Acta Psychiatr Scand 2005: 111: 144–149 Copyright Ó Blackwell Munksgaard 2004

All rights reserved ACTA PSYCHIATRICA

DOI: 10.1111/j.1600-0447.2004.00440.x
SCANDINAVICA

Validation of the Bech–Rafaelsen

Melancholia Scale and the Hamilton
Depression Scale in patients with major
depression; is the total score a valid measure
of illness severity?
Licht RW, Qvitzau S, Allerup P, Bech P. Validation of the Bech– R. W. Licht1, S. Qvitzau2,
Rafaelsen Melancholia Scale and the Hamilton Depression Scale in P. Allerup3, P. Bech4
patients with major depression; is the total score a valid measure of 1
Mood Disorders Research Unit, Aarhus University
illness severity? Psychiatric Hospital, Risskov, 2Pfizer Denmark,
Acta Psychiatr Scand 2005: 111: 144–149. Ó Blackwell Munksgaard 2004. Copenhagen, 3The Danish University of Education,
Copenhagen and 4Psychiatric Research Unit,
Objective: Evaluation of antidepressant drug efficacy requires Frederiksborg General Hospital, Hillerød, Denmark
adequate rating scales for measuring the severity of depression.
However, to measure the illness severity by such a total score, the scale
needs to fulfil criteria of unidimensionality. On this background, we
aimed at comparing the unidimensionality of the Bech–Rafaelsen
Melancholia Scale (MES) and the 17-item Hamilton Depression
Rating Scale (HAM-D17).
Method: A total of 1629 patients aged between 18 and 65 years with a
major depressive episode were treated openly with sertraline at a fixed
oral dose of 50 mg daily during 4 weeks. The HAM-D17 and the MES
were applied at baseline and at weeks 2 and 4. Unidimensionality was
tested with Mokken and Rasch analysis.
Results: Unidimensionality of the HAM-D17 could not be confirmed.
However, the 6-item Hamilton Depression Subscale (HAM-D6), was Key words: major depressive disorder; acute therapy;
accepted by the Rasch analysis both at baseline and after 2 and 4 weeks sertraline; rating scales; validity; antidepressant drug
of therapy. For the MES (as well as for the HAM-D6), a Loevinger trials
coefficient of homogeneity above 0.40 (suggesting acceptance) was Rasmus W. Licht, Mood Disorders Research Unit,
found at week 4. Aarhus University Psychiatric Hospital, 8240 Risskov,
Conclusion: The HAM-D6 and the MES did fulfil criteria for Denmark.
unidimensionality while the HAM-D17 did not. Therefore, the E-mail: rl@psykiatri.aaa.dk
extended use of the HAM-D17 in drug trials may be questioned.
Accepted for publication July 16, 2004

implies that scorings on lower prevalence items

Introduction
Evaluation of antidepressant drug efficacy requires
adequate rating scales for measuring the severity of
depression. When such rating scales are applied,
generally the individual item scores are summed up
to give a total score. However, to measure the
illness severity by such a total score, the scale needs
to fulfil essential criteria of unidimensionality.
Thus, to be additive, the individual items of a
scale must be consistently rank-ordered according
to their relation to the severity of illness. This
presuppose scorings on higher prevalence items.
The statistical analysis for unidimensionality of
rating scales is referred to as item response theory
analysis (1, 2).
Only a few of the available rating scales for
depression have been properly tested for the
fulfilment of criteria of unidimensionality by item
response theory analysis. Among the rating scales,
the Bech–Rafaelsen Melancholia Scale (MES) (3)
has been shown to fulfil such criteria (4–6), whereas
the Hamilton Depression Rating Scale (HAM-D17)

