International Clinical Psychopharmacology 2002, 17:281-285

Relative sensitivity of the Montgomery-Asberg

Depression Rating Scale, the HamiltonDepression rating scale and the Clinical GlobalImpressions rating scale in antidepressantclinical trials

A. Khana.b, S.R. Khana, E.B. Shanklesa and N.L. Polissarc

aNorthwest Clinical Research Centet; Bellevue, WA, bDepartment of Psychiatry andBehavioral Sciences, Duke University Medical Centet; Durham, NC, andcThe Mountain-Whisper-Light Statistical Consulting, Seattle, WA, USA

Correspondence to Arif Khan, 1900 116th Avenue NE #112, Bellevue, WA 98004, USATel: +14254530404; fax: +14254531033; e-mail: akhan@nwcrc.net

Received: 11 January 2002; accepted 5 September 2002

Although early antidepressant clinical trials simply relied on a clinician's judgment as to whether a depressed patient clinicallyimproved or not, the Hamilton Depression (HAM-D) rating scale has become the 'gold standard' to assess the efficacy of newantidepres.4;ants. The alternative Montgomery-Asberg Depres.4;ion Rating Scale (MADRS) has not achieved general acceptance.However, its ease of use \varrants evaluation as to whether it is comparable to HA.~-D in its sensitivity in detecting antidepressant-placebo differences in antidepressant clinical trials. A retrospective chart review was performed on the records of 208 depres.4;ed adultpatients that participated in eight randomized, placebo-controlled, double-blind antidepressant clinical trials at the NorthwestClinical Research Center between 1996 and 2000. We compared the effect sizes of the HAM-D, MADRS and Clinical ImpressionsRating Scale (CGI-S for severity and CGI-I for improvement) for patients assigned to placebo or an established antidepressant. Theeffect size (measured as the mean change in rating with antidepressants minus the mean change for placebo divided by the pooled SDof change, adjusted for age, gender and initial scores) was 0.49 with MADRS, 0.53 with HAM-D, 0.55 with CGI-S and 0.59 withCGI-I. The four rating scales had similar effect sizes regardless of the type of antidepressant evaluated. These data suggest thatMADRS is as sensitive an instrument as HAM-D for detecting antidepressant efficacy in clinical trials. Thus, MADRS may be adesirable tool in large-scale. pivotal antidepressant clinical trials. In! Clin Psychopharmacol17281-285 :!;) 2002 Lippinco!! Williams & Wilkins

Keywords: antidepressants, Clinical Global Impressions rating scale, effeclMontgomery-Asberg Depression Rating Scale, placebo-controlled trials

size, Hamilton Depression rating scale,

INTRODUCTION Although extensively used, the HAM-D has comeunder criticism for multiple reasons. Approximatelyone-half of randomized, double-blind, placebo-con-trolled antidepressant trials fail to show statisticalsuperiority of generally used antidepressants comparedto placebo (Khan et al., 2002). This failure rate ispartly blamed on the deficiencies of HAM-D as acontributing factor. There are several assumptionsbehind this consideration. A commonly held belief isthat HAM-D is neither accurate nor well balanced inassessing various symptoms and signs of depression.For example, insomnia tends to influence total scoredisproportionately compared to many other depressive

The earliest of randomized, double-blind, placebo-controlled antidepressant clinical trials (Bali et al.,1959) simply assessed the proportion of depressedpatients thought to have 'clinically improved', amethod of assessment currently used in the ClinicalGlobal Severity (Guy, 1976) (CGI-S) or Improvement(CGI-I) scales. However, within the next decade, thenewly developed Hamilton Depression (HAM-D)rating scale (Hamilton, 1960) came into use amongantidepressant clinical trials assessing the effectivenessof antidepressants.

2810268-13 2002 Lippincott Williams & Wilkins intern. mal Clinical Psychopharmacology 2002, Vol No

Copyright @ Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

KHAN ET AL.

MATERIALS AND METHODSsymptoms. Also, it is unclear if the primary function ofHAM-D is to assess severity of depression or to assessa change in symptom profile and severity. Addition-ally, HAM-D is believed to be better at detecting

tricyclic antidepressant (TCA) versus placebo differ-ences rather than newer antidepressants, such asselective serotonin reuptake inhibitors (SSRls) orselective serotonin and norepinephrine reuptake in-hibitors (SNRls) versus placebo differences. This is inpart due to the hypnotic property of the TCAs, whichmay be better detected by HAM-D.

