The Bech-Rafaelsen Melancholia Scale (MES) in clinical trials of therapies in depressive disorders: a 20-year review of its use as outcome measure

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. © Blackwell Munksgaard 2002. Objective: To evaluate the psychometric properties of the Bech–Rafaelsen Melancholia Scale (MES) by reviewing clinical trials in which it has been used as outcome measure. Method: The psychometric analysis included internal validity (total scores being a sufficient statistic), interobserver reliability, and external validity (responsiveness in short‐term trials and relapse prevention in long‐term trials). Results: The results showed that the MES is a unidimensional scale, indicating that the total score is a sufficient statistic. The interobserver reliability of the MES has been found adequate both in unipolar and bipolar depression. External validity including both relapse, response and recurrence indicated that the MES has a high responsiveness and sensitivity. Conclusion: The MES has been found a valid and reliable scale for the measurement of changes in depressive states during short‐term as well as long‐term treatment.

Keywords: rating scales; melancholia scale; validity; reliability; acute therapy; long‐term therapy

Although the Bech–Rafaelsen Melancholia Scale (MES) ([1]) was published in the same year as the DSM‐III ([2]) the scale was constructed independently of the DSM‐III universe of symptoms defining the diagnosis of major depression.

Table 1 shows the DSM‐IV ([3]) symptoms included in the algorithm of major depression as well as the MES symptoms with their reference both to the Cronholm–Ottossons' Depression Scale ([4]) and to the Hamilton Depression Scale (HAM‐D) ([5]). For a comparison, the Montgomery–Åsberg Depression Scale (MADRS) ([6]) is also shown in Table 1.

The MES is based on the core items of depression as identified by Bech et al. ([7]) when making the first validation study of the HAM‐D, using a global clinical assessment of the severity of depression as the index of validity. The global assessment was made by experienced psychiatrists on a straight vertical line from 0=no depression to 10=maximum depression. The line was provided with numbers (1=doubtful depression, 2–4= mild depression, 5–7=moderate depression, and 8–10=severe depression). The results showed that this global assessment was associated statistically with only six of the 17 items included in the HAM‐D ([7]). These six items are shown in Table 1.

The MES is also based on the Cronholm–Ottosson Depression Scale [for a comprehensive overview, see Bech ([8])]. The first version of the Cronholm–Ottosson Depression Scale was published in the same year as the first version of the HAM‐D ([9], [10]), but differed from the HAM‐D in its construct. Thus, the Cronholm–Ottosson Depression Scale was designed to rate changes in depressive states during treatment, while the HAM‐D originally was designed for measuring the severity of clinical symptomatology of depressive states. As a consequence, the Cronholm–Ottosson Depression Scale included only nine items in its first version ([9]) and eight items in its final version (Table 1). The HAM‐D versions have even been enlarged from 17 to 28 items to cover the full picture of the depressive symptomatology ([11]).

When validating the Cronholm–Ottosson Depression Scale ([8], [9]) the index of validity was the global improvement assessed by an experienced psychiatrist (0=unimproved; 1=slight improvement; 2=marked improvement, and 3=full remission). Table 1 shows that from the Cronholm–Ottosson Depression Scale the MES, which is based on the HAM‐D6, has included the symptoms of suicidal thoughts, sleep disturbance, poor concentration and emotional retardation.

MADRS, like the Cronholm–Ottosson Depression Scale, was designed to rate changes during treatment. This scale includes most of the MES items, but lacks the important item of motor retardation which in the MES is separated into verbal retardation and motor retardation. On the other hand, the MADRS includes the item of reduced appetite.

In this review of the MES, the psychometric properties of the scale will be shown with special emphasis on its external validity in being sensitive to measuring changes in depressive states during treatment, not only in regard to responsiveness during short‐term antidepressive therapies, but also when recognising relapse during medium‐term treatment or recurrences of depression during long‐term therapy. Results both with the HAM‐D6 and the MES will be taken into consideration.

