Antidepressants in the treatment of depression/depressive symptoms in cancer patients: a systematic review and meta-analysis

Background Over the past thirty years a number of studies have suggested that antidepressants can be effective in the treatment of depressive symptoms in patients with cancer. The aim of this paper was to review randomized controlled trials (RCTs) and to perform a meta-analysis in order to quantify their overall effect. Methods Pubmed and the Cochrane libraries were searched for the time period between 1980 and 2010. Results Nine RCTs were identified and reviewed. Six of them (with a total of 563 patients) fulfilled the criteria for meta-analysis, but exhibited an unclear risk for bias. The estimated effect size was 1.56 with 95% CI: 1.07- 2.28 (p= 0.021). There were no differences in discontinuation rates between antidepressants and placebo groups (RR= 0.86 with 95% CI 0.47- 1.56, p=0.62). Conclusions This meta-analysis suggests that antidepressants can be effective in treating depressive symptoms beside clinical depression. When considering the risk of side effects and interactions and the heterogeneity among the mostly small studies, a general recommendation cannot be made until well-controlled studies are conducted.


Background
The role of depression in physical illness has been recognized and addressed by many authors. Up to one third of physically ill patients attending hospital have depressive symptoms. The diseases with the highest prevalence of major depression have been reported as follows (from higher to lower incidence): inflammatory bowel disease, multiple sclerosis, epilepsy, asthma, back problems, cancer, COPD, migraine, rheumatic arthritis, stroke, Parkinson's disease, diabetes mellitus, and heart disease [1]. Over the last several years, there has been a growing interest for the psychological aspects of cancer due to its severe impact on quality of life (QOL) [2,3]. Physical symptoms that are associated with both depression and cancer can be a confounding factor in the assessment of depression in this population [4]. Several studies evaluated the efficacy of psychological and pharmacological interventions in the treatment of depression. The psychopharmacological interventions and particularly the use of antidepressants are systematically reviewed here and meta-analytical methods are applied to quantify their overall effect.
Cancer is associated with depression and depressive symptoms. A meta-analysis by Mitchel et al. [5] included 94 studies and found that the pooled prevalence of major depression in palliative care settings and haematooncological settings was equal to 16.5% (95% CI: 13.1-20.3%) and 16.3% (95% CI: 13.4-19.5%), respectively. Restricting the analysis to standardized clinical assessment, Ng and colleagues [6] found a prevalence of 10.8% and also a substantial amount of heterogeneity. Even though these rates appear to be high, a consistent difference to a systematically matched sample from general population data is still controversial. The existing data suggest that "survivorship presents ongoing psychological challenges" [7].
There are two main confounding factors in the assessment of depression in cancer patients. First, the distinction between normal sadness or grief and symptoms indicating a depressive episode is not well-defined. Indeed, a phase of reduced mood or depression is considered part of healthy coping with grief (e.g. Kübler-Ross already in 1969 [8]). Further, such reaction patterns may recur as the disease progresses, by treatment failure, or by findings of metastases. Therefore, also time criteria may not capture the dynamics of disease progression.
A second confounder is the lack of specificity of the depressive symptoms. The ICD-10 and DSM-IV criteria for depressive episode include symptoms that are often present in patients with cancer as well, e.g. loss of appetite, low energy levels, or sleep disturbance. Therefore, the definition of depression in cancer patients and in physical illness is ambiguous [4]. It is suggested to identify the patients by their symptoms and not by a clinical syndrome because the ability to detect cases of depressive episode or disorder may be less important than the ability to detect depressive symptoms remediable to treatment [9]. Several approaches have been developed to solve the problem of diagnosing depression in cancer patients [10]. For instance, the substitutive approach suggests that all physical/somatic symptoms (change in appetite/weight, sleep disturbance, fatigue, loss of energy, diminished ability to think or concentrate) are replaced by non-somatic symptoms (tearfulness, depressed appearance, social withdrawal, decreased talkativeness, brooding, self-pity, pessimism, lack of reactivity, blunting) [11]. These are known as the Endicott criteria. Therefore, established criteria for depression may not be better suited to detect therapeutic indications in cancer patients than the presence of depressive symptoms.
Many guidelines for the treatment of cancer recommend that all cancer patients should be screened for depression, pain, and fatigue (e.g. by the National Institute of Health [12]). Multi-item scales are used for screening and diagnosing depression. Quantified results are used to specify the illness severity and to monitor the course of the disease. For detecting depression, ultra brief screening tools have been developed and proven to be reliable. For instance, Chochinov reported that the single question: "Are you depressed?" can be a reliable screening tool [13]. There is clear evidence that the systematic application of screening instruments reduces false negative findings, but the specificity and effects on outcome measures not sufficiently studied [14].
Depression seems to influence the prognosis and even the survival of cancer patients. For instance, Satin et al. conducted a meta-analysis with 27 studies on mortality in cancer patients and depression. A significant effect of depression on mortality was reported (RR: 1.25, P<0.001). The majority of studies measured the effect of depressive symptoms and only three of them included patients with clinical depression. A correlation with disease severity cannot be excluded [15]. In a large cohort study, patients who had recently received a cancer diagnosis had an increased risk for both suicide and death from cardiovascular causes as compared with controls [16]. Trials found that a decrease in severity of depressive symptoms is associated with a prolongation of survival in cancer patients [17,18]. However, these findings remain controversial as the majority of studies failed to replicate them [19][20][21][22].