In 1979, Montgomery and Asberg challenged thisscale and devised a new rating scale, the Montgomery-

Asberg Depression Rating Scale (MADRS) (Kearnset at., 1982), which they considered more sensitive tochange than the HAM-D. The MADRS has gainedpopularity throughout Europe but is not currently astandard for assessing efficacy of antidepressants in theUSA. However, this pattern is changing as the numberof randomized, double-blind, placebo-controlled anti-depressant clinical trials using MADRS as the primaryefficacy variable is growing. Furthermore, it is con-sidered that MADRS may have a better sensitivityindex when SSRls or SNRIs are test antidepressantscompared to HAM-D.

In addition to some interest in the theoreticalconstruct of these depression-rating scales regardingtheir relative merit (Bech et al., 1981; Hearns et al.,1982; Maier and Philipp, 1985; Davidson et at., 1986;Maier et at., 1988), the relative sensitivity of HAM-D,MADRS or CGI-S or Cal-l in antidepressant clinicaltrials has not undergone empirical testing. Wehypothesized that MADRS may have a highersensitivity, as measured by magnitude of effect sizebetween antidepressants and placebo, comparedto HAM-D or Clinical Global Impressionscales (CGI-S or Cal-l). Furthermore, we hypothe-sized that MADRS may have a higher sensitiyity forSSRIs or SNRIs differences from placebo compared toHAM-D.

On the other hand, we hypothesized that HAM-Dmay have a higher sensitivity as measured bymagnitude of effect size between TCAs and placebocompared to SSRls or SNRls and placebo. In order toverify our hypothesis, we compared the effect sizesbetween antidepressant treated (n = 119) and placebo-treated (n = 89) depressed patients using HAM-D,

MADRS or CGI (S and I) who participated inrandomized, double-blind, placebo-controlled antide-pressant clinical trials at the Northwest ClinicalResearch Center. Additionally, we conducted similaranalysis on depressed patients assigned to TCAs orSSRIs and SNRIs, as well as individual antidepressants(n = 7).

The present study consisted of a retrospective chartreview of depressed outpatients treated with placebo oran antidepressant between June 1996 and December2000 at the Northwest Clinical Research Center

located in the greater Seattle area. During this studyperiod, 27 randomized, double-blind, placebo-con-trolled antidepressant clinical trials were conducted inadult outpatients meeting DSM-IV criteria for depres-sion of moderate to severe degree. After the completionof 11 of the 27 trials, the sponsoring companyunblinded the treatment codes. We report data onpatients that received placebo or an active comparatorin the eight out of 11 trials which included both an

antidepressant and a placebo arm. A total of 208patients received either placebo or an active compara-tor antidepressant in these eight trials. The active

comparator antidepressants used were fluoxetine20 mg/day, paroxetine 10--40 mg/day, venlafaxine, ex-tended release, 75--225 mg/day, sertraline 50 mg/day,

citalopram 4Omg/day, imipramine 150 mg/day anddesipramine 200 mg/day.

Each of the studies used the following standardsubject inclusion and exclusion criteria: (i) 18 years ofage or older; (ii) diagnosis of major depressive disorderof moderate to severe degree, according to DSM-IV(n = 7) or Structured Clinical Interview (n = 1); (iii)

signed informed consent form (witnessed); (iv) mini-mum score at screen evaluation of 20 on HAM-D-17(n= 3), 22 on HAM-DI7 (n=3), 22 on HAM-D-21(n= I), or 22 on MADRS (n=I); (v) 7::!:3 days ofplacebo run-in; (vi) less than a 20% decrease in scorebetween screen and baseline on HAM-D (n = 7) or less

than a 25% decrease (n= 1); (vii) not actively suicidalor posing a serious suicidal risk; and (viii) nosignificant concurrent or previous other psychiatricillnesses.

