Meta-analysis of effects and side effects of low dosage tricyclic antidepressants in depression: systematic reviewBMJ 2002; 325 doi: https://doi.org/10.1136/bmj.325.7371.991 (Published 02 November 2002) Cite this as: BMJ 2002;325:991
- Toshi A Furukawa, professor ()a,
- Hugh McGuire, trials search coordinatorb,
- Corrado Barbui, psychiatristc
- a Department of Psychiatry, Nagoya City University Medical School, Nagoya 467-8601, Japan
- b Cochrane Collaboration Depression, Anxiety, and Neurosis, Health Services Research, King's College Institute of Psychiatry, London SE5 8AF
- c Department of Medicine and Public Health, Section of Psychiatry, University of Verona, Ospedale Policlinico 37134 Verona, Italy
- Correspondence to: T A Furukawa
- Accepted 24 June 2002
Objective: To compare the effects and side effects of low dosage tricyclic antidepressants with placebo and with standard dosage tricyclics in acute phase treatment of depression.
Design: Systematic review of randomised trials comparing low dosage tricyclics (100 mg/day) with placebo or with standard dosage tricyclics in adults with depression.
Main outcome measures: Relative risk of response in depression (random effects model), according to the original authors' definition but usually defined as 50% or greater reduction in severity of depression. Relative risks of overall dropouts and dropouts due to side effects.
Results: 35 studies (2013 participants) compared low dosage tricyclics with placebo, and six studies (551 participants) compared low dosage tricyclics with standard dosage tricyclics. Low dosage tricyclics, mostly between 75 and 100 mg/day, were 1.65 (95% confidence interval 1.36 to 2.0) and 1.47 (1.12 to 1.94) times more likely than placebo to bring about response at 4 weeks and 6-8 weeks, respectively. Standard dosage tricyclics failed, however, to bring about more response but produced more dropouts due to side effects than low dosage tricyclics.
Conclusions: Treatment of depression in adults with low dose tricyclics is justified. However, more rigorous studies are needed to definitively establish the relative benefits and harms of various dosages.
What is already known on this topic
What is already known on this topic Tricyclics are still prescribed as often as selective serotonin reuptake inhibitors and other newer antidepressants worldwide
Experts have often claimed that clinicians prescribe tricyclics at less than adequate dosages
What this study adds
What this study adds Tricyclics at dosages below the recommended range are more effective than placebo
They may or may not be as effective as standard dosage tricyclics but result in fewer dropouts due to side effects
The minimum effective dosage and ranges for antidepressants has not been established—a simple set of numbers that every practising doctor and patient would want to know
Despite the growing popularity of selective serotonin reuptake inhibitors and other newer antidepressants, tricyclic andtidepressants are still extensively prescribed worldwide. In the United Kingdom between 1991 and 1996, there was a 460% increase in prescriptions for selective serotonin reuptake inhibitors, but there was also a 40% increase in prescriptions for tricyclics for patients starting treatment, with these new patients still outnumbering those taking selective serotonin reuptake inhibitors by 56%.1 In the United States between 1990 and 1995 antidepressant use increased by 73% mainly because of patients being prescribed selective serotonin reuptake inhibitors, but even today tricyclics are prescribed as often as selective serotonin reuptake inhibitors. 2 3 Other countries show similar trends.4
Evidence for the recommended dosage of tricyclics is poor. 5 6 Many of the existing guidelines recommend dosages greater than 100 mg/day or 125 mg/day, but there is a lack of convincing evidence that lower dosages are not effective. 7 8 This uncertainty casts doubt on the widely held view that depression is undertreated both in primary care and in psychiatric settings. 9 10 It also questions whether selective serotonin reuptake inhibitors should be preferred over tricyclics when controlled trials failed to find differences in effectiveness between the two, because it is easier to achieve “adequate” dosage with selective serotonin reuptake inhibitors.11
We included randomised trials comparing low dosage tricyclics with placebo or with standard dosages of the same tricyclic in the acute phase treatment of adults with depression. Low dosage was defined as 100 mg/day or less of imipramine, amitriptyline, clomipramine, desipramine, doxepin, dothiepin, trimipramine, or lofepramine. We excluded nortriptyline because the standard dosage is debatable. Standard dosage was defined as greater than 100 mg/day. Our trial was to last at least four weeks.
Our primary outcome was the effect of treatment on depression, according to the original authors' definition but usually defined as 50% or greater reduction in severity of depression. The severity of symptoms was measured by either observer rating (preferred) or self report.
