JOURNALOF

PSYCHIATRIC

Journal of Psychiatric Research 38 (2004) 577–582RESEARCH

www.elsevier.com/locate/jpsychires

Derivation of a definition of remission on theMontgomery–Asberg depression rating scale corresponding to

the definition of remission on the Hamilton rating scale for depression

Mark Zimmerman*, Michael A. Posternak, Iwona Chelminski

Department of Psychiatry and Human Behavior, Brown University School of Medicine, Rhode Island Hospital, 235 Plain Street,

Suite 501, Providence RI 02905, USA

Received 14 November 2003; received in revised form 5 March 2004; accepted 15 March 2004

Abstract

During the past decade the Montgomery–Asberg Depression Rating Scale (MADRS) has been used with increasing frequency to

measure outcome in antidepressant efficacy trials (AETs). In characterizing treatment outcome in AETs it is common to define

treatment remission as a score below a predetermined cutoff score on the scale. Various cutoffs have been used to define remission on

the MADRS. The goal of the present paper is to determine the cutoff on the MADRS that most closely corresponds to the cutoff

most frequently used on the Hamilton Rating Scale for Depression to define remission. Three hundred and three psychiatric

outpatients who were being treated for a DSM-IV major depressive episode were rated on the HRSD and the MADRS. A linear

regression equation was computed to estimate MADRS scores from HRSD scores. After deriving the regression equation, we

computed the MADRS score corresponding to an HRSD score of 7. We also examined the sensitivity, specificity and overall

classification rate of the MADRS for identifying remission on the HRSD. Based on the equation from a linear regression analysis

for the entire sample, a MADRS score of 611 would correspond to a score of 67 on the HRSD. We repeated the analysis after

excluding the more severely depressed patients who currently met criteria for MDD, and based on the equation from this regression

analysis a MADRS score of 610 would correspond to a score of 67 on the HRSD. In a complementary analysis, we examined the

sensitivity, specificity and overall classification rate of the MADRS at different cutoff points for identifying remission, and found

that a cutoff of 610 maximized the level of agreement with the HRSD definition of remission. In conclusion, the regression equation

relating HRSD and MADRS scores is dependent, in part, on the range and severity of scores in the sample. To facilitate com-

parisons of studies using the HRSD and MADRS our results suggest that a cutoff of 10 on the MADRS is equivalent to the HRSD

cutoff of 7.

� 2004 Elsevier Ltd. All rights reserved.

Keywords: Depression; Remission; Hamilton depression scale; Montgomery–Asberg depression rating scale

1. Introduction

In antidepressant efficacy trials (AETs) outcome istypically measured on standardized instruments, the two

most common being the Hamilton Rating Scale for

Depression (HRSD) (Hamilton, 1960) and the Mont-

gomery–Asberg Depression Rating Scale (MADRS)

* Corresponding author. Tel.: +1-401-277-0724; fax: +1-401-277-

0726.

E-mail address: mzimmerman@lifespan.org (M. Zimmerman).

0022-3956/$ - see front matter � 2004 Elsevier Ltd. All rights reserved.

doi:10.1016/j.jpsychires.2004.03.007

(Montgomery and Asberg, 1979). For the past thirty

years the HRSD has been the most widely used outcome

measure in AETs (Prien et al., 1991), though during thepast decade the MADRS has been used with increasing

frequency (Khan et al., 2002). The HRSD was intended

as a measure of the severity of depressive symptoms,

whereas the MADRS was designed to be particularly

sensitive to change in patients treated with antidepres-

sant medication.

In describing treatment outcome in AETs it is com-

mon to define treatment response as a 50% or moreimprovement in scores on the HRSD or MADRS, and

Table

1

StudiesoftheassociationbetweentheMontgomery–Asbergdepressionratingscale

(MADRS)andtheHamiltonratingscale

fordepression(H

RSD)

Author

Patients

#Item

son

HRSD

MADRSmean

(SD)

HRSD

mean

(SD)

Correlation

coeffi

cient

Comment

Davidsonet

al.

44Inpatients

a32.2

(8.4)

27.0

(6.8)

0.46

Differentraters

completedHRSD

andMADRS

Dratcuet

al.