Each study had the following design features:(i) primary efficacy measures included HAM-D(n = 7) or MADRS (n = I); (ii) mean total raw scores

and change in scores on the HAM-D, MADRS andCGI were available for calculations and comparisons;

(iii) depressed patients underwent evaluations atweekly intervals during the first month and at leastbiweekly after 4 weeks; and (iv) the dose of theantidepressants was adequate based on standard

practice.For the 208 patients, we recorded baseline demo-

graphics including age, gender, duration of placeboor antidepressant treatment, and whether the patientwas assigned to receive placebo or an antidepressant(Table 1). An independent samples t-test or Fisher'sexact test was carried out to compare the baseline

282 International Clinical Psychopharmacology 2002, Vol No6

Copyright @ Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited

RELATIVE SENSITMTY OF RATING SCALES

Table 1. Demographic and clinical features in 208 depressed outpatients participating in eight randomized placebo-con-trolled, double-blind trials

80.4 (45/56)69.7 (23/33)

NANA

81.8 (63/77}71.4 (30/42}

8633

NANANANA

Number of subjectsMean age :tSO in years (years)Female (0/0)Mean :t SO (range) weeks of participationCompleters (0/0) in:

6-week trials (n = 5)8-week trials (n = 3)

Number of subjects on SSRI or SNRINumber of subjects onTCA

NA, indicates not applicable.

study effect was not statistically significant for any ofthe rating scales (P=O.5--0.8) and it was dropped fromall subsequent analyses.

The effect size is a common method for comparingtreatment effect as measured by different scales(Hedges and Olkin, 1985). For each rating scale, theeffect size for antidepressant versus placebo wascalculated in two steps. First, each patient's changefrom baseline to the last observation was divided by apooled standard deviation to create normalized scores.The pooled SD was the square root of the weightedsum of the variance of change scores for theantidepressant patients and the variance for theplacebo patient, with each variance weighted byappropriate degrees of freedom. Second, the meaneffect size was calculated from analysis of covariancewith the normalized scores as the dependent variable,treatment as a factor and age, gender and the initialvalue of the HAM-D and MADRS scores as covari-ates. In order to achieve a common set of modelsacross the scales, the initial HAM-D and MADRSscores were used as covariates for all rating scales.Confidence intervals for the difference in the HAM-Dand MADRS effect sizes were calculated using

characteristics between the placebo-treated and anti-depressant treated patients.

During the trial period, efficacy was evaluated withthe use of HAM-D-17, MADRS and CGI forImprovement and Severity. If HAM-D-2l or -24 itemscales were used, only the first 17 items were includedin analysis. Efficacy analysis was performed on data for

intent-to-treat patients having at least one dose ofmedication and at least one efficacy rating during thedouble-blind phase. The mean total scores at baseline,as well as the change in score from baseline to lastobservation, were compared between placebo- and

antidepressant-treated patients using analysis of var-iance and covariance (Table 2). A negative change inscore indicates improvement. Patients prematurelyterminating from the trial are assumed to experienceno further improvement and the last measured scores

were used to calculate the change from baseline.The possibility of a study effect was investigated

using analysis of covariance with the change frombaseline to follow-up in a given rating scale as thedependent variable, treatment (antidepressant versusplacebo) and study (eight levels) as factors, and age,gender and initial value orthe rating ascovariates. The

Table 2. Baseline and change in Hamilton Depression (HAM-D) rating scale, Montgomery and Asberg Depression RatingScale (MADRS) and Clinical Global Impressions (CGI) scores

25.3 :t3.0, n = 89

-7.72:t 7.8,n =8832.5:t3.7, n = 72

-9.6 :t 10.5, n = 714.1 :to.3, n = 89

-1.13 :to.98, n = 693.2:t 1.1, n = 88

Mean baseline total HAM-O 17 score :t SOMean change in total HAM-O 17 score at LOCF :t SOMean baseline total MAORS score :t SOMean change in total MAORS score at LOCF :t SOMean baseline total CGI score :t SOMean change in total CGI-S score at LOCF :t SOMean final LOCFCGI-I score :tSO

Change scores for each patient wer~ calculated as the difference between the last observation carried forward (LOCF) andbaseline, such that the negative changes indicate improvement. na indicates not applicable. Effect sizes are adjusted for age,gender and initial value of HAM-D and MADRS.

International Clinical Psychopharmacology 2002, Vol 17 No 6 283

Copyright @ Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

KHAN ET AL

standard errors from repeated measures analysis ofcovariance, with the HAM-D and MADRS normal-ized scores as the within-patient repeated factor andage, gender and initial HAM-D and MADRS scores ascovariates. Effect sizes for patients assigned to SSRlsand SNRls were calculated separately from thoseassigned to TCAs, and then an effect size for allpatients combined was calculated.

difference in effect size between HAM-D and MADRSare rather wide.

DISCUSSION

RESULTS

As shown in Table 1, between the 89 patients assignedto placebo and the 119 to antidepressants, nosignificant differences were noted at baseline forgender, weeks of participation or completion rate.However, patients assigned to antidepressants weresignificantly older (placebo 39.8 years, antidepressants44.4 years; P = 0.04).

