Protocol and registration

This study will build on and extend prior work by our group.^{Reference Weitz, Hollon, Twisk, van Straten, Huibers and David27–Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30} It is registered in the PROSPERO International Prospective Register of Systematic Reviews (PROSPERO registration number: CRD42022341602). Additional important protocol amendments will also be documented in the PROSPERO register. This documentation will include the date of the amendment, a description of the change and the rationale. This article is reported in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) statement.^{Reference Shamseer, Moher, Clarke, Ghersi, Liberati and Petticrew31}

Eligibility criteria

This meta-analysis will include (a) randomised clinical trials (b) comparing cognitive therapy and behavioural activation (c) for the acute-phase treatment of depression (d) in adults. No restrictions will be placed on the years when the study was conducted, publication language, publication date or publication status.

Studies will be included if they directly compare cognitive therapy and behavioural activation among participants who were randomly assigned to these treatments. An intervention will be considered to be cognitive therapy if it is a manualised psychotherapy with cognitive restructuring as the main treatment component.^{Reference Weitz, Hollon, Twisk, van Straten, Huibers and David27} Behavioural techniques (including activity scheduling) will be allowed, as long as they are part of an intervention protocol that is aimed at cognitive restructuring. Beck et al's model^{Reference Beck, Rush, Shaw and Emery6} is considered the cognitive therapy prototype, but other models can also be included. An intervention will be considered to be behavioural activation if the core element of treatment is aimed at increasing positive reinforcement by means of activity scheduling. Inclusion of cognitive restructuring techniques will not be allowed. As such, behavioural activation encompasses the early intervention developed by Lewinsohn,^{Reference Lewinsohn, Friedman and Katz7} which mostly involves mood monitoring and scheduling pleasant activities, as well as newer interventions, such as the one developed by Jacobson et al,^{Reference Jacobson, Martell and Dimidjian8} which additionally includes functional analysis and a focus on reducing avoidance behaviours, and the behavioural activation approach studied extensively by Lejuez et al.^{Reference Lejuez, Hopko, LePage, Hopko and McNeil32} Any cognitive therapy and behavioural activation format will be allowed (e.g. individual, group), as will any delivery method (e.g. telephone, e-mail, videoconferencing) as long as a care professional delivers the therapy. Thus, unguided bibliotherapy or unguided internet interventions will be excluded. Guided bibliotherapy, guided internet therapy or other guided self-help formats will be included if provided by a trained healthcare worker. No restrictions will be placed on the setting where the psychotherapies are conducted, the number of sessions or the duration of follow-up.

Following previous depression treatment IPD meta-analyses,^{Reference Driessen, Abbass, Barber, Connolly Gibbons, Dekker and Fokkema29,Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30} participants will be considered to have depression if they meet specified criteria for a unipolar mood disorder assessed by means of a semi-structured interview or clinician's assessment, or if they present a score above the validated cut-off on an evaluator-assessed, clinician-assessed or self-reported measure of depression. Comorbid mental and somatic disorders will be allowed to increase generalisability of this study's findings to clinical practice. Thus, studies allowing comorbidities or focusing specifically on comorbidities (e.g. depression and substance misuse) will be included. As this study assesses the acute-phase treatment of depression, studies in which cognitive therapy and behavioural activation are examined as maintenance or relapse prevention treatments after the acute treatment phase will be excluded.

Participants must be at least 18 years old. Thus, studies for child or adolescent depression will be excluded, but studies including older adult populations will be considered. Participant criteria will be assessed at study level. In line with prior depression treatment IPD meta-analyses,^{Reference Driessen, Abbass, Barber, Connolly Gibbons, Dekker and Fokkema29,Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30} eligible participants from a study including a wider population (e.g. adults from a study including both adolescents and adults) will not be included, because the integrity of randomisation within the subgroup of eligible participants could be compromised.

Information sources, search strategy and selection process

A database of randomised clinical trials examining the efficacy of psychological treatments for depression will be searched to identify relevant studies (www.metapsy.org). This METAPSY database has been used in a series of published meta-analyses and is updated annually. It is developed through comprehensive literature searches in the bibliographic databases PubMed, PsycINFO, EMBASE and the Cochrane Library. The search strings use a combination of index terms and free-text words indicative of depression and psychotherapies, with filters for randomised clinical trials. The exact terms for the searches are available from https://osf.io/nv3ea/.

