0005~7967/92 $5.00f0.00
Eehou.Rex Thu. Vol. 30,No. 5,pp. 443451, 1992 Printed in Great Britain. All rights reserved
SEVERITY
Copyright Q 1992 Pergamon Press Ltd
OF UNIPOLAR DEPRESSION TREATMENT* PETER MCLEAN
Department
of Psychiatry,
The University
and
AND CHOICE
OF
STEVEN TAYLOR
of British Columbia, Canada V6T 2A1
(Received I December
2255 Wesbrook
Mall, Vancouver,
BC,
1991)
Summary-The National Institute of Mental Health (NIMH) Treatment of Depression Collaborative Research Program (Elkin et al., Archives of General Psychiatry, 46, 971-982; 1989) reported treatment-byseverity interactions favouring pharmacotherapy for more depressed outpatients, on a minority of relevant comparisons. The present study reports secondary analyses from a similar, preexisting data set in which treatment-by-severity interactions are systematically investigated with depressed outpatients treated either with nondirective psychotherapy, behaviour therapy, pharmacotherapy, or relaxation/placebo. Despite multiple severity measures and variable severity cut scores, no treatment was differentially effective in improving more severely depressed patients. Also, there was little difference across symptom severity levels in the proportions of recovered patients between treatment groups. Finally, dynamic cluster analysis demonstrated that the porportion of pharmacotherapy nonresponders (20%) did not differ from the proportion of nonresponders in behaviour therapy or placebo groups. It is concluded that this failure to replicate the NIMH trial findings can not be attributed to treatment differences, populations or statistical power. The suggestion that pharmacotherapy be the treatment of choice for more severely depressed outpatients appears to be unjustified on the basis of available evidence.
The recent large-scale National Institute of Mental Health (NIMH) Treatment of Depression Collaborative Research trial found little evidence of overall, differential treatment efficacy among psychological and pharmacological treatments of major depression (Elkin, Shea, Watkins, Imber, Sotsky, Collins, Glass, Pilkonis, Leber, Docherty, Fiester & Parloff, 1989). However, when patients were dichotomized on the basis of the Hamilton Rating Scale for depression (Hamilton, 1967) and the global assessment scale (Endicott, Spitzer, Fleiss & Cohen, 1976), some 9 out of 24 treatment-by-severity interactions proved statistically significant, with a probability level set at P < 0.10. For the more severely depressed patients, antidepressant medication (imipramine) gave the best result, followed by interpersonal psychotherapy, cognitive behaviour therapy, and the attention control condition. Elkin et al. (1989) did not report treatment-by-severity analyses for their other outcome measures, namely the Beck Depression Inventory (BDI: Beck, Rush, Shaw & Emery, 1979) or the Hopkins Symptom Check-list-90 Total Score (Derogatis, Lipman & Covi, 1973). While only 9 out of 48 possible comparisons supported the superiority of pharmacotherapy, the investigators found the results to be “provocative.” In an editorial immediately following this article’s appearance, the editor (Freedman, 1989) in a surprising media advisory, interpreted these results to mean that for more severely depressed patients I‘. . . drug treatment . . . shows more efficacy on more measures than other treatments” and proceeded to “. . . caution against an exclusion of pharmacotherapies based solely on ideology or limited professional competencies”, when treating depression (p. 983). Despite the modest nature of these findings, they bolster the common belief that more severe, if not all, cases of depression should be treated by pharmacological rather than psychological methods. Gelder (1990), for example suggests, “the clinician needs some guidance about the use of these psychological treatments for depressive disorders. Drug treatment . . . is the first line of treatment for most depressive disorders” (p. 1087). Similarly, Paykel (1989), referring to chronic and resistant cases of depression, suggests that the role of cognitive therapy is limited to patients who are reluctant to take antidepressants. There are few studies that address the issue of the comparative efficacy of psychological and pharmacological therapies as a function of the initial severity of depression. Unlike the NIMH trial *This article
is based on a paper
presented
at the European Congress of Behavioural Therapy, Oslo, Norway, 443
1991.
444
PETER MCLEAN and STEVENTAYLOR
reported by Elkin et al. (1989), Thase, Simons, Cahalance, McGeary and Harden (1991) found cognitive therapy to be equally effective by the end of treatment for depressed patients split into more and less severe subgroups. Blackburn, Bishop, Glen, Whalley and Christie (1981) found cognitive therapy to be more effective than pharmacological treatment in an outpatient population, but equivalent in efficacy to pharmacological treatment in a more severely depressed inpatient sample. In a recent review of psychological treatment studies, Robinson, Berman and Neimeyer (1990) found no evidence of a systematic relationship between initial depression symptom severity and treatment outcome. These results and conclusions bring into question the replicability of the treatment-by-severity interactions reported by Elkin et al. (1989). The purpose of the present study is to attempt to replicate the treatment-by-severity interaction reported by Elkin et al. (1989). Analyses are based on a previously reported study (McLean & Hakstian, 1979, 1990), the design of which is comparable to the Elkin et al. study in the method of selecting unipolar depressed outpatients, and in the number of Ss for whom complete treatment data are available (n = 151 for McLean & Hakstian, 1979; n = 155 for Elkin et al., 1989). In the McLean and Hakstian (1979) study, three of the following four treatments were comparable to those employed by Elkin et al.: nondirective (psychodynamic) psychotherapy, behaviour therapy, pharmacotherapy, and a relaxation/attention control treatment group. The BDI was an outcome measure common to both studies. In the present study the patient sample was stratified on the basis of initial depression severity, in order to investigate the relationship between pretreatment severity and treatment outcome. METHOD
Subjects Patients between 20 and 60 years of age were recruited through newspaper announcements. The diagnosis of moderate to severe unipolar depression was made according to the research criteria specified by Feighner, Robins, Guze, Woodruff, Winokur and Munoz (1972). This was determined by a three-stage assessment: (a) telephone screening to determine whether depressed mood had been present for at least 1 month, (b) a semi-structural clinical interview by a psychologist or psychiatrist to determine whether the S met Feighner criteria for depression, and (c) psychometric evaluation. To meet the latter, the potential S was required to have scores that were at least in the moderately depressed range for two of three measures: a score > 25 for males or 2 29.5 for females on Scale 2 of the MMPI; a score 223 on the BDI; or a score B 14 on the Depression Adjective Check List (DACL: Lubin, 1965). Patients were required to be fluent in English and not in treatment elsewhere for depression. Of 541 Ss screened for the study, 154 were initially assigned to treatment. The analyses reported in this article are limited to 15 1 of these patients, for whom complete pretreatment data was available. Thirty-six of these patients dropped out of treatment, and were replaced in order to control for differential attrition across treatment groups. Thus, our sample consisted of 151 treatment completers and 36 dropouts. For further details on S selection see McLean and Hakstian (1979). Measures Syndromal depression was measured on 28 questionnaire variables, This consisted of the BDI, and measured of 7 rationally-derived dimensions of depression: (a) cognitive functioning (subjective appraisal of memory functioning, decision-making ability, frequency of negative cognitions); (b) coping behaviour (number of productive and wasted hours, amount of procrastination); (c) personal activity (e.g. home duties, hobby involvement); (d) social functioning (frequency of various social activities); (e) somatic indicators (hours sleep per night, ratings of fatigue, ratings of relaxation); (f) genera1 life satisfaction Gob, marital, and housekeeping satisfaction); and (g) mood (DACL, frequency of crying, frequency of laughing). Each of the 7 dimensions consisted of 3-7 items, and referred to events or states within the previous l-7 days. The dimensions assessed important areas of functioning that are often absent or underrepresented in conventional treatment outcome studies, such as the study by Elkin et al. (1989). Patients were also assessed on a number of other depression variables (e.g. number of previous treatment episodes, duration of current episode), demographic variables (e.g. occupation
Severity Table 1. Number
of oatients
of unipolar
in each treatment Criteria
BDI > 28
condition
for defining
BDI > 31
depression and severitv
445 level for each severitv
the more severely depressed
BDI > 34
Mood > median
criterion
group Somatic ind. > median
Treatment condition
L
M
L
M
L
M
L
M
L
M
Psychotherapy Relaxation training Behaviour therapy Pharmacotherapy
20 24 21 I9
17 II I9 20
28 27 28 25
9 8 I2 I4
30 29 32 31
7. 6 8 8
16 14 23 24
21 21 I7 I5
17 I4 24 I9
20 21 I6 20
L = less severely depressed
group, M = more severely depressed
group.