Treatment of depression in patients with cancer
The treatment of depression can mainly be divided into two categories: psychosocial and pharmacological interventions. A meta-analysis found that cognitive behavioral therapy has a positive effect on depression and quality of life in patients with cancer. In the same meta-analysis patient education had a positive influence on quality of life but not on depression [23]. A further meta-analysis of psychotherapeutic and psychopharmacological studies found allover positive effects on depression ratings [24]. Metaanalyses focusing on pharmacological treatment give a less consistent picture.
Previous reviews [6,[25][26][27][28] underpinned the lack of evidence of the adequate effect of pharmacological interventions in the treatment of depression in cancer patients. However, the reviewed studies were heterogeneous as concerns the studied population (e.g. fatigue or pain as eligibility criterion and depression as secondary outcome) and the type of the drug applied (e.g. antidepressants, benzodiazepines, antipsychotics, psychostimulants, etc.). Only limited conclusions could be drawn from these reviews for the effectiveness of antidepressants in this population. A metaanalysis estimated the efficacy of antidepressants in palliative care (patients with cancer, HIV, COPD, etc.; [29]). The overall effect of antidepressants was significantly higher than the effect of placebo. However, only four of the twenty-five studies included cancer patients. A subgroup analysis was not performed for each subpopulation and thus no recommendation can be given for oncological patients. In another meta-analysis, antidepressants were found to be effective in the treatment of major depression with a co-morbid physical illness (RR= 1.42, P<0.0001). Again, only four RCTs were included that studied cancer patients with a diagnosis of major depression and no significant effect of antidepressants on response rates emerged in this subpopulation (RR=1. 26, P=0.19) [30]. In contrast, Hart et al. [24] found a significant effect on depression ratings in the subgroup of four pharmacological studies which was not significantly different from the overall effect of the psychotherapeutic trials. However, this analysis included one placebo group twice and used Hedge's g to quantify the results, which may bias statistics in the pharmacological subgroup. The authors discuss as a limitation that changes in questionnaire ratings may have limited clinical relevance. Depressiveness even without manifest diagnosis of depression may have adverse effects on prognosis and quality of life in cancer patients (see [14]) and, therefore, these patients should be included in intervention trials and subsequent meta-analyses. To overcome the limitations of the previous analyses, the present systematic review and metaanalysis focuses on the event of clinical relevant symptom changes in depressed or depressive patients with cancer.

Search strategy
The aim of the present study was to determine whether antidepressants are effective in the treatment of depression and depressive symptoms in patients with cancer. The inclusion criteria for the studies were: 1. Double-blind randomized-controlled trials (RCTs), which could be placebo-controlled or head-to-head trials. For the purposes of the meta-analysis only placebo-controlled trials were used. Antidepressants are often used for indications other than depression (e.g. fatigue, pain, hot flashes) in patients with malignancy. Thus, we excluded all studies, which had depression as a secondary outcome only.
We searched for studies in the electronic databases Pubmed and the Cochrane Library. We aimed towards higher sensitivity and lower precision in this first selection in order not to miss an appropriate study. In particular, we omitted any search term for therapy or treatment, which could reduce the search sensitivity. This approach is suggested by the "Cochrane Handbook for systematic Reviews of Interventions" ( §6.4.4) [31]. Search terms were: "(depressive OR depression) AND (cancer OR tumor OR neoplasm OR lymphoma OR leukemia)". The applied limits of the search were 1. articles should be published in the time between 01.01.1980 and 31.12.2010; 2. articles should be in English language; and 3. the search term appeared in the title or the abstract of the articles. We further searched through the reference lists of reviews and relative articles to identify any additional studies. Exploratory extensions of search terms (e.g. including 'oncology') did not yield additional studies.

