Chapter  7:  Treatment of Depression: Newer Pharmacotherapies: Evidence Report/Technology Assessment Number 7

Gaps in knowledge

A number of important questions could not be answered with available evidence. No studies compared combinations of newer antidepressants or newer antidepressants plus another psychotropic (e.g., an anxiolytic) with a single antidepressant. Data were insufficient to determine if the combination of newer antidepressants with psychosocial therapies is more effective than antidepressants alone. Data also were insufficient to determine if augmenting agents (e.g., pindolol, lithium) in combination with a newer antidepressant quicken or improve response rates in patients with resistant depression. Whether particular antidepressant agents are more effective than others could not be determined for patients with resistant or refractory depression. Finally, the need for and efficacy of long-term antidepressant therapy needs to be evaluated in more representative populations.

Gaps in knowledge

There is insufficient evidence to establish whether newer antidepressants are effective for subsyndromal (minor) depression or mixed anxiety depression.

Gaps in knowledge

It is not clear if hypericum (St. John's wort) is as effective as standard antidepressive agents. No trial data for two other herbal remedies (valeriana and kava kava) were found^a.

Gaps in knowledge

No trials compared newer agents with educational or supportive counseling. Only one small trial compared psychotherapy directly with a newer agent in adults. These data were too limited to determine if newer antidepressants are more or less effective than psychosocial therapies.

Gaps in knowledge

Gaps in knowledge for selected populations of special interest are substantial. Only two small studies evaluated newer agents in children or adolescents; data were insufficient to guide management of depression in children and adolescents. A small number of studies evaluated newer antidepressants in patients with depression and either alcoholism, chronic fatigue syndrome, human immunodeficiency virus (HIV) disease, ischemic heart disease, renal failure, or stroke. The results are conflicting and insufficient to reliably determine the efficacy of newer agents compared with that of placebo or older agents. Since fewer than 10 percent of trials reported data about participants' ethnic background, data are insufficient to determine whether efficacy differs across ethnic groups.

Gaps in knowledge

Only one small study with a high dropout rate evaluated newer pharmacotherapy in women with major or subsyndromal depression after childbirth. These data are insufficient to determine the efficacy of newer antidepressants in the postpartum setting.

Gaps in knowledge

Marked variability in methods of ascertainment and reporting of common adverse effects makes interpretation difficult. Some adverse effects, such as sexual dysfunction and changes in weight, were reported too infrequently for reliable interpretation.

Sources

English and non-English literature sources were: (1) The Cochrane Collaboration Depression, Anxiety, and Neurosis (CCDAN) Review Group's specialized registry of 8,451 clinical trial articles; (2) references from trial articles and 46 pertinent meta-analyses;^{6, 7, 22, 28, 30, 35-77} and (3) experts. The CCDAN specialized registry includes trials relevant to treatment of depression that are identified from multiple sources. They include electronic databases such as MEDLINE, EMBASE, PsycLIT, LILACS, Psyndex, SIGLE, CINAHL, Biological Abstracts, and The Cochrane Library (Appendix 1.1); handsearches of conference proceedings and 69 psychiatry-related journals (Appendix 1.2); contacts with 30 pharmaceutical companies (Appendix 1.3); and routine reference checking of trial articles and systematic reviews. All records in the CCDAN database have subject headings applied based on the National Library of Medicine's MeSH terms.

Pertinent meta-analyses were identified from MEDLINE using Hunt and McKibbon's search strategy.⁷⁸ Technical experts on the advisory panel were asked to identify additional trials.

Search Methods

Table 3. List of newer antidepressants used in literature (more...)

Table 3. List of newer antidepressants used in literature search

Amisulpride	Milnacipran
Brofaromine	Minaprine
Bupropion	Mirtazapine
Citalopram	Moclobemide
Felsinoxan	Nefazodone
Femoxetine	Paroxetine
Fengabine	Reboxetine
Fluoxetine	Ritanserin
Fluvoxamine	Sertraline
Gepirone	Sulpiride
Hypericum	Tandospirone
Indalpine	Toloxatone
Ipsapirone	Tomoxetine
Kava kava	Valeriana
Litoxetine	Venlafaxine
Medifoxamine	Viloxazine

The search yielded 1,277 records, 97 percent of which were identified from the CCDAN registry. Thirty trials were identified from references of meta-analyses. One additional trial was identified by experts. (Of note, continuing searches of MEDLINE and PubMed were conducted from January 1998 through August 1998 while this report was being prepared and peer reviewed. Twenty-three additional trials were found; pertinent additional trials identified through recent ongoing searching are noted in each section.)

Selection Processes

One person initially screened titles and abstracts for inclusion criteria. Titles and abstracts that were rejected by this person were independently assessed by a second person. If either person had doubt about the appropriateness of excluding the study at the title-abstract stage, it was not excluded. There were 418 titles and abstracts that were rejected as definitely not meeting inclusion criteria.

Of the remaining 859 records, the text of 747 reports was obtained. One of 112 unobtainable reports was an article published in a journal not available through either the United States Interlibrary Loan system or the CCDAN group.⁷⁹ The other 111 unobtainable reports were abstracts from conference proceedings or titles identified from pharmaceutical company records and meta-analyses.^{37, 40, 48, 58, 69, 80-153} It is not known whether final reports are available for these abstracts and titles, but 63 (55 percent) were judged from the information available as very likely to meet inclusion criteria. When author contact information was available (n=34), letters were sent to determine eligibility and obtain a full report. Trials available in abstract form only were not considered eligible. We were unable to ascertain whether 51 abstracts and 1 article met selection criteria, were ongoing versus completed studies, or were duplicative studies.

At least two independent reviewers screened the full text of each of the 747 retrieved articles for final inclusion. There were 458 articles that met selection criteria; reviewers agreed on all selections (100 percent concordance). Most articles were easily excluded for obvious reasons (n=276); eligibility was difficult to ascertain in 13 cases (see pertinent references for specific difficulties).^154-166

Figure 2. Flow Diagram of Selection Process

   Figure 2. Flow Diagram of Selection Process

Figure 2

Duplicate publications are cited in Appendix 1.4. Duplicate publications were identified by an information specialist (nurse with library and reference management expertise) and a physician with clinical knowledge of the topic area and methodologic expertise. The following items were considered in assessing duplicate publications: author names; geographic locations of studies; and study designs including which drugs were studied, numbers of participants, study lengths, and outcome measures that were used. No single item mentioned above was adequate for identifying duplicate publications (e.g., there are examples of results of the same trial being reported by totally different authors). When the information specialist and physician could not determine whether a publication represented a duplication, authors and pharmaceutical companies were contacted for clarification.

Search and Selection for Uncommon but Serious Adverse Effects Question

Articles addressing serious adverse effects were identified with specific keywords and text words for the adverse effects defined by MedWatch and also with nonspecific keywords and text words such as "adverse," "serious," "severe," or "poisoning." Articles were narrowed down to those addressing the relevant drugs with appropriate keywords, text words, and, in the case of MEDLINE, registry numbers. Articles were not limited to the topic of depression. For details of the serious adverse effects search strategies, see Appendix 1.5.

The initial searches retrieved a total of 12,374 records. Forty-eight percent of the records were unique to EMBASE, 8 percent unique to MEDLINE, and 3 percent unique to PsycLIT. The remainder were found in at least two of the databases.

The titles and abstracts of an arbitrary sample of 3,298 records (27 percent of total) were screened by one person. There were 323 of these 3,298 records that were judged as meeting inclusion criteria. A search strategy that captured 93 percent of the thus-far included records was retrospectively devised. The revised strategy was then used to identify remaining articles warranting assessment. All 3,400 records returned by the revised search strategy were assessed for inclusion by one person.

There were 692 studies that met inclusion criteria. These studies included 52 controlled studies, 20 publications of postmarketing databases, and 620 case series or case reports. Complete texts of all controlled studies were dual abstracted by two independent reviewers. The drugs and adverse effects of the case series and reports were tabulated, based on the information presented in the titles and abstracts.

Randomized Controlled Trials

Two independent persons abstracted data from each trial regarding participant and diagnostic descriptors, intervention characteristics, study designs, and clinical outcomes. Abstractors were not blinded to study title or author names. Items related to the internal validity of studies that were assessed included adequacy of randomization (method and concealment of assignment), whether the trial was single or double blind, cointerventions, and numbers of dropouts. A standard code book that explicitly defined each item was used.

Eight persons with pertinent clinical and methodology experience performed abstractions. They were trained and calibrated with each other using a pilot set of 20 trials. Reliability checks of abstractions were assessed in a 25 percent sample. For most items, agreement was greater than 95 percent. For diagnostic categorizations and reasons for dropouts, agreement was lower than 85 percent. Thus, a third independent person (physician with extensive clinical and methodologic expertise) confirmed diagnostic and dropout information. Because of its importance, the physician also verified a 90 percent sample of outcome assessments.

Disagreements were resolved by consensus. In the few instances where consensus was not easily reached, an additional person was consulted for arbitration. Abstractions were entered into an electronic database. Prior to data analysis, checks for outlier values were performed and data entries were corrected as appropriate.

Studies Addressing Uncommon Serious Adverse Effects

Studies addressing uncommon serious adverse effects were abstracted by three physicians with expertise in methodology. Items that were abstracted included study design (case report, case series, case control, cohort, controlled trial) and type of specific adverse effect. Several explicit criteria aimed at assessing drug adverse event causality were assessed such as appropriate temporal relationship, lack of apparent alternative causes, known toxic concentrations of the drug at the time of the appearance of the symptom, disappearance of the symptom with drug discontinuation, dose-response relationship, and reappearance of the symptom if the drug was readministered. Abstractions of the studies were discussed among the three physicians and consensus ratings were used. No formal reliability testing was done.

Publication Bias

Table 4. Known missing records

Table 4. Known missing records

No. of Missing Records by Publication Type	No. Known to be Eligible
1 article	0
51 abstracts	26
60 unpublished reports	35
112 total	61

Table 5. Drugs used in known missing trials meeting (more...)

Table 5. Drugs used in known missing trials meeting criteria^a

Newer Antidepressant	No. of Known 63 Eligible Missing Trials Using this Agent	No. of 315 Eligible Retrieved Trials Using this Agent
Bupropion	1	11
Citalopram	1	11
Fluoxetine	10	103
Fluvoxamine	9	34
Gepirone	1	4
Mirtazapine	10	6
Milnacipran	10	4
Nefazodone	6	13
Paroxetine	14	47
Sertraline	10	24
Venlafaxine	12	14

a

This table does not include trials that were subsequently identified as published through ongoing searches conducted from January 1998 through August 1998.

Figure 4. Funnel plot Newer Agent vs. Older Agent Trials (more...)

   Figure 4. Funnel plot Newer Agent vs. Older Agent Trials

Figure 5. Funnel Plot Newer Agent vs. Newer Agent Trials (more...)

   Figure 5. Funnel Plot Newer Agent vs. Newer Agent Trials

Figure 6. Funnel plot Newer Agent vs. Placebo Trials (more...)

   Figure 6. Funnel plot Newer Agent vs. Placebo Trials

Figures 4

5

6

Unreported or Missing Data

Data were often missing or unreported for the following: sociocultural characteristics of study participants, methods for concealing randomization and allocation, specifics about how interventions were delivered, outcome values, and adverse events. Whether data reporting was selective in a manner that introduced particular biases could not be assessed.

Missing Data About Patient Characteristics

Table 6. Number of trials not reporting sociodemographic (more...)

Table 6. Number of trials not reporting sociodemographic data

Education level	302
Socioeconomic status	309
Ethnicity	287
Age	44
Gender	37

Unspecified Methods of Randomization and Concealment

Table 7. Blinding of 315 trials

Table 7. Blinding of 315 trials

Double blind	302
Single blind	3
Not blinded	3
Not applicable or unclear	7

Unreported Details About Intervention Delivery and Followup

The majority of studies did not report details regarding how interventions were delivered. For example, very few trials reported any information about the type of clinician (e.g., nurse, psychiatrist, research associate) who interacted with patients or whether any type of educational information or supportive therapy was provided to participants during the course of the trial. Whether the persons responsible for delivering the interventions had established therapeutic relationships with trial participants was rarely clear.

Planned followup visits varied from weekly to twice monthly to monthly. Almost all trials followed participants in clinic or hospital settings rather than in the home or by telephone. Trials were not designed to assess whether different intensity of followup was related to adherence, efficacy, or costs. Among trials of 12 weeks or shorter duration, average dropout rates in those with scheduled weekly followup was 31 percent whereas average dropout rates in those with scheduled followup every 2 weeks (bimonthly) was 22 percent. Whether varying intensities of followup schedules were related to differences in discontinuation rates among different drugs could not be determined (see section, Unreported Data for Adverse Events, Adherence, and Dropouts).

Although unplanned cointerventions such as benzodiazepines and psychosocial therapy were described in some trials, most did not report the number of participants who received the cointerventions. The extent to which the outcomes of these studies can be attributed solely to the tested antidepressants is not entirely clear.

Unreported Outcomes

Table 8. Outcome, adverse event, and dropout data not (more...)

Table 8. Outcome, adverse event, and dropout data not reported in 315 trials

No. of trials not reporting outcome values	14
No. of trials not assessing adverse events	21
No. of trials not reporting dropout data	42

Most studies did not report outcomes using true intention-to-treat or last-observation-carried-forward methods. Some studies that used last-observation-carried-forward only did so for patients completing at least 3 weeks of treatment. The results of these studies could be biased by systematic differences in dropout rates between treatment arms.

