- P Hazell, senior lecturer in psychiatrya,
- D O'Connell, senior lecturer in biostatisticsb,
- D Heathcote, research psychologista,
- J Robertson, associate lecturer in pharmacoepidemiologyb,
- D Henry, senior lecturer in clinical pharmacologyb
- Discipline of Psychiatry, Faculty of Medicine and Health Sciences, University of Newcastle, Callaghan, NSW 2308, Australia
- Centre for Clinical Epidemiology and Biostatistics, Faculty of Medicine and Health Sciences, University of Newcastle
- Correspondence to: Dr Hazell.
- Accepted 17 February 1995
Objective: To examine whether tricyclic antidepressants are superior to placebo in the treatment of child and adolescent depression.
Design: Meta-analysis of 12 randomised controlled trials comparing the efficacy of tricyclic antidepressants with placebo in depressed subjects aged 6-18 years.
Main outcome measures: Most studies employed several depression rating scales. For each study the “best available” measure was chosen by using objective criteria, and individual and pooled effect sizes were calculated as the number of standard deviations by which the change scores for the treatment groups exceeded those for the control groups. Where authors had reported numbers “responding” to treatment we calculated individual and pooled ratios for the odds of improvement in treated compared with control subjects.
Results: From the six studies presenting data which enabled an estimation of effect size the pooled effect size was 0.35 standard deviations (95% confidence interval of -0.16 to 0.86) indicating no significant benefit of treatment. From the five studies presenting data on the number of “responders” in each group, the ratio of the odds of a response in the treated compared with the control subjects was calculated and the pooled odds ratio was 1.08 (95% confidence interval of 0.53 to 2.17); again indicating no significant benefit of treatment. The pooled sample had more than an 80% chance of detecting a treatment effect of 0.5 standard deviations or greater. There was an inverse relation between study quality and estimated treatment effect.
Conclusions: Tricyclic antidepressants appear to be no more effective than placebo in the treatment of depression in children and adolescents.
Previous studies, and narrative reviews of the topic, have shown that tricyclic antidepressants are of equivocal benefit in juvenile depression * This meta-analysis of 12 randomised double blind placebo controlled trials found an overall small but clinically non-significant treatment effect
Tricyclic drugs are not recommended as a first line treatment for depression in children and adolescents
Depression is a common but underrecognised problem in young people. Its estimated prevalence is 1.9% in primary school children, rising to 4.7% in adolescents.1 Depression may be present in more than half of child and adolescent psychiatric inpatients.1 Important consequences of depression in this age group include social dysfunction, academic underachievement, and suicidal behaviour. Consequently, adequate detection and treatment of depressed adolescents is an important strategy for curbing the rising rate of suicide in youth seen in many developed countries.2
The available evidence concerning the efficacy of tricyclic antidepressants in child and adolescent depression is equivocal. Individually, treatment trials have been small and variable in quality. Studies have generally shown no effect, or only small and statistically non-significant effects. Concern about the problem is evidenced by there being more reviews on the subject than there are original studies amenable to systematic analysis.3 4 5 6 7 8 9 10 11 12 The apparent lack of efficacy shown by studies of tricyclic drugs in child and adolescent depression contrasts with a widespread clinical conviction that they are useful in at least some patients.8 13 The controversy over the role of these drugs is made more important because of the risks associated with their use by young people. They are potentially lethal in overdose and may be cardiotoxic even in the therapeutic dose range.14 15 16 They may cause “switching” from depression to mania,17 and they may be implicated in the induction of rapid cycling bipolar disorder.18
The equivocal evidence for the efficacy of tricyclic drugs also has implications for the definition of the syndrome and the pharmacology of depression in children and adolescents. That children and adolescents do not respond positively to tricyclic drugs suggests that the biological substrate to depression in this age group may differ from that in adults.