Article selection and review strategy
The selection of studies involved an initial screening of the title and the abstract in order to find studies, which were appropriate according to the inclusion criteria stated above. If it was not clear from the title or the abstract that the study should be rejected, the full text was obtained. The process was conducted independently by both authors in order to reduce the possibility of relevant articles being rejected.
The data were extracted independently by both authors. In case of disagreement, a clinician experienced in psychooncology and liaison psychiatry could be involved to mediate consensual decisions. A structured format was used as the one applied in the presentation of the single studies in the appendix. Dichotomous data were collected for the primary outcomes of this review (responders and nonresponders to treatment). Secondary outcomes were the number of drop outs, the number of patients with adverse effects, and the quality of life.

Statistical methods (meta-analysis)
A random effects model was applied in the metaanalysis because of the assumption that the true effect size was not the same in all studies. Indeed, there were marked differences between the studies regarding the type of cancer, the stage of cancer, the drug used in each study, and the design (intention to treat analysis or completers' analysis). The risk ratio (RR with 95% confidence intervals) was preferred to odds ratio for the computation of the effect size because it has the advantage of being more intuitive [32]. Heterogeneity I 2 was computed in order to assess the percentage of the overall variability attributed to the between studies variability.
The risk of bias in individual studies was evaluated using the Cochrane Collaboration' s domain based tool which assesses allocation concealment, sequence generation, blinding, selective outcome reporting, and other sources of bias. Risk of publication bias was assessed using a funnel plot, i.e. a display of estimated study quality in terms of standard error and the reported effect size. The calculations were performed using standard formulas [32] in MicroSoft Excel (Excel 2003 Edition, MicroSoft, Redmond, CA). The statistical program "Comprehensive meta-Analysis" (2 nd version, Biostat, Englewood, NJ) was used to create forest and funnel plots.

Search results
The electronic searches yielded 5959 references from MEDLINE and 1041 references (clinical trials) from the Cochrane Library. After the initial scanning of the abstracts, a total of 38 reports were detected that may relate to drug trials using anti-depressants. Based on the full-text of these reports, 29 of them were rejected since they did not reported RCTs on anti-depressant treatment in depressive cancer patients Figure 1. From the remaining 9 RCTs, 3 studies were head-to-head trials, i.e. active drugs were compared with each other [33][34][35]. Thus in total 6 randomized placebo-controlled studies fulfilled the criteria for this meta-analysis [36][37][38][39][40][41]. Table 1 provides an overview of the reviewed studies. The complete list of the assessed trials is presented in Appendix A.
Previous reviews and meta-analyses exhibited a larger diversity of study designs [25][26][27][28]. For instance, we did not include 3 trials that had not depression or depressive symptoms as an eligibility criterion even though the primary outcome measure was improvement in depression/ depressive symptoms [42][43][44]. Similarly, we did not include trials which tested the efficacy of antidepressants in preventing depressive symptoms in patients with cancer [45] or in patients with melanoma undergoing therapy with interferon [46]. Appendix D lists all the 38 trials, which were screened and the reasons for in-or excluding them.

Review of RCTs in depression
A. Head-to-head trials Holland et al. [33] studied 38 patients with depression and breast cancer. The selective serotonine reuptake inhibitor (SSRI) fluoxetine was not found to be superior to the tricyclic antidepressant (TCA) desipramine. Similarly, Pezella et al. [34] found no significant differences in efficacy between paroxetine (SSRI) und amitryptiline (TCA) in a sample of 185 patients with breast cancer. In contrast, the noradrenergic and specific serotonergic antidepressant (NaSSA) mirtazapine had a larger effect on depression than imipramine in a sample of 53 cancer patients with depression (Cancurtaran et al. 2008; [35]). B. Placebo-controlled studies Costa et al. [36] and van Heeringen et al. [37] compared the tetracyclic antidepressant mianserin with placebo in 73 and 55 patients with gynecological tumors, respectively. Both publications reported a significant effect on the observed depressive symptoms. In contrast, a trial by Razavi et al. [38] in 91 patients with various types of cancer did not reveal a significant difference between fluoxetine and placebo. Similarly, Fisch et al. [39] failed to demonstrate an advantage of fluoxetine over placebo in a sample of 163 patients with advanced cancer. A reevaluation of the results of the former study with the generalized estimating equation (GEE) method of regression suggested a significant effect of the verum as well. Navari et al. [40] found a significant effect of fluoxetine in 193 depressive patients with breast cancer. Finally, Musselmann et al. [41] could not document a drug effect in a trial with a small number of patients (n= 35) and 3 groups (paroxetine, desipramine, placebo). For the purposes of the current metaanalysis, we created a combined intervention group, which included the patients from both the paroxetine and the desipramine group, as recommended in the Cochrane handbook [31].