Less than 5 percent of the trials reported health-related quality-of-life outcomes. Quality-of-life information would have been particularly useful as it is a global measure of functioning that allows proxy comparisons of both benefits and adverse effects across drugs. For example, dry mouth may occur with a particular agent and headaches or sexual dysfunction with another agent. Without a global measure of functioning, it is difficult to compare whether adverse effects in general are more troublesome with one agent than another.

Unreported Data for Adverse Events, Adherence, and Dropouts

Trials rarely provided information about adherence to assigned therapy. For example, only one large trial provided detailed information about numbers of visits completed and numbers of prescriptions filled.⁴²⁷ Less than 5 percent of the trials reported pill counts. Another 11 percent provided numbers and reasons for dropouts that did not add up. When studies reported how many patients had dropped out for adverse events, they often failed to report what kinds of adverse events had prompted the dropouts. This is of interest because anecdotal evidence suggests that patients may be just as likely to stop taking medications for innocuous, but annoying side effects (e.g., dry mouth) as they are for serious side effects (e.g., hypotension). Letters were sent to 213 authors in an attempt to gather some of the unreported data noted above. Although we had a response rate of 30 percent, very little missing information was retrieved because either investigators had died or data were in storage.

Heterogeneity of Diagnostic Nomenclature

Several different diagnostic schemes (e.g., Diagnostic and Statistical Manual of Mental Disorders [DSM], Research Diagnostic Criteria [RDC], Endicott, Newcastle) and several different revisions of the same scheme (e.g., DSM-III, DSM-III-R, DSM-IV) were used in the trials to diagnose specific depressive disorders. Approximately 70 percent used either DSM-III or DSM-III-R criteria as the primary diagnostic tool. Several studies also required a certain severity of depression (usually determined by the Hamilton Depression Rating Scale) for inclusion criteria.

Some of the diagnostic criteria have evolved over the past two decades as our understanding of depressive disorders has increased. For example, unlike DSM-III, DSM-IV requires that the symptoms for major depression "cause clinically significant distress or impairment in social, occupational, or other important areas of functioning." Although the trend has been to develop more objective diagnostic criteria, there remains subjectivity and heterogeneity in applying them.

High Dropout Rates

Approximately 30 percent of participants in the 281 trials that reported such information dropped out prior to completing followup. Eleven percent dropped out due to adverse events, 9 percent dropped out due to lack of efficacy, and the remaining 11 percent dropped out for other reasons that were usually unspecified. Numbers of dropouts varied across trials, with approximately 30 percent of the trials reporting dropout rates of 20 percent or less, and 20 percent of the trials reporting dropout rates of 40 percent or more. Many experts consider high dropout rates a threat to internal validity, especially if the patients who drop out differ in systematic ways (e.g., in severity of depression, experience of adverse events) from the patients who remain in the study. The trials with high dropout rates rarely compared dropouts with completers for such systematic differences.

Table 9. Dropout rates of the 269 trials reporting (more...)

Table 9. Dropout rates of the 269 trials reporting such information

<20%	84
20-30%	81
30-40%	53
40-50%	36
>50%	15

Multiple Outcome Measures

Table 10. Methods of reporting HAMD, MADRS, or CGI (more...)

Table 10. Methods of reporting HAMD, MADRS, or CGI outcomes in 293 trials

HAMD or MADRS mean, medium, or mean change	271
HAMD or MADRS 50% response rate	133
CGI "Much or Very Much Improved"	87
HAMD or MADRS "any improvement"	57
HAMD or MADRS or CGI "Cure"	53
Neither HAMD or MADRS or CGI	5

HAMD -- Hamilton Depression Rating Scale;MADRS -- Montgomery-Âsberg Depression Rating Scale;CGI -- Clinical Global Impression.

The Hamilton Depression Rating Scale was the single most commonly used outcome. This assessment scale relies on an assessor (usually a clinician) to interview a subject and rate symptoms according to specified criteria. Its reliability may depend on adequate blinding of assessors. Even though almost all of the studies were double blind, maintenance of double blinding was never reported. Fewer than 5 percent of the trials used separate investigators to monitor adverse effects and severity of symptoms. As adverse effects between agents such as SSRIs and TCAs differ, it is possible that investigators who were assessing outcomes were not always adequately blinded. As there is inevitable interest and optimism that new treatments will be beneficial, observer bias may have occurred.⁵⁷

Cultural Differences in Outcome Assessment

This evidence report includes studies conducted in over 20 countries. There may be systematic differences across cultures in the way in which depressive disorders are experienced, reported, and assessed. Outcomes that depend on a provider's or a patient's global assessment of functioning may be interpreted differently by different linguistic and cultural groups. When instruments are translated into different languages, the quality and meaning of the translation can be problematic. Studies, including multicenter trials conducted across several countries, rarely addressed the issue of the linguistic or cultural equivalence of outcome measures.

Inconsistent Reporting of Adverse Events

Table 11. Methods of assessing adverse events in 286 (more...)

Table 11. Methods of assessing adverse events in 286 trials

Method unclear or not stated	120
Rating scale	73
Checklist	44
Generic open-ended question	45
Voluntary report	67

Statistical Concerns

Most trials were small, involving fewer than 100 subjects. Dropout rates were often greater than 30 percent, yielding even smaller effective sample sizes. Small trials were typically underpowered, making it difficult to interpret "negative" findings of individual studies when no statistically significant differences were found. Meta-analyses addressed some of these concerns but were not possible for many areas with few studies such as subsyndromal depression, mixed anxiety depression, and depression with concomitant comorbid medical illness.

Generalizability

Several additional problems could limit ability to apply the evidence to clinical practice. These include the use of restrictive inclusion criteria, prerandomization run-in periods, short trial duration, and methodologic restrictions of randomized controlled trials designed to assess treatment efficacy.

Inclusion and Exclusion Criteria

Most trials used restrictive inclusion and exclusion criteria. To ensure a homogenous patient sample, patients with significant medical and psychiatric comorbid conditions were usually excluded. Two-thirds of the trials explicitly stated exclusion criteria related to alcohol and other substance abuse; in all but three instances, participants with these conditions were excluded. Women who were pregnant or lactating, or who were not using a reliable form of birth control, were typically excluded because of safety concerns. The most severely depressed patients were also often excluded. Even though heterogeneous nomenclatures were used to define patients, the majority of trial participants had major depression of moderate severity. (The section addressing major depression gives detail on the severity of depression in most trial participants.) Many of the trial participants had past histories of depressive disorders and/or had previously received antidepressant therapy. Whether such participants represented a population who were refractory or responsive to prior therapy was usually not clear. Given this lack of data, it is not possible to generally conject whether trial effect sizes are overly optimistic or conservative for patients with a first untreated episode of depression.

Run-ins

Prerandomization run-in periods are often used in pharmacotherapy trials to exclude noncompliant subjects and placebo responders. The rationale is to reduce the number of participants who will respond to placebo and thus increase detectable differences between placebo and active treatments. Of the 214 trials that had run-in designs, 128 excluded placebo responders and other types of patients. Some evidence suggests that results of such trials overestimate benefits and underestimate risks of treatment.⁴⁸⁵ However, meta-analysis of 101 trials included in the 1993 AHCPR guideline,⁶ showed that trials with placebo run-ins did not have lower placebo response rates or higher drug-placebo differences.⁴⁸⁶ Subgroup analysis of major depression trials in this report showed no major differences in effect sizes between trials with run-in periods and those without. (See section in Chapter 3, Quantitative Results of Effects of Design and Methodologic Characteristics on Response Rates)

Trial Length

Table 12. Number of trials of varying lengths

Table 12. Number of trials of varying lengths

6-7 weeks	229
8-9 weeks	41
10-11 weeks	2
12-13 weeks	27
14-16 weeks	1
17+ weeks	15

Types of Randomized Controlled Trials

The majority of studies were designed to assess efficacy -- whether drugs were effective under ideal and standardized conditions. As a result, trials were often rigidly conducted to meet certain methodologic standards. They often required that patients assigned to the same groups be treated exactly the same. Some did not allow drug dosages to be titrated or modified according to an individual subject's response to adverse effects or lack of efficacy, which is rarely the case in actual clinical practice. Moreover, there is a need to study effectiveness of clinical treatments as they are really delivered in clinical practice.⁴⁸⁷ Such trials may be more informative for policymakers than are efficacy trials.

Primary Measure of Treatment Effect

The primary measure of treatment effect was response rate. It was defined as a 50 percent or greater improvement in symptoms as assessed by a depression symptoms rating scale or a rating of much or very much improved as assessed by a global assessment method. This definition of response was chosen because it represents a clinically meaningful improvement and is commonly reported in depression trials. Outcome measures that were most often reported were the Hamilton Depression Rating Scale, the Montgomery-Âsberg Depression Rating Scale, and the Clinical Global Impression. If multiple outcomes were reported, HAMD data were used first, followed by MADRS and CGI data, respectively. Treatment-specific response rates were calculated across studies, based on the total number of responders divided by the total number randomized to that class of antidepressant.

The log risk ratio was used to estimate response rate. The log risk ratio is

where p ₁ and p ₂ are the relative frequencies with which patients in the two comparison groups show a clinically significant response to treatment. These values were transformed back to risk ratios to facilitate clinical interpretation of the results. Since trials rarely collected depression symptom outcomes in patients who dropped out, analyses were not based on true intent-to-treat results. Rather, a modified intent-to-treat method was used.⁴⁸⁸ The denominator for p ₁, i=1,2, was the number randomized to treatment, and the numerator was the number of patients who stayed in treatment and got better. Given that some patients who left treatment may have gotten better, modified intent-to-treat response rates are more conservative or lower than those derived from a true intent-to-treat analysis.

Secondary Measure of Effect

Quantitative change scores on the HAMD and MADRS were examined to confirm primary analyses based on response rates. Standardized mean differences between mean change scores were the effect size measure in these analyses. Hedges' g was used to compute the standardized mean difference for each trial:
,
where, for a given trial i, and are the mean change scores for the two groups being compared and S _pooled is the pooled standard deviation for the difference between the two means.⁴⁸⁹ These estimates were adjusted for small sample bias when the total sample size at the posttreatment assessment was fewer than 50 patients.⁴⁸⁹

Published reports seldom provided estimates of S _pooled three strategies was used to estimate S _pooled when the authors did not directly provide it. First, the standardized mean difference was back calculated from either the test statistic or the p value for the difference between change scores.⁴⁹⁰ If this was not possible, values for S _pooled were imputed from the partial variance information provided in the published trial report.⁴⁹¹ If neither was possible, a mean variance derived from studies of similar size was used.

Partial variance information reported in the trials usually consisted of either the pretreatment standard deviations for the two groups or pre- and posttreatment standard deviations for the two groups. The posttreatment sample size for the two groups was used to calculate S _pooled because it yielded a larger estimate for S _pooled and thus a smaller more conservative estimate of the effect size.

Estimation of the variance of a change score also required an estimate of the "within trial correlation" between the pretreatment and posttreatment scores. "Within trial correlations" were estimated from data provided by some authors and from a current multisite trial. Correlations were pooled to yield a mean estimate (r=0.40) and 95 percent confidence interval.

Empirical Bayes Random-Effect Estimates

We adopted the DerSimonian and Laird empirical Bayes random-effects estimator to estimate the pooled measures of treatment efficacy.^{492, 493} Empirical Bayes estimators are particularly suited for relatively small samples (n=10 trials). If there is no substantial heterogeneity among the trials, the empirical Bayes estimator reduces to the classical fixed-effects estimate. When significant heterogeneity exists, the empirical Bayes estimator moves the point estimates of effect towards the overall mean, thereby decreasing heterogeneity.

The empirical Bayes estimator for each study is

where d ₊ is the DerSimonian and Laird random effects for the overall treatment effect and is the commonly used moment estimator for the "between studies" variance. Its variance is

where We performed all computations with the STATA meta command written by Sharp and Sterne⁴⁹⁶ and Sharp.⁴⁹³

Organization of Comparisons

Comparisons were organized into three major groupings: (1) newer agents versus placebo, (2) newer versus older agents, and (3) newer versus newer agents. Within major groupings, comparisons were arranged according to drug classes such as SSRIs versus TCAs. In trials that compared different doses of a newer agent with placebo or another agent, the group that received the minimum dose within the specified usual dose range was used. For multi-arm trials comparing more than one newer agent with placebo or an older agent, all comparisons were used.

Results for all pairwise comparisons were displayed in standard forest plots. A chi-square test for homogeneity was performed for each set of pairwise comparisons; p<0.20 was considered evidence for significant heterogeneity. Galbraith plots identified outlier studies that might help explain heterogeneity in estimates of treatment efficacy.⁴⁹⁷ Publication bias was assessed with funnel plots using Begg's rank order correlation test and with Egger's regression approach.⁴⁹⁸ STATA 5.0 (STATA Corporation) was used to perform all pairwise analyses and produce the graphical output.⁴⁹⁹ Specifically, the meta command was used to compute and graph empirical Bayes estimates,⁴⁹⁶ the galbar command to access homogeneity and produce Galbraith plots,⁴⁹⁹ and the metabias command to examine publication bias.⁵⁰⁰

Sensitivity Analysis

As previously stated, primary analyses were based on a modified intent-to-treat method. A sensitivity analysis was done based on an endpoint method. In this method, the denominator for the risk ratio was the number of participants who completed followup or whose last observation was carried forward.

