Types of studies

We included randomised controlled trials (RCTs) comparing pharmacological treatments for patients with acute psychotic depression.

As we expected to identify very few RCTs assessing treatment of psychotic depression as the primary focus, we decided a priori to also include RCTs assessing treatment of major depression with or without psychotic features. Effects in the subgroup of participants with psychotic features should then be reported separately, irrespective of whether the subgroup with psychotic features was stratified before randomisation.

We applied no language restrictions for included studies.

Types of participants

Participants of any age in any setting (both inpatients and outpatients) had a major depressive disorder and a current episode with psychotic features (delusions and/or hallucinations appearing in the context of a full major depressive episode) according to the Diagnostic and Statistical Manual of Mental Disorders, Third Edition (DSM‐III)/DSM, Third Edition, Revised (DSM‐III‐R)/DSM, Fourth Edition (DSM‐IV)/DSM, Fourth Edition, Text Revision (DSM‐IV‐TR), Fifth Edition (DSM‐5) (APA 1980; APA 1987; APA 1994; APA 2000; APA 2013), or consistent with International Classification of Diseases (ICD) codes for the same.

We also included studies in which participants had comorbidities, as comorbidity was not a reason for exclusion.

As patients with a major depressive episode in the context of a bipolar disorder (bipolar depression) are at increased risk of switching to mania (Licht 2008), we intended to include only trials that assessed participants with unipolar depression. If trials had included participants with both unipolar and bipolar depression, we decided a priori to include only trials for which results in the unipolar group were reported separately, or for which the percentage of participants with bipolar depression did not exceed 20% of the total study population.

Types of interventions

We included any pharmacological treatment of a current (i.e. acute) episode. Treatment had to be given for at least four weeks with the intention of treating the current episode.

When possible, we considered the following pair‐wise comparisons.

1. Antidepressant versus placebo.

2. Antipsychotic versus placebo.

3. Antidepressant versus antidepressant.

4. Antipsychotic versus antipsychotic.

5. Antidepressant versus antipsychotic.

6. Antidepressant plus antipsychotic versus placebo.

7. Antidepressant plus antipsychotic versus placebo plus antidepressant.

8. Antidepressant plus antipsychotic versus placebo plus antipsychotic.

9. Mifepristone versus placebo.

10. Mifepristone plus antidepressant versus placebo plus antidepressant.

11. Other psychopharmacological agents versus placebo.

12. Other psychopharmacological agents plus antidepressant versus placebo plus antidepressant.

Types of outcome measures

Electronic searches

Search strategies were updated from 2013 onwards. 

Previous searches, conducted up to April 2013, for earlier versions of this review ‐ Wijkstra 2005 and Wijkstra 2015 ‐ can be found in Appendix X, with a description of the Cochrane Common Mental Disorders Group (CCMD) controlled trials register (CCMDCTR) provided in Appendix X.

For this update of this review, the CCDANCTR Studies Register was searched (from 2013 to 12 February 2020) using the following terms.

Condition = (depressi* or “affective disorder*” or “affective symptoms”) 

AND 

Condition or Comorbidity = (psychosis or psychoses or psychotic* or delusion*)

The CCDANCTR Studies Register was searched (all years to 12 April 2013) using the following terms to identify additional untagged references.

Title/Abstract/Keywords = ((depressi* or “affective disorder*” or “affective symptoms”) 

AND

Free‐Text = (psychosis or psychoses or psychotic* or delusion* or hallucin* or antipsychotic* or psychotropic*))

An information specialist with CCMD (previously known as the Cochrane Collaboration Depression, Anxiety and Neurosis Group (CCDAN)) ran an updated search (21 February 2020) of the following electronic databases, using relevant subject headings (controlled vocabularies) and search syntax, as appropriate to each resource. 

1. Ovid MEDLINE.

2. Ovid Embase.

3. Ovid PsycINFO.

4. Cochrane Central Register of Controlled Trials (CENTRAL; February 2020, Issue 2 of 12).

5. Cochrane Common Mental Disorders Controlled Trials Register (CMDCTR).

The international trial registries (ClinicalTrials.gov and World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP)) were searched via CENTRAL, in the Cochrane Library (all years from 2013 to 12 February 2020).