40Patients

–pre-treatm

ent(18inpatients,5

daypatients,17outpatients)

17

38.5

(8.1)

31.2

(6.1)

0.89

10Depressed

patients

after

4weeksoftreatm

ent

12.9

(16.1)

12.0

(12.9)

0.99

Hawleyet

al.

81In-andoutpatients

17

21.8

(12.8)

15.9

(9.1)

0.92

Correlationbasedon107pairsofratingsbecause

26patients

assessedtw

iceatleast

2weeksapart

Korner

etal.

40Inpatients

17

aa

0.82

Maieret

al.

130Inpatients

17

22.9

(11.5)

22.0

(7.3)

[17-item]

0.85

21

25.8

(7.9)

[21-item]

0.83

Mittm

annet

al.

77Outpatients

–pre-treatm

ent

17

29.7

(5.4)

23.9

(4.9)

0.66

49Outpatients

–after

4weeksoftreatm

ent

12.6

(9.5)

10.8

(6.6)

0.93

Mulder

etal.

195Outpatients

after

6weeksoftreatm

ent

17

13.0

(10.4)

8.0

(6.4)

0.92

Muller

etal.(2000)

40Inpatients

17

32.6

(5.0)

24.6

(4.3)

0.70

Muller

etal.(2003)

85Inpatients

17

23.4

(13.2)

18.4

(10.2)

0.94

Senra

52Outpatients

–pre-treatm

ent

17

26.5

(4.4)

20.8

(2.1)

0.68

45Outpatients

–after

24weeksoftreatm

ent

9.8

(6.6)

7.1

(6.1)

0.88

a

578 M. Zimmerman et al. / Journal of Psychiatric Research 38 (2004) 577–582

treatment remission as a score below a predetermined

cutoff score on the scale. Through the years many cutoff

scores have been used on the HRSD to define remission

(Zimmerman et al., 1985); however, since the publica-

tion of the recommendations of Frank et al. (1991) aconsensus has emerged to define remission on the

HRSD as a score of 7 or less.

No such consensus has yet emerged in defining re-

mission on the MADRS. Montgomery (1994) suggested

that a cutoff of 11 on the MADRS was comparable to

the HRSD cutoff of 7; however, investigators have used

cutoffs ranging from 6 to 12 to define remission on the

MADRS, a twofold difference (Forlenza et al., 2001;Guelfi et al., 2001; Kyle et al., 1998; Levine et al., 1989;

Nierenberg et al., 1994; Schweitzer et al., 2001). The lack

of convention in defining remission on the MADRS

makes it difficult to combine and contrast findings with

studies relying on the HRSD to assess remission.

Ten studies have examined the association between

the MADRS and HRSD (Table 1). In all studies, the

two measures were significantly correlated. The onlystudy finding a correlation below 0.65 used separate

interviewers to rate the two scales on different occasions

the same day (Davidson et al., 1986). All three studies

that examined the correlation before and after antide-

pressant treatment found higher correlations after

treatment initiation (Dratcu et al., 1987; Mittmann

et al., 1997; Senra, 1996). This is not surprising because

the size of the correlation between any two measures ofdepression severity will be influenced by the range and

variability of scores of patients included in the analysis.

The data in Table 1 shows that in each of the three

studies with pre- and post-treatment assessments, the

standard deviations of both the MADRS and HRSD

scores were higher in the post-treatment evaluations. In

these three studies, the average correlation between

post-treatment scores was 0.93. Thus, in treatmentstudies of depression, the MADRS and HRSD are

highly correlated, particularly when administered at

follow-up visits.

Three of the 10 studies used regression analyses to

develop formulas to derive MADRS scores from HRSD

scores. Mittmann et al. (1997) collected 262 MADRS-

HRSD pairs of ratings in 77 depressed outpatients, and

derived the following formula (MADRS¼ 1.23�HRSD) 0.30). Based on this formula they suggested

that a MADRS cutoff of 6 8 was equivalent to the

HRSD remission definition of 6 7. However, as the

authors noted, patients contributed multiple data points

to the regression analysis, thus most of the rating

couplets were not statistically independent.

Hawley et al. (1998) collected 107 MADRS-HRSD

pairs of ratings in 81 depressed in- and outpatients, andderived the following conversion formula (MADRS¼1.30�HRSD+0.7). Although they did not suggest a

cutoff for remission on the MADRS, based on this

Inform

ationnotprovided

inarticle.