Subgroup and Multivariate Analyses

Studies varied greatly with respect to a wide range of patient, treatment, design, and setting characteristics. Several were identified a priori as potentially affecting efficacy estimates. These characteristics were age, ethnicity, gender, education, presence of significant comorbid medical illness or concomitant substance abuse, history of treatment failure or recurrence, type and severity of depression, agent class, study duration, run-in design, dropout rates, and setting (outpatient, inpatient, primary care, mental health). As later noted, only age, gender, severity of depression, class of newer agent, dropout rates, and setting could be used in multivariate analyses because of limitations in the reporting of trials.

Primary subgroup analyses were done for trials conducted in older patients, primary care settings, patients with major depression, and patients with dysthymia. Additional analyses were done to examine the effects of study design (run-in phase, dropout rates) on outcomes. These additional analyses were restricted to major depression trials in order to maximize the number of studies in each category. "Run-in" was classified as studies without a run-in phase, studies using a run-in to exclude placebo responders, and studies using a run-in phase without stating if it was used to exclude placebo responders. Dropout rates were analyzed using three separate thresholds: dropout rates greater than 20 percent, 30 percent, and 40 percent. Choice of cut-off did not make a difference in the dropout analyses. Subgroup analyses provided a picture of how heterogeneity impacts treatment efficacy for a single characteristic. However, studies varied with respect to multiple characteristics; it is not possible to manage multiple covariates effectively in subgroup analyses. Thus, multivariate statistical models were used to examine the impact of multiple covariates on measures of treatment response.

Multivariate models were used to examine results across depressive disorders and, separately, for major depression studies only. They were performed separately for newer agent versus placebo studies and newer versus older agent studies. No multivariate models were constructed for newer versus newer agents because of the smaller number of studies in this category. Ethnicity and education were not used in multivariate models because trials rarely reported such data. Presence of significant comorbid illness and history of treatment failure were not used because there were few trials in these categories.

Multivariate models used random-effects meta-regression.⁴⁹² An iterative weighted least-squares method recommended by Morris produced a maximum likelihood empirical Bayes estimate of the regression parameters.⁵⁰¹ The method iterated between estimating the regression coefficients and estimating between study variance until a convergence criterion was reached. The log risk ratio was the response variable. The STATA command, metareg, was used for these analyses.⁴⁹³

Since proportions are nonnormal, dichotomous measures, such as percent female patients, were transformed to the log odds scale. All covariates were represented as deviations from their respective means in the random-effects meta-regressions. By expressing the covariates as deviation scores, the intercept constant estimated the value of the log risk ratio at the mean value for the covariates.

Dropout Rates and Adverse Effects

Comparisons between newer and older agents were done using proportions of participants who discontinued therapy for any reason, for adverse effects, and for lack of efficacy. Such dropout analyses were limited to studies less than 13 weeks in duration because study duration is likely related to dropout percentages and there were few studies longer than 12 weeks.

The relative frequency of 12 commonly reported adverse effects, such as nausea and headache, was examined. Statistical analyses comparing adverse effects were done only for selective serotonin reuptake inhibitors versus tricyclic agents because this comparison was the only one that consistently had pertinent adverse effect data from at least 10 trials.

Context

The efficacy of antidepressants for major depression has been firmly established. However, different costs make the relative efficacy of newer antidepressants highly relevant. In addition, the efficacy of some newer agents has not been systematically reviewed.

The Evidence from Placebo-Controlled Trials (Tables 20, 21, 22, 23)

Eighty-one trials of at least 6 weeks' duration, involving over 10,000 adult patients with major depression, compared newer antidepressants with placebo. Trials were conducted in 15 countries; the majority in the United States (n=54). All studies were sponsored by pharmaceutical companies (n=55 of 55 with identified funding). Participants were recruited from a variety of settings including outpatient (n=48), mixed in- and outpatient (n=7), inpatient (n=3), and not described (n=23). Only four studies specified recruitment that included primary care settings.

On average, the severity of depression was moderate to moderately severe as measured by a standard symptom rating scale (mean=24, standardized to the 17-item Hamilton Depression Rating Scale). Fewer than 1 percent of patients had depression with psychotic features.

The quality of the studies of SSRIs, among newer antidepressants, are the best (n=43 comparisons). SNRIs (n=12), RIMAs (n=10), 5-HT₂ (n=9), 5-HT_1A (n=3), and DopRI (n=3) antidepressants have been evaluated in more than one study. There are single studies of fengabine (GABA) and reboxetine (NRI). Of these, 47 studies reported usable response rates. No studies compared DopAnt with placebo. Almost all trials were short duration, focusing on acute phase treatment. Duration was 6 to 8 weeks in 79 studies and 9 to 12 weeks in 2. Dropout rates were less than 20 percent (n=7), 20 to 30 percent (n=14), >30 percent (n=42), and not given for 18. All trials appeared double blind.

The Findings of Placebo-Controlled Trials (Figure 7)

Response rates were analyzed using a modified intention-to-treat (MIT) approach, which gives a conservative estimate of response. Risk ratios were computed using both the MIT approach and an endpoint analysis (based on the number of patients evaluated at a predefined endpoint). The latter results are reported only when they differ from the MIT approach by at least 10 percent. Fifty-one percent of the participants randomized to active treatment experienced at least a 50 percent improvement in depressive symptoms as assessed by a depressive symptom rating scale or were much to very much improved on the Clinical Global Rating Scale. In contrast, only 32 percent responded significantly to placebo. As a group, newer antidepressants were significantly more efficacious than placebo (risk ratio [RR] 1.6, 95 percent CI 1.5 to 1.7). Studies reporting mean changes in symptom scores rather than response rates showed similar results. There was evidence of positive publication bias for two of the newest classes of antidepressants (SNRI and 5-HT₂)_.

Two classes of newer antidepressants have not been adequately evaluated against placebo. In one small study, fengabine (GABA) showed no effect.³⁸¹ In the single study comparing an NRI (reboxetine) with placebo, reboxetine improved self-reported psychosocial functioning more than placebo, but effects on depressive symptoms were not reported.¹⁹⁶ The effective classes are described below.

Findings of Placebo-Controlled Trials by Drug Class

Evidence from Ongoing Searches

Five, short-term, placebo-controlled trials, involving 965 subjects were identified in searches from January 1998 through August 1998 (venlafaxine n=3; sertraline n=1; nefazodone n=1).^{179, 715, 716, 717, 718} Results of these most recent trials are consistent with past trials; newer antidepressants show consistently positive effects compared with placebo.

The Evidence from Newer Compared with Older Antidepressant Trials (Tables 24, 25, 26, 27)

In 150 studies, involving over 16,000 participants, there have been 154 comparisons between newer and older antidepressants. Studies were conducted predominantly in Europe (n=76) and the United States (n=49). Eighty-four studies with an identified funding source were sponsored by industry; 2 were sponsored by governmental agencies. When the study setting was described, the outpatient setting was most common (n=77), followed by mixed in- and outpatient settings (n=28) and inpatient settings (n=20). For 17 studies, recruitment included primary care settings.

Newer antidepressants were compared with first generation tricyclic antidepressants (n=122), second generation tricyclic drugs (n=8), trazodone (a triazolopyridine; n=8), tetracyclic antidepressants (n=14), or monoamine oxidase inhibitors (n=2). Among newer agents, SSRIs are the most extensively tested with 97 comparisons to older agents. Other classes compared with older agents are SNRIs (n=14), RIMAs (n=26), NRI (n=2), minaprine (Misc; n=3), 5-HT₂ (n=6), 5-HT_1A (n=1), and DopRI (n=5). There were no comparisons to older agents identified for the two classes currently available only outside the United States -- GABA and DopAnt).

Trial duration was 6 to 8 weeks for 146 comparisons, 9 to 12 weeks for 5, and less than 12 weeks for 3. All but seven studies were double blind. Dropout rates were less than 20 percent (n=33), 20 to 30 percent (n=39), more than 30 percent (n=61), and not given for 21.

The Findings from Newer Compared with Older Antidepressant Trials (Figures 8, 9, 10, 11)

Fifty-four percent of the participants randomized to a newer antidepressant experienced at least a 50 percent improvement in depressive symptoms as assessed by a rating scale (e.g., Hamilton Depression Rating Scale) or by "much to very much improvement" on the Clinical Global Rating Scale. An identical proportion (54 percent) responded significantly to an older antidepressant. As a group, newer antidepressants were equally efficacious compared with older antidepressants (RR 1.0, 95 percent CI 0.97 to 1.06). Studies reporting mean changes in symptom scores rather than response rates had similar findings. There was evidence of positive publication bias in the reporting of RIMA versus TCA1, but not for other comparisons of newer versus older agents. Specific drug classes are reviewed below.

The Findings from Newer Compared with Older Antidepressant Trials by Drug Class

Evidence From Ongoing Searches

Three additional trials, involving over 370 subjects, were identified in searches from January 1998 through August 1998 that compared milnacipran (SNRI, n=2) or sertraline (SSRI, n=1) with a TCA1.^{717, 720, 721} These trials are consistent with past trials in showing no major differences in efficacy between newer and older agents. The trial comparing sertraline and amitriptyline showed positive effects for both drugs across a broad array of health-related quality-of-life domains compared with placebo.

The Evidence From Comparisons Between Newer Antidepressants (Tables 28, 29, 30, 31)

Newer antidepressants have been compared head to head in 32 studies, involving over 4,000 participants. Studies were conducted predominantly in Europe (n=19) and North America (n=8). All studies with an identified sponsor were funded by pharmaceutical companies. Participants were recruited from inpatient settings (n=16), mixed in- and outpatient settings (n=7), and outpatient settings (n=2); in seven studies, the setting was not specified. No study reported recruitment from primary care settings.

Comparisons were between an SSRI and another newer agent (n=17), between two SSRI antidepressants (n=14), and between a RIMA and NRI antidepressant (n=1). Twenty-nine studies were 6 to 8 weeks in duration, 1 was 9 to 12 weeks, and 2 were longer than 12 weeks. Of the 32 studies, 30 were double blind. Dropout rates were lower than 20 percent (n=12), 20 to 30 percent (n=11), greater than 30 percent (n=5), and not given (n=4).

The Findings From Comparisons Between Newer Antidepressants (Figure 12)

Fifty-three percent of the participants experienced at least a 50 percent improvement in depressive symptoms as assessed by a rating scale such as the Hamilton Depression Rating Scale or were much to very much improved on the Clinical Global Rating Scale. As a group, SSRI antidepressants were equally efficacious compared with other newer antidepressants. There was no evidence of publication bias for these comparisons. Ten studies reporting response rates compared two SSRI antidepressants, of which eight involved a comparison with fluoxetine. There was a trend for fluoxetine to be slightly less effective (RR 0.9, 95 percent CI 0.8 to 1.0) than the comparator SSRI (citalopram, sertraline, paroxetine). Analysis of the mean changes in depressive symptoms showed no significant difference for fluoxetine compared with other SSRIs.

Evidence from Ongoing Searches

Three trials identified from January through August 1998, involving 446 subjects, compared an SSRI with an SNRI, NRI, or DopRI.^{722, 723, 724} Mirtazapine, administered at levels outside the approved U.S. dosing range (mean=56 mg/d), was more efficacious than fluoxetine; otherwise, treatments were equally efficacious.

Longer Duration Trials

Since the minimum duration of treatment for major depression is 6 months, longer duration studies may offer insights into long-term efficacy and delayed adverse effects. Only four trials, involving 764 participants, were designed to continue initial treatment for at least 24 weeks. Each study involved a different newer antidepressant (SSRI, 5-HT₂, DopRI, DopAnt) compared with a TCA1. There was a single comparison to placebo. One study did not report outcomes.³⁷⁶

Response rates were similar for newer antidepressants (54 percent) and TCA1 (52 percent). These rates are nearly identical to 6-week trials. The most striking finding from these studies was the high overall dropout rate (>50 percent). Dropouts due to adverse effects were 24 percent for newer agents and 25 percent for TCA1. These rates are more than double those seen in shorter duration studies.

Dropout Rates in All Major Depression Trials

Analysis of dropout rates offers direct evidence about the tolerability of antidepressant medications and indirect data about efficacy. For SSRI, the most commonly used class of newer antidepressants, overall dropouts did not differ significantly from TCA1 (risk difference 0.02, 95 percent CI 0.05 to 0.02), TCA2 (risk difference -0.09, 95 percent CI -0.20 to 0.01), or Tetra antidepressants (risk difference -02, 95 percent CI -0.10 to 0.09). Dropout rates due to adverse effects were lower for SSRI compared with TCA1 (risk difference -0.04, 95 percent CI -0.07 to -0.01) and showed a trend towards a lower dropout rate compared with TCA2 (risk difference -0.07, 95 percent CI -0.08 to 0.01) and tetracyclic antidepressants (risk difference -0.04, 95 percent CI -0.08 to 0.0). Additional details about dropout rates are given in another section.

Gaps in Evidence

Significant gaps in evidence on treating major depression with newer antidepressants are as follows: Few studies examined long-term treatment efficacy and adverse effects, few studies reported effects of antidepressants on physical function or quality-of-life outcomes, and few studies were designed to test the effectiveness of antidepressants under usual clinical conditions without the specialized treatment and followup inherent in a randomized trial.