Meta-analysis Effect size
All studies defined a measure of response, i.e. what was considered a meaningful improvement of the depressive symptoms. The overall effect size in the analysis is RR=1.56 with 95%-CI: 1.07-2.28 (p= 0.021), i.e. under the antidepressants a therapeutic response (as defined in the considered studies) is about 50% more likely than in the placebo group (see Table 2). A graphical display of the relative strength of each study is presented in the forest plot ( Figure 2). Four studies found a positive effect of the antidepressants on depressed cancer patients. In the other two studies no significant difference emerged but the 95% confidence intervals were wider than those of the four other studies ( Figure 1). This can be considered as indicative of low precision in the trials with negative finding.

Heterogeneity
The meta-analysis revealed a substantial heterogeneity I 2 = 71% with 95% CI: 54%-82%. For a substantial I 2 (50-90%), the "Cochrane Handbook for systematic Reviews of Interventions" [31] recommends a reanalysis without the outlying studies as part of a sensitivity analysis. Indeed, the RR of the study by Navari et al. [40] is 7,000 potential relevant references identified according to the search criteria 6,962 excluded (as irrelevant)

articles retrieved in full text
For detailed evaluation 29 excluded (reviews, uncontrolled studies-see list in the Appendix) 9 studies included in the review Figure 1 Flow diagram of the study. The electronic searches provided a total of 7000 references from MEDLINE and from the Cochrane Library. After the initial scanning of the abstracts a total of 38 reports remained. These reports were further screened and assessed for eligibility and 29 of them were rejected. The remaining 9 RCTs fulfilled the inclusion criteria for the review and six of them fulfilled the criteria for the meta-analysis.
much higher than all other values (see Figure 2) and the analyses were repeated excluding this study. The effect of the antidepressants remained significantly better in comparison to placebo after excluding the outlying study (RR = 1.39, 95%-CI: 1.09-1.77, p= 0.008) and the heterogeneity decreased to 10% (95%-CI 0-22%). Heterogeneity between 0% and 40% is considered to be of no importance [31]. This finding confirms that the high heterogeneity is most likely due to one outlier which however does not bias the finding.

Adverse effects and dropouts
There was a substantial amount of missing data concerning the adverse effects in these studies. Only three studies reported the total number of patients with side effects. Four studies provided data about the number of drop outs because of side effects in each arm (Table 3). Visual inspection suggested no difference. However due to missing data, we did not perform a meta-analysis for the adverse effects. All but one studies provided information about the number of dropouts in each arm (Table 4). We performed a meta-analysis using relative risk ratios and found no significant difference between dropouts in the verum and placebo groups (mean RR = 0.86, 95%-CI 0.47-1.56, p = 0.62).

Quality of life
Only three of the six studies included an outcome measure for quality of life. Razavi et al. [38] used the Spitzer Quality of Life Index (SQOLI). The increase in the SQOLI scores was significant in both the drug and the placebo group, but the difference between the two groups was not statistically significant. Fisch et al. [39] used the Functional Assessment of Cancer Therapy-General (FACT-G, version 3). There was no significant difference between the fluoxetine and the placebo group in the proportion of responders (six points change). Using the generalized estimating equations (GEE) method of regression (post-hoc), there was a significant improvement in the total FACT-G scores in the fluoxetine group compared with placebo. Navari et al. [40] also  Figure 2 Forest plot of RR with CI for all studies and overall. The overall effect size in the analysis is RR=1.56 with 95%-CI: 1.07-2.28 (p= 0.021). This means that the effect of antidepressants in this population is significant better than the placebo effect. Four studies found a positive effect of the antidepressants on depressed cancer patients. In two studies the antidepressant was not better than the placebo. The 95%-CIs of these two studies were wider than the ones of the other four studies, which is indicative of low precision. RR: relative risk; CI: confidence intervals.
used the FACT-G scale. The number of patients who had a significant improvement in quality of life was statistically significantly higher in the fluoxetine group as compared to the placebo group.

Risk for bias and publication bias
The risk of bias for each study can be determined by assessing the following six domains: 1. sequence generation, 2. allocation concealment, 3. blinding, 4. missing data, 5. selective outcome reporting, and 6. other sources of bias [31]. The group of studies was relative homogenous and the overall risk for bias could be described as "unclear" (Figure 3). The results for every single trial are presented in the Appendix B. The risk for publication bias (i.e. studies with small sample size are more likely not to be published if their effect is small to moderate) is assessed by means of the funnel plot, which displays the relationship between the sample size and the effect size of the studies. The standard error instead of the sample size is usually used in the Y axis. No indication for publication bias can be derived from the present funnel plot; in particular, there was no gap on the bottom left side, which would be indicative of unpublished studies with small to moderate effects ( Figure 4).