The psychometric principle, consisting of internal validity, reliability, and external validity, has been applied. Internal validity contains both content validity (the extent to which the MES covers the DSM‐IV symptoms of major depression) and construct validity (the extent to which the total score of the MES items is a sufficient measure of changes in depressive states). Reliability signifies the interrater reliability or degrees of interrater agreement. External validity covers both responsiveness (the ability of the MES to measure changes in the depressive patients during different antidepressive treatments) and sensitivity (the ability of the MES to discriminate between active therapy and placebo).

Internal validity covers both content and construct validity.

The internal validity of the MES is high as the scale includes all symptoms of major depression according to DSM‐IV, apart from appetite/weight (Table 1). Because the MES is a measure of change during treatment the item of appetite/weight has been excluded. Thus, in our studies with the HAM‐D17 in which both appetite and weight are individual items we found that these items are often indicators of side‐effects of therapy rather than indicators of changes in depressive states ([7]). When measuring changes during treatment with patients with seasonal affective disorders (K.P.J. Martiny, C. Simonsen, M. Lunde, personal communication) we found that the MES was more sensitive than the HAM‐D because these patients are 'atypical' in having an increased appetite and not a reduced appetite.

Table 1 shows also that the MES includes two items not to be found in DSM‐IV, namely psychic anxiety and emotional retardation. In studies with the HAM‐D the item of psychic anxiety has been very important. Thus, we ([12]) and van Praag ([13]) have found that the depressive states are linked to psychic anxiety.

In his last analysis of the HAM‐D, Hamilton ([14]) showed that both in females and males with depressive illness (melancholia) the items of depressed mood, lack of interests, tiredness and psychic anxiety are the most inclusive (core) symptoms.

Concerning the item of emotional retardation it has been found important, when the MES is used, together with the Bech–Rafaelsen Mania Scale (MAS) to cover bipolar states of mania and melancholia ([15][16][17]), ranging from the lack of contact with others to intrusiveness in social contacts.

The assessment made by use of a rating scale is, as discussed by Hamilton ([18]), a skilled perception. Table 2 shows the levels of skilled perceptions as proposed by Dodwell ([19]) in which he refers to the Einstein statement: 'It is the theory which decides what we can observe'. When combining symptoms into states of depression, including changes in depression over time, the flow of information has both a passive and an active component in the skilled perception. The passive component deals with the perception of signals (symptoms) while the active component deals with the combination of signals (syndromes) into a diagnosis or into a dimension of change in the illness over time.

Information theory takes into account the amount of information that each symptom contains. Items with no amount of information have a redundancy in their message (no significant signal). Thus, a redundancy of 50% means that the average number of items per patient is only half of what it could be if all the items of the scale were transmitting significant clinical information. Information theory has several ways of measuring such redundancy. Miller ([20]) was the first to show that human observers are very limited when measuring information in terms of redundancy. They behave, as discussed by Dodwell ([19]), like limited‐capacity information channels. This was the background for Broadbent ([21]) to use algorithms to define mental states. DSM‐IV introduced algorithms for the diagnosis of mental states including major depression. The analysis of the HAM‐D with global ratings ([7]) was followed by a latent structure analysis to test the construct of the HAM‐D6 ([22]). It was shown that the HAM‐D6 was unidimensional (total score a sufficient statistic) which again was confirmed by Bech et al. ([23]) when evaluating the changes in depressive states during therapy with antidepressants.

Table 3 illustrates the principle of the latent structure analysis as found with the HAM‐D6 Miller ([20]), when establishing his information theory, found that the limited capacity for skilled perceptions was seven items plus or minus two. As one of the human constructs of the perspective of time he used the days of the week as an example. Latent structure analysis is a way to test the amount of information in the individual items. If they can be rank‐ordered as shown in Table 3, such as Tuesday is always closer to Monday than to Friday, the total score is a sufficient statistic of a scale.

