The goal of this systematic review and network meta-analysis is to synthesize the relative efficacy of all available psychological treatments in one analysis. We aim to combine both direct and indirect evidence to create effect estimates, even for treatments that have never been directly compared in a trial. Furthermore, we plan to extend the existing knowledge on the topic by examining multiple endpoints, including symptom severity, quality of life, functioning, and relapse rates.

This systematic review and network meta-analysis protocol followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA-P) statement extension for network meta-analysis. ²⁰ Appendix 1 provides the PRISMA-P checklist. This protocol was registered in PROSPERO (CRD42024514287).

Characteristics of studies

We will include (i) randomized controlled trials comparing (ii) a psychological intervention to (iii) a control group (waitlist, treatment as usual, psychological placebo, or other psychosocial or psychological treatments). Psychological placebo will be defined as any condition aimed at controlling the time spent with a therapist, without delivering any bona fide psychotherapeutic content (e.g., recreational therapy, activity groups). We expect that, in the majority of studies, psychological intervention will be provided in addition to standard care, which usually includes antipsychotic medication. If we find studies in which psychological intervention is administered without concomitant pharmacotherapy, we will consider these studies as well.

Characteristics of participants

We will consider trials in patients with a recent first-time diagnosis of psychosis, first-episode schizophrenia, or early phase schizophrenia, as per the inclusion criteria of specific studies. In cases of “early-phase schizophrenia” or “recent-onset psychosis”, we will accept a maximum duration of illness of 5 years. We will also include studies conducted in specialized early intervention centers since they target the population in question. We will include studies with participants aged 16 years or older, allowing us to capture the variability in the age of onset of schizophrenia. Studies focusing on participants over the age of 65 years as part of their inclusion criteria or those investigating late-onset psychosis will be excluded. We will exclude studies that only include patients with bipolar mania, bipolar depression, substance-induced disorders, and psychoses of organic origin. We enforce this criterion because, despite unavoidable overlap, our focus is on schizophrenia and schizophrenia-related disorders (schizoaffective disorder and delusional disorder). We will exclude studies in “at risk of psychosis” or “prodrome” populations because such patients do not experience full-blown psychotic symptoms by definition. If a study included participants with other diagnoses, we include it if participants with psychosis (as defined above) comprised at least 50% of the participants. There will be no restrictions on sex, ethnicity, or setting.

Interventions

We will consider all psychological interventions that meet the classification criteria proposed by the Cochrane Common Mental Disorder Group (former CCDAN). ²¹ These classification criteria have already been employed in a previous meta-analysis on schizophrenia, in which early phase participants were not included. ¹⁶ We do not expect that studies and data will be available for all interventions mentioned in the classification, but if they do, they will be considered. Thus, we will consider psychotherapies (e.g., cognitive behavioral therapy, “third-wave” cognitive behavioral therapies, such as acceptance and commitment therapy, mindfulness and metacognitive training, psychodynamic therapy, art therapy, and music therapy), treatments that are often routinely provided in clinical settings, such as supportive therapy and psychoeducation, interventions focused on functioning (e.g., social skills training), and family interventions. Newly developed treatments that are not included in the list will also be considered for inclusion (e.g., virtual reality and AVATAR therapy). Organization of care-oriented interventions (e.g., case management, assertive community treatment) and supported employment will not be considered, because they have other aims, and it will not be meaningful to analyze their efficacy in contrast to the ones mentioned above.

Primary outcome

Our primary outcome was the severity of schizophrenia symptoms, measured by rating scales such as the Positive and Negative Syndrome Scale (PANSS), Brief Psychiatric Rating Scale (BPRS), or any other validated scale for the assessment of overall symptoms of schizophrenia. Based on our experience, we expect these scales to be used in almost all the trials. If only results of other rating scales are reported, they will only be used if the instrument has been validated in psychosis and published in a peer-reviewed journal, because non-validated schizophrenia scales tend to exaggerate differences. ²²

Relapse will be the primary outcome. For patients in the early phase, it is crucial to prevent relapse because the risk of chronicity increases with every episode. If more measures of relapse are reported, we will prioritize measures according to the following order: (i) operationalized criteria, (ii) hospital admissions due to psychopathology, and (iii) clinical judgement.