We applied no restrictions on language or publication status to these searches.

Searching other resources

The reference lists of all studies, related reviews, and relevant conference proceedings were screened, and key study authors contacted.

Selection of studies

In step 1, all abstracts of identified publications were screened independently by two review authors (JL and JW in 2005 and 2015, JK and EL in 2020), and studies were selected if they met the following criteria.

1. Were randomised controlled trial (RCT).

2. Included participants with a major depressive disorder.

3. Investigated the efficacy of pharmacological treatment.

4. Focused on acute phase treatment (minimum of four weeks' treatment), not on continuation or maintenance treatment.

If any doubt or disagreement arose between the review authors, the publication was included in step 2. Full articles were obtained for the selected abstracts.

In step 2, selected full articles were screened (JL and JW in 2005 and 2015, JK and EL in 2020) according to the following criteria.

1. Participants with psychotic depression were not excluded.

2. Results in the subgroup of psychotic depressed participants were reported separately.

If any doubt arose about an article, it was included in step 4.

In step 3, the reference lists of related reviews and of included publications, conference abstracts, and personal communications were searched.

Finally, in step 4, two review authors (JL and JW in 2005 and 2015, JK and EL in 2020) independently reviewed all identified publications according to the full inclusion criteria of the review. Any disagreement was resolved by consensus discussion with another review author (WN).

Data extraction and management

Extracted data included the following: participant characteristics (age, gender, setting: inpatient/outpatient); diagnosis (diagnostic instrument, system of classification, number of bipolar participants); prior treatment for the current episode; intervention; length of illness; suicide attempts; treatment details (treatment period, washout period, additional medication, blood levels, doses); and outcome measures. Data were extracted independently by two review authors (JL and JW in 2005 and 2015; JK and EL in 2020).

All data (from the 2013 review and recent data) were entered into RevMan 5.4 (Review Manager 2020).

Assessment of risk of bias in included studies

In the original version of this review, we assessed methodological quality of included studies using criteria set out in the Cochrane Handbook for Systematic Reviews of Interventions (Alderson 2004); however, after publication of the revised and expanded Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021), we updated our methods accordingly. Working independently, two review authors (JL and JW in 2005 and 2015; JK and EL in 2020) assessed risk of bias of included studies using the tool described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021b). The following items were assessed.

1. Sequence generation: was the allocation sequence adequately generated?

2. Allocation concealment: was allocation adequately concealed?

3. Blinding of participants, personnel, and outcome assessors for each main outcome or class of outcomes: was knowledge of the allocated treatment adequately prevented during the study?

4. Incomplete outcome data for each main outcome or class of outcomes: were incomplete outcome data adequately addressed?

5. Selective outcome reporting: were reports of the study free of suggestion of selective outcome reporting?

6. Other sources of bias: was the study apparently free of other problems that could put it at high risk of bias?

We included quotations from the text of included studies and comments on how we assessed risk of bias; we judged each study to be at low, unclear, or high risk of bias.

See risk of bias figures (Figure 1; Figure 2).

Figure 1

---

---

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 2

---

---

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

If disputes arose as to which judgement should be given, resolution was achieved after consultation with the third review author (WN).

Measures of treatment effect

Unit of analysis issues

We identified neither cluster‐randomised nor cross‐over trials. However, if found, we would not have included them in our review, as we were interested in differences not between clusters (e.g. clinics) but between drugs or classes of drugs; nor were we interested in the results of a cross‐over phase, as the outcome of the first phase might have had an impact on the outcome of the second (i.e. cross‐over) phase.

In case a study had multiple intervention groups, we included only data for the pair‐wise comparison in question. Further, if a study compared two or more medications of the same type (e.g. venlafaxine, imipramine), we combined data into a single category, for example, the category 'antidepressant'.