Maier and Philipp ([24]) were the first to compare the HAM‐D17, MES and MADRS by latent structure analysis. In this study the MES, but not HAM‐D17 or the MADRS fulfilled the Rasch criteria. This was replicated by Maier et al. ([25]). The unidimensionality of the MES was also confirmed by Chambon et al. ([26]) by using the Rasch analysis in patients with DSM‐III major depression of an age between 18 and 80 years. However, the item of suicidal thoughts had small deviations from the unidimensional model in some of the subanalyses. Table 4 shows that Maier et al. ([25]) found that only HAM‐D6 and the MES were accepted by the latent structure (Rasch) analysis. The lack of undimensionality of MADRS has also been confirmed by Allerup ([27]) Therefore, the MADRS and HAM‐D17 were considered as multidimensional scales.

Table 5 shows the rank order of the MES scale in the French study ([28]). This structure has been compared with the MES‐correspondent items found sensitive to treatment by Nelson et al. ([29]).

Table 6 shows the item‐total score analysis of the MES ([30]) and the MADRS ([31]) which is another way to indicate that the items should be positively correlated with the total score of a scale if the total score is a sufficient statistic of severity ([32]). Davidson et al. ([31]) concluded that the MADRS seems not to be unidimensional, suicidal thoughts, reduced sleep, and appetite lead to lower coefficients.

Factor analyses of the MADRS and the MES have been very limited, probably because the use of factor analysis with the HAM‐D has illustrated that this method is very sensitive to the dispersion of scale scores in the sample of patients under examination ([22]). This has, for instance, been found for the MADRS by Galinowsky and Lehert ([33]) who showed that in depressed patients before therapy more than one factor was identified of which the general factor explained only 32% of the variance. However, after 4 weeks of therapy when the dispersion of the scale score was the largest, the general factor explained 66% of the variance. The only MES study on depressed patients comparable with the MADRS study ([33]) is the study by Chambon et al. ([28]) who examined depressed patients with the same age as included in the MADRS study ([33]), i.e. mean age approximately 42 years. In this MES study, however, the patients were only examined once, namely before therapy, and the first factor was a general factor explaining only 32% of the variance, i.e. of the same order as in the MADRS study ([33]).

The other MES studies with factor analysis have been carried out either on elderly subjects of whom only 20% were depressed ([34]) or on psychotic patients predominantly suffering from acute schizophrenia ([35]). In the latter study confirmatory factor analysis was used, identifying before therapy three factors which descriptively were called 'depressive core symptoms', 'retardation', and 'accessory depressive symptoms'.

Among latent structure analyses the Mokken analysis ([36]) should be considered a more useful test for unidimensionality than factor analysis within the frame of reference to correlation coefficients. Thus the Loevinger coefficient of homogeneity expresses the rank‐ordering of items in terms of transmitted information concerning the dimension under investigation ([37]). The coefficient of homogeneity in the Mokken analysis is the Loevinger coefficient. If the rank order is acceptable the Loevinger coefficient should be 0.40 or higher. For the MES a Loevinger coefficient of 0.47 was obtained while it was 0.30 for the HAM‐D17 and 0.44 for the HAM‐D6 ([38]).

The HAM‐D6 as well as the MES are scales measuring the severity of depressive states. They are not diagnostic (e.g. endogenous vs. reactive depression). In accordance with this, Thase et al. ([39]) found that an index of endogenous depression (the Hamilton Endogenomorphic Scale, HES), including items such as agitation, weight loss, and diurnal variation, correlated very poorly with the HAM‐D6 (0.33).