and education status), and personality variables (e.g. the Eysenck Personality Questionnaire: Eysenck & Eysenck, 1975-1976). For further details see Table 1 of McLean and Hakstian (1979). Design Patients were randomly assigned to one of four treatments: (a) nondirective psychotherapy; (b) relaxation training (attention control condition); (c) behaviour therapy; or (d) pharmacotherapy. Patients were assessed at pretreatment, and at six points over a 27-month follow-up period: immediately after treatment, and 3, 9, 15, 21, and 27 months posttreatment. The results to be reported will be confined largely to outcomes assessed at posttreatment and 3-month follow-up. The long-term treatment main effects have been described previously (McLean & Hakstian, 1990). Treatments All treatments consisted of 10 weekly sessions conducted on an individual outpatient basis by a licensed and experienced psychologist, psychiatrist, or physician. Therapists were selected on the basis of their reputations in applying the particular type of treatment they offered, with the exception of relaxation training, for which therapists were trained. Patients were provided with an explanation for depression that was consistent with the particular treatment that they received. For example, the pharmacotherapy patients were given a biochemical account of depression. Each treatment session was 60 min in duration for the three psychological treatments, and 30 min for the pharmacological treatment. Nondirective psychotherapy used fundamental principles of short-term insight-oriented psychodynamic therapy, as described by Marmor (1973, 1975) and Wolberg (1967). Relaxation training is not usually used as an exclusive treatment for depression, but was intended as a psychotherapy placebo. Behaviour therapy used graduated practice and modelling techniques, with the following targets for intervention: communication, behavioural productivity, social interaction, assertiveness, decision-making, problem solving, and self-control (McLean, 1976). Daily homework assignments were described, and an emphasis was placed on the avoidance of depressive rumination by gainfully interacting with the environment in a manner that would lead to more frequent positive experiences. Pharmacotherapy consisted of amitriptyline, starting at 75 mg/day and graduating over a 10 day period to a fixed treatment dosage of 150 mg/day. After 11 weeks at 150 mg/day, the patients were weaned at a rate of 25 mg/day. Medication was taken as a single dose at bedtime. Compliance was assessed by unannounced blood samples drawn at two random visits over the 1l-week treatment period. Further details of the treatments are provided by McLean and Hakstian (1979). Analyses Since the treatment main effects have been described at length elsewhere (McLean & Hakstian, 1979, 1990), the analyses in the present article are restricted to the treatment-by-severity interaction. A critical issue is the criterion for defining patients as more or less severely depressed. Elkin et al. used a score 220 on the Hamilton Rating Scale for Depression. According to the data presented in their study (Table 1 of Elkin et al., 1989), this corresponds to a BDI score 228. Thus, we divided our patients into more and less severely depressed groups on the basis of a pretreatment BDI of 28, and performed two-way ANOVAs (treatment-by-severity) for each of the following dependent variables, as assessed immediately after treatment and at 3-month follow-up; BDI, mood, social functioning, personal activity, somatic indicators, average satisfaction, cognitive functioning, and coping behaviour.
PETER MCLEAN and STEVENTAYLOR
446
Since a BDI score 2 28 is an arbitrary criterion for defining a group as more severely depressed, we sought to examine the treatment-by-severity interaction by four other severity criteria. We repeated the above-mentioned analysis for each of the following criteria for defining the more severely depressed group: BDI B 3 1, BDI 2 34, mood 2 median, and somatic indicators 3 median. These variables all refer to pretreatment assessment. As before, the dependent variables were the eight outcome measures at posttreatment and 3-month follow-up. We investigated other BDI cutting scores because the optimal demarcation is unknown. We also used mood scores to define the severity criterion in order to examine the generality of the results. Somatic indicators were used as a severity criterion since it may be that patients with more somatic (“endogenous”) features are more likely to show a superior response to pharmacotherapy. Recently, Klein (1990) suggested that psychometric criteria should be used to divide patients into more and less severely depressed groups. Although he did not state what methods should be used, cluster analysis may be the most appropriate of available procedures. Such an analysis was reported by McLean and Hakstian (1979) and there was no significant cluster-by-treatment interaction on any of the outcome measures. In the present study, we performed additional cluster analyses, but did not rely heavily on this method, due to its inherent problems (Morris, Blashfield & Satz, 1981). Instead, we relied mainly on a systematic and comprehensive set of severity criteria, as described above. Our use of multiple severity criteria precluded the use of the usual strategy of protecting experiment-wise Type I error by performing a MANOVA on all of the dependent variables. Instead, we directly calculated interaction effects for 80 ANOVAs (8 dependent variables x 5 severity criteria x 2 assessment periods). To provide some protection against excessive Type I error, we set the alpha level to 0.01 for each test. Significant interactions were followed by simple main effects at each severity level (alpha = 0.01). If the latter were significant, then Tukey comparisons were performed to compare treatments at a given severity level (alpha = 0.01). Given the number of statistical tests involved, this approach is clearly liberal with respect to Type I error. We adopted this approach in order to increase statistical power, and so enhance our chances of replicating the findings of Elkin et al. This approach has the advantage of making null results more interpretable since the analyses are biased to decrease Type II error at the expense of Type I error. Since we found that the pattern of interaction effects did not differ across severity criteria, we used the BDI 2 28 criterion for the remaining analyses. This cutoff was chosen because it is closest to that used by Elkin et al., and provided approximately equal numbers of patients within each severity level. Two sets of analyses were then conducted. First, analyses of the dropout proportions across treatment and severity. Second, following Elkin et al., we analysed the proportions of “recovered” patients (BDI ~9) in each treatment and severity condition. RESULTS Pretreatment
comparisons
The 151 treatment completers (including dropout replacements) were distributed across the treatments in approximately equal proportions: psychotherapy (n = 37) relaxation training (n = 35), behaviour therapy (n = 40), and pharmacotherapy (n = 39). The treatment groups did not differ in age, gender composition, marital status, employment status, education level, neuroticism, number of previous treatment episodes for depression, or duration of current depressive episode (all Ps > 0.08). The groups also did not differ on the pretreatment dependent variables, Pillai F (24, 426) = 1.07, P < 0.37. Treatment
main effects
The main effects of treatment have been described in detail in previous articles (McLean & Hakstian, 1979, 1990). At posttreatment, behaviour therapy was superior to the other treatments on the BDI, measures of mood, social functioning, and average satisfaction. Trends in the same direction were found at 3-month follow-up, with behaviour therapy being significantly superior on measures of mood and social functioning. Psychotherapy performed most poorly on the majority of outcome measures at both evaluation periods. Pharmacotherapy was not superior to behaviour therapy on any outcome measure, and there were no significant differences between pharmacotherapy and relaxation therapy on any outcome measure.
Severity
of unipolar
447
depression
Table 2. Interaction effects (F values) for severity groups defined according Criteria
BDI > 28
for defining
BDlg31
to various
the more severely depressed
BDI > 34
Mood 2 median
criteria
group Somatic ind. a median
Dependent variable
Post
F.U.
Post
F.U.
Post
F.U.
Post
F.U.
Post
F.U.
BDI Mood Social functioning Personal activity Somatic indicators Average satisfaction Cognitive Coping
0.54 0.87 0.18 0.32 2.82 2.78 2.76 I.19
1.60 0.67 0.68 2.82 5.67’ I .99 3.81 3.32
0.89 1.54 0.32 0.30 1.62 I .70 0.66 0.18
2.63 2.05 0.02 3.39 4.70* 2.23 5.101 5.22*
0.64 I .62 0.55 I .02 0.62 I .07 0.55 0.30
2.24 1.80 0.98 0.62 2.91 1.22 3.86 3.34
0.84 2.60 I .03 0.52 0.55 I .78 0.45 0.96
I .93 0.67 3.16 0.82 I .78 I .98 0.90 I .57
I .69 I .68 0.71 0.54 I.15 I .36 I .29 0.92
3.85 1.71 0.60 0.60 2.41 1.24 2.71 3.32
Post = interaclion effect, F(3,143), for posttreatment for 3-month follow-up dependent variables. lP < 0.01.