M. Zimmerman et al. / Journal of Psychiatric Research 38 (2004) 577–582 579

formula a MADRS cutoff of 610 would be equivalent

to a HRSD cutoff of 67.

Muller et al. (2000) rated 40 moderate-severely de-

pressed inpatients on the MADRS and HRSD, and

derived the following conversion formula (MADRS¼0.81�HRSD+12.6). The goal of this study was to de-

termine a MADRS cutoff to distinguish moderate from

severe depression. When their formula is used to derive a

cutoff for remission, then a score of 6 18 on the

MADRS would correspond to the HRSD remission

threshold.

These three studies computed different regression

equations extrapolating MADRS scores from HRSDvalues, resulting in different cutoffs on the MADRS

corresponding to the cutoff of 7 on the HRSD. The

formulas in the two studies that included outpatients

who were rated after treatment initiation were roughly

comparable, and both differed from the formula de-

rived in the study of moderate-severely depressed in-

patients. This suggests that patient severity might

influence the derivation of the regression equation, andraises the question of how to select patients in studies

trying to extrapolate scores on one measure from an-

other. The inclusion of patients who currently meet

criteria for major depression might not be appropriate

for a study attempting to derive a cutoff for remission

on the MADRS that is equivalent to the HRSD cutoff

of 67.

In the present report from the Rhode Island Meth-ods to Improve Diagnostic Assessment and Services

(MIDAS) project, we determined the cutoff on the

MADRS that most closely corresponds to the HRSD

definition of remission in two ways. First, similar to

other research groups, we conducted a regression

analysis to derive an equation relating MADRS and

HRSD scores. We first conducted this analysis with the

entire sample, which includes patients who met fullcriteria for major depression as well as patients who

were in partial and full remission. Second, we repeated

the analysis after excluding patients who met criteria for

MDD. In this way, we were able to examine whether

inclusion of more severely depressed patients impacted

upon the regression equation used to derive a MADRS

equivalent for the HRSD definition of remission. The

second method of determining the MADRS equivalentto the HRSD cutoff of 6 7 used receiver operating

curve analysis.

2. Methods

Participants were 303 psychiatric outpatients who

were being treated for a DSM-IV major depressive ep-isode in the Rhode Island Hospital Department of

Psychiatry outpatient practice. This private practice

group predominantly treats individuals with medical

insurance on a fee-for-service basis, and it is distinct

from the hospital’s outpatient residency training clinic

that predominantly serves lower income, uninsured, and

medical assistance patients. The sample included 114

(37.6%) men and 189 (62.4%) women who ranged in agefrom 18 to 79 years (M ¼ 42:9, SD¼ 12.7). Almost half

of the subjects were married (47.9%, n ¼ 145); the re-

mainder were single (23.4%, n ¼ 71), divorced (19.8%,

n ¼ 60), separated (5.6%, n ¼ 17), widowed (2.0%,

n ¼ 6), or living with someone as if in a marital rela-

tionship (1.3%, n ¼ 4). The racial composition of the

sample was 86.8% (n ¼ 263) white, 2.6% ðn ¼ 8Þ black,4.3% ðn ¼ 13Þ Hispanic, 0.7% ðn ¼ 2Þ Asian, and 5.6%ðn ¼ 17Þ other. The Rhode Island Hospital institutional

review committee approved the research protocol, and

all patients provided informed, written consent.

Diagnoses were based on the Structured Clinical In-

terview for DSM-IV (First et al., 1995). The patients

were rated by the first two authors on the MADRS and

the 17-item HRSD. Inter-rater reliability on the

MADRS and HRSD was obtained in 16 patients, withone of the authors interviewing the patient while the

other observed and made independent ratings. For the

MADRS and HRSD the intraclass correlation

coefficients of reliability were 0.96 and 0.97, respectively.