144

(7) and the Montgomery-Åsberg Depression scale
(8) have been shown not to fulfil such criteria (4, 6,
9). However, a subscale of the HAM-D17, the
HAM-D6 (10), consisting of the items of depressed
mood, psychic anxiety, guilt, work and interests,
tiredness and motor retardation, was found to be
unidimensional (11). What is mentioned here about
the validity of the HAM-D17 can be generalized to
the 21-item version (the HAM-D21), which include
additional four items of diurnal variation, dereal-
ization (or depersonalization), paranoid symptoms
and obsessional symptoms (12).
Aims of the study
Despite the consistency in the reports cited above,
a major limitation in these studies is the relatively
small sample sizes and the heterogeneity of the
samples. On this background, we aimed at com-
paring the unidimensionality of the MES, the
HAM-D17 and the HAM-D6 in a large sample of
patients with major depression, all treated with
sertraline at a fixed daily dose of 50 mg in the
context of a drug trial.
Material and methods
Patients
The patients were originally collected for the
purpose of a drug trial (13). To be included, a
patient should be between 18 and 65 years of age,
suffering from a major depressive episode without
psychotic symptoms within a major depressive
disorder according to DSM-IV (14) and have a
HAM-D17 total score of 18 or more. The exclusion
criteria were: current substance abuse; risk of
suicide; treatment with any antidepressant or
lithium in the last 2 weeks (4 weeks in the case of
fluoxetine); currently receiving and unable to
discontinue any psychotropic apart from hypnotics
and oxazepam; previous adequate treatment with
sertraline without a clinically significant effect; and
serious somatic illness. According to these criteria,
a total of 1629 patients were recruited from 30
Danish and nine Icelandic centres between Novem-
ber 1996 and January 2000. All patients gave
informed, written consent to participate in the
study, which was approved by all relevant Ethical
Committees and by the National Board of Health
in Denmark and Iceland.
Design
The design of the drug trial has been described in
details elsewhere (13). Briefly, the drug trial was
Internal validity of two depression rating scales
divided into two parts. In the first part, the patients
were treated openly with sertraline at a fixed oral
dose of 50 mg daily during the first 4 weeks.
Patients not responding (achieving a 50% reduc-
tion or more in the HAM-D17 total score) after
these 4 weeks of therapy were treated for two
additional weeks with 100 mg sertraline daily. In
the second part, the patients who had still not
responded after the 6 weeks of treatment were now
randomly and double-blindly allocated to one of
the following three arms over the next 5 weeks:
100 mg sertraline daily plus 30 mg mianserin,
100 mg sertraline daily plus placebo-mianserin or
200 mg sertraline daily plus placebo-mianserin.
Rating scales
In the trial, the HAM-D17 was used as the primary
outcome scale, and the MES was used as the
secondary outcome scale.
The HAM-D17 consists of 17 items, nine of
which are scored on a five-category Likert scale
(from 0 to 4) and the remaining eight on a three-
category Likert scale (from 0 to 2), giving a
theoretical total score range from 0 to 52. From
the HAM-D17, the total score on the HAM-D6
(defined in the Introduction) was extracted, ranging
theoretically from 0 to 22. The MES consists of 11
items which are all scored on a five-category Likert
scale giving a theoretical total score range from 0
to 44. These instruments were applied at baseline
and at week 2, 4, 6, 7, 8 and 11. For each patient,
the clinical assessments were made by the same
clinician at the same time of the day, preferably
between 8 and 11 am. The assessment of inter-rater
reliability with respect to the total rating scale
scores was achieved through joint ratings of video-
taped patient ratings.
As the responding patients by design were
withdrawn from the study after the first 4 weeks
of treatment and as the dose was increased (in non-
responders) hereafter, only the ratings from the
first 4 weeks were used in the present psychometric
analyses.
Statistical analysis
Both the one-parameter Rasch analysis (1) and the
non-parametric Mokken analysis (2) have been
used to test rating scales by item response theory
models. The one-parameter Rasch analysis (1, 15)
is based on a logistic function analysis of the
response probabilities for each patient–item com-
bination. By this procedure, item-specific curves,
describing the relationship of each item to the
latent dimension (e.g. severity of depression) are
145
Licht et al.