Two raters will independently screen all records based on titles, abstracts and keywords, will assess full-text papers for METAPSY database eligibility and will rate the treatment comparison(s) examined. Disagreements will be resolved through consensus. Next, two raters will independently assess all full-text papers of studies marked as comparing a psychotherapy monotreatment condition against another active monotreatment condition, for meeting the eligibility criteria for this work. Disagreements will again be resolved through consensus. In addition, prior behavioural activation for depression meta-analyses^{Reference Simmonds-Buckley, Kellett and Waller12–Reference Ciharova, Furukawa, Efthimiou, Karyotaki, Miguel and Noma14} and reference lists of the included studies will be checked for studies meeting the eligibility criteria that might have been missed.

Data collection process

Authors of the included studies will be invited to participate in this project by using a strategy that has been successful in soliciting participation in previous depression treatment IPD meta-analyses.^{Reference Weitz, Hollon, Twisk, van Straten, Huibers and David27–Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30} Authors sharing their IPD will be offered co-authorship for all publications resulting from the use of these data, inasmuch as they meet internationally accepted criteria for authoring scientific articles (www.icmje.org). In addition, the aggregated database will be made available to investigators who share IPD to examine other research questions, provided that the authors of the original studies approve the use of their data for this purpose.

To invite authors, a multi-step contact protocol will be applied, which has previously been described^{Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30} and has proven to be successful in reaching authors for prior depression treatment IPD meta-analyses. More specifically, contact details of all corresponding authors will be collected from the relevant publications, through internet searches or personal contacts with other researchers. Corresponding authors will be sent an email invitation outlining the project's goals and asking for their collaboration by sharing their study's IPD. If the corresponding author does not respond after 3 weeks, a second and third email will be sent. If no response is received from the corresponding author, the other authors will be contacted in the same way, in this order: first, last, second, third, fourth, etc. If none of the authors respond to the emails, a letter will be mailed to the corresponding author (again with three attempts). If no response to these letters is received, the corresponding author will be contacted by telephone. If the corresponding author does not respond, the other authors will be contacted by letter and telephone. If none of the authors responds to these efforts, other ways will be sought to contact one of the authors (e.g. via colleagues who might know them). A study's data will be considered unavailable only if (a) all of these attempts fail or (b) an author either indicates that the IPD were not retained or declines to share these data.^{Reference Driessen, Abbass, Barber, Connolly Gibbons, Dekker and Fokkema29,Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30}

Data items

The following participant-level data items will be requested: treatment condition, all outcome variables assessed during and after treatment (with item-level data for depression outcome measures) and all potential moderator variables assessed in the study. Moderator variables are defined as all demographic (e.g. age, gender), clinical (e.g. depressive episode duration, comorbid anxiety disorder) and psychological (e.g. personality, coping style) participant characteristics assessed in the study before the start of treatment. The primary study's authors will anonymise the participant-level data-set before transferring it.

The following study-level characteristics will be extracted from the publications: country where the study took place, recruitment method (e.g. community, clinical), target group (e.g. adults in general, students), depression inclusion criteria, number of cognitive therapy/behavioural activation sessions, cognitive therapy/behavioural activation format (e.g. individual, group) and assessment time points. Cognitive therapy and behavioural activation treatment quality will be examined with regard to use of a treatment manual, provision of therapy by trained therapists and verification of treatment integrity. In addition, effect size data will be extracted from the published articles. If information on study-level characteristics or treatment quality, or effect size data, is not reported in the publications, it will be requested from the authors.

Data integrity checks

Upon receiving the data-set, three data integrity checks will be performed. First, it will be checked whether the data-set includes the full intention-to-treat sample (i.e. all participants randomised to treatment) and otherwise matches the data reported in the published article. To this end, all baseline characteristics, and observed mean pre-treatment and post-treatment scores reported in the article, will also be calculated from the data-set and both will be compared. Second, it will be checked whether all outcome and potential moderator variables reported in the article are included in the data-set. Third, the outcome and moderator variables will be checked for inconsistent, invalid or out-of-range items. Discrepancies resulting from these data integrity checks will be resolved with the authors.