Dealing with missing data

We used intention‐to‐treat (ITT) response data in the analyses, as ITT data are less biased, and because they address a more pragmatic and clinically relevant question (Higgins 2021). When necessary, we converted response data from trials into ITT response data by using the total number of randomly assigned participants per group who had started with treatment as the denominator. So participants who had started with medication but were withdrawn before endpoint were assumed not to have experienced response.

When data were missing, we contacted study authors to request the required data.

We used no other imputation techniques to deal with missing data.

Assessment of heterogeneity

First, we evaluated whether clinical homogeneity could be assumed by evaluating any between‐study dissimilarities regarding participants, interventions, and outcome measures. We excluded from the meta‐analysis studies that were considered to threaten the clinical homogeneity assumption.

We used the I² statistic supplied with a 95% CI to assess the magnitude of statistical heterogeneity when values exceeding 0.40 were considered possibly relevant. We did not perform the Q test to determine heterogeneity because of its low power in our meta‐analysis resulting from low numbers of studies in all of our comparisons.

We planned to conduct re‐stated subgroup analyses to explore sources of heterogeneity. We performed no meta‐regression.

Assessment of reporting biases

When data from at least 10 studies became available, we assessed the presence of publication bias by using contour‐enhanced funnel plots in which treatment effects expressed as RR (risk ratio) from individual studies were plotted against each study's sample size. We did not perform Egger's regression test in view of low statistical power in our meta‐analyses, again resulting from the low number of included studies.

It must be noted that asymmetry of funnel plots does not necessarily indicate publication bias but may result from other biases such as inflated results in smaller studies due to poorer methodological quality, or true heterogeneity.

Data synthesis

We used the Mantel‐Haenszel fixed‐effect method to calculate pooled risk ratios with 95% confidence intervals. Risk ratios are preferred over odds ratios because of their more straightforward interpretation (i.e. the number of times the outcome is more likely to occur given one treatment over another). We used fixed‐effect models as most meta‐analyses consisted of a small number of studies, therefore there were insufficient data to estimate random‐effects models.

Subgroup analysis and investigation of heterogeneity

If sufficient data were available, we planned to conduct the following subgroup analyses.

1. Participants who were non‐refractory to prior treatment(s) during the current episode.

2. Participants with mood congruent psychotic features only.

Because all psychotically depressed patients are considered to be severely depressed, it was not considered appropriate to evaluate baseline severity in a subgroup analysis.

Sensitivity analysis

If sufficient data were available, we planned to perform the following sensitivity analyses.

1. Studies focusing on psychotic depression only.

2. Studies in which participants with psychotic depression were separately randomised.

3. Studies of lower methodological quality to assess robustness of results.

4. Smaller versus larger studies.

Summary of findings and assessment of the certainty of the evidence

 We created 'Summary of findings' tables, in which we summarised findings of studies comparing:

1. antidepressant versus placebo;

2. antipsychotic versus placebo;

3. antidepressant versus antidepressant;

4. antipsychotic versus antipsychotic;

5. antidepressant versus antipsychotic;

6. antidepressant plus antipsychotic versus placebo;

7. antidepressant plus antipsychotic versus placebo plus antidepressant; and

8. antidepressant plus antipsychotic versus placebo plus antipsychotic.

We have presented a separate 'Summary of findings' table for each comparison group. We included the following outcomes: depression response, overall dropout, depression remission, change in depression severity, quality of life, and dropout due to adverse effects as measured between baseline and end of study.

We used the five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness, and publication bias) to assess the certainty of a body of evidence as it relates to studies that contributed data to meta‐analyses for prespecified outcomes. We used methods and recommendations described in Chapter 14 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2020), along with GRADEpro software (GRADEpro GDT). We justified all decisions to downgrade the certainty of evidence by using footnotes, and we made comments to aid the reader's understanding of the review when necessary.

Two review authors (BV, LR) independently assessed the certainty of evidence and resolved disagreements through discussion or by consultation with a third review author (WN). We justified, documented, and incorporated judgements into reporting of results for each outcome.