generated. In the one-parameter model, the items eigenvalue could be observed, indicating unidimen-
only differ by their threshold parameter (the sionality (21).
location of the item with respect to the latent
dimension), assuming that all items in the scale
Results
under examination have the same slope. The Rasch
analysis is used to evaluate how well the data Table 1 shows the mean and standard deviation of
obtained with a rating scale fit into the model. The the scores of the three outcome scales during the
one-parameter Rasch Model is equivalent to total 4 weeks of treatment with a daily oral dose of
scores being sufficient statistics, i.e. exhaust all 50 mg sertraline. Of the total of 1629 included
item- and patient-information which, if rejected, patients, 170 (10.4%) patients dropped out during
have practical negative consequences for the prac- this period. The mean score of the three outcome
tical use of simple sum scores across the items (16, scales after 4 weeks of therapy was still at the level
17). The two-parameter alternative may be consid- of clinical depression as defined by the various
ered, especially if prior knowledge to the item scales. The response rate after 4 weeks (number of
discriminations (slopes of the item-specific curves), patients achieving a 50% reduction in the total
being the second parameter, is available. In this score, in percentage of included patients) with the
case weighted item scores are sufficient statistics MES and the HAM-D17 was 28 and 26%,
and are the means for comparing patients. The respectively.
fundamental property of objective comparisons is, Table 2 shows the internal validity of the three
however lost with the introduction of item dis- outcome scales in period 1 using baseline data as
crimination parameters (18). The so-called three- well as data after 2 and 4 weeks of therapy. After
parameter Rasch Model includes a parameter for 4 weeks of therapy, factor analysis still identified
guessing and therefore for the purpose of this five factors in the HAM-D17, whereas in the HAM-
study, it is not considered as a relevant alternative D6 only one factor was identified. In the MES, two
to the one-parameter Rasch Model. factors were identified after 4 weeks. However, the
The non-parametric Mokken analysis (2, 19) is first factor in the MES explained approximately
concerned with a weaker property of the scale items 50% of the variance. The second factor of the five
compared with the requirements set by the Rasch HAM-D17 factors was an agitation-anxiety factor
Model. In fact this type of analysis aims at assessing and the third a somatic factor including the sleep
the level of simple ordinal scaling by the items, i.e. items and gastrointestinal symptoms.
whether the items create a simple ordering.
The Rasch analysis has been performed by the Table 1. Mean values (standard deviation) of total scores on the MES, the HAM-
method described by Allerup (9). In the analysis, a D6 and the HAM-D17
level of rejection of P < 0.01 has been used. The
N MES HAM-D6 HAM-D17
Mokken analysis is based on a Loevinger coeffi-
cient of homogeneity, which assess the degree of Baseline 1629 21.1 (3.6) 12.2 (1.8) 23.3 (3.2)
rank-ordering of items in terms of transmitted 2 weeks 1581 17.8 (4.6) 10.5 (2.6) 19.4 (5.0)
4 weeks 1459 14.0 (6.3) 8.4 (3.5) 15.1 (6.5)
information concerning the dimension under
investigation (20). The level of acceptance is a
Loevinger coefficient of homogeneity of 0.40 or
Table 2. The internal validity of the MES, the HAM-D6 and the HAM-D17
higher, while a coefficient of 0.30 to 0.39 is
considered to be only just acceptable. A coefficient MES HAM-D6 HAM-D17
between 0.20 and 0.29 is doubtful.
Cronbach alpha
Classical psychometric means for testing internal Baseline 0.56 0.20 0.14
validity of a rating scale, i.e. the Cronbach coef- 2 weeks 0.77 0.59 0.67
ficient alpha and factor analysis, have been inclu- 4 weeks 0.88 0.80 0.83
ded for illustrative purposes. In particular the issue Loevinger coefficent
Baseline 0.13 0.06 0.01
of unidimensionality was addressed in these ana- 2 weeks 0.28 0.24 0.13
lyses. A coefficient alpha between 0.60 and 0.80 has 4 weeks 0.49 0.49 0.28
been used to indicate internal consistency (21). The Number of factors in factor analysis
Baseline 4 3 7
factor analysis approach was run as a standard
2 weeks 3 3 6
Principal Component Analysis with orthogonal 4 weeks 2 1 5
varimax rotation of the components. A key point Explained variance by first factor (%)
was to study the sequence of ordered eigenvalues Baseline 23.7 23.8 14.6
2 weeks 32.2 34.3 17.5
(considering only those above 1.0) to see, if a 4 weeks 46.8 50.9 27.8
substantial shift between the first and the second

