To address this evidence gap, we will conduct a systematic review and synthesis of IPD from randomized trials in the Vivli ²¹ database that compared antipsychotics with each other or placebo in individuals with acute schizophrenia. The aim is to examine sex differences in efficacy, tolerability and dosing using robust meta-analytic methods and input from a multidisciplinary team including clinicians, statisticians, and people with lived experience. The findings may inform sex-specific recommendations in treatment guidelines.

We will focus on subgroup differences based on biological sex, typically categorized as female or male, which is a key variable influencing drug effects. ²² Gender, as a sociocultural construct, may also affect treatment outcomes; however, it is beyond the primary aims of this project, and most included trials were likely conducted before distinctions between sex and gender were widely recognized or systematically captured in clinical research.

Studies meeting the following eligibility criteria in terms of study design, participants and interventions will be included in the systematic review.

We will include blinded and open randomized controlled-trials (RCTs) comparing eligible antipsychotic drugs, doses, or formulations (see “Interventions”) with each other or with placebo as monotherapy for acute schizophrenia. Trials must have a minimum treatment duration of three weeks. ²⁴

The first phase of crossover trials will be considered to avoid carryover effects. ²⁵ We will exclude studies with high risk of bias in the randomization process (see “Study risk of bias”), and cluster-randomized trials due to unit-of-analysis concerns. ²⁵

There will be no other restrictions in terms of publication year, publication status, sponsorship, language or country.

Population

Adults with acute schizophrenia, schizophreniform disorder, or schizoaffective disorder will be eligible. There will be no upper age limit or restrictions based on sex, setting, ethnicity, or diagnostic criteria. Studies in which fewer than 20% of participants have a different mental health condition will be included, and all participants with schizophrenia-spectrum disorders will be eligible for the IPD analysis.

We will exclude studies focusing specifically on the following subgroups: children or adolescents, advanced age, comorbid somatic illness or substance use disorders, first-episode psychosis, treatment resistance, predominant negative symptoms, or clinical stability or remission at baseline. This approach is consistent with previous meta-analyses ¹ to safeguard the transitivity assumption. However, if enough studies in patient subgroups are identified, we will request access to these for potential secondary analyses and separate publications.

We will include monotherapy with second-generation antipsychotics (i.e., amisulpride, aripiprazole, asenapine, brexpiprazole, cariprazine, clozapine, iloperidone, lumateperone, lurasidone, olanzapine, quetiapine, paliperidone, risperidone, sertindole, ziprasidone, and zotepine), three first-generation antipsychotics (i.e., chlorpromazine, haloperidol, and perphenazine), and placebo. Other first-generation antipsychotics will be excluded, as they have been studied in smaller, lower-quality trials and are no longer in use in most industrialized countries. ^{1, 26}

There will be no dose restrictions, as the aim is to examine whether women require lower doses than men and because optimal dosing in women remains unclear. ^{3, 27} We will include both fixed- and flexible-dose trials, in which investigators titrate to the appropriate dose. For flexible-dose trials, we will use the median of the randomized dose range in the main analysis and conduct sensitivity analyses using alternative estimates and by excluding these studies.

Oral and long-acting injectable antipsychotics will be considered the same intervention in the main analysis and as separate interventions in a sensitivity analysis. We will exclude short-acting injectables used for the management of acute agitation, as well as studies involving augmentation or combination strategies with antipsychotics.

Study identification

We will search for all studies available in Vivli, ²¹ a major data-sharing platform that includes published and unpublished trials conducted by both pharmaceutical companies and academic institutions. Studies from YODA, ²⁸ another data-sharing platform primarily containing Johnson & Johnson trials, are also accessible through Vivli. The search will be conducted through the NIAID Data Ecosystem by downloading all available studies in Vivli, and data requests will be submitted via Vivli and YODA as appropriate.

We will not search for trials that are not already available in Vivli, as the hundreds of potentially eligible RCTs, ¹ would require resources beyond the scope of this project. However, if feasible, we will consider complementing our analyses with data requested from the Dutch regulatory authority, as these were used in a previous similar analysis. ¹⁴

Study selection

Two independent reviewers will screen all titles and abstracts of available records in Vivli and retrieve full texts for studies considered relevant or unclear. This process will be facilitated by our previous meta-analyses of group-level data from acute schizophrenia trials. ¹ Discrepancies will be resolved through discussion or, if needed, by a third senior reviewer. If uncertainty persists, we will contact Vivli for clarification. Study selection will be presented in a PRISMA flowchart. ^{29, 30}

Data harmonization and integrity

IPD from eligible RCTs will be requested and obtained through Vivli, harmonized, and merged into a common dataset accompanied by a data dictionary. We will consider data on study identification, methodology, population and interventions relevant for study description and analysis (see Table 1 for a tentative list of variables), as well as outcome data across different time points (see “Outcomes”) and risk of bias assessments (see “Study risk of bias”).