We considered it timely to submit the existing research on the use of tricyclic drugs in juvenile depression to a meta-analysis, since a recognised indication for meta-analysis is a series of studies with conflicting or inconclusive results.19 The aim of this meta-analysis was to pool the results of randomised controlled trials in order to determine whether tricyclic drugs are superior to placebo in the treatment of child and/or adolescent depression.
We searched the literature by using the CDROM databases Silver Platter Medline (1966-92) and Excerpta Medica (June 1974-92). Terms used for the Medline search were the exploded terms child and depression; the MeSH (medical subject headings) terms antidepressant drugs, tricyclic, and affective disorders; individual tricyclic drugs by name; names of well known researchers in the field; and school phobia. A similar search strategy was used for Excerpta Medica, with the exception that individual authors were not entered. Abstracts in English (of English and non-English papers) were reviewed. Bibliographies of previously published reviews and papers describing original research were cross-checked. Current Contents was screened for recent publications. We contacted authors of abstracts describing “work in progress” identified in conference proceedings of the American Academy of Child and Adolescent Psychiatry to determine whether they held data that could be included in the meta-analysis.
CRITERIA FOR INCLUSION
Studies were included in the meta-analysis if they described subjects between 6 and 18 years who were identified as suffering from a depressive illness and if they randomised subjects to a tricyclic antidepressant (and no other pharmacological intervention) or placebo. Studies of mixed adolescent and adult subjects were not included because it was not possible to separate out the data on the adolescents. Studies were also excluded if subjects had IQs less than 80.
Each of the studies included in the meta-analysis was assessed for quality by using a modified version of the scheme suggested by Chalmers et al.20 For this purpose “quality” is defined in terms of the measures taken by the investigators to minimise bias in the study. The studies were firstly scored independently by two of the authors, with discrepancies resolved by consensus. Studies were rated in the range 0-3 on each of the following features:
degree to which randomisation was truly blind;
inclusion of data from subjects who subsequently withdrew from the study (intention to treat);
degree to which assessors of outcome were blind to the treatment allocation;
whether subjects were assessed to determine if they had accurately guessed their treatment status;
statement of criteria for improvement;
use of multiple informants for the assessment of outcome;
method of determining dose of tricyclic;
assessment of compliance;
whether concurrent treatment was held constant;
length of baseline observation;
control for previous treatment;
control for comorbidity.
Scores for the individual quality items were summed for the purposes of analysis. The total possible score was 36 points.
Most studies used multiple outcome measures. For the purposes of pooling results, a single “best available” outcome measure was chosen for each study. The order of selection was determined by the rating of each instrument over the following five criteria: appropriateness to children and adolescents; reliability; construct validity; agreement with clinical interview; track record in psychopharmacological research. Most of the data for this rating were obtained from a review by Petti.21
On the basis of number of criteria met, instruments were ranked for selection for analysis. When possible we used the schedule for affective disorders and schizophrenia for school-age children, combined child and parent report (five criteria met), and then the children's depression rating scale (four criteria met), the Bellevue index of depression, children's depression inventory, or Hamilton depression rating scale (three criteria met), and then the depressive adjective checklist (two criteria met).
For studies reporting either baseline and follow up scores from one of the six listed instruments, or a change in scores between baseline and follow up, the effect size was calculated as the number of standard deviations by which the change in score for the actively treated group exceeded that of the placebo group. The standard deviation for change in scores was calculated by using the correlations reported in the paper for that study22 or a value of 0.9, as estimated from the placebo group in Kramer and Feiguine23 and from each group in Petti and Law.24 A negative change indicated improvement (a lower score at follow up than at baseline). A positive difference between treatment and placebo (a positive effect size) indicated that the effect was greater (that is, there was more improvement) in the actively treated group. For each effect size 95% confidence intervals were calculated by using the method described in Hedges and Olkin.25 Pooling of effect sizes was based on both the fixed effects and random effects models,26 and a test of homogeneity was performed.