Other Relevant Evidence

Several literature syntheses have addressed the efficacy of antidepressant drug therapy for major depression. These reviews differed in scope from the present evidence report in that they were limited to older agents or SSRIs compared with other agents. All prior reviews have found antidepressants to be more efficacious than placebo and have found similar efficacy between antidepressants.^{7, 28, 29, 30} Estimates of dropout rates have ranged from no differences to clinically small (<4 percent), but statistically significant differences were found in favor of SSRI.^{28, 29, 30}

Context

Chronic, mild depressive disorder (dysthymia) is found in approximately 3 percent of community populations and causes significant impairment. Because the symptoms are less intense but more chronic (more than 2 years), dysthymia may respond differently to treatment than major depression.

The Evidence (Tables 32, 33, 34, 35)

Nine trials of at least 6 weeks' duration involving 1,420 adult patients with dysthymia compared newer antidepressants with placebo (n=5), first generation tricyclics (n=5), an antipsychotic (n=1), or another newer antidepressant (n=1).^{182, 202, 282, 295, 366, 415, 430, 445, 458} Six of the nine trials were conducted in Europe, and pharmaceutical companies sponsored the four with an acknowledged funding source. All studies were conducted in outpatients. Settings were mental health specialty clinics (n=6), primary care clinics (n=1), and not described (n=2).

Selective serotonin reuptake inhibitors (n=3), a dopamine antagonist (n=1), and a 5-HT₂ receptor antagonist (n=1) were compared with placebo. Sertraline (SSRI, n=1), ritanserin (5HT₂, n=2), amisulpride (DopAnt, n=1), and minaprine (Misc, n=1), were compared with first generation tricyclic antidepressants. A single small trial compared ritanserin with an antipsychotic.

All trials examined acute phase treatment; duration was 6 to 8 weeks in four studies and 12 weeks in five studies. Dropout rates were lower than 20 percent (n=4) and 20 to 30 percent (n=5). All trials were double blind.

The Findings (Figure 13)

Response rates were analyzed using a modified intent-to-treat approach, which gives a conservative estimate of response. Fifty-nine percent of the participants randomized to active treatment experienced at least a 50 percent improvement in depressive symptoms on a rating scale such as the Hamilton Depression Rating Scale or much to very much improved on the Clinical Global Rating Scale. Response rates for participants given placebo or first generation tricyclic antidepressants were 37 percent and 59 percent, respectively. Newer antidepressants were significantly more efficacious than placebo (RR 1.7, 95 percent CI 1.3 to 2.3) but similar to first generation tricyclic antidepressants (RR 1.0, 95 percent CI 0.9 to 1.2). There was no evidence of publication bias for these comparisons.

In a single study comparing two newer agents, the efficacy of amisulpride (DopAnt) was similar to that of fluoxetine (SSRI). A small study in patients with chronic tension or mixed tension/migraine headaches and dysthymia (n=38) compared ritanserin (5HT₂) with amitriptyline (TCA).³⁶⁶ Both groups showed similar but modest improvements in pain ratings and depressive symptomatology. Another study that only reported mean changes in depressive symptomatology showed similar responses for patients treated with ritanserin (5HT₂) or imipramine (TCA); both drugs were superior to placebo.¹⁸²

Overall dropout rates for newer antidepressants, first generation tricyclics, and placebo were 15 percent, 17 percent, and 19 percent, respectively. Four percent of participants given newer antidepressants dropped out because of side effects compared with 9 percent given tricyclics (risk difference -0.05, 95 percent CI -0.10 to -0.001) and 2 percent for placebo (risk difference 0.02, 95 percent CI -0.01 to 0.04).

Gaps In Evidence

Relative to major depression, there are few trials examining newer pharmacotherapies for dysthymia. No trials report the effect of treatment on social functioning, work performance, or ability to carry out daily activities. The longest trials are only 12 weeks, leaving significant gaps in knowledge about long-term efficacy. Although response rates are similar to those seen for major depression, a large proportion do not respond to initial treatment. Studies are needed to examine treatments designed to enhance the initial response rate and treatments for refractory patients.

Other Relevant Evidence

A Cochrane Collaboration review, with broader inclusion criteria (e.g., major depression with concurrent dysthymia included), examined 15 placebo-controlled trials. Results were concordant with the current review showing similar efficacy for multiple agents compared with placebo.⁶¹ A Cochrane Collaboration protocol titled, "A comparison of active drugs for the treatment of dysthymia," is in progress.⁶⁰

Context

Depression and anxiety are common coexisting conditions; the most effective treatment for persons suffering from both conditions has not been established.

The Evidence (Tables 36, 37, 38)

Three trials studied 1,237 adults with mixed anxiety depression.^{327, 359, 394} All participants had depression and significant anxiety as assessed by an anxiety symptom scale. The trials were conducted in primary care settings and were sponsored by industry. Patients with alcohol abuse or "serious" concomitant medical illness were excluded.

One large international trial of 12 weeks' duration (n=1,019) compared paroxetine (SSRI) at 20 to 40 mg/d with clomipramine (TCA) at 75 to 150 mg/d. One much smaller European trial (n=112) compared fluvoxamine (SSRI) at 300 mg/d with lorazepam (anxiolytic) at 6 mg/d. The other small European trial (n=106) compared sertraline (SSRI) at 50 to 150 mg/d and clomipramine (TCA) at 50 to 150 mg/d. Dropout rates were 24 percent for the large trial and 13 percent for each of the smaller trials.

The Findings (Figure 14)

In the largest trial, 85 percent of patients given paroxetine and 83 percent of patients given clomipramine experienced at least a 50 percent improvement in symptoms as measured by the Montgomery-Âsberg Depression Scale. In one small trial, 80 percent and 82 percent of patients receiving fluvoxamine and lorazepam, respectively, were much or very much improved on the Clinical Global Impression Scale. In the other small trial, 58 percent of patients given sertraline and 61 percent of patients given clomipramine improved as measured by the Hamilton Depression Rating Scale, and 58 percent and 57 percent (sertraline and clomipramine, respectively) improved as measured by the Hamilton Anxiety Rating Scale. None of the comparisons within the three trials was significantly different.

Dropout rates due to adverse effects were significantly greater with clomipramine (17 to 18 percent) compared with paroxetine (11 percent) and sertraline (4 percent). Administration of fluvoxamine (5 percent) and lorazepam (4 percent) resulted in similar dropout rates.

Gaps In Evidence

There are inadequate data to guide drug management for patients with mixed anxiety depression. Specifically, there are no placebo-controlled trials and there is a dearth of trials of any type in persons with mixed anxiety depression.

Other Relevant Evidence

No other recent relevant systematic review was found.

Context

The efficacy of antidepressant drug therapy has been established for major depression. Less severe forms of depression, sometimes called "minor," "subthreshold," or "subsyndromal" depression, cause impairment and are common in primary care settings. The efficacy of antidepressant drug therapy for these disorders has been unclear.

The Evidence (Tables 39, 40, 41, 42)

Three trials evaluating newer agents have involved 392 adult patients with subsyndromal depression.^{379, 457, 441} All were conducted in Europe and were sponsored by pharmaceutical companies. All studies were conducted in outpatients. Recruitment settings were mental health specialty (n=1), mixed primary care and mental health (n=1), and not described (n=1).

Only one study, in older patients (mean age=71), was placebo controlled. Two studies compared newer agents (paroxetine [SSRI] and moclobemide [RIMA]) with older agents. Depressive illnesses were defined as DSM-III subsyndromal or neurotic depression (n=1); International Classification of Disease, 9^th ed. (ICD-9) prolonged depressive reaction to a stressor (n=1); and Research Diagnostic Criteria (RDC) for minor depression (n=1). Although these diagnoses are not exactly equivalent, they all describe patients with depressed mood or anhedonia and fewer than the five symptoms required for major depressive disorder. The three trials examined acute phase treatment; duration was 6 weeks in two studies and 13 weeks in the other. Dropout rates were lower than 20 percent (n=1), 20-30 percent (n=1), and >30 percent (n=1).

The Findings (Figure 15)

The single placebo-controlled trial was conducted in older adults (age 60 to 85) with a "prolonged depressive reaction." It showed "improvement" for 60 percent of patients randomized to minaprine 200 mg/d compared with 39 percent of patients given placebo as measured by a Clinical Global Assessment. Compared with placebo, minaprine improved social functioning. There were no dropouts due to adverse effects.

The largest trial involved 544 patients with major (n=298) or subsyndromal (n=245) depression. Among those with subsyndromal depression, a 50 percent reduction in Hamilton Rating Depression scores was obtained in 77 percent of patients on paroxetine (20 to 40 mg/d) compared with 69 percent of patients on maprotiline (100 to 150 mg/d). (RR 1.1, 95 percent CI 0.95 to 1.25). For the entire sample, dropout rates due to adverse effects were similar for the two drugs (13 percent and 14 percent).

A small trial (n=17) compared moclobemide (RIMA) with amitriptyline (TCA) administered at a low dose (75 mg/d). Overall clinical response rates were very good or good in seven of eight patients on moclobemide and four of nine patients on amitriptyline. Dropout rates were high (25 percent and 44 percent) for the two drugs, respectively. This study was too small to detect clinically significant differences between treatment groups, and the dose of amitriptyline may have been inadequate.

Gaps in Evidence

There are inadequate data to guide management for patients with subsyndromal depression. Data are insufficient to determine if therapy with newer antidepressant drugs is better than placebo for acute phase treatment. There are no trials comparing newer antidepressant drug therapy with any type of psychosocial therapy, a reasonable and attractive alternative for some patients with milder forms of depression. The data are inadequate to guide drug selection for clinicians who choose antidepressant drug therapy for patients with subsyndromal depression. Finally, no studies have examined the effects of treatment on patients' ability to perform their normal daily activities.

Other Relevant Evidence

None was found.

Context

Approximately 50 percent of patients who suffer from an initial depressive episode are likely to suffer a recurrence of their symptoms. Whether newer antidepressants are effective in treating such patients is unclear.

The Evidence (Tables 43, 44, 45)

Most of the 315 trials addressed in this evidence report included patients with histories of depression. For example, 75 percent of the trials had study populations that included more than 50 percent of participants with histories of depression. Thus, much of the data reviewed in the section addressing major depression are derived from patients with one or more prior depressive episodes. Only eight studies specifically stated that they were studying recurrent depression.^{225, 238, 276, 298, 425, 468, 504, 505} (Three additional studies that address long-term maintenance treatment in patients with recurrent depression are discussed in another section.) These are the only trials reviewed here, though it is clear many of the patients in the other trials had recurrent depression. The eight trials involved 1,624 patients. Most patients had recurrent major depression. Five of the studies were conducted in North America and two in Europe.

The trials included four comparisons of newer agents with placebo, five of newer agents with first generation tricyclic agents (TCAs), and one comparison each of bupropion (DopRi) and venlafaxine (SNRI) with trazodone (Triazo). Dropout rates ranged from 12 percent to 55 percent, with only one trial having a dropout rate lower than 20 percent. Study duration was 6 weeks for all studies.

The Findings (Figure 16)

Of patients with recurrent depression given a newer antidepressant, 47 percent had at least a 50 percent improvement in symptoms as measured by a depressive symptom rating scale. Only 28 percent responded significantly to placebo. For the three relevant trials reporting response rates, newer antidepressants were significantly more efficacious than placebo (RR 1.5, 95 percent CI 1.2 to 1.9).

Of patients with recurrent depression given an older antidepressant, 45 percent responded. In the six relevant trials reporting response rates, newer antidepressants were equally efficacious as older agents.

Gaps in Evidence

There are significant gaps in knowledge concerning the treatment of recurrent depression. More trials are needed to reliably determine whether response rates to antidepressants in such patients are lower than those seen in patients suffering their first episode of depression. Trials are also needed to determine whether particular patients such as those with multiple prior episodes of depression are more refractory to treatment than those with only one prior episode. Whether adverse effects of antidepressants are higher in persons with recurrent depression is also not well known.

Other Relevant Evidence

None was found.

Context

Approximately 50 percent of people suffering from major depression will have significant alleviation of symptoms with antidepressant therapy. Few guides help patients and clinicians select particular treatments when first attempts at therapy are not successful.

The Evidence (Tables 46, 47, 48, 49)

Five European trials compared newer and older antidepressant agents in adults with refractory depression.^{177, 296, 305, 438, 477} Four were conducted in mental health settings; three were limited to inpatients.

These trials involved 396 patients with major depression who had persistent symptoms despite antidepressant therapy. Patients with alcohol abuse and serious medical conditions were not studied.

Two trials compared paroxetine at 30 to 40 mg/d (SSRI) with either amitriptyline or imipramine at 150 mg/d (TCA). One compared brofaromine (RIMA) at 100 mg/d with tranylcypromine (MAOI) at 20 mg/d. The other two trials compared augmentation strategies of lithium with brofaromine or fluoxetine (SSRI) against lofepramine (TCA) or maprotiline (Tetra).

Four trials were 6 weeks in duration and one was 12 weeks. Dropout rates ranged from 19 percent to 49 percent.

The Findings (Figure 17)

The two small trials that involved lithium showed no consistent benefits; they are described in a later section that addresses augmentation. Response rates for agents evaluated in the other three trials ranged from 66 percent to 80 percent. In these few trials with few participants, none of the comparisons varied significantly from agent to agent.