At the time of the HRSD and MADRS assessments,

the patients were also rated on a 6-point rating scale

reflecting remission status according to DSM-IV. This

rating was based on the number of DSM-IV criteria fora major depressive episode and the level of psychosocial

impairment present during the past week. Slightly more

than one-third of the sample met MDD criteria at the

time of the evaluation ðn ¼ 114Þ, slightly less than one-

third were in partial remission ðn ¼ 87Þ, and one-third

were in remission ðn ¼ 102Þ.A linear regression equation was computed to esti-

mate MADRS scores from HRSD scores. This was doneseparately for the entire sample, and for patients who

were in partial or full remission. After deriving the re-

gression equation, we computed the MADRS score

corresponding to an HRSD score of 7.

We also examined the ability of the MADRS to

identify patients who were in remission according to the

HRSD across the range of MADRS cutoff scores by

conducting receiver operating curve (ROC) analyses(Hsiao et al., 1989). An ROC curve is a plot of a

measure’s sensitivity versus one minus specificity at each

cut-off score. The area under the curve (AUC) is the

evaluative measure, which can range from 0.5 (random

performance) to 1.0 (perfect performance).

3. Results

The mean score on the MADRS for the entire sample

was 17.2 (SD¼ 12.7). The mean score for the HRSD

580 M. Zimmerman et al. / Journal of Psychiatric Research 38 (2004) 577–582

was 11.4 (SD¼ 8.4). The scattergram plot indicates

there was a linear relationship between the two measures

(Fig. 1). The Pearson correlation between the MADRS

and HRSD was 0.94 for the entire sample.

The linear regression analysis for the entire samplewith the MADRS as the dependent variable resulted in

the following equation: MADRS¼ 1.43�HRSD+

0.87. Based on this equation a MADRS score of 11

would correspond to a score of 7 on the HRSD. We

HRSD score403020100

MA

DR

S sc

ore

60

50

40

30

20

10

0

Fig. 1. Scatterplot of individual subjects’ pairwise co-ordinates be-

tween the Hamilton rating scale for depression (HRSD) and the

Montgomery–Asberg depression rating scale (MADRS).

1.00.75.50.250.00

1.00

.75

.50

.25

0.00

1 – Specificity (False Positive Rate)

Sensitivity (True Positive Rate)

Fig. 2. Receiver operating characteristic (ROC) curve of the Mont-

gomery–Asberg depression scale in depressed psychiatric outpatients

using the cutoff of 7 on the Hamilton depression scale to define

remission.

repeated the analysis after excluding the more severely

depressed patients who currently met criteria for MDD

and derived the following equation: MADRS¼ 1.36�HRSD+0.33. Based on this equation a MADRS score

of 10 would correspond to a score of 7 on the HRSD.In the Receiver Operating Curve analysis, the area

under the curve was significant (AUC¼ 0.98, p < 0:001)(Fig. 2). The sensitivity, specificity and overall classifi-

cation rate of the MADRS for identifying remission

according to the HRSD threshold of 67 is presented in

Table 2. A cutoff of 610 on the MADRS maximized the

level of agreement with the HRSD definition of

remission.

4. Discussion

The MADRS and HRSD are the two most frequently

used clinician-rated depression symptom severity scales.

Despite some differences in content and scaling of items,

the two scales are highly correlated. As seen in thesummary of studies in Table 1, MADRS scores tend to

be higher than HRSD scores; thus, the cutoffs to define

remission on the two scales would not be expected to be

equivalent.

In the absence of studies calibrating the MADRS

relative to the HRSD, investigators have had to estimate

the MADRS cutoff that would correspond to the HRSD

threshold. Two studies reporting remission rates usingboth the MADRS and HRSD found higher rates of

remission on the MADRS when a cutoff of 12 was used.

Guelfi et al. (2001) reported the results of an 8-week

study comparing venlafaxine and mirtazepine in severely

depressed inpatients. Remission was defined as a score

67 on the HRSD and 612 on the MADRS. From

figure 3 in their paper it appears that the remission rate

is about 10–15% lower in the two treatment groups ac-cording to the HRSD results. In contrast, the response

rate, defined as a 50% improvement in scores, was

similar according to both scales. Nierenberg et al. (1994)

reported response and remission rates according the

MADRS and 21-item HRSD in 70 patients treated with

venlafaxine. Remission was defined as a score 68 on the

HRSD and 612 on the MADRS. Response was defined

as a 50% decrease in scores along with final scores abovethe threshold used to define remission. At each time

point the remission rate was higher on the MADRS and

the response rate was higher on the HRSD. The higher

remission rates on the MADRS in these two studies are

not surprising in the context of the present study which

suggested that a cutoff of 10 on the MADRS most

closely approximates the HRSD cutoff of 7 to define

remission.Our results suggest that the regression equation

relating HRSD and MADRS scores is dependent, in

part, on the range and severity of scores in the sample.