Data integrity will be assessed by checking for missing values, duplicates, outliers, and verifying the appropriateness of randomization. To ensure internal consistency, IPD from each study will be summarized into aggregate form and cross-checked against summary statistics from published reports, which are expected to be available and were double-extracted in our previous meta-analysis. ¹ Any discrepancies will be addressed through discussion or by contacting Vivli.

Study risk of bias

Two independent reviewers will evaluate risk of bias using the Risk of Bias 2 (RoB2) ³¹ tool, following the approach used in our previous meta-analyses. ^{1, 33} RoB2 evaluates potential biases in the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of reported results, and provides an overall risk-of-bias judgment. ³¹ The domain of missing outcome data will be assessed based on the IPD after imputation, rather than relying on published reports.

We will consider a comprehensive set of efficacy, acceptability, and tolerability outcomes relevant for investigating sex differences, selected based on previous meta-analyses of antipsychotics ¹ and refined with input from clinicians and individuals with lived experience.

Primary outcome

The primary outcome will be overall symptom of schizophrenia measured with the Positive and Negative Syndrome Scale (PANSS) ³⁴ or Brief Psychiatric Rating Scale (BPRS) ³⁵ or any other validated scale. ³⁶ To have the same outcome measure across trials in the IPD analysis, BPRS data will be converted to PANSS by a validated method of equipercentile linking. ³⁷ If other scales are used, data will be synthesized using standardized mean differences (see “Effect sizes”), a method that will also be applied to other outcomes assessed with different instruments.

Secondary outcomes

We will examine the following secondary outcomes: -

Response to treatment defined by ≥50% reduction of PANSS/BPRS, which corresponds to at least “much improvement” according to the global impression of clinicians, and is recommended in acutely-ill patients. ³⁸

Effect sizes

For continuous outcomes, we will use mean differences (e.g., total PANSS scores, weight in kg) or standardized mean differences (SMD) when multiple scales are used to assess the same construct.

For dichotomous outcomes, we will use odds ratios (OR) due to their favourable mathematical properties in meta-analysis, ⁴² and convert them into absolute and relative risks for better interpretability, using the pooled absolute risk in the placebo group as the control event rate. ⁴³

For time-to-event outcomes, we will use hazard ratios if estimable from the available data, otherwise, we will treat them as dichotomous.

Timing

The time point of measurement will be as close to 6 weeks as possible, which is the typical duration in acute schizophrenia trials, ¹ with a minimum follow-up of 3 weeks (see “Study design”). Adverse events may occur at any time during the follow-up period.

We will use data from baseline and all available time points in the statistical modelling or for multiple imputation, and we will consider longer-term outcomes if available.

Synthesis approach

The main aim of the analysis is to examine differences between women and men across outcomes when comparing antipsychotic to placebo, analysed according to randomized allocation. As noted in the introduction, our a priori hypothesis is women may be more susceptible to side-effects and may require lower doses to achieve comparable efficacy.

We will adopt a multi-step strategy, progressing from simpler to more fine-grained synthesis methods. We will begin by presenting sex-disaggregated summary data for each study, followed by pairwise meta-analyses comparing antipsychotics with placebo, dose-response meta-analyses, and if feasible, dose-response network meta-analyses. ⁴⁴

Sex differences will be explored initially for antipsychotics as a group, and if data permit, at the level of individual drugs and/or classes according to their receptor-binding profiles. ³² Similarly, we will first assess sex differences irrespective of antipsychotic dose. If sufficient data are available, we will conduct dose-specific analyses by categorizing doses into groups (e.g., informed by previous dose-response meta-analysis ¹⁹ or expert consensus) ²⁷ or, preferably, by modelling dose-response relationships using restricted cubic splines, with knot placement informed by quantiles of the available dose distributions. ⁴⁵ If needed when pooling different antipsychotics, their doses will be converted to a common scale using dose equivalents. ¹⁹

Evidence for sex differences is expected to be drawn primarily from comparisons within trials that include both sexes; however, data from trials that enrolled only one sex will also be considered. We will first conduct stratified analyses for females and males. If sufficient data are available, we will perform regression models that include sex both as a prognostic factor (i.e., associated with outcomes irrespective of treatment) and as an effect modifier (i.e., influencing the relative treatment effect). Adjustment for additional prognostic factors will be considered in sensitivity analyses, particularly for age (accounting for potential nonlinear relationships related to menopausal status), baseline severity, and body size (e.g., weight and height, which may influence plasma drug levels), as well as other sociodemographic or clinical characteristics, depending on availability across studies (see Table 1). ^{2– 4}

The final selection of the synthesis approach, including the choice between one- or two-stage IPD meta-analysis, Bayesian or frequentist frameworks, regression model specifications, and variable standardization, will be determined a posteriori, based on data availability across studies, their clinical relevance, and input from clinicians, statisticians, and individuals with lived experience.