Major efficacy variables are presented in Table 2.

Overall, the antidepressant-treated patients showed astatistically significant greater degree of improvementon aNour rating scales than placebo-treated patients

(P<O.OOI).Table 3 shows the effect sizes for the four rating

scales. The effect sizes were very similar for all fourscales in the trials involving SSRI/SNRI antidepres-sants. Effect sizes are also quite similar in the TCAtrials, though the effect size for MADRS in somewhatsmaller than for the other three scales, The effect sizesfor all trials combined are quite similar across the fourrating scales. However, confidence intervals for the

The aim of our study was to assess the relativesensitivity of HAM-D compared to MADRS inrelation to CGI in antidepressant clinical trials. Ourresults empirically support our hypothesis thatMADRS is just as sensitive as HAM-D in differentiat-ing between antidepressants and placebo. The CGI-Iand CGI-S scales are also comparable to each otherand to the HAM-D and MADRS scales.

Earlier comparisons between the MAD RS and theHAM-D also support the notion that these scales aresimilar. Kearns et al. (1982) noted that MADRS andHAM-D performed well in assessing severity ofdepression, whereas other scales such as the BeckDepression Inventory or the Wakefield Inventory werenot as good. Furthermore, Maier et al. (1988) notedthat MADRS and HAM-D had comparable reliability,validity and ability to detect change. However, thecurrent report is first to note that MADRS iscomparable, if not superior to HAM-D, in assessing

antidepressant-placebo differences in antidepressantclinical trials.

These data lend themselves to further assess thestrengths and weaknesses of HAM-D and MADRS inthe clinical setting while conducting antidepressantclinical triats. Overall, MADRS is easier to useclinically because its language is simple and straight-

Table 3. Treatment effect sizes (mean standardized change) by antidepressant type

Effect sizes

HAM-D17 MADRS CGI-I-

n Effect size SE nDrug type

CGI-S

Effect size SEn

-0.40-1.12-1.13-0.28-1.29-0.60

0.240.400.500.450.540.19

7827221614

114

-0.44NA

-1.07NA

-1.31-0.58

0.23NA0.52NA0.570.19

-0.031.450.740.33-1.750.53

0.330.370.600.460.450.24

41

27

22

16

14

77

-0.40-1.38-0.53-0.28-1.78-0.59

0.220.350.680.460.370.19

7827221614

114

.1.11

.0.26

.0.42

.0.53

0.44 230.45 300.30 530.17 161

.1.03

0.09

0.31

0.49

0.45 230.35 300.26 530.16 161

-0.85

-0.12

0.44

-0.55

0.51 230.36 300.27 530.19 124

.0.78

.0.29

.0.46

.0.59

0.59 230.41 300.31 530.17 161

Section A: SSRI/SNRIFluoxetineParoxetine*Venlafaxine ERSertraline*

CitalopramPooled SSRI/SNRlaSection B: TCA

ImipramineDesipraminePooled TCA b

Pooled SSRI/SNRI,TCAc

Effect sizes are adjusted for age, gender, initial value of MADRS, and initial value of HAM-D. *Effect sizes for paroxetine andsertraline studies are adjusted for age, gender, and initial value of HAM-D. aOifference of MADRS and HAM-D effect sizes is-0.02, 95% confidence interval (CI) (-0.13 to 0.10). bOifference of MADRS and HAM-D effect sizes is -0.11, 95% CI (-0.36 to0.15). cDifference of MADRS and HAM-D effect sizes is -0.04, 95% CI ( -0.16 to 0.08).

284 ntemational Clinical Psychopharmacology 2002, Vol U No 6

Copyright @ Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited

78NA22NA14114

RELATIVE SENSITIVITY OF RATING SCALES

could not be included either because there were noactive control antidepressants in the design of the trialor because we were not unblinded to patient assign-ment. Additionally, seven of the eight clinical trialsused HAM-D as the primary outcome measure and asthe tool to assess pre-randomization severity of illness.Thus, it is expected that pre-treatment severity asmeasured by HAM-D matched between antidepressanttreated and placebo-treated patients. However, the

antidepressant treated group had a higher pre-treat-ment score on MADRS compared to those givenplacebo. This may in part limit generalizations of our

findings.In conclusion, our results suggest that MADRS is as

sensitive an instrument as HAM-D to detect anti-

depressant efficacy. Thus, MADRS may be a desirabletool in large-scale, pivotal antidepressant clinical trials.