Secondary outcomes

We will also consider a broad collection of outcomes that may be informative for patients, clinicians, and guideline makers if available in the primary studies. 1.

Premature discontinuation (“dropout”) due to any reason: All-cause discontinuation (‘dropping out’) due to any reason combines efficacy, tolerability, and other factors and can therefore been considered as a measure of ‘acceptability of treatment’. ²³ We will also separately report the number of dropouts due to inefficacy of the treatment and worsening of clinical conditions.

We searched the most important electronic databases for relevant publications: BIOSIS, CINAHL, Cochrane CENTRAL, Dissertation Abstracts, EMBASE, LILACS, MEDLINE, PsycINFO, and Pubmed), including clinical trial websites such as Clinicaltrials.gov and the International Clinical Trials Registry Platform of the World Health Organization ( http://apps.who.int/trialsearch). Additionally, we will search previous reviews, statements of guidelines, and specific relevant journals concerning psychological treatments for schizophrenia to retrieve other studies. We will contact the first authors of the included studies in case of missing information. The search strings are provided in the Supplementary Material (Appendix 2). ⁵⁰

Two reviewers from our team will independently screen the search results for general inclusion criteria using the online tool Rayyan. ²⁶ Disagreements will be resolved by discussion, and, where doubts remain, we will acquire the full article for further inspection. Once the full articles were obtained, the two reviewers will independently decide whether the studies meet the review criteria. If disagreement cannot be clarified by discussion, we will consult a senior researcher (SL, IB) or seek further information from the authors.

Two reviewers will independently extract data from all selected trials using specifically developed forms in a Microsoft Access database. The forms will be piloted on a random sample of ten randomized controlled trials (RCTs). Disagreements will be resolved by discussion with a third senior researcher (SL, IB). The authors will be contacted if needed to resolve disagreements. The following data will be extracted from each included study: (i) general study information, (ii) information on methodology, (iii) characteristics of study participants, (iv) characteristics of psychological interventions, and (v) outcome measures.

Measurement of treatment effect

For continuous outcomes, we will calculate the standardized mean difference (SMD) as the effect measure, since we expect studies to use different rating scales for the measurement of overall schizophrenia symptom severity. We will prefer results obtained with imputation methods to handle missing data over complete data; results from mixed models of repeated measurement (MMRM) or multiple imputations will be preferred over last observation carried forward (LOCF). In case of missing summary statistics data, we will contact the authors of the primary study. When standard errors (SEs) are presented instead of standard deviations (SDs), the former are converted to standard deviations. If both are missing, we estimate SDs from confidence intervals, t-values, or p-values, as described in the Cochrane Handbook for Systematic Reviews. ²⁷ If none of these options is viable, we will contact the study authors. When no information could be obtained, we derive SDs from those of the other studies using a validated imputation technique. ²⁸

Odds ratios will be used to calculate dichotomous outcomes. Everyone allocated to the intervention will be counted to determine whether they have completed the follow-up. For beneficial outcomes, patients (e.g., responses) with missing outcome data will be assumed to be non-cases. All effect sizes are presented with 95% confidence intervals (CIs).

The primary time point of interest is 12 months, which, according to Cochrane, can be considered a medium term. The primary outcomes will also be collected and analyzed for up to six months (short-term) and more than 12 months (long-term).

Risk of bias assessment

Two reviewers will independently assess the risk of bias of the primary outcomes using the Cochrane Risk of Bias tool, version 2 (RoB 2.0). ²⁹ Disagreements between the two reviewers will be discussed with a third senior reviewer (SL, IB). We will not include studies whose sequence generation was judged to show a high risk of bias in the data analyses. The impact of the risk of bias will be analyzed using a sensitivity analysis.