When a study reported the numbers or proportions who had “improved” in each of the treatment groups, we estimated the ratio of the odds of improvement in the actively treated group compared with that in the placebo group. An odds ratio greater than one indicates that a larger proportion improved in the actively treated group than in the placebo group. Because of small cell sizes, exact 95% confidence intervals were calculated for the study specific odds ratios by using StatXact.27 Pooling was based on the fixed effects model by using exact methods as implemented by Mehta and Patel,27 and a test of homogeneity was performed.
Twelve studies22 23 24 28 29 30 31 32 33 34 35 36 fulfilled the criteria for inclusion in the meta-analysis. Table I summarises descriptive information for each study, and quality scores. Six of these studies presented data as change in scores or baseline and follow up scores, using at least one of the instruments listed in the methods section (table II). All studies but one32 suggested a larger improvement in the actively treated than in the control group, but the treatment-placebo difference was significantly different from zero in only one study.23 There was significant heterogeneity across the studies, so the pooled effect size and 95% confidence interval from the random effects model is reported. Overall there was no significant difference in levels of improvement in the actively treated group over that in the placebo group (pooled effect size=0.35, 95% confidence interval -0.16 to 0.86). In other words, the improvement in scores in the actively treated group was greater than that in the placebo group by 0.35 standard deviations, but the 95% confidence interval includes the null value of zero.
Five studies presented results expressed as numbers improved in the treatment and placebo groups (table III, with footnote giving the definition of improvement used in each of the studies). Two papers by Geller et al32 33 reported data on the same subjects and so were included only once in the analysis. For all studies analysed, the 95% confidence interval for the odds ratios were very wide and included one, indicating no significant differences in the rates of improvement in the two groups. The pooled odds ratio was 1.08 (0.53 to 2.17), indicating no significant improvement in the treated group over the placebo group.
Usable data could not be extracted from papers by Berney et al28 and Preskorn et al.30 The paper by Kashani et al29 was not included in the pooling as this was a report of a crossover trial so that the treatmentplacebo difference was a paired comparison rather than the comparison of two independent groups. In two of these studies28 30 there was a non-significant trend for a more favourable response to treatment than to placebo. The study of Kashani et al29 found a 43% improvement on scores on the child depression rating scale with treatment, compared with a 35% improvement with placebo. This difference was statistically significant (P<0.05).
Mean scores for individual quality items across the nine studies included in the analysis were as follows (maximum=3, minimum=0): randomisation, 1.3; handling of withdrawals, 2.0; blinding of assessors, 1.9; blinding of subjects, 1.9; improvement criteria, 2.3; use of multiple raters, 2.5; dose determination, 2.1; compliance, 2.2; control for concurrent treatment, 1.6; baseline observations, 2.0; control for previous treatment, 1.2; control for comorbidity, 1.9. Scores ranged from 0 to 3 for randomisation and 1 to 3 for the other variables.
Study quality and effect size were negatively correlated (r=-0.83, df=4, P<0.05) indicating a possible bias away from the null in poorer quality studies. Study quality was not related to the size of the odds ratio (r=0.08, df=3, P>0.10).
INTERPRETATION OF FINDINGS
Nine of 12 studies examining the efficacy of tricyclic antidepressants in the treatment of child and adolescent depression were amenable to analysis with a meta-analytic approach. The 95% confidence interval indicated that any improvement in depression rating scores in the treated group compared with controls is likely to be less than 0.86 standard deviations. The pooled sample size had a better than 80% chance of detecting a significant treatment effect of 0.5 standard deviations or greater at (alpha)=0.05 (two sided).