Treatment-by-severity
dependent
variables,
F.U. = interaction, F(3,136),
interactions
The number of Ss on each treatment and severity level for each severity criterion are present in Table 1. Here it can be seen that the cell sizes were adequate for statistical purposes. The cell sizes for the BDI 228 and median-split criteria were comparable to those in the Elkin et al. (1989) study. Table 2 shows the interaction effects for the univariate ANOVAs for each dependent variable and severity criterion. Only four interactions were significantly at alpha = 0.01, and only one of the corresponding simple main effects was significant, F(3,38) = 4.84, P < 0.006. The latter was for the somatic indicators at 3-month follow-up in the more severe condition, defined by BDI 2 3 1. Here, Tukey comparisons revealed that the psychotherapy group had significantly more somatic complaints than the pharmacotherapy group, (P < 0.01). There was also a trend (P < 0.05) for the relaxation training and behaviour therapy groups to have fewer somatic complaints than the psychotherapy group. Pharmacotherapy, behaviour therapy, and relaxation training did not differ from one another (P > 0.05). Following Elkin et al., the analyses were repeated with an “intend to treat” group, defined as the treatment completers (n = 115) plus dropouts (n = 36), and not including dropout replacements (n = 36). Thus, the sample consisted of patients who were initially admitted to treatment, and dropouts were assigned their pretreatment score as their posttreatment value. The results for this group can be considered as an estimate of the overall performance of the treatments, including their abilities to retain patients in treatment. The pattern of results was similar to that reported in Table 2, and failed to reveal an interaction effect favouring pharmacotherapy at any severity level. In summary, despite the use of a significance level that was liberal with respect to Type I error, there was no support for the hypothesis that pharmacotherapy is a superior treatment for more severely depressed patients. Indeed, pharmacotherapy tended to be statistically undistinguishable from the placebo, relaxation training (McLean & Hakstian, 1979). These conclusions did not change when results were analysed at 9-, 15-, 21- and 27 month follow-up. Treatment
dropouts
The analyses reported in the previous section show that the pattern of treatment-by-severity interactions did not change appreciably over severity criteria. The results of the analyses of dropouts were also consistent over severity criteria, and so we confine ourselves to the BDI 228 criterion, which was comparable to the criterion used by Elkin et al. Behaviour therapy was associated with the fewest overall dropouts (7.5%) and pharmacotherapy had the most (33.3%). The proportion of dropouts within each treatment and severity condition are presented in Table 3. The proportion of dropouts across severity levels and treatment groups was not homogeneous, x2(3, n = 36) = 75.00, P < 0.001. To identify the sources of heterogeneity, analyses of residuals were calculated, and the alpha level was set at 0.01. There were significantly more dropouts in the more depressed group relative to the less depressed group for patients receiving psychotherapy, d = 2.93, P < 0.003, and relaxation training, d = 5.97, P < 0.001. The proportion of dropouts did not differ in the behaviour therapy group, although there was a trend for more dropouts in the less severe group, d = 2.04, P < 0.02. This result must be interpreted with
PETER MCLEAN and STEVEN TAYLOR
448 Table
3. Dropouts
within each treatment severity level
condition
and
Table 4. Patients “recovered” at termination BDI < 9)
(posttreatment
Severity*
seventy*
MOK
Less
Treatment condition
Treatment condition
Psychotherapy Relaxation training Behaviour therapy Pharmacotherapy ‘More
severe groups:
15.0 8.3 9.5 52.6
3 2 2 IO
BDI (pretreatment)
47. I 63.6 53 15.0 >28.
8 7 I 3
Psychotherapy Relaxation traming Behaviour therapy Pharmacotherapy *More severe groups:
%
,t
%
n
35.0 37.5 51. I 52.6
I 9 I2 IO
29.4 34.4 51.9 25.0
5 4 II 5
BDI (pretreatment)
3 28.
caution due to the small number of dropouts in the group receiving behaviour therapy (Table 3). Interestingly, the pharmacotherapy group had significantly more dropouts in the less depressed group, d = 7.76, P < 0.001. When the severity groups were defined according to other criteria, the dropout rates still did not favour pharmacotherapy at any severity level, and behaviour therapy had the consistently lowest rate of attrition. Proportion
of “recovered”
patients
As suggested by Elkin et al. (1989), patients were defined as “recovered” if they had a posttreatment BDI ~9. The number and proportion of patients within each treatment group and severity condition are presented in Table 4. Here, the more severely depressed group was defined by BDI 828. The table shows that there was generally little difference across severity levels in the proportions of recovered patients. However, there was a trend toward an heterogeneity of cell proportions, x2(3, n = 151) = 7.03, P < 0.075. Analyses of residuals revealed that for the pharmacotherapy group there were fewer recovered patients in the group that was initially more depressed, d = 2.61, P < 0.005. For the other treatments there was no significant difference between the proportions of recovered patients across severity levels. Further analyses, using the other severity criteria, failed to support the hypothesis that pharmacotherapy is the most efficacious treatment, or that pharmacotherapy exerts greater treatment effects on more, rather than less, severely depressed patients. Cluster analyses In the original report, McLean and Hakstian (1979) performed a cluster analysis on several of the pretreatment characteristics of the patients, obtaining a four-cluster solution. The clusters differed on a variety of dimensions, including that of depression severity. There was no significant interaction between cluster and treatment. In the final set of analyses to be reported in the present article, we take the question of differential treatment effects one step further by using a method known as dynamic clustering (Prochaska, Velicer, Guadagnoli, Rossi & DiClemente, 1991). This is an application of cluster analysis where Ss are grouped on the basis of the similarity of their scores on a given variable over time. Thus, instead of clustering Ss on the basis of their scores on a number of variables at a given time, Ss are clustered on the basis of their scores on a single variable measured at several different times. We performed dynamic clustering on the patients’ BDI scores taken at pretreatment, posttreatment, and at five follow-up points (3,9, 15,21, and 27 months). This analysis enabled us to identify patterns of treatment response, and to determine whether a particular form of treatment was more likely to be associated with a given treatment response. For example, on the basis of the Elkin et al. study, a three-cluster solution would be predicted. One cluster would be characterized by patients who initially had moderate levels of depression, and who made good treatment gains regardless of the type of treatment. Thus, this cluster would contain approximately equal proportions of patients from each treatment group. The second cluster would consist of patients who were severely depressed at intake, and who did not respond to treatment. Elkin et al.‘s findings lead to the prediction that this cluster would be composed mainly of patients who received psychotherapy, relaxation training, or behaviour therapy, and relatively few patients who were treated with amitriptyline. The third cluster should consist of severely depressed patients who made good treatment gains. This cluster is predicted to consist mainly of patients receiving amitriptyline. If
449
Severity of unipolar depression
30 25 20
9
15
21
27
Follow-up (months)
Fig. I. BDI scores at pretreatment and follow-up for each cluster of depressed patients.
there is no treatment by severity interaction, then a two-cluster solution might be expected, with one cluster consisting of treatment responders and another consisting of nonresponders, with patients from each of the treatment groups distributed approximately equally over clusters. Four-cluster analyses were performed in order to rule out any artifactual results arising from the type of clustering algorithm or choice of assessment points (cf. Morris et al., 1981). Two clusterings were performed using Ward’s method, and two using the average linkage method. For each method, two clusterings were performed using either pre- and posttreatment BDI scores as the cluster variables, or BDI scores obtained at all 7 time points. All solutions were similar, indicating a two-cluster solution in each case. Therefore, only the results for one analysis will be reported. The analysis using all 7 time points was selected since it provides a broader picture of the long-term outcome for each cluster, although there was some sacrifice in sample size due to sample attrition (complete data across the 7 time points was available for 121 patients). Figure 1 shows the BDI scores for each cluster. Here it can be seen that Cluster A can be described as the treatment nonresponders, who were slightly more depressed at intake than Cluster B, the treatment responders. The number of Ss in each treatment condition and cluster type are presented in Table 5. In Cluster A the proportion of patients from each type of treatment were as follows: psychotherapy (1 l%), relaxation training (20%), behaviour therapy (18%), and amitriptyline (20%). Thus, with the exception of psychotherapy, there were approximately equal proportions in each cluster. The results of Table 5 fail to support the view that severely depressed patients are selectively responsive to pharmacotherapy. It appears that the nonresponders (Cluster A) failed to respond to any of the treatments. Considering both the cluster results and the previous analyses of the treatment main effects (McLean & Hakstian, 1979, 1990) it appears that although the Cluster B patients responded to all of the treatments, the behaviour therapy group tended to respond best. Cluster A patients tended to respond poorly to any treatment in the study. The nature of these clusters and their distinctions is beyond the scope of the present article, and will be the subject of a subsequent paper.