Gaps in Evidence

There are large gaps in evidence concerning the treatment of refractory depression. There are very few trials. None is long enough to evaluate appropriate duration of therapy for patients with refractory depression. None has evaluated therapy with newer agents in the event of failed psychosocial therapy.

Other Relevant Evidence

No other recent high-quality systematic reviews were found.

Context

Psychosocial therapies, including psychotherapy, group exercise, supportive counseling, and interventions to improve social supports, are preferred by some patients. Among these, psychotherapy is the best established alternative to pharmacotherapy for the treatment of nonpsychotic, nonbipolar depressive disorders in outpatients. To aid informed decisionmaking, we addressed the relative efficacy of psychosocial therapies compared with that of pharmacotherapy.

The Evidence (Tables 50, 51, 52)

Only one small trial (n=31) has compared the efficacy of a newer agent, fluoxetine (SSRI) directly with a psychosocial therapy in adults.²⁵⁰ The trial was conducted in the United States, and the funding source was not identified. The psychosocial intervention involved individual cognitive therapy; the credentials and educational levels of the therapists were not clear. All subjects in the trial had dysthymia. The trial examined acute phase treatment, and the intervention lasted 16 weeks. There was a 29 percent dropout rate.

The Findings (Figure 18)

Both the cognitive therapy group and the fluoxetine group showed improvement over time. The number of patients was too small to detect clinically significant differences between treatment groups. Six of the 18 fluoxetine patients and 3 of the 13 cognitive therapy patients dropped out of treatment or were unavailable for end-of-treatment assessment.

Gaps in Evidence

The paucity of studies comparing psychosocial interventions to newer pharmacotherapies represents a significant gap in the literature. A variety of psychosocial therapies such as general support and counseling have not been compared with newer pharmacotherapies. The only pertinent study had a small sample size and inadequate followup assessments. No trials of maintenance or relapse prevention are available. Data are insufficient to determine which patients will best respond to psychotherapy and which will respond to pharmacotherapy.

Other Relevant Evidence

Prior systematic reviews and meta-analyses have shown that certain types of depressive-specific psychotherapies (e.g., cognitive or cognitive-behavioral therapy and interpersonal therapy) are as effective as pharmacotherapy for major depression.^{6, 8, 21} Trials addressed in these reviews have primarily involved older antidepressant agents. The relative effectiveness of the different varieties of psychotherapies is not entirely clear.^{506, 507}

Context

Extracts of the plant Hypericum perforatum L., popularly known as St. John's wort, have long been used in folk medicine and are licensed in Germany for the treatment of depression. The efficacy and tolerability of hypericum compared with standard antidepressants as well as placebo have been questioned.

The Evidence (Tables 53, 54, 55)

Fourteen trials of at least 6 weeks' duration involving 1,417 adult participants compared hypericum with placebo (n=8) or first generation tricyclic antidepressants (n=6).^{193, 240, 292, 297, 314, 317, 375, 397, 418, 436, 465, 466, 470, 475} None compared hypericum with newer agents. Most trials were conducted in Germany and were funded by industry. Multiple preparations, standardized based on content of hypericins (which may not be the only active ingredient), were studied. Doses of hypericum extract varied from 300 to 1800 mg/d. Three trials evaluated combinations of hypericum with other extracts such as valeriana. First generation tricyclic agents compared were amitriptyline (30 to 150 mg/d), desipramine (100 mg/d) and imipramine (75 mg/d).

Participants in the studies were recruited by psychiatrists, general practitioners, internists, and gynecologists. All were outpatients. They had multiple depressive disorders. All trials were described as double blind, though hypericum extracts have a characteristic taste and some degree of unblinding is possible. Dropout rates were low (approximately 10 percent).

The Findings (Figure 19)

Sixty-two percent of the participants who were given single preparations of hypericum extracts experienced a 50 percent improvement in symptoms as measured by a depressive symptoms scale such as the Hamilton Depression Rating Scale. Response rates for participants given placebo and older tricyclic antidepressants at low doses were 38 percent and 61 percent, respectively. Hypericum extracts were significantly more effective than placebo (RR 1.9, 95 percent CI 1.2 to 2.8) but similar to older tricyclic agents given at low doses (RR 1.2, 95 percent CI 1.0 to 1.4). Publication bias analysis was positive, suggesting positive results of hypericum are systematically reported. Two trials of hypericum and valeriana that compared combination preparations with tricyclic agents found 68 percent and 50 percent response rates, respectively (RR 1.5, 95 percent CI 0.8 to 2.9).

Six percent of persons given hypericum preparations dropped out of the trials compared with 7 percent of those given older tricyclics at lower doses (risk difference -0.01, 95 percent CI -0.05 to 0.02). Proportions of patients reporting "any" side effect were 25 percent for hypericum preparations compared with 40 percent with tricyclic agents (risk difference -0.15, 95 percent CI -0.24 to 0.06).

Gaps in Evidence

Evidence regarding the efficacy of hypericum is lacking in many areas. Neither the most appropriate preparation nor the most effective dose is known. There are not enough comparisons with older agents given at adequate doses, and there are no comparisons with newer antidepressant agents. There are no long-term maintenance or prevention of relapse data. Whether hypericum is effective for certain depressive disorders and not others is still unknown.

Other Relevant Evidence

A recent Cochrane review, which will be continually updated, summarizes 27 short-term trials (including trials of less than 6 weeks' duration) of hypericum and is consistent with the above findings.⁶²

Context

Although newer antidepressants are widely used in adults, their safety and efficacy in depression management for children and adolescents are unclear.

The Evidence (Tables 56, 57, 58, 59)

Only two trials (n=80) evaluated the use of newer antidepressants in children and/or adolescents.^{340, 426} One of the trials was conducted in the United States and the other was conducted in Canada. Funding sources were not identified in either study.

One trial compared fluoxetine (SSRI), at 20 to 60 mg, with placebo. The other trial compared a combination of venlafaxine (SNRI) and psychotherapy with a combination of placebo and psychotherapy. Venlafaxine was titrated to low doses of 12.5 mg 3 times per day (t.i.d.) for children and 25 mg t.i.d. for adolescents.

Patients in both trials had nonpsychotic major depression. Both trials examined acute phase treatment of 6 to 8 weeks' duration. Overall dropout rates were 18 percent and 25 percent, respectively. Few data on adverse effects were given.

Gaps in Evidence

Gaps in the literature about efficacy of antidepressants in children and adolescents are substantial. There are no direct comparisons of newer agents alone with psychosocial therapy, older agents, or other newer agents. Only one trial compares a newer agent with placebo (see below for update on one additional trial). The two trials reviewed above are small and short. Adequate data about adverse events are lacking.

Other Relevant Evidence

A recent systematic review summarized results of six randomized trials evaluating cognitive behavioral therapy in 231 adolescents with depression.⁵⁰⁸ Adolescents receiving cognitive behavioral therapy had fewer depressive symptoms than those assigned to control groups. Ongoing searches conducted through August 1998 identified one additional trial. This recent 8-week double-blind trial compared fluoxetine (20 mg/d) with placebo in 96 outpatients age 7 to 17 years.⁷²⁶ All had severe persistent nonpsychotic major depressive disorder. Overall discontinuation rates were greater among participants assigned to placebo than to fluoxetine (45 percent versus 29 percent). Discontinuation rates due to lack of efficacy with placebo were 40 percent compared with 16 percent with fluoxetine. At study exit, 56 percent of patients assigned to fluoxetine were much or very much improved on the CGI scale compared with 33 percent assigned to placebo (p=0.02).

Context

Frequent comorbid medical illness and concomitant treatment with other medications has led to the suspicion that antidepressant therapy is less effective in older persons than in middle-age persons.

The Evidence (Tables 60, 61, 62)

There were 27 trials involving 3,929 persons over 60 years of age.^{220, 249, 251, 254, 259, 260, 283, 287, 291, 298, 299, 306, 313, 320, 321, 339, 352, 365, 371, 377, 379, 386, 390, 414, 443, 447, 509} Nineteen were conducted in Europe or Scandinavia, 6 in the United States, and 2 in South America. Of 17 trials with an identified funding source, pharmaceutical companies funded all.

Ten comparisons of newer agents versus placebo were made. Most other comparisons were made with selective serotonin reuptake inhibitors (SSRI, n=10), reversible inhibitors of monoamine oxidase (RIMA, n=4), or serotonin norepinephrine reuptake inhibitors (SNRI, n=2) versus tricyclic antidepressants. All but two of the tricyclic comparisons were with first generation drugs such as amitriptyline, imipramine, and dothiepine. Newer antidepressants generally were given in the usual dose range, whereas 8 of the 17 comparison groups given tricyclic agents used low doses.

Most participants in the trials were outpatients with major depression. Two small trials (n=134) involved physically ill inpatients,^{254, 390} and one included 46 frail elders receiving home nursing care.³¹³ Persons with cognitive impairment were often excluded; however, three trials involving 961 patients specifically recruited patients with dementia.^{306, 414, 443}

Few trials described the actual treatment environment or characteristics of providers who delivered treatment. Decisions regarding antidepressant management (initial dose, dosage changes, treatment discontinuation, or switches) generally followed set protocols. Adherence in terms of number of scheduled visits attended, pill counts, or numbers of prescriptions filled generally were not reported.

All trials studied acute phase treatment. Treatment duration was typically 6 to 8 weeks (n=24). Dropout rates were <20 percent (n=5), 20 to 30 percent (n=8), >30 percent (n=11), and unclear (n=3). Eleven trials reported cognitive status outcomes; none reported quality-of-life or physical function outcomes.

The Findings (Figure 20)

Approximately 50 percent of the older participants who received newer antidepressants had at least a 50 percent improvement in symptoms as measured by a depressive symptoms rating scale. Response rates did not vary significantly from newer to newer agents except for sulpiride (DopAnt, also used as an antipsychotic drug), where one small study rendered a response rate of only 14 percent. The average response rate for older patients given placebo was 30 percent. Average response rates for older agents was 54 percent.

Newer antidepressants excluding sulpiride were significantly more effective than placebo (RR 1.4, 95 percent CI 1.2 to 1.6). Newer antidepressants were similar in efficacy to tricyclic agents (RR 1.0, 95 percent CI 0.9 to 1.1). No specific classes of newer agents, including SSRI, RIMA, and NRI, were superior to tricyclic agents. Subgroup analyses showed that newer antidepressants were not superior to either tricyclics at low doses or first generation tricyclics only. There was no evidence of publication bias, and statistical tests for heterogeneity were not significant for above comparisons. Endpoint analyses showed identical results to the modified intent-to-treat analyses presented above.

Total dropout rates did not vary significantly between newer and older antidepressants. Dropout rates due to adverse effects were placebo 1 percent, SSRI 16 percent, TCA1 17 percent, NRI 11 percent, and RIMA 7 percent. One study had a 26 percent dropout rate in patients given nortriptyline (second generation tricyclic), which was titrated based on therapeutic drug levels. In general, dropouts due to adverse events did not vary significantly between newer agents and first generation tricyclics (risk difference -0.03, 95 percent CI -0.07 to 0.02).

The three trials conducted in cognitively impaired older adults compared moclobemide (RIMA) at 400 mg/d with placebo, fluoxetine (SSRI) at 10 mg/d with amitriptyline at 25 mg/d, and paroxetine (SSRI) at 20 to 40 mg/d with imipramine (TCA) at 50 to 100 mg/d. The large trial (n=726) found that moclobemide significantly improved depressive symptoms and cognitive status scores compared with placebo. The two small trials (total n=55) found no differences in depression or cognitive status between agents.

Three small trials in physically frail elders compared fluoxetine (SSRI) at 20mg/d with placebo, sulpiride (DopAnt) at 200 mg/d with placebo, and fluvoxamine (SSRI) at 150 mg/d with dothiepine (TCA) at 150 mg/d. None of these trials showed significant differences between groups.

Gaps in Evidence

Significant gaps in the literature on treating depression in older adults are as follows: few trials in U.S. general practice and nursing home settings, few trials in non-major depression, no trials with physical function or quality-of-life outcomes, no trials comparing newer pharmacotherapies with psychosocial therapy, no trials of relapse prevention, and few trials in frail elders with significant comorbid illness.

Other Relevant Evidence

A recent systematic review of 37 controlled studies found tricyclic agents were superior to placebo in treating depression in older ambulatory patients.⁶⁴ Few trials of newer agents were included. Six small studies suggested behavioral therapy was more effective than usual care but not statistically significantly better than nonspecific interventions that provided similar attention to that received by behavioral treatment groups. Another recent meta-analysis addressing late-life depression found no differences in efficacy between selective serotonin reuptake inhibitors, reversible inhibitors of monoamine oxidase A, tricyclics, or other newer agents such as mianserin.⁵¹⁰

Ongoing searches through August 1998 identified a recent double-blind trial comparing milnacipran (at 50 mg two times per day) with imipramine (at 50 mg two times per day) in 219 older persons with major depression. One-third of the participants dropped out of the 8-week trial that showed no significant differences in efficacy between the two agents.⁷²⁷

Context

Patients with significant comorbid medical conditions often develop depression, but the efficacy of antidepressant pharmacotherapy is not well known in these populations. It is often thought that side effects and drug interactions may impair already compromised functioning in such patients.