After the integrity checks have been performed, the data-sets will be standardised and merged in the IPD meta-analysis database. For this purpose, relevant variables will be extracted from each trial's raw data file, recoded and copied into a single database in which each participant is identified by a study identifier and a unique individual participant identifier. After all data files have been recoded and entered, the data for each study will be checked with the original data file for accuracy.^{Reference Driessen, Abbass, Barber, Connolly Gibbons, Dekker and Fokkema29,Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30}

Outcomes and prioritisation

Depressive symptom level at treatment completion will be the primary outcome of this study, as symptom reduction is considered to be the main aim of cognitive therapy and behavioural activation in the acute-phase treatment of depression. Depressive symptom level at treatment completion is operationalised as a participant's score on the primary continuous depression scale administrated at the primary post-treatment time point, both as defined by the study's authors. Continuous rather than dichotomous measures (e.g. remission) were chosen as primary outcome because of the increased variance that might facilitate moderator identification. Secondary outcomes will be depressive symptom level at follow-up, as well as all other outcomes at post-treatment or follow-up that are assessed in at least two studies (e.g. anxiety symptoms, interpersonal functioning, quality of life).

Individual studies are expected to use different instruments to assess depressive symptoms as well as the secondary outcome domains. Outcomes will therefore be standardised by converting raw scores into z-scores within each study.^{Reference Driessen, Abbass, Barber, Connolly Gibbons, Dekker and Fokkema29,Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30} Sensitivity analyses will be conducted with unstandardised scores for each specific depression measure that is assessed in the majority of studies included in the meta-analysis.^{Reference Driessen, Abbass, Barber, Connolly Gibbons, Dekker and Fokkema29,Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30} The meta-analyses’ principal measures of effect will be Cohen's d effect sizes for analyses with z-scores as outcome measure, and mean differences for analyses with unstandardised scores as outcome measure. Cohen's d effect sizes of 0–0.32, 0.33–0.55 and 0.56–1.2 will be considered small, moderate and large, respectively.^{Reference Lipsey and Wilson33}

Potential moderators will include all demographic, clinical and psychological participant characteristics that are assessed in at least two studies. These are likely assessed differently in individual studies and will be standardised as well; for instance, by converting scores into z-scores for continuous variables or by recoding variables into similar categories for categorical variables.

Risk of bias in individual studies

Risk of bias in the included studies will be assessed with the Cochrane risk-of-bias tool for randomised trials.^{Reference Higgins, Savovic, Page, Elbers, Sterne, Higgins, Thomas, Chandler, Cumpston, Li, Page and Welch34} Two raters will independently assess this tool at outcome level. Disagreements will be resolved by consensus. Ratings will be primarily based on information reported in the publications, although selection bias and reporting bias will be assessed with the IPD. As studies are expected to be included that were published before the universal adoption of reporting guidelines for randomised clinical trials,^{Reference Begg, Cho, Eastwood, Horton, Moher and Olkin35} requisite information will be requested from the authors if it is not reported in the publications.

Data synthesis

To facilitate the comparison of this study's findings to prior work, a data-analysis strategy will be adopted that has been previously applied in depression treatment efficacy IPD meta-analyses, and is described in more detail elsewhere.^{Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30,Reference Driessen, Dekker, Peen, Van, Maina and Rosso36} This strategy is based on the approach recommended by Twisk et al,^{Reference Twisk, Bosman, Hoekstra, Rijnhart, Welten and Heymans37} and will be used because it adequately accounts for baseline values and has favourable properties concerning missing data (allowing participants with only a baseline value but missing post-treatment and/or follow-up assessments to remain included in the analyses).

One-stage IPD meta-analyses will be conducted in intention-to-treat samples, using mixed-effects models with restricted maximum likelihood estimation. Because additional help-seeking usually cannot be controlled during the follow-up period, follow-up data will be excluded from post-treatment analyses. Heterogeneity will be assessed by calculating the variance between studies as a proportion of the total variance (i.e. I²-statistic). The normality assumption will be checked by inspecting histograms of residuals.