Interrater reliability is often shown as the intraclass coefficient when several raters are interviewing the same group of patients (joint ratings). Table 7 shows the interrater reliability as found by Maier et al. ([25]) when comparing the MES with the HAM‐D17, and the MADRS, both before treatment and after 3 weeks of therapy. In the DUAG (Danish University Antidepressant Group) studies the intraclass coefficient of the MES has typically been evaluated in depressed patients with a MES score of around 20 and an intraclass coefficient of 0.75 or higher have been obtained, equal to the coefficients found for the HAM‐D17 ([40]). In the first study on the interrater reliability of an MES the intraclass coefficient was 0.82 ([41]). Kørner et al. ([42]) found an interclass coefficient of 0.81, while Smolka and Stieglitz ([43]) found an intraclass coefficient of 0.87. In patients with depression comorbid with Parkinson's Disease, Årsland et al. ([44]) found an intraclass coefficient of 0.90, while Rossi et al. ([17]) found an intraclass coefficient of 0.92 in bipolar states.

In a judgement analysis study with the MES a Spearman coefficient of 0.98 was found when testing the test–retest reliability ([45]).

While only one version of the MES or the MADRS exist, [e.g. Schutte and Malouff ([46])], several different versions of the HAM‐D have been published over the years [e.g. Hooijer et al. ([47])]. Hamilton published two different versions of the HAM‐D ([5], [10]), of which the 1967 version is the one that is used most widely (e.g. the item of agitation measured on a Likert scale from 0 to 4, and not from 0 to 2). In drug trials the American version ([48]) is most frequently used internationally ([49]), probably because the American Food and Drug Association (FDA) prefers this version. However, Hamilton ([50]) never accepted this version. The version used in the DUAG (Danish University Antidepressant Group) ([51]) has been accepted by Hamilton ([12]) and has, as stated by Hooijer et al. ([47]), been considered as the standard version for Europe. Compared with the DUAG version, the total HAM‐D17 ([5]) was scored approximately one point higher in the study by Hooijer et al. ([47]). The American version includes more than 17 items and this often constitutes a problem when interpreting the severity of depression assessed with this version.

Figure 1 shows the terminology of response, remission, relapse, and recovery in the treatment of a major depressive episode as outlined by Frank et al. ([52]) or Kupfer ([53]). Figure 1 is a modification of the version by Bech ([54]) in which 'early improvement 'is included and the percentage of change has a reference to the MES.

1.  The terminology of response, remission, relapse and recovery in accordance with Kupfer (52), but modified to include 'early improvement' and percentage of changes with reference to the Melancholia Scale (55).

The HAM‐D is the standard rating scale for measuring severity of depressive states. Paykel ([55]) has formulated the world‐wide familiarity with the HAM‐D: '...We all know the approximate meaning of a Hamilton score of 17, 13, or 26 ...' Over the years the HAM‐D scores have been associated with the following clinical concepts: major depression (a HAM‐D17 score of 18 or more), response (a 50% reduction of the pretreatment HAM‐D17 score), relapse or of recurrence (a HAM‐D17 of 18 or more). Kørner et al. ([42]) have shown that a MES score of 15 approximately equals a HAM‐D17 score of 18 and a MADRS score of 24.

The first study comparing the sensitivity to change of the HAM‐D17, MES, and MADRS using the global clinical improvement after three weeks of therapy as index of validity ([25]) showed a coefficient of validity for the HAM‐D17 of 0.69, for the MES of 0.77, and for MADRS of 0.63. The difference between MES and MADRS was statistically significant ([25]).

Figure 1 shows the MES scores as typically seen in out‐patients with major depression, i.e. with a mean MES score before treatment of 20; an improvement score after 2 weeks of therapy of 15 (25%); and a response score after 4 weeks of therapy of 10 (50%). Only in patients with a response score of 10 or less it is meaningful to evaluate relapse during the continuation therapy. A relapse is defined as a MES score of 15 or more.

Table 8 shows the results with the MES in randomized clinical trials in samples of patients representing the whole spectrum of depression, from delusional depression ([56], [57]) over non‐delusional in‐patients with major depression ([51], [58]) to out‐patients with major depression ([59], [60]) and out‐patients with mixed minor/major depression ([61]). In these trials the serotonine‐specific reuptake inhibitors or SSRIs (paroxetine or citalopram) have been compared with tricyclics (imipramine, clomipramine, or maprotiline).