The Evidence (Tables 63, 64, 65, 66)

Ten trials involving 484 adults have evaluated newer antidepressants for treating depression in patients with specific comorbid medical and psychiatric illnesses.^{171, 172, 190, 198, 224, 229, 345, 410, 444, 461} Four trials were conducted in the United States and six in European and Scandinavian countries. Of six trials that identified funding sources, five were funded by pharmaceutical companies. Six additional trials studied elders with physical illness or cognitive impairment^{254, 306, 313, 390, 414, 443} They are described in the section addressing treatment in older patients.

Newer agents evaluated have included selective serotonin reuptake inhibitors, reversible inhibitors of monoamine oxidase A, and serotonin norepinephrine reuptake inhibitors.

A variety of depressive disorders have been studied, including major depression, dysthymia, adjustment disorder with depressed mood, atypical disorder, and other unspecified types of depressions.

Two studies had a mixed patient group that included more than one disorder. There have been eight comparisons with placebo (one in combination with psychotherapy), four comparisons with older agents, and no direct comparisons of newer agents. A number of medical and psychiatric conditions were examined, including alcoholism, chronic fatigue syndrome, HIV, ischemic heart disease, renal failure, and stroke. There were no trials of at least 6 weeks' duration that addressed depressed patients with other major comorbid medical conditions such as diabetes or cancer.

Six trials examined acute phase treatment lasting 6 to 8 weeks, and four trials were 12 weeks. Total dropout rates for eight of the nine trials reporting this information were 20 percent or lower.

The Findings

Gaps in Evidence

There are several major gaps relevant to the efficacy of newer antidepressants in treating persons with concomitant depression and medical illness. First, there are few relevant trials. Important common comorbid medical illnesses such as cancer and diabetes have not been studied. Most available trials are small and lack power to detect clinically important differences. Most trials are too short to assess sustainability of treatment benefits as comorbid illness continues or worsens. Few trials assess comprehensive outcomes including physical, mental and social function. Only one trial compares the efficacy of newer pharmacotherapies with that of a psychosocial therapy, which may be an attractive option in populations where side effects and drug interactions are a particular concern.

Other Relevant Evidence

No recent systematic reviews evaluating the efficacy of newer antidepressants in patients with comorbid illness were found. Three recent trials in HIV-positive patients have been identified in ongoing searches conducted through August 1998. One 8-week trial has not reported outcomes according to groups who were randomized to either fluoxetine or placebo.⁷²⁸ A 12-week trial in 75 HIV-positive outpatients with major depression has compared paroxetine with imipramine and placebo. Fewer than 50 percent of the participants completed the trial, which found no significant differences in efficacy between paroxetine and imipramine, but there was a higher dropout rate from adverse effects with imipramine compared with paroxetine (48 percent versus 20 percent). A third recent 7-week trial compared fluoxetine plus group psychotherapy with group psychotherapy alone in 47 HIV-seropositive men with major depression. More participants assigned to fluoxetine plus psychotherapy versus psychotherapy alone had a 50 percent or greater response in depressive symptoms as measured by the Hamilton Depression Rating Scale.⁷³⁰

Context

Most prior evidence addressing efficacy of antidepressant drug therapy has been derived from psychiatric settings. The efficacy of antidepressant therapy for the wide spectrum of patients seen in primary care has been unclear.

The Evidence (Tables 67, 68, 69)

Twenty-seven trials evaluating newer agents involved 5,540 adult primary care patients. ^{178, 184, 189, 202, 213, 223, 227, 242, 254, 282, 299, 313, 318, 327, 328, 333, 359, 360, 361, 380, 394, 412, 427, 429, 450, 471, 472} Twenty-six trials were conducted in Europe and 1 in the United States. Of the 20 trials with an identified funding source, pharmaceutical companies sponsored all.

All trials had recruitment and inclusion criteria that specified participants as patients of primary care providers or general practitioners. Few described the actual treatment environment or characteristics of providers that delivered treatment. Decisions regarding antidepressant management (initial dose, dosage changes, treatment discontinuation, or switches) generally followed set protocols except in the U.S. health maintenance organization trial where patients and their physicians made these decisions.⁴²⁷

Multiple newer agents including SSRIs, SNRIs, RIMAs, and DopAnts were evaluated. Most trials compared newer agents with first or second generation tricyclic agents. All but one trial used drug doses that were in usual dose ranges, though many were toward the lower range of standard doses.

Multiple depressive disorders were studied: major depression (n=12), "depression requiring treatment" as judged by the primary care provider (n=4), dysthymia (n=2), mixed anxiety depression (n=3), endogenous depression (1), and heterogeneous patient groups with depressive illness that included more than one disorder (n=4). Patients with alcohol abuse, cognitive impairment, and "serious" concomitant medical illnesses typically have been excluded. Exceptions were a large trial from a U.S. staff model health maintenance organization that did not exclude patients with concomitant medical illness and two small trials that specifically evaluated older frail patients with medical illness.^{254, 313}

All trials studied acute phase treatment. Treatment duration was typically 6 to 8 weeks (n=22), although three trials were 12 weeks and two were 20 to 24 weeks. Followup was generally conducted with outpatient visits, though the U.S. trial relied primarily on telephone followup assessments. Most trials did not clearly specify whether research assistants or primary care providers conducted followup assessments. Content of followup assessments appeared oriented primarily toward assessment of depressive symptoms using standardized rating scales. Whether general supportive measures and counseling occurred during followup visits was not specified.

Adherence to therapy generally was not measured except in the U.S. health maintenance organization trial. This trial reported numbers of patients completing telephone followups (91 percent) and percentages of patients who received at least 90 days of adequate treatment among those originally assigned to fluoxetine (61 percent), desipramine (49 percent), or imipramine (48 percent). Minimal "adequate treatment" doses were fluoxetine at 10 mg/d, desipramine 75 mg/d, and imipramine 74 mg/d.

Dropout rates in the trials were less than 20 percent (n=12), 20 to 30 percent (n=11), and more than 30 percent (n=3). Only four trials measured functional status or quality of life.

The Findings (Figure 23)

Approximately 60 percent of the primary care participants who received newer antidepressants had at least a 50 percent improvement in symptoms as measured by a depressive symptoms rating scale (e.g., Hamilton Depression Rating Scale) or were much to very much improved as measured by a global assessment method (e.g., Clinical Global Impression Scale). Response rates did not vary significantly from newer agent to newer agent except for minaprine, where two studies showed lower response rates of approximately 34 percent. The average response rate for patients given placebo was 35 percent. The average response rate for patients given tricyclic agents was 60 percent. Newer antidepressant agents were significantly more effective than placebo (risk ratio 1.6, 95 percent CI 1.2 to 2.1) but similar to tricyclic agents (risk ratio 1.0, 95 percent CI 0.9 to 1.1) in efficacy. Risk ratios were virtually identical in modified intent-to-treat and endpoint analyses. Response rates appeared similar across depressive disorders.

Dropouts due to adverse effects occurred in 2 percent, 8 percent, and 13 percent of patients given placebo, newer agents, and tricyclic agents, respectively. Newer agents had significantly fewer dropouts from adverse effects than tricyclic agents (rate difference -0.04, 95 percent CI -0.07 to -0.00). There was some variability among newer agents with two venlafaxine (SNRI) studies showing higher (18 percent) dropout rates from adverse effects and one minaprine (Misc) study showing lower dropout rates of 2 percent.

In the 6-month U.S. health maintenance organization trial,⁴²⁷ patients assigned to fluoxetine (SSRI) reported fewer adverse events, were more likely to continue their original medication, and were more likely to reach "adequate" doses than those assigned to either desipramine (TCA2) or imipramine (TCA1). Depressive symptoms, quality of life, and total health care costs did not differ significantly between groups.

Gaps in Evidence

Significant gaps in the literature on the primary care treatment of depression are as follows: few trials in U.S. general practice settings, few trials in patients without major depression, few trials with functional status or quality-of-life outcomes, no trials comparing newer pharmacotherapies with psychosocial therapies, no trials of relapse prevention, and few trials in patients with significant comorbid illness.

Other Relevant Evidence

A recent evidence-based guideline also showed no major differences in efficacy between newer and older antidepressants in the primary care setting.³⁰ Ongoing searches through August 1998 identified one additional trial. This double-blind multicenter trial in 400 patients with major depression compared sertaline (at 50 to 150 mg/d) with citalopram (at 20 to 60 mg/d). Approximately 75 percent of parTicipants completed the 24-week trial. No statistically significant differences in efficacy as measured by the Montgomery-Âsberg Depression Rating Scale scores were found.⁷³¹

Context

Depression occurring postpartum has been identified as a significant problem for both mothers and their babies. Because pregnant and lactating women have often been excluded from clinical trials, little is known about the safety and efficacy of antidepressant drugs in this population.

The Evidence (Tables 70, 71, 72)

Only one trial in the postpartum setting was identified. ¹⁷⁶ It was conducted in Great Britain. The funding source was not identified.

Participating women were diagnosed with either major or minor depression 6 to 8 weeks after childbirth. It was not clear whether their depression had a postpartum onset. The women were randomized to one of four treatments: fluoxetine (SSRI) or placebo plus one or six sessions of cognitive behavioral therapy delivered by nonspecialists in mental health.

The trial examined acute phase treatment lasting 12 weeks. The participants had a 26 percent dropout rate.

The Findings

Fluoxetine compared with placebo showed significant improvements in depressive symptoms as assessed by three different rating scales. Likewise, six sessions of counseling were superior to one in improving depressive symptoms. A synergistic interaction between fluoxetine and counseling was not apparent.

Thirty percent of the participants dropped out for unknown reasons (n=14), dislike of drug or adverse effects (n=9), or lack of improvement (n=3). The authors also noted that 101 of 188 recruited women with depression refused to participate in the trial because they were reluctant to take medication and they expected to improve without treatment.

Gaps in Evidence

There are major gaps in knowledge about treatment of depression in the postpartum setting. They include no direct comparisons of newer agents with placebo, older agents, or psychosocial therapy. No trials involving pregnant women are available. The one trial conducted in the postpartum period did not report infant outcomes or general well-being and quality-of-life outcomes in mothers.

Other Relevant Evidence

No other recent systematic reviews on the treatment of depression in the postpartum setting were found.

Context

Depressed patients often receive both antidepressants and psychosocial therapies in clinical practice. Although some feel that the two approaches are complementary and may be especially helpful for chronic cases of depression, others suggest that the two treatments may interfere with each other. For example, pharmacotherapy is seen by some as a superficial treatment that does not address the underlying cause of depression. Another concern is that early symptom relief by pharmacotherapy may lead to premature termination of psychotherapy. Others argue that psychotherapy counteracts the effect of pharmacotherapy by encouraging the patient to focus on upsetting issues. Pharmacotherapy may reduce symptoms more rapidly and psychotherapy may provide mood management skills that reduce likelihood of relapse or recurrence.

The Evidence (Tables 73, 74, 75)

Only two trials involving 107 adults have evaluated combined treatments of newer antidepressants and psychosocial therapies to treat remission.^{176, 444} One of the trials was conducted in the United States and the other in Great Britain. Only one trial identified a funding source (nonpharmaceutical).⁴⁴⁴

Both studies compared a combination of a psychosocial therapy and fluoxetine with combined psychosocial therapy and placebo. Neither study compared the combination treatment with a newer agent alone. The British trial involved 87 postpartum women with major or minor depressive disorder.¹⁷⁶ Psychosocial therapy was individual counseling based on cognitive-behavioral therapy.¹⁷⁶ Each session was structured to offer reassurance and advice. Therapists were general practitioners who were not mental health specialists. This trial had four arms and compared one versus six sessions of counseling.

The U.S. trial involved 20 HIV-positive men with either major depression or adjustment disorder with depressed mood.⁴⁴⁴ The psychosocial intervention involved a combination of elements from behavior therapy to problem-solving therapy. It was delivered as structured group therapy that emphasized active skills in coping. Sessions included relaxation training and problem-solving skills focused on specific HIV-related concerns. Therapists were psychiatric residents.

Both trials examined acute phase treatment lasting 12 weeks. Thirty percent of the postpartum women in the British trial dropped out; 10 percent of the HIV-positive men in the U.S. trial dropped out.

The Findings

In the trial of 87 postpartum women, fluoxetine was significantly better than placebo in alleviating symptoms, and six sessions of counseling were significantly better than a single session. No significant interactions between drugs and counseling were seen. In the second trial of 20 HIV-positive men, fluoxetine plus psychotherapy was associated with similar improvements compared with psychotherapy alone.

Gaps in Evidence

The lack of studies comparing combinations of newer agents and psychosocial therapy with monotherapy is problematic and leaves many questions unanswered. The two identified trials have methodologic problems and do not directly address whether combination therapy is superior to newer antidepressant therapy alone. One trial lacked adequate power to detect clinically significant differences between treatment groups and used relatively inexperienced therapists. Both trials were conducted in special populations that are not easily generalized to most patients with depression. No trials addressing combination therapy (newer pharmacotherapy plus psychosocial therapy) for maintenance or relapse prevention were found.

Other Relevant Evidence

A few systematic reviews of the literature examining the efficacy of combined treatments suggest that combinations of psychotherapy and pharmacotherapy regimens are only slightly superior to either alone.^{6, 22, 23,} (See also the Cochrane Depression, Anxiety, and Neurosis Review Group's forthcoming systematic review of this literature.⁴¹ )

Context

Augmenting agents such as pindolol and lithium are sometimes added to primary antidepressant medication. They are used for patients who fail to respond to initial treatment and to speed rate of response.