Table 2

Sensitivity and specificity of the Montgomery–Asberg depression rating scale (MADRS) for identifying remission according to the Hamilton rating

scale for depression (HRSD) threshold of 6 7 in 303 depressed outpatients a

MADRS cutoff Cumulative N Sensitivity % Specificity % Overall % correct rate Kappa

0 23 19.5 100.0 68.6 0.23

61 34 28.8 100.0 72.3 0.33

62 52 44.1 100.0 78.2 0.49

63 60 50.8 100.0 80.9 0.56

64 68 57.6 100.0 83.5 0.62

65 78 66.1 100.0 86.8 0.70

66 88 73.7 99.5 89.4 0.77

67 96 79.7 98.9 91.4 0.81

68 99 81.4 98.4 91.7 0.82

69 102 83.9 98.4 92.7 0.84

610 106 87.3 98.4 94.0 0.87

611 116 89.8 94.6 92.7 0.85

612 121 89.8 91.9 91.1 0.81

613 126 90.7 89.7 90.1 0.79

614 135 95.8 88.1 91.1 0.82

615 146 98.3 83.8 89.4 0.79

616 148 98.3 82.7 88.8 0.77

617 155 98.3 78.9 86.5 0.73

618 164 100.0 75.1 84.8 0.70

619 176 100.0 68.5 80.8 0.63

620 178 100.0 67.6 80.2 0.62aResults are shown up to a MADRS cutoff score of 20. Beyond a score of 20, the sensitivity remains 100% and the specificity and overall level of

agreement continue to decline.

M. Zimmerman et al. / Journal of Psychiatric Research 38 (2004) 577–582 581

The mean HRSD and MADRS scores in the present

study are closest to those of Hawley et al. (1998), and

the regression equations in the two studies were sim-

ilar. To approximate the analysis by Muller and col-

leagues, we repeated the regression analysis in patients

who met MDD criteria. The equation computed for

these patients (MADRS¼ 1.04�HRSD+10.13), and

the MADRS score corresponding to an HRSD cutoffof 7 (i.e., 617), replicated Muller et al.’s findings. The

variability in results as a function of who is included

in the analyses suggests that the ROC approach,

which directly examines which MADRS score most

accurately corresponds to the HRSD definition of re-

mission, is a more valid method of addressing this

issue.

While the HRSD cutoff of 67 has been widelyadopted, there is little data supporting this threshold.

Rather it has been used for many years (DUAG, 1986;

Reisby et al., 1977), and was the recommended thresh-

old of a consensus conference to operationalize the

concepts of remission, relapse, and recovery-terms that

had acquired diverse definitions across studies. Addi-

tional recommendations from this panel were to em-

pirically validate the suggested operational definitions.Because the HRSD cutoff has gained widespread ac-

ceptance, it is important to determine the comparable

score on the MADRS, the second most frequently used

depression symptom severity outcome measure. The

findings of the present study do not, however, address

the question of which cutoff on the MADRS is the most

valid threshold for defining remission.

References

Davidson J, Turnbull CD, Strickland R, Miller R, Graves K. The

Montgomery–Asberg depression scale: reliability and validity. Acta

Psychiatrica Scandinavica 1986;73:544–8.

Dratcu L, da Costa Ribeiro L, Calil HM. Depression assessment in

Brazil. The first application of the Montgomery–Asberg depression

rating scale. British Journal of Psychiatry 1987;150:797–800.

Danish University Antidepressant Group (DUAG). Citalopram:

clinical effect profile in comparison with clomipramine. A

controlled multicenter study. Psychopharmacology 1986;90:131–

138.

First MB, Spitzer RL, Gibbon M, Williams JBW. Structured clinical

interview for DSM-IV axis I disorders – Patient edition (SCID-I/P,

version 2.0). New York: Biometrics Research Department, New

York State Psychiatric Institute; 1995.