Effect size is sensitive to change, but is also the most difficult measurement to interpret. The literature suggests that a magnitude of change of two standard deviations is necessary for treatment effects to be considered clinically significant.37 This is consistent with the criterion for improvement stated in several of the studies included in this meta-analysis. For example, the studies of Boulos et al36 and Hughes et al35 sought a 50% reduction in depression rating scores from baseline to follow up, representing a difference of approximately two standard deviations. This criterion for magnitude of difference is often adopted in studies of adult depression—for example, Berne.38
We acknowledge that there is lack of consensus over what constitutes a clinically meaningful effect size in the treatment of depression.39 Our results suggest an average treatment gain over placebo of 0.35 standard deviations, with the probability that the true effect is less than 0.86 standard deviations, falling well short of the criterion of two standard deviations. There was a high response rate to placebo across the studies, in several cases exceeding the two standard deviation threshold. We consider that the small additional effect afforded by treatment in comparison with placebo is unlikely to be clinically important in most patients. This conclusion is reinforced by the analysis of numbers of subjects in the treated and control groups who were considered to have improved. The results suggest that the true odds ratio for improvement in treated versus control subjects is unlikely to be greater than 2. In addition, the inclusion of one in the confidence interval indicates the probability of no effect. Because of the high rate of improvement in the placebo group (37%), a trial would require 133 subjects per group to detect an odds ratio for improvement of 2.0 with a power of 80%. The high response rate to placebo and the relatively small numbers in this pooled sample mean that the data do not exclude a significant treatment response, but the data are not encouraging.
Any review of the literature is vulnerable to the effect of publication bias, but such bias usually exaggerates the positive effect of treatment because small “negative” studies are not published. This is not an issue in our study as the meta-analysis does not show a positive treatment effect. We excluded three published studies from the meta-analysis. One study showed a positive treatment effect; the other two did not. Their inclusion is unlikely to have significantly altered our findings.
The placebo response rate of upward of 50% in some studies is also worth further comment. The rate was lower in the recent studies conducted by Geller and her colleagues,31 32 34 presumably because subjects who responded in the placebo washout phase were excluded from the main study. The strong placebo response in most studies should inform clinicians about the treatment of juvenile depression. Children and adolescents can be expected to respond to strategies such as hospitalisation, removal from stressors, the development of a treatment alliance, and treatment planning, even in the absence of other “specific” therapies.
QUALITY OF STUDIES
The quality of the studies deserves some comment. Previous reviews7 of randomised placebo controlled trials of tricyclic drugs in the treatment of child and adolescent depression have highlighted several problems that may contribute to negative findings. Our meta-analysis partly overcomes one of these problems, small sample size, since the pooled sample had sufficient power to detect as significant at least moderate treatment effects. Our quality assessment focused on attempts by the investigators to minimise bias. Randomisation was done poorly in many studies, as was blinding. Weaknesses in these areas have been shown to contribute to a systematic bias away from the null.40 We found an inverse relation between study quality and effect size, and therefore we consider it unlikely that poor study quality accounts for the negative findings in these studies.
There are several plausible neuropharmacological explanations why tricyclic drugs are not efficacious.10 The neurotransmitter systems involved in the control of effect are incompletely mature in children. The noradrenergic system does not develop fully until early adulthood, while the more rapid hepatic metabolism of tricyclic compounds in children shifts the ratio of noradrenergic to serotonergic activity in the direction of noradrenergic activity. In addition, adolescents have high ketosteroid levels, which also affect noradrenergic transmitter systems. In theory at least, selective serotonergic compounds may be expected to have greater efficacy than noradrenergic compounds in juveniles. The hormonal milieu of the adolescent brain may also influence neurotransmitter activity, but the mechanism is unknown. Jensen et al have raised the possibility that childhood onset depressive illness is aetiologically distinct from adult onset depressive illness, and that adult depressives with a childhood onset of their disorder may also be relatively non-responsive to tricyclic drugs.7
Further replication studies using “traditional” tricyclic drugs with mixed noradrenergic and serotonergic activity are probably not warranted. Pharmacological research should probably be directed to new generation selective serotonergic agents, which may have greater efficacy. In view of the need for a large sample size a multicentre trial approach is needed. Treatment research should also examine other widely adopted strategies, such as family therapy, supportive psychotherapy, and specific psychotherapies.