Table 5. Number of patients in each treatment condition and cluster Cluster Treatment condition
A
B
Psychotherapy Relaxation training Behaviour therapy Pharmacotheraov
3 7 6 5
25 28 27 20
450
PETERMCLEAN and STEVENTAYLOR
DISCUSSION
The present study failed to replicate findings of Elkin et al. (1989) despite comparable sample sizes and the use of a liberal alpha level. Our use of multiple severity criteria suggest that the results were not simply an artifact of an arbitrary criterion for defining patients as more or less severely depressed. There are a number of possible explanations as to why our results failed to replicate those of Elkin et al. First, it may be argued that the difference in results was due to a difference in antidepressants. However, reviews of the comparative efficacy of tricyclic antidepressants have found amitriptyline to be as effective as imipramine when equivalent doses are administered (Davis & Glassman, 1989; Morris & Beck, 1974). Moreover, Davis and Glassman (1989) claimed that 240 mg of imipramine has an equivalent antidepressant effect to 150 mg of amitriptyline, suggesting that the latter is a more potent antidepressant. These findings suggest that the doses of amitriptyline administered in the present study (150 mg/day) were probably as effective as the doses of imipramine administered in the study by Elkin et al. (M = 185 mg/day). However, the therapies differed in duration, with the Elkin et al. patients receiving an average of six more treatment sessions than patients in the present study. In our study 43% of pharmacotherapy patients had posttreatment BDI scores in the recovered range, whereas 69% recovered in the Elkin et al. study. If the drug treatment in the present study was less effective than that in Elkin et al’s study, then it is possible that the treatment-by-severity interaction effects were attenuated. Even if this were the case, it is notable that we did not find a trend in the direction of favouring pharmacotherapy for more severely depressed patients. The two drugs differ in other ways. Amitriptyline is considered to have greater sedative and anticholinergic effects (Hyman & Arana, 1987). The greater anticholinergic effects of amitriptyline made no difference in the level of attrition in the present study relative to that in the Elkin et al. study (33% in both cases). The greater sedative effects of amitriptyline might make this drug more efficacious than imipramine in agitated depression. However, there is no reason why this difference in sedative effects should account for our failure to replicate Elkin et al’s findings. Thus, we conclude that our failure to replicate their findings is unlikely to have been due to differences in drugs. Another possible explanation for our failure to replicate Elkin et al’s interaction concerns the differences in the types of cognitive-behavioural treatments that were administered. Compared to the cognitive therapy used in the Elkin et al. study, the behavioural treatment in the present study was considerably more focused on increasing patients’ level of activity, and less concerned with cognitive restructuring. As mentioned earlier, the therapies also differed in duration, with the Elkin et al. patients receiving an average of six more treatment sessions. Our behavioural treatment, which included cognitive techniques, appears to have been approximately equal in efficacy to Elkin et al’s cognitive therapy, as suggested by the proportions of recovered patients (respectively, 57 vs 65% on the posttreatment BDI). Thus, differences in the nature of the two therapies are unlikely to account for the failure to replicate their findings. It could be argued that we treated a different sample to those in the Elkin et al. study, since we diagnosed unipolar depression according to Feighner et al. (1972) criteria, whereas Elkin et a/. used Research Diagnostic Criteria (Spitzer, Endicott & Robins, 1978). It seems unlikely that these differences would account for our failure to find a treatment-by-severity interaction, since the two sets of diagnostic criteria have more similarities than differences. Another possible explanation is that we did not replicate Elkin et al’s interaction because the latter was a spurious finding based on an unduly liberal alpha level. Although Elkin et al. claimed that their analyses were conservative, it is clear from their paper that the alpha level for their analyses of interaction effects was greater than 0.10. Regardless of which explanation best accounts for the lack of replicability, our results demonstrate clearly that severe depression is neither necessarily a contra-indication for cognitive-behavioural therapy, or a specific indication for pharmacotherapy. Our findings showed that the therapies did not vary in their comparative efficacy across depression severity. This conclusion holds for depressed outpatients. However, it remains to be seen whether this also applies to inpatients. The results of this study indicate that the preference for pharmacotherapy over cognitive-behavioural therapies lacks convincing support.
Severity
of unipolar
depression
451
REFERENCES Beck, A. T. (1967). Depression. New York: Harper & Row. Beck, A. T., Rush, A. J., Shaw, B. F. & Emery, G. (1979). Cognifive therupy of depression. New York: Guilford. Blackburn, I. M., Bishop, S., Glen, A. I. M., Whalley, L. J. & Christie, J. E. (1981) The efficacy of cognitive therapy in depression: A treatment trial using cognitive therapy and pharmacotherapy, each alone and in combination. British Journal of Psychiatry, 139, 181L189. Davis, J. M. & Classman, A. H. (1989). Antidepressant drugs. In Kaplan, H. I. & Sadock, B. J. (Eds), Comprehensive fextbook of psychiatry (5th Edn, pp. 1627-1655). Baltimore: Williams & Wilkins. Derogatis, L. R., Lipman, R. S. & Covi, L. (1973). SCL-90: An outpatient psychiatric rating scale-preliminary report. Psychopharmacology Bulletin, 9, 13-28. Elkin, I., Shea, T., Watkins, J. T., Imber, S. D., Sotsky, S. M., Collins, J. F., Glass, D. R., Pilkonis, P. A., Leber, W. R., Docherty, J. P., Fiester, S. J. & Parloff, M. B. (1989). National Institute of Mental Health Treatment of Depression Collaborative Research Program: General effectiveness of treatments. Archives of General Psychiatry, 46, 971-982. Endicott, J., Spitzer, R. L., Fleiss, J. L. &Cohen, J. (1976). The Global Assessment Scale: A procedure for measuring overall severity of psychiatric disturbance. Archives of General Psychiatry, 33, 766-771. Eysenck, H. J. & Eysenck, S. B. G. (197551976). Eysenck personality questionnaire. San Diego, CA: Educational and Industrial Testing Service. Fieghner, J. P., Robins, E., Guze, S. B., Woodruff, R. A., Winokur, R. A. & Munoz, R. (1972). Diagnostic criteria for use in psychiatric research. Archives of General Psychiatry, 26, 5763. Freedman, D. X. (1989). Editorial note (especially for the media). Archives of General Psychiatry, 46, 983. Gelder, M. G. (1990). Psychological treatment for depressive disorder. British Medical Journal, 300, 1087-1088. Hamilton, M. A. (1967). Development of a rating scale for primary depressive illness. British Journal of Social and Clinical Psychology, 6, 278-296. Hyman, S. E. & Arana, G. W. (1987). Handbook of psychiatric drug therapy. Boston: Little, Brown & Co. Klein, D. F. (1990). NIMH collaborative research on treatment of depression (letter). Archives of General Psychiatry, 47, 682-684. Lubin, B. (1965). Adjective checklists for measurements of depression. Archives of General Psychiatry, 12, 5762. Marmor, J. (1973). Psychiatry in transition. New York: Brunner/Mazel. Marmor, J. (1975). The nature of the psychotherapeutic process revisited. Canadian Psychiatric Association Journal, 20, 557-565. McLean, P. D. (1976). Decision-making in the behavioural management of depression. In Davidson, P. 0. (Ed.), Behavioural management of anxiety, depression and pain (pp. 5489). New York: Brunner/Mazel. McLean, P. D. & Hakstian, A. R. (1979). Clinical depression: Comparative efficacy of outpatient treatments, Journal of Consulting and Clinical Psychology, 47, 818-836. _ McLean, P. D. & Hakstian, A. R. (1990). Relative endurance of unipolar depression treatment effects: Longitudinal follow-up. Journal qf Consulting and Clinical Psychology, 58, 482488. Morris, J. B. & Beck, A. T. (1974). The efficacy of antidepressant drugs. Archives of General Psychiatry, 34 667474. Morris, R., Blashfield, R. & Satz, P. (1981). Neuropsychology and cluster analysis: Potentials and problems. Journal of Clinical Neuropsychology, 3, 79-99. Paykel, E. S. (1989). Treatment of depression: The relevance of research for clinical practice. British Journal of Psychiatry, 155, 754-763. Prochaska, J. O., Velicer, W. F., Guadagnoli, E., Rossi, J. S. & DiClemente, C. C. (1991). Patterns of change: Dynamic typology applied to smoking cessation. Mulfiuuriute Behavioral Research, 26, 83-107. Robinson, L. A., Berman, J. S. & Neimeyer, R. A. (1990). Psychotherapy for the treatment of depression: A comprehensive review of controlled outcome research. Psychological Bulletin, IO8, 30-49. Spitzer, R. L., Endicott, J. & Robins, E. (1978). Research diagnostic criteria: Rationale and reliabilitv. Archives of General Psychiatry, 35, 773-782. Thase, M. E., Simons, A. D., Cahalance, J., McGreary, J. & Harden, T. (1991). Severity of depression and response to cognitive behavior therapy. American Journal of Psychiatry, 148, 784-789. Wolberg, L. R. (1967). Short-term psychotherapy. New York: Grune & Stratton.