Identification of trials
We electronically searched the Cochrane Collaboration depression, anxiety, and neurosis controlled trials register up to November 2000 for any trials in which tricyclics were given. This database incorporates results of group searches of Medline (1966 onwards), Embase (1980 onwards), CINAHL (1982 onwards), PsycINFO (1974 onwards), PSYNDEX (1977 onwards), and LILACS (1982-99). We also hand searched the major psychiatric and medical journals. Two reviewers (HM and TAF) then manually examined the potential papers to see if they were randomised trials comparing low dosage tricyclics with placebo or with standard dosage for any form of depression. All potential identified papers were then checked according to the strict eligibility criteria by two independent reviewers (TAF and CB).
To identify further reports TAF checked the references of this preliminary list of selected studies along with references of other relevant review papers. To identify more recent reports HM subjected nine of the most representative studies to SciSearch. TAF contacted authors of major papers and other experts in the specialty.
Quality assessment and data extraction
TAF and CB assessed the methodological quality of the selected studies. The criteria for quality assessment were based on the recommendations of the Cochrane Collaboration Handbook and focused on concealment of allocation and double blinding.12 HM and TAF independently extracted data from the original reports using data extraction forms. Disagreements between the two reviewers were resolved by consensus.
Data were entered twice into Review Manager (version 4.1) using the duplicate data entry facility. For dichotomous outcomes, we calculated relative risks and their 95% confidence intervals with a random effects model because they may be more generalisable and more easily interpreted than those obtained with fixed effects models, odds ratios, or risk differences.13 14 We assessed heterogeneity between studies with the Q statistic and by visual inspection of the results. For continuous outcomes, we calculated standardised weighted mean differences with a random effects model.
We first performed per protocol analysis according to the values reported by the original authors. When data on dropouts were included by the last observation carried forward method, we analysed them according to the primary studies. We also performed a worst case scenario intention to treat analysis whereby dropouts were considered non-responders in the active treatment group but as responders in the placebo group. This extreme scenario was to guard against favouring active drugs that could be the more harmful.
We performed a funnel plot analysis to check for publication bias. To examine the robustness of the findings we performed two sensitivity analyses, by limiting the included studies to those using operational diagnostic criteria for major depression and to those in which the dosage was 75 mg/day or less.
Subgroup analyses should be performed and interpreted with caution because multiple analyses lead to false positive conclusions. We did, however, perform two subgroup analyses, where possible: for older people (age 65 or more) separately, because these people may be more vulnerable to side effects associated with tricylics and a decreased dosage is often recommended for them; and for psychiatric patients and primary care patients separately, because it is sometimes believed that results obtained from either of these settings may not be straightforwardly applicable to the other setting.
Study inclusion and characteristics
Of the 2418 citations originally identified in our electronic search, 141 were potentially relevant and were assessed for strict eligibility and quality. The inter-rater reliability of the two reviewers for this first stage of study selection was good (agreement 97%, κ=0.61). After the reference search, SciSearch, and personal contacts, we ultimately agreed on 35 studies (2013 participants) that compared low dosage tricyclics with placebo, and six studies (551 participants) that compared low dosage tricyclics with standard dosage tricyclics. Two of these had three arms of standard dosage tricylics, low dosage tricylics, and placebo (fig 1). The inter-rater reliability for this second stage of assessment for eligibility and validity was excellent, with weighted kappas between 0.58 and 0.86. The inter-rater reliability of the two validity criteria was also satisfactory, with weighted kappas of 0.58 and 0.79, respectively.
Sixteen studies used amitriptyline as active drugs and 13 used imipramine. The remaining randomised controlled trials studied clomipramine (3 trials), doxepine (3), dothiepin (2), trimipramine (2), and lofepramine (1). Five studies focused on depression in people aged 65 or more. Ten studies were conducted in primary care and 12 studies in psychiatric settings. Six studies dealt with depression seen in patients with comorbid physical conditions such as migraine or rheumatoid arthritis. All of the included studies were randomised controlled trials with both patients and doctors blinded (table; the complete list of included studies is also available in the Cochrane Library). However, only four studies reported enough details on their randomisation procedure.
Low dosage tricyclics versus placebo
Low dosage tricyclics, on average between 75 and 100 mg/day, were 65% (36% to 100%; random effects model), 47% (12% to 94%), and 114% (41% to 226%) more likely than placebo to bring about response at 4 weeks, 6-8 weeks, and 3-12 months, respectively. On average 45% of patients taking low dosage tricyclics responded at 4 weeks, 59% at 6-8 weeks, and 53% at 3-12 months. Heterogeneity was noted only for the outcome at 6-8 weeks (fig 2).