Conventional pairwise meta-analyses

First, we will perform conventional pairwise meta-analyses for specific comparisons covered by the included trials. If the requirements for a network meta-analysis are not met, we present only the findings from pairwise syntheses. Pairwise meta-analyses will be based on a random-effects model with τ 2 quantified using the restricted maximum likelihood estimator. ³⁰ The confidence intervals of the pooled effect will be adjusted using the method described by Hartung and Knapp. ^{31 , 32}

Assessment of heterogeneity

Heterogeneity will be quantified using the estimated between-study heterogeneity variance ( τ 2 ) and will be presented using 95% prediction intervals. In the network meta-analysis model, heterogeneity variance is assumed to be common across the various treatment comparisons, and the empirical distributions are used to characterize the amount of heterogeneity as low, moderate, or high using the first and third quantiles. ^{33 , 34} Potential reasons for heterogeneity will be explored using subgroup analysis and meta-regression.

Network meta-analysis

Network meta-analysis allows the combination of direct and indirect evidence for all relative treatment effects, and can therefore provide effect estimates with maximum power and increased precision. ³⁵ The NMA models will be implemented using the frequentist graph-theoretical framework developed by Rücker. ³⁶ NMA models are based on the transitivity assumption, which is checked by investigating whether potential clinical and methodological effect modifiers (see subgroup analysis) are similarly distributed across studies. We will use the design-by-treatment interaction test that evaluates inconsistency from all possible sources in the network jointly, ³⁷ as well as the SIDE test (separating indirect evidence from direct evidence), ³⁸ assessing the agreement of indirect and direct evidence for every possible comparison in the network. In the case of evidence of inconsistency or intransitivity, we investigated possible sources. Small or moderate amounts of inconsistency will be further explored by network meta-regression and subgroup analyses, using the potential effect modifiers listed below. We estimated the probability for each intervention to be ranked at each possible place, given the relative effect sizes estimated in the NMA. We obtain a hierarchy of competing interventions using a frequentist extension of the surface under the cumulative ranking curve (P-score), as well as the mean ranks. ³⁹

Subgroup analysis and meta-regression

We plan to conduct the following subgroup analyses and network meta-regressions of the primary outcomes to investigate the impact of potential effect modifiers: a) age, b) patients clearly experiencing a first episode versus broader criteria (e.g., time criteria or second episode), c) duration of illness, d) number of sessions, e) manualized vs. non-manualized interventions, f) dose of antipsychotic treatment, and g) baseline severity of symptoms.

Sensitivity analyses

We plan to conduct the following sensitivity analyses of the primary outcomes by excluding: a) studies that did not use a blind outcome assessor (open studies); b) studies presenting only completer analyses; c) overall high risk of bias studies; d) studies with researchers’ allegiance; and e) studies that did not use operationalized diagnostic criteria.

Small study effects and publication bias

We will explore the association between study size and effect size using a comparison-adjusted funnel plot method. ⁴⁰ A comparison of more than 10 studies will be plotted in a contour-enhanced funnel plot. Any asymmetry observed can be attributed to systematic differences between small and large studies (e.g., differences in the risk of bias or recruitment), true heterogeneity, or publication bias. If significant small trial effects are identified, we will account for them via network meta-regression using a measure of study uncertainty (e.g., variance) as a covariate. ^{40 , 41} The possibility of reporting bias across the entire network will be assessed using the RoB-MEN framework. ^{42 , 43}

Evaluating the confidence in estimates

The confidence in estimates of the main outcomes will be evaluated with the framework Confidence in Network Meta-Analysis (CINeMA), ⁴⁴ an adaptation of the Grading of Recommendations Assessment, Development, and Evaluation framework (GRADE) specifically developed for NMA. The possibility of reporting bias across the entire network will be assessed using the RoB-MEN framework. ⁴² The margin of equivalence for continuous endpoints will be an SMD of 0.2 ⁴⁵ and an OR of 0.8 (1.25) for dichotomous outcomes (cf. Ref. 46).

Statistical software

Analyses will be conducted in R. A pairwise meta-analysis will be performed using the “meta” package. ⁴⁷ The frequentist NMA model will be fitted using the “netmeta” package. ⁴⁸ Network meta-regression will be performed in a Bayesian framework using models implemented in JAGS, ⁴⁹ assuming a weakly informative prior for the between-study heterogeneity variance τ 2 .