Eehou.Rex Thu. Vol. 30,No. 5,pp. 443451, 1992 Printed in Great Britain. All rights reserved
SEVERITY
Copyright Q 1992 Pergamon Press Ltd
OF UNIPOLAR DEPRESSION TREATMENT* PETER MCLEAN
Department
of Psychiatry,
The University
and
AND CHOICE
OF
STEVEN TAYLOR
of British Columbia, Canada V6T 2A1
(Received I December
2255 Wesbrook
Mall, Vancouver,
BC,
1991)
Summary-The National Institute of Mental Health (NIMH) Treatment of Depression Collaborative Research Program (Elkin et al., Archives of General Psychiatry, 46, 971-982; 1989) reported treatment-byseverity interactions favouring pharmacotherapy for more depressed outpatients, on a minority of relevant comparisons. The present study reports secondary analyses from a similar, preexisting data set in which treatment-by-severity interactions are systematically investigated with depressed outpatients treated either with nondirective psychotherapy, behaviour therapy, pharmacotherapy, or relaxation/placebo. Despite multiple severity measures and variable severity cut scores, no treatment was differentially effective in improving more severely depressed patients. Also, there was little difference across symptom severity levels in the proportions of recovered patients between treatment groups. Finally, dynamic cluster analysis demonstrated that the porportion of pharmacotherapy nonresponders (20%) did not differ from the proportion of nonresponders in behaviour therapy or placebo groups. It is concluded that this failure to replicate the NIMH trial findings can not be attributed to treatment differences, populations or statistical power. The suggestion that pharmacotherapy be the treatment of choice for more severely depressed outpatients appears to be unjustified on the basis of available evidence.
The recent large-scale National Institute of Mental Health (NIMH) Treatment of Depression Collaborative Research trial found little evidence of overall, differential treatment efficacy among psychological and pharmacological treatments of major depression (Elkin, Shea, Watkins, Imber, Sotsky, Collins, Glass, Pilkonis, Leber, Docherty, Fiester & Parloff, 1989). However, when patients were dichotomized on the basis of the Hamilton Rating Scale for depression (Hamilton, 1967) and the global assessment scale (Endicott, Spitzer, Fleiss & Cohen, 1976), some 9 out of 24 treatment-by-severity interactions proved statistically significant, with a probability level set at P < 0.10. For the more severely depressed patients, antidepressant medication (imipramine) gave the best result, followed by interpersonal psychotherapy, cognitive behaviour therapy, and the attention control condition. Elkin et al. (1989) did not report treatment-by-severity analyses for their other outcome measures, namely the Beck Depression Inventory (BDI: Beck, Rush, Shaw & Emery, 1979) or the Hopkins Symptom Check-list-90 Total Score (Derogatis, Lipman & Covi, 1973). While only 9 out of 48 possible comparisons supported the superiority of pharmacotherapy, the investigators found the results to be “provocative.” In an editorial immediately following this article’s appearance, the editor (Freedman, 1989) in a surprising media advisory, interpreted these results to mean that for more severely depressed patients I‘. . . drug treatment . . . shows more efficacy on more measures than other treatments” and proceeded to “. . . caution against an exclusion of pharmacotherapies based solely on ideology or limited professional competencies”, when treating depression (p. 983). Despite the modest nature of these findings, they bolster the common belief that more severe, if not all, cases of depression should be treated by pharmacological rather than psychological methods. Gelder (1990), for example suggests, “the clinician needs some guidance about the use of these psychological treatments for depressive disorders. Drug treatment . . . is the first line of treatment for most depressive disorders” (p. 1087). Similarly, Paykel (1989), referring to chronic and resistant cases of depression, suggests that the role of cognitive therapy is limited to patients who are reluctant to take antidepressants. There are few studies that address the issue of the comparative efficacy of psychological and pharmacological therapies as a function of the initial severity of depression. Unlike the NIMH trial *This article
is based on a paper
presented
at the European Congress of Behavioural Therapy, Oslo, Norway, 443
1991.
444
PETER MCLEAN and STEVENTAYLOR
reported by Elkin et al. (1989), Thase, Simons, Cahalance, McGeary and Harden (1991) found cognitive therapy to be equally effective by the end of treatment for depressed patients split into more and less severe subgroups. Blackburn, Bishop, Glen, Whalley and Christie (1981) found cognitive therapy to be more effective than pharmacological treatment in an outpatient population, but equivalent in efficacy to pharmacological treatment in a more severely depressed inpatient sample. In a recent review of psychological treatment studies, Robinson, Berman and Neimeyer (1990) found no evidence of a systematic relationship between initial depression symptom severity and treatment outcome. These results and conclusions bring into question the replicability of the treatment-by-severity interactions reported by Elkin et al. (1989). The purpose of the present study is to attempt to replicate the treatment-by-severity interaction reported by Elkin et al. (1989). Analyses are based on a previously reported study (McLean & Hakstian, 1979, 1990), the design of which is comparable to the Elkin et al. study in the method of selecting unipolar depressed outpatients, and in the number of Ss for whom complete treatment data are available (n = 151 for McLean & Hakstian, 1979; n = 155 for Elkin et al., 1989). In the McLean and Hakstian (1979) study, three of the following four treatments were comparable to those employed by Elkin et al.: nondirective (psychodynamic) psychotherapy, behaviour therapy, pharmacotherapy, and a relaxation/attention control treatment group. The BDI was an outcome measure common to both studies. In the present study the patient sample was stratified on the basis of initial depression severity, in order to investigate the relationship between pretreatment severity and treatment outcome. METHOD
Subjects Patients between 20 and 60 years of age were recruited through newspaper announcements. The diagnosis of moderate to severe unipolar depression was made according to the research criteria specified by Feighner, Robins, Guze, Woodruff, Winokur and Munoz (1972). This was determined by a three-stage assessment: (a) telephone screening to determine whether depressed mood had been present for at least 1 month, (b) a semi-structural clinical interview by a psychologist or psychiatrist to determine whether the S met Feighner criteria for depression, and (c) psychometric evaluation. To meet the latter, the potential S was required to have scores that were at least in the moderately depressed range for two of three measures: a score > 25 for males or 2 29.5 for females on Scale 2 of the MMPI; a score 223 on the BDI; or a score B 14 on the Depression Adjective Check List (DACL: Lubin, 1965). Patients were required to be fluent in English and not in treatment elsewhere for depression. Of 541 Ss screened for the study, 154 were initially assigned to treatment. The analyses reported in this article are limited to 15 1 of these patients, for whom complete pretreatment data was available. Thirty-six of these patients dropped out of treatment, and were replaced in order to control for differential attrition across treatment groups. Thus, our sample consisted of 151 treatment completers and 36 dropouts. For further details on S selection see McLean and Hakstian (1979). Measures Syndromal depression was measured on 28 questionnaire variables, This consisted of the BDI, and measured of 7 rationally-derived dimensions of depression: (a) cognitive functioning (subjective appraisal of memory functioning, decision-making ability, frequency of negative cognitions); (b) coping behaviour (number of productive and wasted hours, amount of procrastination); (c) personal activity (e.g. home duties, hobby involvement); (d) social functioning (frequency of various social activities); (e) somatic indicators (hours sleep per night, ratings of fatigue, ratings of relaxation); (f) genera1 life satisfaction Gob, marital, and housekeeping satisfaction); and (g) mood (DACL, frequency of crying, frequency of laughing). Each of the 7 dimensions consisted of 3-7 items, and referred to events or states within the previous l-7 days. The dimensions assessed important areas of functioning that are often absent or underrepresented in conventional treatment outcome studies, such as the study by Elkin et al. (1989). Patients were also assessed on a number of other depression variables (e.g. number of previous treatment episodes, duration of current episode), demographic variables (e.g. occupation
Severity Table 1. Number
of oatients
of unipolar
in each treatment Criteria
BDI > 28
condition
for defining
BDI > 31
depression and severitv
445 level for each severitv
the more severely depressed
BDI > 34
Mood > median
criterion
group Somatic ind. > median
Treatment condition
L
M
L
M
L
M
L
M
L
M
Psychotherapy Relaxation training Behaviour therapy Pharmacotherapy
20 24 21 I9
17 II I9 20
28 27 28 25
9 8 I2 I4
30 29 32 31
7. 6 8 8
16 14 23 24
21 21 I7 I5
17 I4 24 I9
20 21 I6 20
L = less severely depressed
group, M = more severely depressed
group.