This advantage of low dosage tricyclics was not maintained when we undertook the strict intention to treat analyses based on the worst case scenario. Effectiveness was, however, corroborated by secondary analyses based on continuous measures. People taking low dosage tricyclics had scores for severity of depression that were 0.29 (0 to 0.59; random effects model), 0.59 (0.30 to 0.87), 0.59 (0.20 to 0.99), and 0.89 (0.10 to 1.68) standard deviations lower than those taking placebo at 2 weeks, 4 weeks, 6-8 weeks, and 3-12 months, respectively. Heterogeneity was noted for all these time periods (fig 3).
No difference was found in total number of dropouts between low dosage tricyclics and placebo groups (relative risk 1.08, 0.93 to 1.26). Overall, 439 of 1840 (24%) enrolled participants dropped out by the end of the trial. People taking low dosage tricyclics, however, were 111% (35% to 228%) more likely than those taking placebo to drop out due to side effects. People taking low dosage tricylics were also 63% (36% to 95%) more likely to experience at least one side effect.
Funnel plot analysis and sensitivity analyses
The funnel plot showed some publication bias because the five smallest studies reported large relative risks in favour of low dosage tricyclics. These studies mainly dated from the 1960s and ‘70s and involved patients recruited outside a clinical setting. When we omitted these studies the plot was no longer asymmetrical and the relative risk decreased only slightly from 1.65 (1.36 to 2.00) to 1.58 (1.31 to 1.90) at 4 weeks, and from 1.47 (1.12 to 1.94) to 1.28 (1.05 to 1.55) at 6-8 weeks. The outcome at 6-8 weeks was no longer heterogeneous. The pooled standardised mean difference for the continuous outcome changed to −0.31 (−0.47 to −0.15) at 4 weeks and −0.32 (−0.49 to −0.15) at 6-8 weeks; these results were also no longer heterogeneous.
When we limited the included studies to those that used operational diagnostic criteria for depression, the results were essentially identical. When we limited the included studies to patients taking less than 75 mg/day of tricyclics they were still more likely to show response than those taking placebo at 4 weeks (relative risk 1.63, 1.29 to 2.07). The corresponding standardised mean difference in measurements of depression was −0.44 (−0.72 to −0.17). Patients receiving this minimal dosage were still more likely to drop out due to side effects (relative risk 2.17, 1.05 to 4.50) or to experience at least one side effect (relative risk 2.18, 1.28 to 3.73) than those taking placebo.
Based on these sensitivity analyses, we did not need to restrict ourselves to studies employing operational diagnostic criteria or to those that administered strictly low dosage tricyclics to arrive at conclusions generalisable to present day patients. The following subgroup analyses therefore deal with studies treating any depression.
Only five studies explicitly dealt with depression in people aged 65 or more (n=265). Due to lack of power, the meta-analysis of these five studies produced only the following significant findings. The people taking low dosage tricyclics were more likely to show response at 6-8 weeks than those taking placebo, but they were also more likely to experience at least one side effect (relative risk 1.52, 1.09 to 2.11 and 1.26, 1.10 to 1.45, respectively). The point estimates of the obtained relative risks and standardised mean differences were in accordance with the overall results.
Primary care settings—Five studies recruited patients with depression in primary care settings; five further studies included patients with various physical conditions such as migraine, low back pain, and rheumatoid arthritis, presumably in primary care settings.
Overall there were 558 participants. Although there were no significant findings in the random effects estimates of our primary outcome, the resulting 95% confidence intervals were compatible with the overall findings. For example, at 4 weeks participants taking low dosage tricyclics were 28% (−2% to 68%; random effects model) more likely to show response than those taking placebo; at 6-8 weeks, participants taking low dosage tricyclics were 23% (−3% to 55%) more likely to do so. The continuous outcomes were again supportive of the overall conclusions because the standardised mean difference was −0.29 (−0.58 to 0.01) at 4 weeks and −0.41 (−0.62 to −0.19) at 6-8 weeks.
Psychiatric settings— We also performed a meta-analysis on only such studies that made clear that they were conducted with patients seen in psychiatric settings and not comorbid with other major psychiatric disorders such as eating disorders or Alzheimer's disease. Twelve studies (n=912) were available; two with inpatients and the others with outpatients.