To compare the efficacy of cognitive therapy and behavioural activation, a model will be estimated including a main effect for time (categorical; represented by dummy variables) and a time×treatment interaction,^{Reference Twisk, Bosman, Hoekstra, Rijnhart, Welten and Heymans37} a random intercept with respect to study (to account for the clustering of participants within studies) and a random intercept with respect to participants (to account for the clustering of repeated measures within participants). The starting model contains fixed slopes, but the –2 log likelihood change will be evaluated to decide whether to add a random slope for the time×treatment interaction at study level. The regression coefficient of the time×treatment interaction will be considered to indicate the magnitude of the treatment effect.

To examine moderator effects, first, a model will be estimated for each potential moderator variable. This model will comprise the model described above, with an additional main effect for the moderator and a time×moderator×treatment three-way interaction. To deal with the issue of multiple analyses, all potential moderators with a P-value <0.10 for the three-way interaction's regression coefficient will next be included in a model simultaneously. A significant (P < 0.05) regression coefficient of the three-way interaction in this final model will be considered to indicate a moderating effect. For categorical moderator variables, separate treatment effects will be estimated for each moderator category if a significant moderating effect is found.

Because sample sizes can vary between variables depending on the data available, effect sizes for each outcome and moderator will also be examined independently of their P-values, to consider whether type 1 or type 2 errors might have been made. For each outcome and potential moderator, the strength of the body of evidence will be assessed based on the number of included studies and participants, as well as the quality of the included studies.

Sensitivity analyses

Six sensitivity analyses will be conducted with regard to the primary efficacy analysis. First, as described previously, the analysis will be repeated with unstandardised scores for each specific depression measure that is assessed in the majority of studies included in the meta-analysis.^{Reference Driessen, Abbass, Barber, Connolly Gibbons, Dekker and Fokkema29,Reference Driessen, Cohen, Weissman, Markowitz, Weitz and Hollon30} Second, to examine the robustness of the findings to those eligible for cognitive therapy and behavioural activation for depression in general mental healthcare, the analysis will be repeated in studies including general adult populations meeting diagnostic criteria for a mood disorder and providing face-to-face cognitive therapy and behavioural activation. Third, to examine the impact of risk of bias in the primary studies, risk-of-bias items will be added as dichotomous covariates to the mixed-effects model. Fourth, the impact of cognitive therapy and behavioural activation quality will be examined in the same way. Fifth, to examine potential differences in treatment effects, the analysis will be conducted separately in the subgroup of studies excluding behavioural techniques in cognitive therapy and the subgroup of studies including behavioural techniques in cognitive therapy. Sixth, to differentiate between Beckian cognitive therapy and other approaches, the analysis will be conducted separately in the subgroup of studies that cite the Beck treatment manual for cognitive therapy^{Reference Beck, Rush, Shaw and Emery6} and the subgroup of studies that refer to other treatment manuals.

In addition, one sensitivity analysis will be conducted with regard to the moderation analyses. To examine the possibility that false conclusions are drawn about what might be important moderators today based on data collected at a time in the past, when some of these factors may have had a different level of importance, we will repeat the moderator analyses excluding studies that were completed in the 1970s and 1980s.

Meta-biases

Following the recommendations proposed by Sterne et al,^{Reference Sterne, Sutton, Ioannidis, Terrin, Jones and Lau38} potential publication bias will be assessed by examining asymmetry in contour-enhanced funnel plots with Egger's test for analyses including ten or more trials. Potential data availability bias will be assessed by comparing characteristics between studies for which IPD were and were not obtained with t-tests for continuous variables and χ ²-test analyses for categorical variables. In addition, Cohen's d effect sizes will be calculated based on the effect size data extracted from the publications. Conventional meta-analysis ‘subgroup analyses’ will then be conducted – applying a fully random-effects analysis and pooling study-to-study variance across subgroups – to compare effect sizes between studies for which IPD were and were not available.

Ethics statement

This study did not require institutional review board (IRB) approval. Depending on their institution's policies, IRB approval may be required for the authors to share their IPD. If their institution's policies require them to do so, it is the authors’ responsibility to obtain IRB approval. With signing the data-sharing agreement, investigators sharing IPD declare that these IPD were collected and are transferred to our research group in accordance with all applicable local and international laws and regulations. Furthermore, they declare that all IPD will be anonymised, so that no personal data are transferred.