and education status), and personality variables (e.g. the Eysenck Personality Questionnaire: Eysenck & Eysenck, 1975-1976). For further details see Table 1 of McLean and Hakstian (1979). Design Patients were randomly assigned to one of four treatments: (a) nondirective psychotherapy; (b) relaxation training (attention control condition); (c) behaviour therapy; or (d) pharmacotherapy. Patients were assessed at pretreatment, and at six points over a 27-month follow-up period: immediately after treatment, and 3, 9, 15, 21, and 27 months posttreatment. The results to be reported will be confined largely to outcomes assessed at posttreatment and 3-month follow-up. The long-term treatment main effects have been described previously (McLean & Hakstian, 1990). Treatments All treatments consisted of 10 weekly sessions conducted on an individual outpatient basis by a licensed and experienced psychologist, psychiatrist, or physician. Therapists were selected on the basis of their reputations in applying the particular type of treatment they offered, with the exception of relaxation training, for which therapists were trained. Patients were provided with an explanation for depression that was consistent with the particular treatment that they received. For example, the pharmacotherapy patients were given a biochemical account of depression. Each treatment session was 60 min in duration for the three psychological treatments, and 30 min for the pharmacological treatment. Nondirective psychotherapy used fundamental principles of short-term insight-oriented psychodynamic therapy, as described by Marmor (1973, 1975) and Wolberg (1967). Relaxation training is not usually used as an exclusive treatment for depression, but was intended as a psychotherapy placebo. Behaviour therapy used graduated practice and modelling techniques, with the following targets for intervention: communication, behavioural productivity, social interaction, assertiveness, decision-making, problem solving, and self-control (McLean, 1976). Daily homework assignments were described, and an emphasis was placed on the avoidance of depressive rumination by gainfully interacting with the environment in a manner that would lead to more frequent positive experiences. Pharmacotherapy consisted of amitriptyline, starting at 75 mg/day and graduating over a 10 day period to a fixed treatment dosage of 150 mg/day. After 11 weeks at 150 mg/day, the patients were weaned at a rate of 25 mg/day. Medication was taken as a single dose at bedtime. Compliance was assessed by unannounced blood samples drawn at two random visits over the 1l-week treatment period. Further details of the treatments are provided by McLean and Hakstian (1979). Analyses Since the treatment main effects have been described at length elsewhere (McLean & Hakstian, 1979, 1990), the analyses in the present article are restricted to the treatment-by-severity interaction. A critical issue is the criterion for defining patients as more or less severely depressed. Elkin et al. used a score 220 on the Hamilton Rating Scale for Depression. According to the data presented in their study (Table 1 of Elkin et al., 1989), this corresponds to a BDI score 228. Thus, we divided our patients into more and less severely depressed groups on the basis of a pretreatment BDI of 28, and performed two-way ANOVAs (treatment-by-severity) for each of the following dependent variables, as assessed immediately after treatment and at 3-month follow-up; BDI, mood, social functioning, personal activity, somatic indicators, average satisfaction, cognitive functioning, and coping behaviour.
PETER MCLEAN and STEVENTAYLOR
446
Since a BDI score 2 28 is an arbitrary criterion for defining a group as more severely depressed, we sought to examine the treatment-by-severity interaction by four other severity criteria. We repeated the above-mentioned analysis for each of the following criteria for defining the more severely depressed group: BDI B 3 1, BDI 2 34, mood 2 median, and somatic indicators 3 median. These variables all refer to pretreatment assessment. As before, the dependent variables were the eight outcome measures at posttreatment and 3-month follow-up. We investigated other BDI cutting scores because the optimal demarcation is unknown. We also used mood scores to define the severity criterion in order to examine the generality of the results. Somatic indicators were used as a severity criterion since it may be that patients with more somatic (“endogenous”) features are more likely to show a superior response to pharmacotherapy. Recently, Klein (1990) suggested that psychometric criteria should be used to divide patients into more and less severely depressed groups. Although he did not state what methods should be used, cluster analysis may be the most appropriate of available procedures. Such an analysis was reported by McLean and Hakstian (1979) and there was no significant cluster-by-treatment interaction on any of the outcome measures. In the present study, we performed additional cluster analyses, but did not rely heavily on this method, due to its inherent problems (Morris, Blashfield & Satz, 1981). Instead, we relied mainly on a systematic and comprehensive set of severity criteria, as described above. Our use of multiple severity criteria precluded the use of the usual strategy of protecting experiment-wise Type I error by performing a MANOVA on all of the dependent variables. Instead, we directly calculated interaction effects for 80 ANOVAs (8 dependent variables x 5 severity criteria x 2 assessment periods). To provide some protection against excessive Type I error, we set the alpha level to 0.01 for each test. Significant interactions were followed by simple main effects at each severity level (alpha = 0.01). If the latter were significant, then Tukey comparisons were performed to compare treatments at a given severity level (alpha = 0.01). Given the number of statistical tests involved, this approach is clearly liberal with respect to Type I error. We adopted this approach in order to increase statistical power, and so enhance our chances of replicating the findings of Elkin et al. This approach has the advantage of making null results more interpretable since the analyses are biased to decrease Type II error at the expense of Type I error. Since we found that the pattern of interaction effects did not differ across severity criteria, we used the BDI 2 28 criterion for the remaining analyses. This cutoff was chosen because it is closest to that used by Elkin et al., and provided approximately equal numbers of patients within each severity level. Two sets of analyses were then conducted. First, analyses of the dropout proportions across treatment and severity. Second, following Elkin et al., we analysed the proportions of “recovered” patients (BDI ~9) in each treatment and severity condition. RESULTS Pretreatment
comparisons
The 151 treatment completers (including dropout replacements) were distributed across the treatments in approximately equal proportions: psychotherapy (n = 37) relaxation training (n = 35), behaviour therapy (n = 40), and pharmacotherapy (n = 39). The treatment groups did not differ in age, gender composition, marital status, employment status, education level, neuroticism, number of previous treatment episodes for depression, or duration of current depressive episode (all Ps > 0.08). The groups also did not differ on the pretreatment dependent variables, Pillai F (24, 426) = 1.07, P < 0.37. Treatment
main effects
The main effects of treatment have been described in detail in previous articles (McLean & Hakstian, 1979, 1990). At posttreatment, behaviour therapy was superior to the other treatments on the BDI, measures of mood, social functioning, and average satisfaction. Trends in the same direction were found at 3-month follow-up, with behaviour therapy being significantly superior on measures of mood and social functioning. Psychotherapy performed most poorly on the majority of outcome measures at both evaluation periods. Pharmacotherapy was not superior to behaviour therapy on any outcome measure, and there were no significant differences between pharmacotherapy and relaxation therapy on any outcome measure.
Severity
of unipolar
447
depression
Table 2. Interaction effects (F values) for severity groups defined according Criteria
BDI > 28
for defining
BDlg31
to various
the more severely depressed
BDI > 34
Mood 2 median
criteria
group Somatic ind. a median
Dependent variable
Post
F.U.
Post
F.U.
Post
F.U.
Post
F.U.
Post
F.U.
BDI Mood Social functioning Personal activity Somatic indicators Average satisfaction Cognitive Coping
0.54 0.87 0.18 0.32 2.82 2.78 2.76 I.19
1.60 0.67 0.68 2.82 5.67’ I .99 3.81 3.32
0.89 1.54 0.32 0.30 1.62 I .70 0.66 0.18
2.63 2.05 0.02 3.39 4.70* 2.23 5.101 5.22*
0.64 I .62 0.55 I .02 0.62 I .07 0.55 0.30
2.24 1.80 0.98 0.62 2.91 1.22 3.86 3.34
0.84 2.60 I .03 0.52 0.55 I .78 0.45 0.96
I .93 0.67 3.16 0.82 I .78 I .98 0.90 I .57
I .69 I .68 0.71 0.54 I.15 I .36 I .29 0.92
3.85 1.71 0.60 0.60 2.41 1.24 2.71 3.32
Post = interaclion effect, F(3,143), for posttreatment for 3-month follow-up dependent variables. lP < 0.01.