The Evidence (Tables 76, 77, 78)

Five trials evaluating various augmentation strategies were found.^{194, 296, 305, 383, 449} They involved 347 patients. Four trials were conducted in Europe and one in the United States.

Most participants in the trials had major depression. Almost all were recruited from outpatient mental health settings. Two trials specifically studied patients with resistant or refractory depression.^{296, 305}

Three trials compared the addition of pindolol with a serotonin reuptake inhibitor (SSRI) with the SSRI alone.^{194, 383, 449} One trial compared maprotiline (Tetra) plus lithium with brofaromine (RIMA) plus lithium.⁶⁵³ A four-arm trial compared the addition of lithium versus placebo in patients randomized to lofepramine (TCA) or fluoxetine (SSRI).³⁰⁵

All trials had treatment durations of 6 weeks. Dropout rates ranged from 19 percent to 35 percent.

The Findings (Figure 24)

Pindolol plus an SSRI was associated with response rates of 43 percent to 75 percent compared with response rates of 55 percent to 65 percent with an SSRI alone. In this small set of trials, augmentation therapy was not significantly better than treatment with an SSRI alone. (risk ratio 1.0, 95 percent CI 0.7 to 1.4). Pindolol also was not consistently associated with more rapid time to response than SSRIs.

No significant differences were found in the small trial of refractory depressed outpatients treated with brofaromine plus lithium versus maprotiline plus lithium. The small trial that added lithium for patients who had failed to respond to fluoxetine or lofepramine also showed no significant differences.

Gaps in Evidence

Many more trials with augmenting agents are needed to determine whether recovery from a first episode of depression can be hastened and whether refractory or resistant depression can be treated successfully. The sustainability of any improvements will need to be established. Whether augmenting agents result in more adverse effects and greater numbers of dropouts than antidepressants agents alone also needs study.

Other Relevant Evidence

An earlier meta-analysis of five small heterogeneous trials found that augmentation with lithium was superior to placebo.⁵¹¹ Most of the trials used older antidepressant agents and had significant methodologic problems; results were not consistent among the trials.

Context

Because major depression has a high relapse rate, continued treatment is recommended beyond initial recovery. For treatments with established short-term efficacy, a critical test of overall efficacy is the ability to maintain remission long term.

The Evidence: Randomized Discontinuation Trials (>Tables 79, 80, 81, 82)

The best information on long-term efficacy comes from trials that study patients who respond to initial treatment and who then are randomized to continued antidepressant treatment or placebo. If the antidepressant medication is effective long term, fewer patients on active treatment will relapse. Six studies using this design involving 910 adults with major depression compared citalopram (n=2), fluoxetine (n=2), paroxetine (n=1), or sertraline (n=1) with placebo.^{206, 241, 261, 356, 358, 407} Of these, two studies did not report outcomes.^{206, 261} Two additional studies involving 128 patients compared fluvoxamine with sertraline or lithium.^{276, 277} All studies reporting outcomes were conducted in Europe, and the four with identified funding sources were sponsored by industry.

Patients were recruited from inpatient mental health settings (n=2), outpatient mental health (n=2), mixed mental health and primary care (n=1), and not described settings (n=3). In three studies, participants met criteria for recurrent major depression, placing them at higher risk for relapse.

All but one of the trials were double blind. Seven lasted at least 24 weeks with two lasting 2 years. Given the long-term followup, dropout rates were high. Dropout rates were lower than 20 percent (n=2), 20 to 30 percent (n=2), greater than 30 percent (n=1), and not given (n=3).

The Findings: Randomized Discontinuation Trials

Ten percent of participants randomized to active treatment experienced a relapse within 24 weeks, defined by a greater than or equal to 50 percent increase in depressive symptoms on an interviewer-administered rating scale (e.g., Montgomery-Âsberg) or a clinical severity equal to 1 or 2 on the Clinical Global Rating Scale. Relapse rates for participants given placebo antidepressants were 35 percent. Newer antidepressants were significantly more effective in preventing relapse than placebo (risk ratio 0.3, 95 percent CI 0.2 to 0.5). For the two placebo-controlled studies with longer followup, the advantage of antidepressant drug versus placebo was maintained at 44 and 52 weeks.

In a study of patients who had maintained recovery for 4 months prior to randomization, two SSRIs, fluvoxamine at 200 to 300 mg/d and sertraline at 100 to 200 mg/d, showed equal effectiveness. No patient had relapsed at week 24; at 104 weeks 7/32 (22 percent) sertraline patients and 6/32 (19 percent) fluvoxamine patients had relapsed (risk ratio 0.86, 95 percent CI 0.5 to 1.4). In the single study comparing fluvoxamine at 200 mg/d with lithium at 600 to 900 mg/d, fluvoxamine was more effective at preventing relapse during the 2-year followup (13 percent versus 25 percent; risk ratio 0.5, 95 percent CI 0.3 to 0.9).

The optimal dose for continuation and maintenance phase treatments is important. Among the placebo-controlled trials, three of the four studies reporting outcomes continued treatment at or above the dose to which the patient had initially responded. The one study continuing treatment at a lower dose had a higher, though not significantly different, relapse rate.

Gaps in Evidence

Efficacy for continuation phase treatment for up to 6 months has been shown for three SSRI antidepressants but not for other newer antidepressants. There is little information addressing the efficacy of newer pharmacotherapies for long-term maintenance phase treatment beyond 6 months. There are no studies comparing newer antidepressants with older agents or with psychosocial therapies.

Other Relevant Evidence

None was identified.

Context

Selections of particular antidepressant agents are often made based on anticipated tolerability and adverse effects.

The Evidence

Adverse events were assessed and reported in multiple ways. Assessment methods included checklists, scales, spontaneous reporting, and chart review. Few trials differentiated emergent from preexisting symptoms. Trials presented results in different ways: total number of symptoms, numbers of patients experiencing specific symptoms, and only symptoms occurring over a specified frequency; all assessed symptoms regardless of frequency.

Table 83. Trials reporting numbers of patients who (more...)

Table 83. Trials reporting numbers of patients who experienced the common adverse events of interest

Drug Class Comparison	Numbers of Trials
RIMA vs. SSRI	3
RIMA vs. 1st TCA	6
SNRI vs. 1st TCA	5
SSRI vs. SSRI	5
SSRI vs. 1st TCA	31
SSRI vs. 2nd TCA	3
SSRI vs. Tetra	4

Fewer than 10 percent of the trials explicitly reported suicide attempts and suicides. In all, 19 suicide attempts and 15 suicides were described in the 315 trials involving over 40,000 participants. Because of small numbers, these events were not compared between different classes. More information on suicide risk that was abstracted from literature other than randomized controlled trials is provided in the section addressing serious adverse events.

Though sexual dysfunction has been noted as an important side effect of antidepressant agents, such effects were reported in only 11 percent of the trials. The 34 trials assessed and reported sexual dysfunction in various ways including nonspecific sexual complaints, ejaculatory abnormality, decreased libido, male impotence, erectile dysfunction, and anorgasmia. Only six trials gave information by gender. Sexual dysfunction was reported with 14 different drugs across seven antidepressant classes. Moreover, data were insufficient to estimate incidence rates of types of sexual dysfunction and whether symptoms vary by agent.

The Findings

Table 84. Pooled rates and rate differences comparing (more...)

Table 84. Pooled rates and rate differences comparing SSRIs and first generation TCAs for common adverse events

Adverse Event	SSRIRate	TCA Rate	Rate Diff.a	95% CI
Anxiety	11%	9%	2%	-0.02 to 0.06
Blurred vision	6%	10%	-4%	-0.05 to -0.01
Constipation	8%	21%	-12%	-0.14 to -0.07
Diarrhea	12%	3%	10%	0.53 to 0.13
Dizziness	8%	19%	-11%	- 0.13 to -0.06
Dry mouth	18%	48%	-30%	- 0.33 to -0.23
Headache	15%	11%	3%	0.002 to 0.04
Insomnia	13%	6%	7%	0.32 to 0.08
Nausea	19%	9%	10%	0.06 to 0.11
Tremors	7%	11%	-4%	- 0.05 to -0.01
Urinary disturbance	3%	8%	-5%	- 0.08 to -0.01

a

As the rate difference is a weighted pooled estimate, it is not exactly equivalent to the difference between the two pooled rates.

Gaps in Evidence

Available randomized controlled trials provide a very limited view of adverse events associated with antidepressant use. Trials both assess and report adverse events in many inconsistent ways. Some adverse effects such as sexual dysfunction are uncommonly assessed. Trials are too short to assess long-term adverse effects. They are too small to assess uncommon but serious adverse effects.

Other Relevant Evidence

A recent meta-analysis of 162 trials compared adverse events between SSRIs and TCAs.²⁸ Findings were similar to those given above, except anxiety was more common with SSRIs than TCAs.

Context

Antidepressant treatment is not always successful. Patients sometimes discontinue or drop out of treatment because of side effects or because their depressive symptoms do not improve or even worsen. Data from short-term efficacy trials may help estimate such discontinuation rates but should be generalized cautiously to real practice settings.

The Evidence

Eighty-seven percent of the trials reported data about numbers of trial participants who discontinued treatment or dropped out. Information was usually given for total numbers of dropouts as well as for numbers dropping out due to lack of efficacy and adverse effects. Other reasons for dropouts were usually not explained.

Table 85. Numbers of trials meeting analysis restrictions (more...)

Table 85. Numbers of trials meeting analysis restrictions that reported dropout information

	Numbers of Trials
Drug Class Comparison	Adverse Events	Lack of Efficacy	Total Dropouts
RIMA vs. TCA1	22	21	22
SSRI vs. TCA1	66	58	68
SSRI vs. TCA2	10	-	10
SSRI vs. Tetra	13	11	15

The Findings

Table 86. Dropout rates due to adverse events^a

Table 86. Dropout rates due to adverse events^a

Class/Rate	Class/Rate	Diff.	CI
RIMA/5%	TCA1/11%	-5%	-0.07 to -0.03
SSRI/11%	TCA1/16%	-4%	-0.06 to -0.02
SSRI/7%	TCA2/13%	-6%	-0.12 to 0.01
SSRI/6%	Tetra/9%	-2%	-0.05 to 0.01

a

Since the rate difference is a weighted pooled estimate, it is not exactly equivalent to the difference between the two pooled rates.

Table 87. Dropout rates due to lack of efficacy^a

Table 87. Dropout rates due to lack of efficacy^a

Class/Rate	Class/Rate	Diff.	CI
RIMA/10%	TCA1/6%	3%	0.01 to 0.04
SSRI/7%	TCA1/6%	1%	0.00 to 0.02
SSRI/5%	TCA2/4%	-	-
SSRI/4%	Tetra/4%	-0%	-0.03 to 0.02

a

Since the rate difference is a weighted pooled estimate, it is not exactly quivalent to the difference between the two pooled rates.

Table 88. Total dropout rates and rate differences (more...)

Table 88. Total dropout rates and rate differences^a

Class/Rate	Class/Rate	Diff.	CI
RIMA/26%	TCA1/27%	-1%	-0.05 to 0.02
SSRI/29%	TCA1/32%	-3%	-0.06 to 0.00
SSRI/16%	TCA2/24%	-7%	-0.16 to 0.02
SSRI/19%	Tetra/23%	-2%	-0.05 to 0.02

a

Since the rate difference is a weighted pooled estimate, it is not exactly quivalent to the difference between the two pooled rates.

Only one of the drug class comparisons had statistically significant differences in dropout rates due to lack of efficacy. RIMAs had a 10 percent dropout rate due to lack of efficacy compared to with6 percent dropout rate for first generation tricyclics in 21 trials.

Total dropout rates ranged from 16 percent to 32 percent. There were no statistically significant differences in any of the four drug class comparisons in total dropout rates.

Gaps in Evidence

Gaps in the evidence include the following: not enough trials to reliably estimate dropouts in most of the drug class comparisons, incomplete reporting regarding reasons for dropouts, and little information about dropouts in patients taking therapy for longer than 12 weeks.

Other Relevant Evidence

Other meta-analyses have compared discontinuation rates of SSRIs with those of TCAs and related agents.^{29, 66, 70} The results have generally indicated that dropouts due to adverse effects are lower with SSRIs than with first generation tricyclics.

Context

There are at least three reasons that published rates of adverse effects of newer agents may underestimate the true rates that will be seen in practice. First, most of the data are from premarketing studies performed by pharmaceutical companies in carefully controlled research settings. When used in routine community practice settings, drugs may produce adverse effects more commonly. For example, no cases of serotonin syndrome occurred in 3,082 patients in premarketing trials of venlafaxine.⁵¹² However, several cases were reported after marketing.⁵¹³ The reasons for this may be several, including the greater numbers of patients exposed, the more likely it is that uncommon side effects will become known. Also, community physicians may be treating patients with more complex illnesses who have greater risk of adverse effects. Alternatively, practitioners may use the drugs for more indications, or they may not withhold the drugs from certain high-risk subgroups, which may lead to higher overall rates of adverse effects. For instance, Feighner and coauthors reported that the serotonin syndrome occurred in 75 percent of patients already taking a monoamine oxidase inhibitor (MAOI) when started on fluoxetine;⁵¹⁴ this combination of drugs is now considered contraindicated.