forward. A less experienced rater (e.g. a clinician whohas worked exclusively in the outpatient setting) mayreliably estimate accurate score with MADRS becausethere are understandable anchor points for each item.As each of the 10 items on the MADRS is equallyweighted, the scale is easy to follow and use.Additionally, the presence or absence of insomnia orhypersomnia does not markedly influence overall

rating.On the other hand, HAM-D has some distinct

advantages. Experienced raters can accurately rate abroad range of patients; those suffering from psychoticdepression to mild dysthymia. Hypersomnia, hyper-phagia, worthlessness, hopelessness and diurnal moodswings as distinctive phenomena can be captured moreaccurately, albeit, the rater needs expertise to fullyassess such factors. In other words, the HAM-D ismultidimensional. For example, identical trial scoresfrom two different patients may reflect a different

combination of individual item scores (i.e. have adifferent clinical meaning).

Thus, for a straightforward pivotal efficacy anti-depressant clinical trial, MADRS may be an optimalrating instrument. However, HAM-D may be anoptimal rating instrument for clinical trials thataddress specific hypothesis or plan to assess variousaspects of depressive spectrum disorder. For example,some of the early trials evaluating new mechanisms ofantidepressant action warranted the use of a longerversion of HAM-D as a secondary efficacy variable.This may provide a profile of specific symptom relief ofnew antidepressants.

Our data do not support our secondary hypothesisthat HAM-D may be more sensitive in differentiating

tricyclic antidepressants from placebo whereasMADRS may be more sensitive differentiating SSRIISNRI from placebo. Although not conclusive, thesimilarity of pattern for drug-placebo differences(Tables 2 and 3) among the four scales in this sampleraise doubts as to whether the blame for failedantidepressant trials can be placed entirely on the useof HAM-D as the primary variable. Replacing HAM-D with MADRS may not solve the riddle of failedantidepressant trials.

The strength of our database is that it came from asingle centre with only one physician psychiatrist(A.K.) evaluating all the patients for CGI scores.Furthermore, the same physician psychiatrist (A.K.)trained and supervised all qualified raters for HAM-Dand MADRS. The other strength is the similarity ofthese eight clinical trials, allowing for a combinedreview. On the other hand, the data for the additional19 clinical trials that were conducted at this centre

REFERENCES

Ball JRB, Kiloh LG (1959) A controlled trial ofimipramine in treatment of depressive state. Br Med J2:1052-1055.

Bech P, AIlerup P, Gram LF, Reisby N, Rosenberg R,Jacobsen 0, Nagy A (1981) The Hamilton depressionscale. Evaluation of objectivity using logistic models.Acta Psychiatr Scand 63:290-299.

Davidson J, Turnbull CD, Strickland R, Miller R,Graves K (1986) The Montgomery-Asberg DepressionRating Scale: reliability and validity. Acta Ps}.chiatrScand 73:544-548.

Guy W. ECDEUAssessrnent Manualfor Psychopharmacol-ogy, rev. edn. US Department of Health, Education, andWetfare (ADM) 76-338. Rockville, MD. NationalInstitute of Mental Health, pp. 218-222.

Hamilton M (1960) A rating scale for depression. J NeurolNeuro.\"urg Psychiatry 23:56-62.

Hedges L V, 01kin I. Statistical Methods for Meta-Analysis. New York, NY: Academic Press, 1985.

Kearns NP, Cruickshank CA, McGuigan KJ, Riley SA,Shaw SP, Snaith RP (1982) A comparison of depressionrating scales. Br J P.I')'chiatry 141:45-49.

Khan A, Leventhal RM, Khan SR, Brown W A (2002)Severity of depression and response to antidepressantsand placebo: an analysis of the FDA database. J ClinPsychopharrnacoI2:40--45.

Maier W, Philipp M (1985) Comparative analysis ofobserver depression scales. Acta Psychiatr Scand 72:239-245.

Maier W, Philipp M, Heuser I, Schlegel S, Buller R,Wetzel H (1988) Improving depression severity assess-ment- I. Reliability, internal validity and sensitivity tochange of three observer depression scales. J PsychiatrRes 22:3-12.

Montgomery SA, Asberg M (1979) A new depression scaledesigned to be sensitive to change. Br J Psychiatry134:382-389.

International Clinical Psychopharnlacology 2002, Vol17 No 6 285

Copyright @ Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.