Discussion
This is the first systematic review with meta-analysis which focuses exclusively on the psychopharmacological treatment in cancer patients with depression. Treatment with SSRI or tetracyclic antidepressants was found to improve depressive symptoms more than placebo. The small number of studies and patients included, as well as the questionable risk of bias, however, points out the demand for well-conducted trials before general recommendations can be derived. The diagnosis and treatment of depression is of high importance in this population of patients because of the high risk for suicide [16], its impact on the quality of life [2,3], and its influence on anticancer treatment adherence and compliance [47].
Critical for an antidepressive treatment in this group of patients is a well-funded base of evidence since elevated risks of adverse effects and interactions must be expected (see adverse effects and side effects below). Interestingly in some cancer patients, subsyndromal depressive symptoms (with DSM-IV or ICD-10 diagnosis) may improve under the antidepressants as well. Direct comparisons between classical tricyclic antidepressants and SSRIs revealed no differences in two of the three reviewed head-to-head trials. The higher effect size of the tetracyclic agent mianserin in comparison to SSRIsas seen in the subgroup analysismust therefore be considered exploratory. At this point no specific recommendation concerning effectiveness for a substance class can be made. Several studies and meta-analyses have reported on the effectiveness of antidepressants in patients with depression and physical illness. Van der Feltz-Cornelis et al. [48] showed in a meta-analysis that pharmacological interventions are effective in reducing depressive  symptoms in patients with diabetes mellitus. Price et al. [49] found in another meta-analysis a significant effect of antidepressants in treating depression in patients with neurological disorders. There are RCTs which show a positive effect of SSRIs on depression in patients with asthma [50,51]. SSRIs were also found to be effective in the treatment of depression in patients with coronary artery disease [52]. Although the prevalence of depression in patients with inflammatory bowel disease (IBD) is high [1], there are no RCTs which assess the treatment of depression in this population [53].

Depression and depressive symptoms
There is a trend towards identifying depressive symptoms instead of trying to define exact diagnoses of concrete depressive syndromes [9,10]. There are also many simple single-or two-item screening tools, which can detect depression with high specificity and sensitivity. As shown in the subgroup analysis, patients with depressive symptoms can benefit from the use of antidepressants exactly like the patients diagnosed with major depression. On the one hand, this conclusion is of significant clinical importance because it addresses a practical issue and can motivate physicians to screen for depression with simple and easy to use tools. On the other hand, it must be taken into consideration that there was only one study in one of the two compared groups in the subgroup analysis [39]. The authors of this study reported a better response in the patients who had a score higher than 4 in the TQSS, suggesting that a minimum of symptom severity may be required for the antidepressive action of antidepressants. Thus, the results of the subgroup analysis should be interpreted very carefully and not be misinterpreted as an excuse for an unreasonable use of antidepressants by physicians or for the limitation of the role of the consultationliaison psychiatry in oncology.

Side effects and interactions
When using an antidepressant, one should pay attention on possible side effects such as pro-emetic effects of SSRIs and anticholinergic effects of TCA. Nausea is a common adverse effect among cancer patients undergoing chemotherapy and  Figure 3 Risk of bias graph. The semaphore colors provide a visual impression of the quality of the study reports for meta-analysis; green: condition is fulfilled; yellow: condition is questionable and; red: condition is not fulfilled and risk of bias is present. The allover quality is unclear and indications for risk of bias can be derived. Therefore the meta-analysis cannot provide a high degree of level of evidence. can be worsened by SSRIs. Cognitive impairment or acute psychiatric conditions such as delirium can also get worse through the anticholinergic properties of TCAs. Adverse effects such as agranulocytosis with mianserin [54] should be taken into consideration in the treatment of cancer patients who receive chemotherapy. There are also many interactions between antidepressants and drugs used in the treatment of cancer. The best studied interactions are these between SSRIs and tamoxifen (a Selective Estrogen Receptor Modulator or SERM), which is metabolized by CYP2D6 into its active form endoxifen [55]. Antidepressants such as paroxetine and duloxetine can inhibit the CYP2D6 cytochrome and thus the formation of the active metabolite endoxifen [56].