Second, adverse effects of newer drugs may be underreported, leading to underestimation of the true frequency. In the premarketing period, serious adverse effects must be reported to the FDA, and although the FDA may require that the adverse effect be added to the drug's labeling, the case and its details may never be published. In the postmarketing period, clinicians may not report every adverse effect they encounter. Studies have estimated that practitioners report to regulators approximately one serious adverse effect for every 100⁵¹⁵ to 4,600⁵¹⁶ that they see. An even smaller proportion of the less serious adverse events that clinicians find in practice are reported.⁵¹⁷

Thirdly, many of the published reports of premarketing trials do not describe study methods and results in detail. This lack of detail hinders the assessment of methodologic strengths and avoidance of bias, which in turn limits confidence in estimates of adverse effects.

Evidence about Uncommon but Serious Adverse Effects

This section addresses the following serious adverse effects: suicidality, complications of pregnancy, seizures, extrapyramidal effects, hyponatremia and the syndrome of inappropriate antidiuretic hormone secretion (SIADH), serotonin syndrome, immunologic and allergic effects, psychiatric and cognitive effects, cardiac dysrhythmias, drug interactions, hematologic effects, and a few additional isolated reports of even rarer effects. Results are presented only for agents approved by the Food and Drug Administration by January 1998 (several SSRIs, venlafaxine, bupropion) and hypericum (St. John's wort). There were 727 studies that were original reports of serious adverse effects; 63 were cohort studies or trials that quantified rates or rate differences. Detailed results of the 63 controlled studies are presented in Appendix 5.

The evidence of association between newer approved drugs and serious adverse effects is presented, starting with the more methodologically rigorous evidence from controlled studies and proceeding through case series and individual case reports. Reported associations cannot be automatically assumed to mean that the drug causes the adverse effect -- chance, coincidence, or confounding explanations are possible. When published case reports are examined carefully, 20 percent to 45 percent have been classified as showing "probable" causation and 50 percent to 75 percent as "possible"' causation.⁵¹⁸ Alternative causes are well excluded in less than 20 percent of reports.⁵¹⁷

In the absence of information from controlled studies, the number of cases reporting an association between a drug and an adverse effect may be of some use in judging causation. For instance, for rare conditions such as agranulocytosis, which occurs spontaneously at a rate of 6 per million people per year, even a single report of association with a new drug suggests that the association is unlikely to be due to chance alone. The more commonly a condition occurs spontaneously, the more case reports would be needed before chance association is unlikely. For instance, liver injury occurs spontaneously in about 24 per 100,000 people, so that at least 3 case reports of its occurring with a new drug would be needed before chance association would be unlikely.⁵¹⁹ To help the reader use this perspective in making decisions about causation, numbers of cases that have been reported for particular putative adverse effects are given.

Of note, uncommon but serious adverse effects from SSRIs have been studied in detail, whereas only one pharmaceutically sponsored meta-analysis has studied venlafaxine and only one cohort of 3,250 patients has systematically sought adverse effects from St. John's wort.⁵²⁰ Although 1.5 percent of the cohort stopped taking St. John's wort because of adverse effects, none of the adverse effects was severe.

The Findings Regarding Uncommon, but Serious, Adverse Events

Table 89. Definite uncommon but serious adverse effects (more...)

Table 89. Definite uncommon but serious adverse effects of newer antidepressants

Adverse Effect	Drug(s)	Magnitude of effect	Level of Evidence
Seizures	SSRIs, Bupropion	<0.05 to 0.26%	Trends and dose response relationships in controlled studies
Hyponatremia/SIADH	SSRIs	<1%	One case control study, systematic review of case reports with resolution on withdrawal and rechallenge recurrence
Serotonin syndrome	SSRIs (combined with other serotoninergic or during overdose)	Case reports with resolution on withdrawal
Extrapyramidal effects: Dystonias Parkinsonism Akathisia	SSRIs	0.4 to 13% depending on underlying risk factors.	Case reports with resolution on withdrawal
Mania in unipolar depression (as compared with placebo)	SSRIs	<1%	Significant meta- analyses
Bradycardia	SSRIs	Unknown	Case reports with one report of recurrence on rechallenge
Bleeding	SSRIs	<0.73%	Controlled study, case report with recurrence on rechallenge
Granulocytopenia	SSRIs	Unknown	Case report with positive withdrawal and rechallenge
Hepatotoxicity	SSRIs	Unknown	Case reports with resolution on withdrawal

Table 90. Equivocal uncommon but serious adverse effects (more...)

Table 90. Equivocal uncommon but serious adverse effects

Adverse Effect	Drug(s)	Magnitude of Effect	Level of Evidencea
Spontaneous abortions	SSRIs	<6%	Trends in small controlled studies
Fetal malformations	SSRIs	<4%	Trends in small controlled studies
Fetal prematurity	SSRIs	<8%	Trend in one small controlled study
Mania in unipolar depression (as compared with tricyclic antidepressants)	SSRIs	<1%	Trends in meta-analyses
Hyponatremia/SIADH	Bupropion	Unknown	Case reports
Extrapyramidal effects	Bupropion	Unknown	Case report
Pulmonary fibrosis	SSRIs	Unknown	Case reports
Tardive dyskinesia	SSRIs	Unknown	Case reports
Parkinsonism	Bupropion	Unknown	Case report

a

Trends in controlled studies were defined as increased rates compared with controls that were not statistically significant at the p<0.05 level.

Table 91. Uncommon but serious adverse effects not (more...)

Table 91. Uncommon but serious adverse effects not related to SSRIs

Adverse Effect	Drug(s)	Level of Evidence
Suicidal acts and ideation	SSRIs, bupropion	Controlled studies
Mania in bipolar depression	SSRIs	Controlled studies

Table 92. Uncommon but serious adverse effects insufficiently (more...)

Table 92. Uncommon but serious adverse effects insufficiently studied

Adverse Effect	Drug(s)
Fetal effects	Bupropion

The Findings for Specific Adverse Effects

Gaps in Evidence

Substantial gaps in evidence concerning uncommon but serious adverse effects of newer antidepressants are likely because of inadequate reporting and surveillance measures and few controlled studies.

Other Relevant Evidence

No other comprehensive rigorous systematic reviews evaluating uncommon but serious adverse effects of newer antidepressants were found.

Methodology Reviewer

• Jesse Berlin, ScD

• University of Pennsylvania School of Medicine

• Philadelphia, PA

Criticism Editor

• Patricia Huston, MD, MPH

• Ottawa Heart Institute

• Ottawa Ontario CANADA

Project Leaders

• Cynthia D. Mulrow, MD, Msc

• EPC Center Director

• John W. Williams, Jr., MD, MHS

• EPC Project Director

• Elaine Chiquette, PharmD

• EPC Project Manager

• Madhukar Trivedi, MD

• EPC Specific Content Expert

Project Staff and Consultants

• Christine Aguilar, MD, MPH

• Robert Badgett, MD

• Valerie Lawrence, MD, MSc

• Kelly Montgomery, MPH

• Polly Hitchcock Noël, PhD

• W. Scott Richardson, MD

Abstractors

• Catherine Vriend, PhD

• Jayme Trott, PharmD

• Anita McQuillen, PhD

• Lisa Archiniega, PhD

Statisticians

• John E. Cornell, PhD

• Michael Luther, MS

• Gilbert Ramirez, DrPH

• Shuko Lee-Borgess, MS

Informatics

• Michael Brand, RN

Technical Writer

• Karen Stamm

Librarians

• Gina Harris, MLS

• Martha Harris

• Linn Morgan

Computer Resources

• Richard Moreno

• Francisco Jose Gonzalez

• Dave Mullins

Administration

• Liz Oyen

• Annie M. Almanza

Andersen 1994

Arminen 1994

Bakish 1992

Bakish D, Bradwejn J, Nair N, McClure J, Remick R, Bulger L. A comparison of moclobemide, amitriptyline and placebo in depression: A Canadian multicentre study [published erratum appears in Psychopharmacology (Berl) 1993;111(3):389-90]. Psychopharmacology (Berl). 1992; 106(Suppl): S98–S101. [Free Full text in PMC] [PubMed]

Bakish D, Wiens A, Ellis J, Alda M, Lapierre YD. A double-blind placebo-controlled comparison of moclobemide and amitriptyline in the treatment of depression. Can J Psychiatry. 1992; 37(Suppl 1): 12–7.

Kusalic M, Engelsmann F, Bradwejn J. Thyroid functioning during treatment for depression. J Psychiatry Neurosci. 1993; 18(5): 260–3. [Free Full text in PMC] [PubMed]

Bakish 1993

Berman 1997

Berzewski 1997

Besancon 1993

Bowden 1993

Bremner 1996

Burke 1996

Cassano 1986

Chouinard 1983

Chouinard 1993

Claghorn 1992

Cohn 1990

Cornelius 1997

D'Amico 1990

Dalery 1992

Doogan 1994

Dubini 1997

Fabre 1987

Fabre 1995

Fairweather 1993

Fava 1995

Feiger 1996

Feighner 1989

Franchini 1994

Geisler 1992

Geisler A, Mygind S, Knudsen OR, Sloth-Nielsen M. Ritanserin and flupenthixol in dysthymic disorder: A controlled double blind study in general practice. Nord J Psychiatry. 1992; 46(4): 237–43.

Geisler A, Mygind S, Riis Knudsen O, Sloth-Nielsen M. A controlled double-blind study in general practice of ritanserin and flupenthixol in dysthymic disorder. Psychopharmacology (Berl). 1991; 103: –. [Free Full text in PMC]

Gillin 1997

Haffmans 1996

Harris 1991

Heiligenstein 1993

Jakovljevic 1996

Jenkins 1990

Judd 1993

Katona 1995

Katona 1998

Keegan 1991

Keller 1995

Kiev 1997

Kragh 1995

la Pia 1992

Lecrubier 1997

Levine 1989

Lonnqvist 1994

Lydiard 1989

Marttila 1995

Mertens 1989

Moller 1992

Moller 1993a

Moller 1993b

Montgomery 1991

Montgomery 1993

Moon 1994

Nyth 1992

Ohrberg 1992

Tollefson 1995

Tome 1997

Versiani 1990

Wernicke 1988

Wilcox 1995

Zapletalek 1990

1.5.1 MEDLINE (Uncommon but serious adverse effects)

• 1

  bupropion/ or bupropion.tw. or bupropion.rw. or 34841-39-9.rn. or wellbutrin.tw.

• 2

  fluoxetine/ or fluoxetine.tw. or fluoxetine.rw. or 54910-89-3.rn. or prozac.tw.

• 3

  fluvoxamine/ or fluvoxamine.tw. or fluvoxamine.rw. or 54739-18-3.rn. or luvox.tw.

• 4

  mirtazapine/ or mirtazapine.tw. or remeron.tw.

• 5

  nefazodone/ or nefazodone.tw. or nefazodone.rw. or 83366-66-9.rn. or serzone.tw.

• 6

  paroxetine/ or paroxetine.tw. or paroxetine.rw. or 61869-08-7.rn. or paxil.tw.

• 7

  sertraline/ or sertraline.tw. or sertraline.rw. or 79617-96-2.rn. or zoloft.tw.

• 8

  venlafaxine/ or venlafaxine.tw. or venlafaxine.rw. or 93413-69-5.rn. or effexor.tw. or "effexor xr".tw.

• 9

  ("john$ wort" or hyperic$ or johanniskraut$ or johannesort).tw.

• 10

  exp serotonin uptake inhibitors/

• 11

  or/1-10

• 12

  (ae or po or to or mo or ci or de or et or co or sc).fs.

• 13

  11 and 12

• 14

  exp Inappropriate ADH Syndrome/

• 15

  hyponatremia/ or hyponatrem$.tw.

• 16

  seroton$ syndrome.tw. or serotonin/

• 17

  exp monoamine oxidase inhibitors/

• 18

  exp seizures/

• 19

  exp suicide/

• 20

  serum sickness/

• 21

  exp fetal development/ or exp fetal death/ or exp fetal diseases/

• 22

  exp fetal heart/ or exp abortion/ or exp infant, low birth weight/

• 23

  delirium/

• 24

  exp basal ganglia diseases/ or exp basal ganglia/

• 25

  (parkinson$ or extrapyramidal or akathisia or dystonia or tardive).tw.

• 26

  exp angioneurotic edema/ or exp tongue/ or exp tongue diseases/

• 27

  or/14-26

• 28

  11 and 27

• 29

  13 or 28

• 30

  risk factors/

• 31

  exp survival analysis/

• 32

  exp death/

• 33

  exp drug interactions/

• 34

  critical illness/

• 35

  exp mortality/

• 36

  abnormalities, drug-induced/

• 37

  or/30-36

• 38

  11 and 37

• 39

  29 or 38

• 40

  ..l/39 hu=y

• 41

  (malignan$ or rare or surviv$ or risk#).tw.

• 42

  (death or fatal$ or grave or die or lethal$ or mortal$).tw.

• 43

  (adverse or serious or severe or poison$ or patholog$ or toxic$).tw.

• 44

  (threaten$ or abnormal$ or failure event# or hazard$).tw.

• 45

  or/41-44

• 46

  11 and 45

• 47

  39 or 46

• 48

  ..l/47 hu=y