146

For the MES (as well as for the HAM-D6), a
Loevinger coefficient of homogeneity above 0.40
was found at week 4. For the HAM-D17 the
Loevinger coefficient of homogeneity was below
0.30, even after 4 weeks. Furthermore, by the
Rasch analysis, the HAM-D17 was neither accep-
ted at baseline nor after 2 or 4 weeks of therapy,
whereas the Rasch analysis showed that the HAM-
D6 was accepted both at baseline and after 2 and
4 weeks of therapy. In the Rasch analysis of the
MES, two of the MES items (suicidal thoughts and
sleep disturbances) were not accepted by the one-
parameter Rasch analysis.
Discussion
This study has definitely shown that both the
HAM-D6 and the MES did fulfil criteria for
unidimensionality while the HAM-D17 did not.
Thus, after 4 weeks of treatment with sertraline, an
acceptable Loevinger coefficient was obtained for
the HAM-D6 and the MES but not for the HAM-
D17. Not surprisingly, the coefficients were lower at
baseline where the dispersion of scale scores was
more limited than after 4 weeks of treatment.
Therefore, these findings robustly show that sum-
ming up the scores of individual items to a total
score is valid for the MES and the HAM-D6 but
not for the HAM-D17. At least this may be
generalized to moderately depressed populations
of outpatients receiving SSRIs. However, the
superiority of the MES in comparison with the
HAM-D17 in terms of unidimensionality has also
been shown in populations of depressed inpatients
receiving other antidepressant than SSRIs (4, 6)
and the unidimensionality of the MES has been
confirmed in depressed in-patients receiving no
treatment (5).
Given the fact that the HAM-D17 has been used
in over 95% of all controlled trials with SSRIs (22),
these findings are quit remarkable. One way of
understanding this, is that the HAM-D17 may
simply be used by tradition, i.e. due to the fact that
most clinicians (and reviewers) are familiar with
this instrument. This was the background for
choosing the HAM-D17 as the primary outcome
measure in the present drug trial. Using a specific
rating scale by tradition is also what is seen in the
field of mania, where the Young Mania Rating
Scale (23) is the most frequently used rating scale
despite its potential limitations concerning unidi-
mensionality (24). Another potential reason for the
wide use of the HAM-D17 despite its lack of
unidimensionality is that unidimensionality may be
considered only to have theoretical interest. How-
ever, it is well known that another aspect of the
Internal validity of two depression rating scales
validity of a scale, namely its responsiveness, i.e.
the ability of a scale to detect changes over time,
has substantial impact on clinical research. It has
been found that the HAM-D6 had a higher effect
size than the HAM-D17 when all placebo-con-
trolled trials with fluoxetine were analysed (0.38 vs.
0.30) (25), and subsequently, that it would there-
fore require one-third fewer patients in studies
using the HAM-D6 than in studies using the
HAM-D17 to detect a given difference (26).
Generally, the responsiveness of the MES and
the HAM-D6 is considered to be higher than that
of the HAM-D17, as recently reviewed by Bech
(27). Only a few studies have compared the
responsiveness of the MES directly with that of
the HAM-D17 (4, 6). In a psychometric reanalysis
of the pooled results of four randomized compar-
isons of moclobemid with a tricyclic (nortriptyline
or clomipramine) the responsiveness of the MES
was found to exceed that of the HAM-D17 in terms
of effect sizes (showing superiority of the tricyc-
lics), with values of 0.53 and 0.47 for the MES and
the HAM-D17, respectively (6).
Unfortunately, the database of the present drug
trial did not allow a direct comparison of the scales
in terms of responsiveness, as the patients who did
respond (showing at least a 50% reduction of the
HAM-D17 total score) at 4 weeks or at 6 weeks
were withdrawn from the study by design, and
beyond these points in time they were not followed
with ratings. However, the overall response pat-
terns may provide some relevant information.
When the three treatment groups, i.e. patients
receiving 100 mg sertraline daily plus 30 mg
mianserin, 100 mg sertraline daily plus placebo-
mianserin or 200 mg sertraline daily plus placebo-
mianserin (from week 6 and on) were pooled, the
accumulated rates of response (defined as at least a
50% reduction of total rating scale score from
baseline to end-point) after 11 weeks were 73 and
71% with the MES and the HAM-D17, respect-
ively. However, when comparing the three treat-
ment groups (intention-to-treat population) in
terms of response (with the HAM-D17 as the
primary outcome measure), which was the aim of
the study, continuing the treatment with 100 mg/
day of sertraline resulted in response in 70% of the
patients still non-responding after the first 6 weeks
of treatment, similar to the response rate (67%)
obtained in the patients who had mianserin added,
whereas increasing the sertraline dose to 200 mg/
day resulted in a lower response rate at 56%
(P < 0.05) (13). Similar results were seen in the
completers. When comparing the three groups on
the secondary measure (response defined as at least
a 50% reduction of the MES total score from
