Researcher

allegiance and meta-analysis: the case of cognitive therapy for

depression. J. Consult. Clin. Psychol. 63, 966–980.

Kass, R.E., Raftery, A.E., 1993. Bayes factors. J. Am. Stat. Assoc.

90, 773–795.

Kazdin, A.E., Bass, D., 1989. Power to detect differences between

alternative treatments in comparative psychotherapy outcome

research. J. Consult. Clin. Psychol. 57, 138–147.

Luborsky, L., Singer, B., Luborsky, L., 1975. Comparative studies

of psychotherapies; is it true that beverybody has won and all

must have prizesQ? Arch. Gen. Psychiatry 32, 995–1008.

Luborsky, L., Rosenthal, R., Diguer, L., 2002. The dodo bird verdict

is alive and well—mostly. Clin. Psychol. Sci. Pract. 9, 2–12.

Muthen, B., 2001. Second-generation structural equation modeling

with a combination of categorical and continuous latent

variables: new opportunities for latent class-latent growth

modeling. In: Collins, L.M., Sayer, A.G. (Eds.), New Methods

for the Analysis of Change. American Psychological Associa-

tion, Washington, DC, pp. 291–322.

Muthen, L.K., Muthen, B.O., 1998–2004. Mplus. Statistical

Analyses with Latent Variables. User’s Guide, 3.0 ed. Muthen

& Muthen, Los Angeles.

Muthen, B., Muthen, L.K., 2000. Integrating person-centered and

variable-centered analysis: growth mixture modeling with latent

trajectory classes. Alcohol., Clin. Exp. Res. 24, 882–891.

Muthen, B.O., Shedden, K., 1999. Finite mixture modeling with

mixture outcomes using the EM algorithm. Biometrics 55,

463–469.

Muthen, B.O., Brown, C.H., Masyn, K., Jo, B., Khoo, S.T., Yang,

C.C., 2002. General growth mixture modeling for randomized

preventive interventions. Biostatistics 3, 459–475.

Nagin, D.S., 1999. Analyzing developmental trajectories: a

semiparametric, group-based approach. Psychol. Methods 4,

139–177.

Reich, J., 2003. The effects of axis II disorders on the outcome of

treatment of anxiety and unipolar depressive disorders: a review.

J. Pers. Disord. 17, 387–405.

Reynolds III, C.F., Frank, E., Perel, J.M., Imber, S.D., Cornes, C.,

Miller, M.D., 1999. Nortriptyline and interpersonal psychother-

apy as maintenance therapies for recurrent major depression: a

randomized controlled trial in patients older than 59 years.

JAMA 281, 39–45.

Schwartz, 1978. Estimating the dimension of a model. Ann. Stat. 6,

461–464.

Scott, J., 2001. Cognitive therapy for depression. Br. Med. Bull. 57,

101–113.

Shadish, W.R.J., Sweeney, R.B., 1991. Mediators and moderators in

meta-analyses: there’s a reason we don’t let dodo birds tell us

which psychotherapies should have prizes. J. Consult. Clin.

Psychol. 59, 883–893.

Ter Smitten, M.H., Smeets, R.M.W., Van den Brink, W., 1998.

Composite International Diagnostic Interview (CIDI), version

2.1, 12-months. World Health Organization, Amsterdam/Gene-

va. (in Dutch).

Van Balkom, A.J., Bakker, A., Spinhoven, P., Blaauw, B.M.,

Smeenk, S., Ruesink, B., 1997. A meta-analysis of the treatment

of panic disorder with or without agoraphobia: a comparison of

psychopharmacological, cognitive–behavioral, and combination

treatments. J. Nerv. Ment. Dis. 185, 510–516.

Van Straten, A., Tiemens, B., Hakkaart, L., Nolen, W.A., Donker,

M., submitted for publication. Effectiveness of stepped care for

mood and anxiety disorders.

Williams, J.W.J., Oxman, T., Frank, E., Katon, W., Sullivan, M.,

Cornell, J., 2000. Treatment of dysthymia and minor depression

in primary care: a randomized controlled trial in older

adolescents. JAMA 284, 1519–1526.

Wittchen, H.U., 1994. Reliability and validity studies of the WHO-

Composite International Diagnostic Interview (CIDI): a critical

review. J. Psychiatr. Res. 28, 57–84.

World Health Organization, 1994. Composite International Diag-

nostic Interview (CIDI), Version 2.1, 12-Months. World Health

Organization, Geneva. Wittchen.