However, in the two trials on delusional depression the patients have been treated with a combination of ECT and imipramine or paroxetine ([56]), while amitriptyline has been combined with haloperidol when compared with risperidone alone ([57]).

Table 8 can be considered as the responsiveness of the MES in the spectrum of depressive states. In all the in‐patient trials (the four toplisted trials in Table 8) tricyclics are significantly superior to the SSRIs or risperdone after 4 weeks of therapy. In the DUAG ([58]) study the difference between clomipramine and paroxetine is already statistically significant after 2 weeks of therapy. This is in accordance with the meta‐analysis published by Anderson and Tomenson ([62]) or Anderson ([63]). In‐patients with major depression have a higher proportion of 'endogenous' or 'somatic' features. These endogenous features include items outside the universe of MES (such as distinct quality of depression, lack of reactivity, depression regularly worse in the morning, or significant anorexia and weight loss). When reviewing trials in which depressed patients have been subgrouped as having 'endogenous' features, Parker ([64]) showed that tricyclic antidepressants are superior to the SSRIs.

In such meta‐analyses the most frequently used statistic for outcome is the effect size which also has been used by Anderson ([63]). We have shown that effect size is a valid statistic when compared with other forms of statistics in meta‐analyses ([65]). The effect size is defined as the mean change between baseline and endpoint (typically 4–6 weeks of therapy) minus the mean change for the reference drug, divided by the pooled standard deviation of these differences.

As the HAM‐D17 has been used in over 95% of all controlled trials with SSRIs ([54]) it is the effect size of this scale that was used by Anderson and Tomenson ([62]) and Anderson ([63]), although the effect size is independent of the scale used in the trials. We found ([65]) that the HAM‐D6 had a higher effect size than HAM‐D17 when all placebo‐controlled trials with fluoxetine were analysed (0.38 vs. 0.30). Faries et al. ([66]) have shown that as a consequence of this difference in effect size it would require approximately one‐third fewer patients in studies using HAM‐D6 than in studies using HAM‐D17.

The effect size of 0.30 in a comparison of fluoxetine with placebo in patients with major depression ([65]) equals a response rate difference of between 15 and 20% on the HAM‐D17, the same magnitude as seen in other controlled trials with old or new antidepressants.

This limited difference in the efficacy of antidepressants compared with placebo in patients with mild to moderate degrees of depression has been discussed by Moncrieff ([67]). Among the problems Moncrieff raises is the use of HAM‐D17 as an outcome measure because this scale contains too many non‐specific or redundant depression items. Antidepressants with a strong antihistaminic effect, (e.g. amitriptyline or mirtazapine) are 'multidimensional' drugs as they have an impact on all the items of the 'multidimensional' HAM‐D17. In the out‐patient trials with mirtazapine compared with amitriptyline and placebo it has been found ([68]) that on the HAM‐D6 both drugs are superior to placebo, with effect sizes of 0.40 or higher.

In several studies, Angst and Stassen have shown that the delayed onset of action of antidepressants is 'something of a myth' ([69]). They introduced the concept of early improvement as a decrease in baseline rating scores of 20% or more, as the day to day fluctuations in the baselines HAM‐D17 scores, i.e. before treatment, are around 20%. Table 8 shows that the various types of antidepressive therapies have an early improvement of approximately 20% or more, but that statistical significance is first reached after 4 weeks of therapy, except for the DUAG ([58]) study.

Table 8 includes the study by Licht and Quitzau ([60]) in which patients with major depression in the private psychiatric practice setting were treated with a fixed dose of 50 mg daily of sertraline. Patients were then classified as responders or non‐responders after 6 weeks of therapy. The non‐responders went into a follow‐up trial over another 6 weeks. In total, 78% completed the 6 weeks open trial. Of these patients 24% were non‐responders. Already after 2 weeks of therapy the responders had improved significantly better than the non‐responders on the MES (Table 8).