Limitations
The heterogeneity between studies was substantial (I 2 = 71% with 95% CI: 54%-82%). As recommended in the Cochrane guidelines (see above), the meta-analysis was recalculated excluding an outlying value (as part of a sensitivity analysis). The efficiency remained significant even after removing an extreme high value (M*= 1.39, p= 0.008). The heterogeneity fell to 10% which is considered to be of no importance. Thus, heterogeneity affects psychopharmacological studies in cancer patients. Nevertheless, the overall therapeutic effects seem to be consistent across studies.
The reliability of these results is limited by the small number of randomized controlled studies. A larger number of studies are needed to get safe conclusions. The small number of studies and participants can be attributed among other reasons to the preference of the patients and the clinicians for non-blinded treatment, as reported by Musselmann et al. [41]. This may explain the large number of open label studies found in the search in the databases (which are not presented here). Another factor which limits the validity of our results is the quality of the studies. The average risk for bias in these studies could be described as "unclear." As shown in the risk for bias graph (Figure 3), the group of studies was relatively homogenous as regards to this issue. Other limitations of this meta-analysis are the use of different depression rating scales and the different response criteria used by the authors.

Conclusions
Considering the high prevalence of depression and its impact on mortality and quality of life in cancer patients, it is a matter of concern that only a few trials assessing antidepressant efficacy are available. Given this limitation, we found that antidepressants are effective in the treatment of depression or depressive symptoms in patients with cancer. A minimum of depressive symptoms' severity may be required for the patients to benefit from the use of antidepressants. Though a larger effect size of mianserin in comparison to SSRIs in the subgroup analysis has been shown, no recommendation can be made for one antidepressant type over another. A quantification of tolerability, as ascertained by comparing the number of patients with adverse effects, was not possible because of the missing data. The number of drop outs did not differ significant between the intervention and the control group.
There are difficulties in defining the diagnosis of clinical depression in cancer patients. Symptoms such as fatigue, sadness, worry, and pain are reported by depressed patients as well as in patients with advanced disease. Practical issues such as the ability of physicians to recognize patients with depression should also be considered. Reliable single-or two-item questionnaires have been developed for this purpose. The detection of depressive symptoms might be more important than the exact diagnosis of clinical depression. The current metaanalysis suggests further that antidepressants could be effective in treating sub-clinical depression. However, studies with larger samples are needed in order to verify such conclusion before general clinical recommendations can be derived.

Appendix A
Listing of the studies After scanning 7,000 references and screening 38 articles that seemed appropriate 9 trials could be identified to fulfill the study inclusion criteria. They were further categorized into 2 clusters: A. Head-to-head trials Three RCTs compared two antidepressants with each other (head-to-head trials). These studies did not include a placebo control group. Response criteria: a statistically significant baseline to endpoint change in HAM-D. Design: double blind RCT with ITT analysis, LOCF approach.
Results: There was a significant improvement in the HAM-D scores in both group as evidenced by baseline-to-endpoint changes (p<0.001). ANOVA showed no significant difference between fluoxetine and desipramine. Similar improvement was also observed in the HAM-A. There were no significant differences in the dropout rates in the fluoxetine and desipramine group (   "No"

Selective outcome reporting
Protocol is not available. All prespecified outcomes of interest are reported in the pre-specified way.

"Yes"
Other sources of bias The study appears to be free of other sources of bias. "Yes"

Sequence generation
Randomized study, randomization method not described.

"Unclear"
Allocation concealment Method is not described. "Unclear" Blinding of participants, personnel, and outcome Double blind study. More frequent anticholinergic effects in the fluoxetine group. Issue is not sufficiently addressed by the authors. "Unclear"

Incomplete outcome data
The main outcome war the raw baseline to endpoint differences in HAM-D. There were 6 dropouts in the fluoxetine group, all due to AEs. There were 7 dropouts in the desipramine, 4 due to AEs, 3 to unknown reasons. Analysis according to ITT principle. Missing data imputation method: LOCF. "No"

Selective outcome reporting
Study protocol is not available. All pre-specified outcomes are reported in the pre-specified way. No response criteria were defined. Improvement was not pre-specified.

"No"
Other sources of bias The study seems to be free from other sources of bias.
"Yes" Response criteria: The authors reported on the statistical significance of the rating differences between baseline and endpoint in each group. Design: double blind RCT (not blinded for those who denied medication and received only psychotherapy), completers' analysis.
Results: The patients in the mirtazapine group improved significantly after the six weeks in the total HADS scores (p=0.014), the anxiety subscale (p=0.04) and the depression subscale (p=0.008).

Sequence generation
The randomization method is not described. "Unclear" Allocation concealment Exact method is not described. "Unclear" Blinding of participants, personnel, and outcome Double blind trial. The issue is not adequately addressed. "Unclear" Incomplete outcome data The main outcome was the HDRS score. 7 dropout in the mianserin group (MG) and 15 in the placebo group (PG). Reasons for withdrawal: 1 in each group due to AEs, 2 in PG due to lack of efficacy, the treatment by one patient in the PG was interrupted by the investigator, 2 in MG and 4 MG ended the anticancer treatment, 2 in MG due to cancer complications, 1 in MG and 2 in PG due to temporary withdrawal from the anticancer treatment, 2 in PG refused anticancer therapy and were dismissed, 3 in PG had problems at home and 1 in MG died. The authors used an ITT analysis. Missing data imputation method: LOCF approach. Proportionally about one third (30%) of the patients were dropouts, which can induce bias in intervention effect estimate.