Treatment-by-severity
dependent
variables,
F.U. = interaction, F(3,136),
interactions
The number of Ss on each treatment and severity level for each severity criterion are present in Table 1. Here it can be seen that the cell sizes were adequate for statistical purposes. The cell sizes for the BDI 228 and median-split criteria were comparable to those in the Elkin et al. (1989) study. Table 2 shows the interaction effects for the univariate ANOVAs for each dependent variable and severity criterion. Only four interactions were significantly at alpha = 0.01, and only one of the corresponding simple main effects was significant, F(3,38) = 4.84, P < 0.006. The latter was for the somatic indicators at 3-month follow-up in the more severe condition, defined by BDI 2 3 1. Here, Tukey comparisons revealed that the psychotherapy group had significantly more somatic complaints than the pharmacotherapy group, (P < 0.01). There was also a trend (P < 0.05) for the relaxation training and behaviour therapy groups to have fewer somatic complaints than the psychotherapy group. Pharmacotherapy, behaviour therapy, and relaxation training did not differ from one another (P > 0.05). Following Elkin et al., the analyses were repeated with an “intend to treat” group, defined as the treatment completers (n = 115) plus dropouts (n = 36), and not including dropout replacements (n = 36). Thus, the sample consisted of patients who were initially admitted to treatment, and dropouts were assigned their pretreatment score as their posttreatment value. The results for this group can be considered as an estimate of the overall performance of the treatments, including their abilities to retain patients in treatment. The pattern of results was similar to that reported in Table 2, and failed to reveal an interaction effect favouring pharmacotherapy at any severity level. In summary, despite the use of a significance level that was liberal with respect to Type I error, there was no support for the hypothesis that pharmacotherapy is a superior treatment for more severely depressed patients. Indeed, pharmacotherapy tended to be statistically undistinguishable from the placebo, relaxation training (McLean & Hakstian, 1979). These conclusions did not change when results were analysed at 9-, 15-, 21- and 27 month follow-up. Treatment
dropouts
The analyses reported in the previous section show that the pattern of treatment-by-severity interactions did not change appreciably over severity criteria. The results of the analyses of dropouts were also consistent over severity criteria, and so we confine ourselves to the BDI 228 criterion, which was comparable to the criterion used by Elkin et al. Behaviour therapy was associated with the fewest overall dropouts (7.5%) and pharmacotherapy had the most (33.3%). The proportion of dropouts within each treatment and severity condition are presented in Table 3. The proportion of dropouts across severity levels and treatment groups was not homogeneous, x2(3, n = 36) = 75.00, P < 0.001. To identify the sources of heterogeneity, analyses of residuals were calculated, and the alpha level was set at 0.01. There were significantly more dropouts in the more depressed group relative to the less depressed group for patients receiving psychotherapy, d = 2.93, P < 0.003, and relaxation training, d = 5.97, P < 0.001. The proportion of dropouts did not differ in the behaviour therapy group, although there was a trend for more dropouts in the less severe group, d = 2.04, P < 0.02. This result must be interpreted with
PETER MCLEAN and STEVEN TAYLOR
448 Table
3. Dropouts
within each treatment severity level
condition
and
Table 4. Patients “recovered” at termination BDI < 9)
(posttreatment
Severity*
seventy*
MOK
Less
Treatment condition
Treatment condition
Psychotherapy Relaxation training Behaviour therapy Pharmacotherapy ‘More
severe groups:
15.0 8.3 9.5 52.6
3 2 2 IO
BDI (pretreatment)
47. I 63.6 53 15.0 >28.
8 7 I 3
Psychotherapy Relaxation traming Behaviour therapy Pharmacotherapy *More severe groups:
%
,t
%
n
35.0 37.5 51. I 52.6
I 9 I2 IO
29.4 34.4 51.9 25.0
5 4 II 5
BDI (pretreatment)
3 28.
caution due to the small number of dropouts in the group receiving behaviour therapy (Table 3). Interestingly, the pharmacotherapy group had significantly more dropouts in the less depressed group, d = 7.76, P < 0.001. When the severity groups were defined according to other criteria, the dropout rates still did not favour pharmacotherapy at any severity level, and behaviour therapy had the consistently lowest rate of attrition. Proportion
of “recovered”
patients
As suggested by Elkin et al. (1989), patients were defined as “recovered” if they had a posttreatment BDI ~9. The number and proportion of patients within each treatment group and severity condition are presented in Table 4. Here, the more severely depressed group was defined by BDI 828. The table shows that there was generally little difference across severity levels in the proportions of recovered patients. However, there was a trend toward an heterogeneity of cell proportions, x2(3, n = 151) = 7.03, P < 0.075. Analyses of residuals revealed that for the pharmacotherapy group there were fewer recovered patients in the group that was initially more depressed, d = 2.61, P < 0.005. For the other treatments there was no significant difference between the proportions of recovered patients across severity levels. Further analyses, using the other severity criteria, failed to support the hypothesis that pharmacotherapy is the most efficacious treatment, or that pharmacotherapy exerts greater treatment effects on more, rather than less, severely depressed patients. Cluster analyses In the original report, McLean and Hakstian (1979) performed a cluster analysis on several of the pretreatment characteristics of the patients, obtaining a four-cluster solution. The clusters differed on a variety of dimensions, including that of depression severity. There was no significant interaction between cluster and treatment. In the final set of analyses to be reported in the present article, we take the question of differential treatment effects one step further by using a method known as dynamic clustering (Prochaska, Velicer, Guadagnoli, Rossi & DiClemente, 1991). This is an application of cluster analysis where Ss are grouped on the basis of the similarity of their scores on a given variable over time. Thus, instead of clustering Ss on the basis of their scores on a number of variables at a given time, Ss are clustered on the basis of their scores on a single variable measured at several different times. We performed dynamic clustering on the patients’ BDI scores taken at pretreatment, posttreatment, and at five follow-up points (3,9, 15,21, and 27 months). This analysis enabled us to identify patterns of treatment response, and to determine whether a particular form of treatment was more likely to be associated with a given treatment response. For example, on the basis of the Elkin et al. study, a three-cluster solution would be predicted. One cluster would be characterized by patients who initially had moderate levels of depression, and who made good treatment gains regardless of the type of treatment. Thus, this cluster would contain approximately equal proportions of patients from each treatment group. The second cluster would consist of patients who were severely depressed at intake, and who did not respond to treatment. Elkin et al.‘s findings lead to the prediction that this cluster would be composed mainly of patients who received psychotherapy, relaxation training, or behaviour therapy, and relatively few patients who were treated with amitriptyline. The third cluster should consist of severely depressed patients who made good treatment gains. This cluster is predicted to consist mainly of patients receiving amitriptyline. If
449
Severity of unipolar depression
30 25 20
9
15
21
27
Follow-up (months)
Fig. I. BDI scores at pretreatment and follow-up for each cluster of depressed patients.
there is no treatment by severity interaction, then a two-cluster solution might be expected, with one cluster consisting of treatment responders and another consisting of nonresponders, with patients from each of the treatment groups distributed approximately equally over clusters. Four-cluster analyses were performed in order to rule out any artifactual results arising from the type of clustering algorithm or choice of assessment points (cf. Morris et al., 1981). Two clusterings were performed using Ward’s method, and two using the average linkage method. For each method, two clusterings were performed using either pre- and posttreatment BDI scores as the cluster variables, or BDI scores obtained at all 7 time points. All solutions were similar, indicating a two-cluster solution in each case. Therefore, only the results for one analysis will be reported. The analysis using all 7 time points was selected since it provides a broader picture of the long-term outcome for each cluster, although there was some sacrifice in sample size due to sample attrition (complete data across the 7 time points was available for 121 patients). Figure 1 shows the BDI scores for each cluster. Here it can be seen that Cluster A can be described as the treatment nonresponders, who were slightly more depressed at intake than Cluster B, the treatment responders. The number of Ss in each treatment condition and cluster type are presented in Table 5. In Cluster A the proportion of patients from each type of treatment were as follows: psychotherapy (1 l%), relaxation training (20%), behaviour therapy (18%), and amitriptyline (20%). Thus, with the exception of psychotherapy, there were approximately equal proportions in each cluster. The results of Table 5 fail to support the view that severely depressed patients are selectively responsive to pharmacotherapy. It appears that the nonresponders (Cluster A) failed to respond to any of the treatments. Considering both the cluster results and the previous analyses of the treatment main effects (McLean & Hakstian, 1979, 1990) it appears that although the Cluster B patients responded to all of the treatments, the behaviour therapy group tended to respond best. Cluster A patients tended to respond poorly to any treatment in the study. The nature of these clusters and their distinctions is beyond the scope of the present article, and will be the subject of a subsequent paper.