The background for developing scales with a high degree of responsiveness is also to predict remission as early as possible during the first weeks of therapy. In the acute therapy of manic states it has recently been shown that the Bech–Rafaelsen Mania Scale (MAS) can predict remission after 11 days of therapy ([70]). This is in accordance with the overview of the psychometric properties of the MAS ([15]). In the acute therapy of depressive states the early improvement model (2 weeks) seems adequate as well (Table 8). Thus, using Kaplan–Meier curves, Angst and Stassen ([69]) have shown that the early onset of improvement after 2 weeks of therapy is highly predictive of later outcome. They have found that only around 20–25% of the patients are non‐responders. In regard to those patients who responded, Angst and Stassen ([69]) claimed that the effectiveness of antidepressants or placebo merely appears to trigger and maintain the conditions necessary for response.

In non‐responders to antidepressive medication, lithium augmentation was introduced by de Montigny et al. ([71]). They showed that the HAM‐D6 was more sensitive than the HAM‐D17 to measure response in therapy‐resistant patients with major depression. This effect of lithium has also been found for the SSRIs ([54]).

In the Standardized Stepwise Drug Treatment Regimen (SSTR) used since 1990 at the Department of Psychiatry at the Free University of Berlin ([72]) the algorithm for non‐response to treatment is a reduction of the MES scores by less than 25% after 2 weeks of therapy. In such cases the dose of the antidepressant should be increased. If the baseline MES has not been reduced by 50% or more after 4 weeks of therapy lithium is added. Bschor et al. ([73]) have analysed those depressed patients who in the period from 1990 to 1996 have received lithium augmentation according to the SSTR protocol ([72]). In total, 37 patients (or 52%) responded to the lithium augmentation which is equal to that found by Katona et al. ([74]) in a controlled clinical trial. Bschor et al. ([73]) demonstrated that the lithium responders were more severely depressed on the MES than the non‐responders.

Of the various augmentations (e.g. lithium, thyroid hormones, or pindolol) in therapy‐resistant depression lithium augmentation is still the most well‐documented medication ([75]).

As to the combination of antidepressants, Maes et al. ([76]) showed that the combination of fluoxetine and mianserin was superior to the combination of fluoxetine and pindolol. In a series of trials combining mianserin with imipramine ([77], [78]) or mianserin with fluoxetine ([79]) it has been shown that mianserin improves the efficacy of the imipramine or fluoxetine therapy. In these trials the MES has been found superior to HAM‐D17 in discriminating between the treatments.

When reviewing the randomized clinical trials with antidepressants ([54]) it was demonstrated that there have been conducted very few trials in the continuation phase (Fig. 1), i.e. the medium‐term treatment following the initial, acute treatment phase. According to the European Guidelines for Clinical Investigations of Antidepressants ([80]) a substance is accepted as an antidepressant if it can be shown to be more effective than placebo, not only in the acute phase but also in the medium‐term (continuation) phase.

Moncrieff ([67]) has criticised the medium‐term or relapse prevention trials because many of them rely on a discontinuation design. In these studies, patients who have responded to treatment with antidepressants are randomized to continuing on the active drug or being withdrawn to an inert placebo. Table 9 shows the two trials in which the MES has been used to demonstrate either a relapse prevention after ECT or after light therapy. In both trials the criticism raised by Moncrieff ([67]) has, thus, been taken into consideration.

The percentage of relapses in depression over 6 months after ECT in patients with a full remission has been used to evaluate the prophylactic effect of lithium or imipramine ([81]). The first study to use this model for serotonin selective reupdate inhibitors was published by Lauritzen et al. ([56]) where paroxetine was compared with imipramine not only during the acute phase in combination with ECT (Table 8), but also in the continuation phase after ECT. In this study, a relapse was defined either as a score of 18 or more on HAM‐D17 or as a score of 15 or more on the MES. In Table 9 the study by Lauritzen et al. ([56]) has been reanalysed to show the percentage of relapse for the two scales separately. The results indicated that only for the MES the difference between paroxetine and imipramine was statistical significant (P < 0.05). However, for the HAM‐D17 the level of statistically significance in favour of paroxetine was 0.06.