Forlenza OV, Almeida OP, Stoppe Jr A, Hirata ES, Ferreira RCR.

Antidepressant efficacy and safety of low-dose sertraline and

standard-dose imipramine for the treatment of depression in older

adults: results from a double-blind, randomized, controlled clinical

trial. International Psychogeriatrics 2001;13:75–84.

Frank E, Prien RF, Jarrett RB, Keller MB, Kupfer DJ, Lavori PW,

et al. Conceptualization and rationale for consensus definitions of

terms in major depressive disorder. Archives of General Psychiatry

1991;48:851–5.

Guelfi JD, Ansseau M, Timmerman L, Korsgaard S, Group M-VS.

Mirtazapine versus venlafaxine in hospitalized severely depressed

patients with melancholic features. Journal of Clinical Psycho-

pharmacology 2001;21:425–431.

Hamilton M. A rating scale for depression. Journal of Neurology

Neurosurgery and Psychiatry 1960;23:56–62.

Hawley CJ, Gale TM, Smith V, Sen P. Depression rating scales can be

related to each other by simple equations. International Journal of

Psychiatry in Clinical Practice 1998;2:215–9.

Hsiao JK, Bartko JJ, Potter WZ. Diagnosing diagnoses: receiver

operating characteristic methods and psychiatry. Archives of

General Psychiatry 1989;46:664–7.

582 M. Zimmerman et al. / Journal of Psychiatric Research 38 (2004) 577–582

Khan A, Khan SR, Shankles EB, Polissar NL. Relative sensitivity of

the Montgomery–Asberg depression rating scale, the Hamilton

depression rating scale and the clinical global impressions rating

scale in antidepressant clinical trials. International Clinical Psy-

chopharmacology 2002;17:281–5.

Kyle CJ, Petersen HEH, Overo KF. Comparison of the tolerability

and efficacy of citalopram and amitriptyline in elderly depressed

patients treated in general practice. Depression and Anxiety

1998;8:147–53.

Levine S, Deo R, Mahadevan K. A comparative trial of a new

antidepressant, fluoxetine. International Clinical Psychopharma-

cology 1989;4:41–5.

Mittmann N, Mitter S, Borden EK, Herrmann N, Naranjo CA, Shear

NH. Montgomery–Asberg severity gradations. American Journal

of Psychiatry 1997;154:1320–1.

Montgomery S. Clinically relevant effect sizes in depression. European

Neuropsychopharmacology 1994;4:283–4.

Montgomery SA, Asberg M. A new depression scale designed to be

sensitive to change. British Journal of Psychiatry 1979;134:382–9.

Muller MJ, Himmerich H, Kienzle B, Szegedi A. Differentiating

moderate and severe depression using the Montgomery–Asberg

depression rating scale (MADRS). Journal of Affective Disorders

2003;77:255–60.

Muller MJ, Szegedi A, Wetzel H, Benkert O. Moderate and severe

depression gradations for the Montgomery–Asberg depression

rating scale. Journal of Affective Disorders 2000;60:137–

40.

Nierenberg AA, Feighner JP, Rudolph R, Cole JO, Sullivan J.

Venlafaxine for treatment-resistant unipolar depression. Journal of

Clinical Psychopharmacology 1994;14:419–23.

Prien RF, Carpenter LL, Kupfer DJ. The definition and operational

criteria for treatment outcome of major depressive disorder.

Archives of General Psychiatry 1991;48:796–800.

Reisby N, Gram L, Bech P, Nagy A, Petersen G, Ortmann J, et al.

Imipramine: clinical effects and pharmacokinetic variability. Psy-

chopharmacology 1977;54:263–72.

Schweitzer I, Burrows G, Tuckwell V, Polonowitz A, Flynn P, George

T, et al. Sustained response to open-label venlafaxine in drug-

resistant major depression. Journal of Clinical Psychopharmacol-

ogy 2001;21:185–9.

Senra C. Evaluation and monitoring of symptom severity and change

in depressed outpatients. Journal of Clinical Psychology

1996;52:317–24.

Zimmerman M, Coryell W, Pfohl B. The treatment validity of

DSM-III melancholic subtyping. Psychiatry Research 1985;

16:37–43.