Criteria for considering studies for this review

Search methods for identification of studies

(1) Electronic search of CCDAN REGISTERS
We will search the Cochrane Collaboration Depression, Anxiety and Neurosis Registers (CCDAN REGISTERS) in December 2004. This comprehensive register is updated regularly adding the results on searches of The Cochrane Library, MEDLINE (1966‐), EMBASE (1980‐), CINAHL (1982‐), PsycINFO (1974‐), PSYNDEX (1977‐) and LILACS (1982‐1999) and handsearches of major psychiatric, medical journals and conference proceedings. It contains more than 24,000 records on trials comparing treatment options within the scope of the CCDAN.
Moreover, the register is being coded continuously with respect to characteristics of studies such as the kind of interventions and their concomitant use, by looking through the full article of relevant studies manually. The studies already coded are being stored in CCDANCTR‐Studies (Table 1), and the others in CCDANCTR‐References (Table 2). Therefore, we will search the relevant studies in these registers.

(2) Complementary electronic search
We will undertake a search of the following electronic databases:
MEDLINE (1966 to December Week 2 2004) (Table 3)
EMBASE (1980 to December Week 2 2004) (Table 4)
PsycINFO (1974 to December Week 2 2004) (Table 5)
The Cochrane Central Register of Controlled Trials (CENTRAL) (Table 6)

(3) Reference search
We will inspect the references of all selected studies for more published reports and citations of unpublished research.
In addition, we will check relevant review papers.

(4) SciSearch
All the selected studies will be sought as a citation in the SciSearch in order to identify more studies.

(5) Personal communication
To ensure all RCTs are being identified, we will contact the authors of the trials included, those of significant papers and other experts in the field.

Data collection and analysis

(1) Selection of trials
Two reviewers (NW and TAF) will examine titles and abstracts of references identified by the electronic search strategies described above in order to see if the study is likely to be relevant. Full articles of all the studies identified by either of the two reviewers will be then obtained and inspected by the same two reviewers for trials meeting the following broad and simple criteria:
(a) Randomised trial;
(b) Major depression, minor depression or dysthymia as diagnosed by operationalized criteria; or inclusion criteria for the primary study contain any depressive status assessed with a depression severity measure;
(c) Any psychotherapy versus non‐treatment, waiting‐list control, or treatment as usual
Although some studies may use 'attention placebo' as the control condition, studies will be included only when the content of 'attention placebo' is similar to TAU as described in 'Types of intervention'.
The inter‐rater reliability of the two raters will be calculated. Any disagreement will be resolved by consensus between the two reviewers and, if need be, with another reviewer. All the candidate papers will then be subjected to the next stage of study selection according to the strict eligibility criteria.
Two independent reviewers (NW and TAF) will examine the strict eligibility of these trials. The inter‐rater reliability of these eligibility criteria will be calculated. Any disagreement will be resolved by consensus between the two reviewers and, if need be, with other reviewers. The references of other relevant review papers will be checked by NW and TAF. All of these studies will be subjected to SciSearch in order to identify more recent reports.

(2) Quality assessment
Three independent reviewers (NW, VH and RC) will assess the methodological quality of the selected studies.
The criteria for the quality assessment are based on the recommendations of the Cochrane Collaboration Handbook. The Handbook is characterised to urge reviewers to consider the following two points in terms of the quality assessment;
(a) Quality of allocation concealment in randomisation to treatment conditions. There is empirical evidence that poor quality of randomisation can produce biased results (Chalmers 1983; Moher 1998; Schulz 1995).
(b) Quality of assessors' blinding. It is often difficult to keep both participants and clinicians blind to their treatment group in psychotherapy trials. In these trials, we need blind assessors for outcome measurement in order to minimize detection bias. A self‐report questionnaire can be a "blind" measurement, only when the participants were kept blind to the allocated treatment, for example, by credible attention placebo. There is some evidence to suggest that trials that did not describe themselves "double‐blind" yield biased results (Coldiz 1989; Schulz 1995).
Taking these two points into consideration, the overall quality of the trials will be scored according to important elements of design and conduct. Study quality will be examined using 23 criteria (Churchill 2001), and each item will be scored between 0 and 2 with a maximum total score of 46. The total point of each study will be utilized in sensitivity analyses for the purpose of looking at robustness of the findings. In addition, we will employ a quality rating scale for quality of conduction studies especially on psychotherapy (Foa 1997). Although this rating scale has seven items and most of them appear to overlap with the 23‐item criteria, we will use it in a sensitivity analysis in order to examine robustness of the results.
The inter‐rater reliability of the total score of study quality assessment will be examined with ANOVA ICC. Any disagreement will be resolved by consensus between the two raters and, if need be, with another reviewer.
In addition to these methodological qualities of the selected trials, we will assess the type and adequacy of the psychotherapy according to the following two criteria.
(1) First, we will determine, regardless of the designation by the original investigators, to which category of psychotherapy (CBT, behavioural therapy, cognitive therapy, interpersonal therapy, problem‐solving therapy, play therapy or psychodynamic therapy) it belonged, according to the definitions described "Types of intervention" section.
(2) Secondly, we will determine if the administered psychotherapy had demonstrable quality in each study.
A: when the conduct of psychotherapy was examined by a third reviewer through audiotapes etc, and when the authors gave sufficiently detailed description of the therapy procedure
B: when the study fulfills either of the two conditions as stated above; and
C: when none of these applied.