This 'ECT model' for relapse prevention has been used by Martiny et al. (personal communication) to compare citalopram with placebo in patients with mild to moderate depression (seasonal affective disorder) who had remitted after 1 week of light therapy. The relapse criteria in this study on less severely depressed patients compared with the ECT study ([56]) were scores for the MES of 11 or more and for the HAM‐D17 of 13 or more. The sensitivity of the MES (Table 9) was higher than that of the HAM‐D17 when citalopram was compared with placebo.

The analysis of the antidepressive properties of the SSRIs have mainly been made in non‐bipolar (unipolar) patients with recurrent major depressive episodes ([54]).

The MES has been compared with HAM‐D17 in a study on the prevention of depressive states after stroke ([82]). All patients were followed for up to 1 year after the stroke. The criteria for depression were similar to the criteria used for the MES and the HAM‐D17, as shown in Table 9. Sertraline was found statistically superior to placebo on the MES after 16 weeks of therapy, while on the HAM‐D17 this difference first was seen after 21 weeks of therapy.

In a double‐blind, placebo‐controlled study, the prophylactic effect of citalopram has been evaluated in elderly patients with recurrent major depression ([83]). This study was, in principle, a discontinuation study in which the patients before randomization to placebo in an open trial had responded to citalopram both in the acute treatment phase and in the continuation phase. In the maintenance phase the patients were followed over 52 weeks after the randomization. In this double‐blind maintenance part of the trial half of the patients received citalopram (n=60) and the other half placebo (n=61). The patients were evaluated every 4 weeks during the 52 weeks of therapy with the MES, MADRS, and the HAM‐D17.

Table 10 shows the results when the corresponding cut‐off scores were used for the three scales ([8], J. Bent‐Hansen, M. Lunde, K. Solstad, personal communication). The most clear discrimination between citalopram and placebo was seen for the MES showing that only 7% had a recurrence on citalopram against 38% on placebo. This is similar to the findings in Table 9 concerning the relapse‐prevention of the SSRIs.

The long‐term mood‐stabilizing effect in bipolar manic‐melancholic patients has shown the superiority of lithium over anticonvulsants ([15]). However, only few trials have been published and in these trials the Bech–Rafaelsen Mania Scale (MAS) has most frequently been used in combination with the HAM‐D.

In a double‐blind trial over 2 years, lithium was compared with carbamazepine in bipolar patients and both the MES and the MAS scales were used as outcome measures ([84]). The results showed that lithium was superior to carbamazepine in preventing new episodes.

In another study both MES and MAS were used when comparing daily lithium dosing to lithium dosing every second day. The results showed that 50% developed new episodes of depression or mania in the every second day dosing regimen while only 10% developed new episodes in the once daily dosing ([85]).

In their monograph on rating scales for the measuring of health and disease, McDowell and Newell ([86]) have concluded that '... compared with the Hamilton Scale the MADRS is easily administered and provides clearer guidelines to the rater. In turn, however, the Bech–Rafaelsen modification of the Hamilton Scale (MES) seems to offer further improvement...' This improvement in relation to the HAM‐D or the MADRS is, as discussed in the following, to be found in the psychometric triangle of validity, interrater reliability, and external validity (Fig. 2).

2.  Internal validity, reliability and external validity of the Bech–Rafaelsen Melancholia Scale, illustrated by the psychometric triangle.

The Bech–Rafaelsen Melancholia Scale (MES) has been shown to cover the same universe of symptoms as the DSM‐IV checklist for major depression. Symptoms with high diagnostic validity may, however, not necessarily have a high validity in measuring changes in severity of mental illnesses ([87], [88]). However, this review over empirical studies on the validity of the MES has shown that the 11 items constitute a unidimensional measure of depressive states. Both the Mokken analysis and the Rasch analysis have confirmed that the MES items have an internal hierarchy that is more consistent than the HAM‐D17 or the MADRS.