"No"
Selective outcome reporting No study protocol available. All pre-specified outcomes are reported in the pre-specified way. The response criteria are not pre-specified.

"No"
Other sources of bias The study seems to be free of other sources of bias. "Yes" Sequence generation Randomization method is not described. Patients who refused to take medication formed a control group. The allocation by preference of the participants is problematic in randomized studies. "No"

Allocation concealment
Not described for the two drug groups. No blinding for the control group. "No" Blinding of participants, personnel, and outcome No blinding for the control group. "No"

Incomplete outcome data
The main outcome was the HAM-D score and single symptom scales score for nausea, pain, vomiting. 4 Dropouts in the mirtazapine group, 4 in the imipramine group, ten dropouts in the control group. No ITT analysis. Missing data imputation method: completers' analysis. "No"

Selective outcome reporting
No study protocol available. All prespecified outcomes are reported in the pre-specified way. No prespecified criteria for response or improvement.

"No"
Other sources of bias The study seems to be free from other sources of bias.

"Yes"
The patients in the other two groups did not show any significant improvement.

B. Placebo controlled trials
The pharmacological agents that were used in these six studies were: mianserin (two trials), fluoxetine (three trials) and paroxetine (one trial). Musselman et al. used an additional third group with cancer patients receiving desipramine.
Mianserin Costa 1985 and van Heeringen 1996 compared mianserin with placebo. Both reported a significant improvement of the depressive symptoms. Mianserin is a tetracyclic antidepressant agent and is considered as the predecessor of mirtazapine.

Allocation concealment
The exact method is not described. "Unclear" Blinding of participants, personnel, and outcome Double blind trial. The authors do not address this issue. "Unclear" Incomplete outcome data The main outcome was the number of patients with success criteria (HADS score≤8) and with response criteria (≥50% improvement in HADS score). There were 15 dropouts in the fluoxetine group (FG) and 7 dropouts in the placebo group (PG). The reasons for dropouts in the FG were: 7 due to AEs, 3 decided to interrupt their participation for unknown reasons, 1 due to alcohol abuse, and 4 for other reasons: "Unclear" (Non-compliance, investigator's decision, lost to follow-up). The reasons for dropouts in the PG were: 2 due to concomitant events, 4 for unknown reasons, 1 for psychiatric reasons. The authors used an ITT basis for the success and response rates. The exact missing data imputation method is not being reported.
Selective outcome reporting No study protocol available. All pre-specified outcomes are reported in the pre-specified way. "Yes" Other sources of bias The study seems to be free of other sources of bias. "Yes"

Sequence generation
The randomization method is not described "Unclear"

Allocation concealment
The exact method is not described. "Unclear" Blinding of participants, personnel, and outcome The study is defined as double blind. The issue is not addressed by the authors. "Unclear" Incomplete outcome data The main outcome was the HDRS score. There were 6 drop outs in the mianserin group and 15 dropouts in the placebo group. 2 patients in the mianserin group and 11 in the placebo group withdrew due to lack of efficacy. 2 dropouts in the mianserin group and 4 in the placebo group due to AEs. ITT analysis. Missing data imputation method: LOCF approach. Over one third of the patients withdrew from the study (38%), No study protocol available. All prespecified outcomes are reported in the pre-specified way. "No" Selective outcome reporting No study protocol available. All pre-specified outcomes are reported in the pre-specified way. "Yes" Other sources of bias The study seems to be free of other sources of bias.

Sequence generation
The randomization method is not described. "Unclear"

Allocation concealment
The exact method is not described. "Unclear" Blinding of participants, personnel, and outcome Double blind trial. The issue is not addressed by the authors. "Unclear" Incomplete outcome data The main outcome was the scores on FACT-G and BZSDS. 193 patients enrolled in the study, 183 were available at the first follow-up and 180 at the second. The reasons for dropouts are not reported. The authors used a completers' analysis, which was not pre-specified in the description of the study.

"No"
Selective outcome reporting The scores of the FACT-G and BZSDS are not reported. The results are presented as numbers of patients with significant improvement, which is not pre-specified in the description of the study. The AEs are also not reported.