147
Licht et al.
baseline to end-point), response rates of 65, 67 and
64% appeared. Although the overall responsive-
ness of the MES and the HAM-D17 in terms of
accumulated response rates seem comparable, the
differential responsiveness of the HAM-D17 in
the three treatment groups may be a result of the
multidimensional nature of the HAM-D17, as
suggested by the present factor analysis and
confirmed by the latent structure analysis, as
serotonergic side-effects my be captured on the
agitation-anxiety dimension or on the somatic
dimension (including the sleep items and gastroin-
testinal symptoms) leading to an artificially
increased HAM-D17 total score. This is an example
of a drug-specific (or dose specific) reduced respon-
siveness due to lack of unidimensionality of a scale.
This phenomenon may also explain that the HAM-
D6 had a higher effect size than the HAM-D17
when all placebo-controlled trials with fluoxetine
were analysed as mentioned above. On the other
hand, lack of unidimensionality may also lead to a
falsely increased responsiveness. As recently poin-
ted out and exemplified by Möller (28), the HAM-
D17 (and the HAM-D21) includes three items of
sleep and one of weight, which may lead to an
inherent bias in favour of measuring antidepres-
sant response to tricyclic antidepressants with a
sedative profile when using the HAM-D17 (or the
HAM-D21) total score as outcome measure. Based
on a psychometric reanalysis of the results of two
randomized trials comparing sertraline and ami-
triptyline (a sedative tricyclic) in depressed in-
patients (29) and in depressed outpatients (30), it
was shown that the apparent tendency towards a
superior efficacy of amitriptyline could be
explained by psychometric bias due to the multi-
dimensional structure of the the HAM-D21 (28). As
pointed out by Möller (28), such bias would be
most likely to occur with sedative antidepressants,
and with any drug given in doses associated with
considerable side-effects as illustrated by the pre-
sent data set. Another example of a potential
psychometrically induced bias is a randomised 5-
week comparison of citalopram and clomipramine
showing clear superiority of clomipramine in terms
of complete response defined on the basis of the
HAM-D17 total score (31). As mentioned by the
authors, at the third to fifth week, the three sleep
items accounted for about 40% of the drug
difference on the HAM-D17.
As mianserin is a sedative agent, one could
expect that using the HAM-D17 as outcome
measure in the present drug trial would favour
the treatment arm containing this drug. However,
no superiority of the combination of sertraline and
mianserine could be found (13).
148
As to the relation between unidimensionality
and responsiveness, i.e. the two major aspects of
internal and external validity, respectively, the first
characteristic is a prerequisite of the latter. Lack of
unidimensionality may as exemplified above and
by the present data set lead to an artificially
increased (biased) responsiveness (as in the case of
evaluation of sedative tricyclics) or to an artificially
reduced responsiveness (as in the case of evaluation
of SSRIs).
A major implication of the present findings is
that the HAM-D17 total score definitely cannot be
considered as a valid measure of the severity of
depression, which generally may question the
extended use of this scale in drug trials. As the
unidimensionality of the MES and the HAM-D6
could be confirmed, these scales are to be preferred
over the HAM-D17 in trials evaluating drugs for
depression. The most appropriate use of HAM-D17
data is to present the profile scores of the various
dimensions of the scale.
Acknowledgements
The authors thank all the investigators for providing data on
individual patients.
References
1. Bond TG, Fox CH. Applying the Rasch Model. London:
Lawrence Erlbaum, 2001.
2. Sijtsma K, Molemaar IW. Introduction to nonparametric
item response theory. London: Sage Publications, 2002.
3. Bech P, Rafaelsen OJ. The use of rating scales exemplified
by a comparison of the Hamilton and the Bech-Rafaelsen
Melancholia Scale. Acta Psychiatr Scand 1980;62(suppl.
285):128–132.
4. Maier W, Philipp M, Heuser I, Schlegel S, Buller R, Wetzel
H. Improving depression severity assessment – I. Reliabil-
ity, internal validity and sensitivity to change of three
observer depression scales. J Psychiatr Res 1988;22:3–12.
5. 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. Pharma-
copsychiatry 1990;23:243–245.
6. Bech P, Stage KB, Nair NP, Larsen JK, Kragh-Sorensen P,
Gjerris A. The Major Depression Rating Scale (MDS).
Inter-rater reliability and validity across different settings
in randomized moclobemide trials. Danish University
Antidepressant Group. J Affect Disord 1997;42:39–48.
7. Hamilton M. A rating scale for depression. J Neurol Neu-
rosurg Psychiatry 1960;23:56–62.
8. Montgomery SA, Asberg M. A new depression scale de-
signed to be sensitive to change. Br J Psychiatry 1979;
134:382–389.
9. Allerup P. Statistical analysis of rating scales. Copenhagen:
Danish Institute of Educational Research, 1986.
10. Bech P, Gram LF, Dein E, Jacobsen O, Vitger J, Bolwig TG.
Quantitative rating of depressive states. Acta Psychiatr
Scand 1975;51:161–170.