(3) Data extraction
Three reviewers (NW, VH and RC) will independently extract data from the original reports using data extraction forms. Although we will define primary outcomes as dichotomous data ("response or remission" or not), for studies where exact numbers of participants with response were not presented in the original reports but only their means and standard deviations of continuous global depression scales, we will impute the number of responders by using a validated statistical method (Furukawa 2005) in order to conduct intention‐to‐treat (ITT) analyses as described below. Any disagreement will be resolved by consensus between the two or, where necessary, between all reviewers.

(4) Data synthesis
Data will be entered into Review Manager 4.2 by NW twice using the duplicate data entry facility.
For dichotomous outcomes, we will calculate the relative risks and their 95% confidence intervals using the random effects model, because the random effects model RR has been shown to be favourable in clinical interpretability and external generalisability than fixed effect models and odds ratios or risk differences (Furukawa 2002). Heterogeneity between studies will be assessed by the Q statistic, I‐squared test and by visual inspection of the results. When heterogeneity is found to be significant, sources will be investigated.
We will use two levels of ITT analyses in terms of dichotomous data. Dropouts will be always included in this analysis. In the first analysis, when dropouts were excluded from any assessment in the original studies (for example, those who never returned for assessment after randomisation), they will be considered non‐responders both in the active psychotherapy group and in the control group.
In the second, more stringent analysis (ITT analyses based on worst case scenario), all the dropouts will be considered non‐responders in the active treatment group but as responders in the control group. It will help readers see how robust the results of the present review are.
For continuous outcomes, the standardized weighted mean difference (SMD) will be calculated using the random effects model. Scores from continuous outcomes will not be analyzed on an ITT basis. This approach is not feasible as studies perform only LOCF or endpoint analyses. Continuous outcomes will therefore be analyzed on an endpoint basis, including only participants with a final assessment or with a last observation carried forward to the final assessment.

(5) Subgroup analyses
Subgroup analyses should be performed and interpreted with caution because multiple analyses will lead to false positive conclusions (Oxman 1992). However, we will perform the following subgroups analyses, where possible.
(a) For each type of psychotherapy separately (CBT, behavioural therapy, cognitive therapy, interpersonal therapy, problem‐solving therapy, play therapy, supportive therapy or psychodynamic therapy)
(b) For the age of participants (e.g. 6 to 12 years or older according to when children and adolescents were included in the study), because the effects of psychotherapy may vary from age to age due to childrens' developing cognitive functions.
(c) For the severity of depression (e.g. mild, moderate or severe), defined by the authors of the original studies

(6) Funnel plot analysis and sensitivity analyses
We will perform funnel plot analyses to check for publication bias.
The following sensitivity analyses are planned a priori. By limiting the studies to be included to those of higher quality we will examine if the results change and intend to check for the robustness of the observed findings.
(a) By limiting studies to those that employed standardised criteria for major depression (e.g. Feighner criteria, Research Diagnostic Criteria, DSM‐III, DSM‐III‐R, DSM‐IV or ICD‐10)
(b) By excluding studies with comorbid physical or mental disorders that were not a primary diagnosis (if possible, excluding only participants that had comorbid diagnosis)
(c) By limiting studies where the status of the control group was treatment as usual
(d) By limiting studies to those that did not allow concomitant use of psychotropic agents
(e) By limiting studies where the administered psychotherapy had a demonstrative quality rates as A or B according to the above criteria in the Quality Assessment section
(f) By limiting studies to those where study quality is assessed as "high" by using 23‐item quality rating scale (Churchill 2001), namely total score of 20 or more
(g) By limiting studies to those satisfying all the items in the 7‐item quality rating scale (Foa 1997)