"No"
Other sources of bias The study seems to be free from other sources of bias. "Yes"

Allocation concealment
The exact method is not described. "Unclear" Blinding of participants, personnel, and outcome "The issue is not addressed by the authors. "Unclear" Incomplete outcome data 163 patients were randomized and 159 allocated to receive medication. The patients were included in the analysis if they provided data at least two assessments (baseline and one of the next four). 64 patients were evaluable in the fluoxetine group and 65 in the placebo group.
The reasons for dropouts are not fully presented. The authors used a modification of completers' analysis. The missing data imputation method was the best change score, which is defined as the difference between the baseline score and the average of the best consecutive scores. According to the authors' opinion this is a valid statistical method for longitudinal data. To our opinion the best change score belongs to the inappropriate imputation methods. "No" Selective outcome reporting No study protocol available. All pre-specified outcomes are reported in the pre-specified way. "Yes" Other sources of bias There were many loss of data especially at the fourth assessment. This could influence the intervention effect estimate.
"No"  Figure 5 Risk of bias graph for all 9 studies reviewed. Most studies did not describe the methods used to generate and conceal the allocation sequence. All studies were defined as double blind, but the exact blinding method was not described in any of them. No study used an appropriate method to address the issue of missing data. As one can see the studies were relative uniform as far as the issue of risk of bias is concerned.

Sequence generation
The randomization method is not described "Unclear"

Allocation concealment
The exact method is not described. "Unclear" Blinding of participants, personnel, and outcome Double blind study. The issue is not addressed by the authors. "Unclear" Incomplete outcome data The main outcome was the number of patients with response (≥50% improvement in the HAM-D scale) and with remission (HAM-D≤7). There were 14 dropouts in a total of 40 participants (40%). Reasons for dropouts were: AEs (2 in paroxetine group, 1 in desipramine group and 2 in placebo group), lack of efficacy (2 in paroxetine and 2 in placebo group), patients' wish to discontinue (2 in placebo group), one was lost to follow-up and one from the placebo group could not ingest any medication. The analysis was done on an ITT base. The missing data imputation method was the LOCF. "No" Selective outcome reporting No study protocol available. All pre-specified outcomes are reported in the pre-specified way. "Yes" Other sources of bias Small number of participants. "No" administered Depression Rating Scale (BZDRS) with 11 items. Inclusion Criteria: A TQSS score of 2 or greater. Patients with a major depressive episode were excluded. Response criteria: A best change score of at least −3 in the BZDRS. A best change score is defined as the difference between the baseline score and the average of the best consecutive scores. Design: Double blind RCT. Computations were made only on patients who completed the baseline questionnaires (n=159) and at least one follow up assessment (n=129). The authors used additionally the generalized estimating equation (GEE) method of regression. Results: There were data for 129 patients at the end point (64 in the fluoxetine group and 65 in the placebo group). 48% (n=31) patients in the fluoxetine group and 36% (n=23) in the placebo group were responders. This difference was not statistically significant. Reevaluating the data with the GEE method of regression the authors found a significant improvement in the fluoxetine group. There were not any concrete data about the number of dropouts or reported adverse effects in each group.

Navari et al., 2008
Number of patients: 203 patients out of the 357 screened patients qualified for enrolment in this study. From the 193 who enrolled in the study the authors reported on the 180 patients who completed the six month assessment. Type of cancer/Stages: Breast cancer, stages I, II Duration: six months Screening tool: The Two Question Screening Survey (TQSS) was used to assess mood and anhedonia. The two questions are: "During the past month have you been bothered by feeling down, depressed or hopeless?" and "During the past month have you been bothered by having little interest or pleasure in doing things?" There five possible answers which are assigned values from 0 to 4: 0 not at all, 1 a little bit, 2 somewhat, 3 quite a bit, 4 very much.  The quality of life was estimated using the Functional Assessment of Cancer Therapy-General (FACT-G) and the depression was estimated with the Brief Zung self administered Depression Rating Scale (BZDRS). Inclusion Criteria: A TQSS score of 2 or greater. Patients with a major depressive episode were excluded. Response criteria: "significant improvement" in the BZSRS (not further defined). Design: double blind RCT, completers' analysis. Results: 71 patients from the fluoxetine group and 23 patients from the placebo group had a significant improvement in depressive symptoms (P<0.0005). There were not any data about adverse effects. Paroxetine One study compared paroxetine with desipramine and placebo in depressed cancer patients.   patients) in the desipramine group and 55% (6/11 patients) in the placebo group (p= 0.91). The remission rates were: 23% (3/11 patients) in the paroxetine group, 45% (5/11 patients) in the desipramine group and 36% (4/11 patients) in the placebo group (p=0.55). There were 14 dropouts by week six: 5 in the paroxetine group, 4 in the desipramine group and 5 in the placebo group.