11. Bech P, Allerup P, Gram LF et al. The Hamilton Depres-
sion Scale. Evaluation of objectivity using logistic models.
Acta Psychiatr Scand 1981;63:290–299.
12. Hamilton M. Development of a rating scale for primary
depressive illness. Br J Soc Clin Psychol 1967;6:278–296.
13. Licht RW, Qvitzau S. Treatment strategies in patients with
major depression not responding to first-line sertraline
treatment. A randomised study of extended duration of
treatment, dose increase or mianserin augmentation. Psy-
chopharmacology (Berl) 2002;161:143–151.
14. American Psychiatric Association. Diagnostic and statis-
tical manual of mental disorders; (DSM-IV), 4th edn.
Washington, DC: American Psychiatric Association, 1994.
15. Rasch G. Probabilistic models for some intelligence and
attainment tests. Studies in mathematical psychology. I.
Copenhagen: Pedagogic Institute in Denmark, 1960.
16. Fischer G, Molenaar IW. Rasch models: foundations,
recent developments and applications. Berlin: Springer,
1994.
17. Allerup P. Rasch measurement, Theory of. In: Husen T,
Postlethwaite N. The International Encyclopedia of
Education, Vol. 8. London: Pergamon, 1994:4902–4913.
18. Allerup P. Statistical analysis of data from the IEA reading
literacy study. In: Rost J, Langeheine R, eds. Applications
of latent trait and latent class models in social sciences.
Münster: Waxman, 1997:50–58.
19. Mokken RJ. Theory and procedure of scale analysis. Berlin:
Monton, 1971.
20. de Jong A, Molenar IW. An application of Mokken’s model
for stochastic, cumulative scaling in psychiatric research.
J Psychiatr Res 1987;2:137–149.
21. Nunnally JC, Bernstein LH. Psychometric theory. New
York: McGraw-Hill, 1994.
22. Bech P. Pharmacological treatment of depressive disorders:
a review. In: Maj M, Sartorius N, eds. Depressive disorders.
Internal validity of two depression rating scales
WPA series evidence and experience in psychiatry. Chi-
chester: John Wiley & Sons, 1999:89–127.
23. Young RC, Biggs JT, Ziegler VE, Meyer DA. A rating scale
for mania: reliability, validity and sensitivity. Br J Psychi-
atry 1978;133:429–435.
24. Licht RW, Jensen J. Internal validation of the Bech Ra-
faelsen Mania Scale using latent structure analysis. Acta
Psychiatr Scand 1997;96:367–372.
25. Bech P, Cialdella P, Haugh MC et al. Meta-analysis of
randomised controlled trials of fluoxetine v. placebo and
tricyclic antidepressants in the short-term treatment of
major depression. Br J Psychiatry 2000;176:421–428.
26. Faries D, Herrera J, Rayamajhi J, DeBrota D, Demitrack M,
Potter WZ. The responsiveness of the Hamilton Depres-
sion Rating Scale. J Psychiatr Res 2000;34:3–10.
27. Bech P. The Bech-Rafaelsen Melancholia Scale (MES) in
clinical trials of therapies in depressive disorders: a 20-year
review of its use as outcome measure. Acta Psychiatr Scand
2002;106:252–264.
28. Möller HJ. Methodological aspects in the assessment of
severity of depression by the Hamilton Depression Scale.
Eur Arch Psychiatry Clin Neurosci 2001;251(suppl. 2):
II13–II20.
29. Möller HJ, Gallinat J, Hegerl U et al. Double-blind,
multicenter comparative study of sertraline and amitript-
yline in hospitalized patients with major depression.
Pharmacopsychiatry 1998;31:170–177.
30. Möller HJ, Glaser K, Leverkus F, Gobel C. Double-blind,
multicenter comparative study of sertraline versus ami-
triptyline in outpatients with major depression. Pharma-
copsychiatry 2000;33:206–212.
31. Danish University Antidepressant Group. Citalopram:
clinical effect profile in comparison with clomipramine.
A controlled multicenter study. Psychopharmacology
(Berl) 1986;90:131–138.
149