Table 5. Number of patients in each treatment condition and cluster Cluster Treatment condition
A
B
Psychotherapy Relaxation training Behaviour therapy Pharmacotheraov
3 7 6 5
25 28 27 20
450
PETERMCLEAN and STEVENTAYLOR
DISCUSSION
The present study failed to replicate findings of Elkin et al. (1989) despite comparable sample sizes and the use of a liberal alpha level. Our use of multiple severity criteria suggest that the results were not simply an artifact of an arbitrary criterion for defining patients as more or less severely depressed. There are a number of possible explanations as to why our results failed to replicate those of Elkin et al. First, it may be argued that the difference in results was due to a difference in antidepressants. However, reviews of the comparative efficacy of tricyclic antidepressants have found amitriptyline to be as effective as imipramine when equivalent doses are administered (Davis & Glassman, 1989; Morris & Beck, 1974). Moreover, Davis and Glassman (1989) claimed that 240 mg of imipramine has an equivalent antidepressant effect to 150 mg of amitriptyline, suggesting that the latter is a more potent antidepressant. These findings suggest that the doses of amitriptyline administered in the present study (150 mg/day) were probably as effective as the doses of imipramine administered in the study by Elkin et al. (M = 185 mg/day). However, the therapies differed in duration, with the Elkin et al. patients receiving an average of six more treatment sessions than patients in the present study. In our study 43% of pharmacotherapy patients had posttreatment BDI scores in the recovered range, whereas 69% recovered in the Elkin et al. study. If the drug treatment in the present study was less effective than that in Elkin et al’s study, then it is possible that the treatment-by-severity interaction effects were attenuated. Even if this were the case, it is notable that we did not find a trend in the direction of favouring pharmacotherapy for more severely depressed patients. The two drugs differ in other ways. Amitriptyline is considered to have greater sedative and anticholinergic effects (Hyman & Arana, 1987). The greater anticholinergic effects of amitriptyline made no difference in the level of attrition in the present study relative to that in the Elkin et al. study (33% in both cases). The greater sedative effects of amitriptyline might make this drug more efficacious than imipramine in agitated depression. However, there is no reason why this difference in sedative effects should account for our failure to replicate Elkin et al’s findings. Thus, we conclude that our failure to replicate their findings is unlikely to have been due to differences in drugs. Another possible explanation for our failure to replicate Elkin et al’s interaction concerns the differences in the types of cognitive-behavioural treatments that were administered. Compared to the cognitive therapy used in the Elkin et al. study, the behavioural treatment in the present study was considerably more focused on increasing patients’ level of activity, and less concerned with cognitive restructuring. As mentioned earlier, the therapies also differed in duration, with the Elkin et al. patients receiving an average of six more treatment sessions. Our behavioural treatment, which included cognitive techniques, appears to have been approximately equal in efficacy to Elkin et al’s cognitive therapy, as suggested by the proportions of recovered patients (respectively, 57 vs 65% on the posttreatment BDI). Thus, differences in the nature of the two therapies are unlikely to account for the failure to replicate their findings. It could be argued that we treated a different sample to those in the Elkin et al. study, since we diagnosed unipolar depression according to Feighner et al. (1972) criteria, whereas Elkin et a/. used Research Diagnostic Criteria (Spitzer, Endicott & Robins, 1978). It seems unlikely that these differences would account for our failure to find a treatment-by-severity interaction, since the two sets of diagnostic criteria have more similarities than differences. Another possible explanation is that we did not replicate Elkin et al’s interaction because the latter was a spurious finding based on an unduly liberal alpha level. Although Elkin et al. claimed that their analyses were conservative, it is clear from their paper that the alpha level for their analyses of interaction effects was greater than 0.10. Regardless of which explanation best accounts for the lack of replicability, our results demonstrate clearly that severe depression is neither necessarily a contra-indication for cognitive-behavioural therapy, or a specific indication for pharmacotherapy. Our findings showed that the therapies did not vary in their comparative efficacy across depression severity. This conclusion holds for depressed outpatients. However, it remains to be seen whether this also applies to inpatients. The results of this study indicate that the preference for pharmacotherapy over cognitive-behavioural therapies lacks convincing support.
Severity
of unipolar
depression
451
REFERENCES Beck, A. T. (1967). Depression. New York: Harper & Row. Beck, A. T., Rush, A. J., Shaw, B. F. & Emery, G. (1979). Cognifive therupy of depression. New York: Guilford. Blackburn, I. M., Bishop, S., Glen, A. I. M., Whalley, L. J. & Christie, J. E. (1981) The efficacy of cognitive therapy in depression: A treatment trial using cognitive therapy and pharmacotherapy, each alone and in combination. British Journal of Psychiatry, 139, 181L189. Davis, J. M. & Classman, A. H. (1989). Antidepressant drugs. In Kaplan, H. I. & Sadock, B. J. (Eds), Comprehensive fextbook of psychiatry (5th Edn, pp. 1627-1655). Baltimore: Williams & Wilkins. Derogatis, L. R., Lipman, R. S. & Covi, L. (1973). SCL-90: An outpatient psychiatric rating scale-preliminary report. Psychopharmacology Bulletin, 9, 13-28. Elkin, I., Shea, T., Watkins, J. T., Imber, S. D., Sotsky, S. M., Collins, J. F., Glass, D. R., Pilkonis, P. A., Leber, W. R., Docherty, J. P., Fiester, S. J. & Parloff, M. B. (1989). National Institute of Mental Health Treatment of Depression Collaborative Research Program: General effectiveness of treatments. Archives of General Psychiatry, 46, 971-982. Endicott, J., Spitzer, R. L., Fleiss, J. L. &Cohen, J. (1976). The Global Assessment Scale: A procedure for measuring overall severity of psychiatric disturbance. Archives of General Psychiatry, 33, 766-771. Eysenck, H. J. & Eysenck, S. B. G. (197551976). Eysenck personality questionnaire. San Diego, CA: Educational and Industrial Testing Service. Fieghner, J. P., Robins, E., Guze, S. B., Woodruff, R. A., Winokur, R. A. & Munoz, R. (1972). Diagnostic criteria for use in psychiatric research. Archives of General Psychiatry, 26, 5763. Freedman, D. X. (1989). Editorial note (especially for the media). Archives of General Psychiatry, 46, 983. Gelder, M. G. (1990). Psychological treatment for depressive disorder. British Medical Journal, 300, 1087-1088. Hamilton, M. A. (1967). Development of a rating scale for primary depressive illness. British Journal of Social and Clinical Psychology, 6, 278-296. Hyman, S. E. & Arana, G. W. (1987). Handbook of psychiatric drug therapy. Boston: Little, Brown & Co. Klein, D. F. (1990). NIMH collaborative research on treatment of depression (letter). Archives of General Psychiatry, 47, 682-684. Lubin, B. (1965). Adjective checklists for measurements of depression. Archives of General Psychiatry, 12, 5762. Marmor, J. (1973). Psychiatry in transition. New York: Brunner/Mazel. Marmor, J. (1975). The nature of the psychotherapeutic process revisited. Canadian Psychiatric Association Journal, 20, 557-565. McLean, P. D. (1976). Decision-making in the behavioural management of depression. In Davidson, P. 0. (Ed.), Behavioural management of anxiety, depression and pain (pp. 5489). New York: Brunner/Mazel. McLean, P. D. & Hakstian, A. R. (1979). Clinical depression: Comparative efficacy of outpatient treatments, Journal of Consulting and Clinical Psychology, 47, 818-836. _ McLean, P. D. & Hakstian, A. R. (1990). Relative endurance of unipolar depression treatment effects: Longitudinal follow-up. Journal qf Consulting and Clinical Psychology, 58, 482488. Morris, J. B. & Beck, A. T. (1974). The efficacy of antidepressant drugs. Archives of General Psychiatry, 34 667474. Morris, R., Blashfield, R. & Satz, P. (1981). Neuropsychology and cluster analysis: Potentials and problems. Journal of Clinical Neuropsychology, 3, 79-99. Paykel, E. S. (1989). Treatment of depression: The relevance of research for clinical practice. British Journal of Psychiatry, 155, 754-763. Prochaska, J. O., Velicer, W. F., Guadagnoli, E., Rossi, J. S. & DiClemente, C. C. (1991). Patterns of change: Dynamic typology applied to smoking cessation. Mulfiuuriute Behavioral Research, 26, 83-107. Robinson, L. A., Berman, J. S. & Neimeyer, R. A. (1990). Psychotherapy for the treatment of depression: A comprehensive review of controlled outcome research. Psychological Bulletin, IO8, 30-49. Spitzer, R. L., Endicott, J. & Robins, E. (1978). Research diagnostic criteria: Rationale and reliabilitv. Archives of General Psychiatry, 35, 773-782. Thase, M. E., Simons, A. D., Cahalance, J., McGreary, J. & Harden, T. (1991). Severity of depression and response to cognitive behavior therapy. American Journal of Psychiatry, 148, 784-789. Wolberg, L. R. (1967). Short-term psychotherapy. New York: Grune & Stratton.