As discussed elsewhere ([89]) rating scales with a strong internal structure fulfilling the Rasch criteria for unidimensionality have also an adequate reliability. Empirical studies with the MES have shown that the interrater reliability is adequate with intraclass coefficients between 0.75 and 0.92, both in unipolar and bipolar patients. These coefficients are as high as those found for the DSM‐III or DSM‐IV diagnosis of major depression.

In this review the external validity of the MES has focused on early improvement, response and relapse or recurrence of major depressive episodes. When compared with the HAM‐D17 or the MADRS, the MES has been found superior especially in the trials on relapse and recurrence of major depressive episodes.

In the few studies with the MES on different biological markers of depression the scale has also been found superior to the HAM‐D17 ([91], [92]).

However, in studies on the plasma level vs. clinical response neither the MES nor the HAM‐D17 have show correlations of statistical significance ([51], [58], [93]).

As discussed by Hamilton ([18]) many scales have been developed to predict response to treatment. Such predictive scales have included what Hamilton refers to as 'fixed' items, for example, items such as family history, number of previous episodes, length of illness and types of personality. However, as stated by Hamilton ([18]) none of these scales have come into extensive use.

In a study on patients with chronic non‐malignant pain the MES has been found predictive for the outcome of antidepressants ([94]). Thus, a score of 15 or more was by Loldrup et al. ([94]) found to be associated with a high degree of remission with clomipramine when compared with placebo.

The MES trials included in the external validity analysis (Table 8) have not been able to conclude that the tricyclic antidepressants and the SSRIs are equally effective. In the acute therapy of depression the tricyclic antidepressants are superior to the SSRIs in hospitalized patients with severe degrees of depression. On the other hand, in the relapse prevention of patients who already have responded to ECT paroxetine was found superior to imipramine. This signifies that in the full‐blown, severe state of depression, the re‐establishment of the serotonin function is not enough, i.e. other neurotransmitters may be involved in the severe depressive states. However, after remission, in the relapse–prevention phase, the function of the SSRIs to maintain the serotonine balance seems sufficient. This theory is confirmed not only in the ECT study ([56]) but also in the light therapy study (K.P.J. Martiny, M. Lunde, C. Simonsen, personal communication) as well as in the recurrence‐prophylactic studies ([83] J. Bent‐Hansen, M. Lunde, K. Solstad, personal communication). The MES has been recommended among the clinician‐based scales to measure efficacy in the treatment of major depressive disorder ([95]).

A questionnaire to be completed by the patients themselves based on the HAM‐D17/MES version ([12]) has been developed by Bent‐Hansen et al. ([96]). They found a correlation coefficient of the MES self‐rating version of 0.68 when compared with the observer version of MES. This is within the limits of 0.40 and 0.75 often found when the Beck Depression Inventory (BDI) ([97]) or the Zung Self‐rating Depression Scale (SDS) ([98]) have been compared with HAM‐D17.

The MES is the first depression scale to fulfil the basic psychometric criteria of internal validity, reliability, and external validity. The MES covers approximately the DSM‐IV checklist of major symptoms of depression (content validity) and the scale is unidimensional, i.e. the total score is a sufficient statistic to assess severity of illness (construct validity). The interobserver reliability has been found to be acceptable in various settings.

The external validity of the MES has been tested both in regard to sensitivity and responsiveness. The sensitivity has been demonstrated in trials on the relapse prevention of depressive episodes in which a high MES discrimination was shown between active therapy and placebo when compared with other depression scales.

The responsiveness of the MES has been found high in short‐term trials in which the MES discrimination between TCAs and SSRIs was of statistical significance. Moreover, the improvement measured after 2 weeks of treatment was predictive for the final response.