Among the psychiatric patients, the relative risk for showing response with low dosage tricyclic rather than with placebo was 2.40 (1.11 to 5.16; random effects model) at 2 weeks, 1.89 (1.24 to 2.87) at 4 weeks, 1.66 (0.87 to 3.15) at 6-8 weeks, and 2.06 (1.34 to 3.17) at 6 months. The standardised mean difference also supported the effectiveness of low dosage tricyclics in psychiatric patients with depression. Exclusion of the outliers lessened heterogeneity associated with some of the results but did not substantially affect the relative risks or standardised mean differences. The tricyclic was more likely to cause dropouts due to side effects or at least one side effect (relative risk 3.80, 1.63 to 8.86 and 1.43, 1.19 to 1.73, respectively) than placebo.
Low dosage tricyclics versus standard dosage tricyclics
Standard dosage tricyclics were not significantly more effective at achieving response than low dosage tricyclics at 1-8 weeks (fig 4): relative risk 0.89 (0.74 to 1.07) at 4 weeks and 1.11 (0.76 to 1.61) at 6-8 weeks. In terms of the standardised mean difference, standard dosage tricyclics outperformed the low dosage tricyclics at 4 weeks only (standardised weighted mean difference 0.29, 0.08 to 0.50).
Overall there was no difference in the acceptability of the treatments when measured by leaving study early for any reason (relative risk 0.95, 0.75 to 1.20). Low dosage regimens, however, were 55% (24% to 73%) less likely than standard dosage regimens to cause dropouts due to side effects.
Low dosage tricyclic antidepressants between 75 and 100 mg/day and possibly below this range brings about more reduction in depression at 4-8 weeks of treatment and beyond, as well as more dropouts due to side effects and more people with at least one side effect than placebo in both primary care and psychiatric settings. Given the average event rates for controls in the included studies, the number needed to treat to bring about response in depression was between 4 and 6 at 1-6 months of treatment, and the number needed to harm to produce one dropout due to side effects was around 24. Standard dosage tricyclics, however, may or may not be able to bring about more reduction in depression than low dosage tricyclics, although they cause more dropouts due to side effects than placebo (number needed to harm around 11).
Reaching definitive conclusions from these data, however, is not straightforward. The strength of our conclusions is compromised by several factors. Firstly, the quality of the included studies was not ideal. The success of blinding was not ascertained in any. Many studies did not employ operational diagnostic criteria and interview schedules to diagnose depression. Some studies used ad hoc outcome measures of unknown reliability and validity. Although the dropout rates were not high overall, as our worst case scenario intention to treat analyses showed, they were large enough to hamper drawing definitive conclusions. The dropout is always a problem but here it is even more prominent because, in the case of low dosage tricyclics, there is a trade-off between response and dropouts. If dropouts are not dealt with appropriately, the higher dosage always wins because it increases response at the expense of dropouts. Secondly, the quality of reporting in the included studies was not ideal. We are uncertain whether the random allocation was adequately concealed in most of the studies. Some studies failed to report standard deviations for their outcome measures. Thirdly, and perhaps due to the above factors, we noted heterogeneity for some of the pooled results. A few studies were extreme outliers, all in favour of the low dosage regimen. Lastly, most of the included studies lasted up to eight weeks only.
We evaluated the seriousness of these shortcomings with several sensitivity analyses. Omitting the positive small studies removed heterogeneity of the pooled analyses and yet showed little changes in relative risks and standardised mean differences. Limiting the studies to those that employed modern operational diagnostic criteria or those that used strictly low dosage regimens did not materially affect the pooled estimates of effect sizes.
These sensitivity analyses greatly strengthen the inferences that in the treatment of depression tricyclics at dosages lower than the usually recommended range are more effective than placebo but possibly a little bit less effective than standard dosage tricyclics although with fewer side effects. The evidence suggests that academicians have been on weak ground in criticising clinicians' use of low dose tricyclics. Every trial protocol should include strategies for ensuring follow up of all the participants even if they stop the prescribed drug, because it is the only way to adhere to the intention to treat principle and to produce results permitting strong inferences about treatment effects. Only then can the relative benefits and harms of various dosages be definitively established.
This systematic review was conducted within the framework of the Cochrane Collaboration Depression, Anxiety, and Neurosis Group. We thank Gordon Guyatt and David Streiner for their helpful comments on earlier drafts.
Contributors: TAF developed the idea of this systematic review and did the literature search, study selection, data extraction, and data analysis. HMcG did the literature search, study selection, and data extraction. CB did the study selection and helped in the interpretation of the results. All the investigators contributed to the writing of the paper and will act as guarantors.
Funding St Luke's Life Science Institute, Tokyo, Japan provided 700 000 yen (£3659; $5696; 5835).
Competing interests TAF has received fees for speaking from several pharmaceutical companies, some of which manufacture various types of antidepressants including paroxetine, fluroxamine, milnacipran.