Heterogeneity, transitivity assumption and incoherence

We will conduct random-effects meta-analyses to account for between-study heterogeneity, with heterogeneity quantified using the between-study variance (τ-squared), assumed to be common across treatment comparisons in network meta-analysis, ²⁵ and 95% prediction intervals. In case of rare events, we will consider appropriate methods for their analysis. ⁴⁶

In network meta-analysis, the transitivity assumption is essential for valid indirect comparisons. ^{25, 47} To support this, we will apply strict eligibility criteria to ensure that included studies are sufficiently similar and that interventions could, in principle, be jointly randomized. We will further assess transitivity by examining the distribution of potential effect modifiers (e.g., age, sex, country, age of onset, duration of illness, baseline severity) across treatment comparisons. Incoherence, i.e., statistical disagreement between direct and indirect evidence, will be evaluated using both local (i.e., separate indirect and direct evidence) and global approaches (i.e., design-by-treatment interaction test). ⁴⁷

Missing outcome and covariate data

Missing outcome and covariate data will be addressed using multiple imputation, accounting for trial-specific stratification of patients and assuming data are missing at random unless there is evidence suggests otherwise. ⁴⁸ Multiple imputed datasets will be generated and analysed separately, with results pooled using Rubin’s rules. ⁴⁹

For multiple measurements, mixed models of repeated measurements will be considered.

Sensitivity analyses

We will conduct sensitivity analyses to assess the robustness of the findings for the primary outcome by: a) adjusting for additional prognostic factors that may influence sex differences; b) excluding flexible-dosing studies to rule out potential bias introduced by clinicians prescribing lower doses to women or using alternative estimates of the dosing; c) excluding studies that did not use operationalized diagnostic criteria; d) excluding open-label and single-blind studies; e) excluding studies with a high overall risk of bias; and f ) treating oral and long-acting injectable formulations of the same antipsychotic as distinct interventions.

Reporting and data availability bias

Sex differences in antipsychotic effects are not expected to be a source of publication or data availability bias. However, we will assess small-study effects using funnel plots, ⁵⁰ and by including study variance as a covariate in the model. Potential data availability bias will be explored by comparing the characteristics of studies available in Vivli with those that are not, using aggregate data already available previous meta-analyses. ¹

Confidence in the evidence

We will assess the confidence in the evidence for sex differences in the effects of antipsychotics for the following outcomes: overall symptoms, weight gain, prolactin levels, QTc interval prolongation, extrapyramidal side-effects, akathisia, sedation, and anticholinergic side-effects.

The credibility of subgroup effects will be assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach for subgroup analyses ⁵¹ and the Instrument to assess the Credibility of Effect Modification Analyses (ICEMAN). ⁵² This approach applies an algorithm based on signalling questions related to the design and analysis of the subgroup comparison, the consistency of effect modification across trials, and other relevant factors. Based on these considerations, the credibility of subgroup effects will be rated as very low, low, moderate, or high, corresponding to judgments ranging from “very likely no effect modification” to “very likely effect modification”. ^{51, 52}

If the credibility of sex differences is judged as moderate or high, we will present sex-specific treatment effects and evaluate the confidence in the evidence using the Confidence in Network Meta-Analysis (CINeMA) framework, ⁵³ considering six domains of bias: within-study bias (assessed using RoB2), ³¹ indirectness (expected to be low given the strict eligibility criteria of the included studies), reporting bias (see “Reporting and data availability bias“), heterogeneity, imprecision and incoherence (evaluated by comparing the 95% confidence/credible intervals and 95% prediction intervals against predefined equivalence ranges of |SMDs|<0.1 and ORs 0.83-1.20).

If the credibility of sex differences is judged as very low or low, we will present the overall effects alongside the sex-specific treatment effects for completeness and indicate that treatment decisions should primarily be guided by the overall population estimates, as established by previous comprehensive meta-analyses of aggregate data. ¹

Statistical software

The analysis will be conducted in R statistical software ⁵⁴ using appropriate packages such as meta, ⁵⁵ netmeta, ⁵⁶ dosresmeta ⁵⁷ and crossnma, ⁵⁸ and self-programmed routines in JAGS ^{59, 60} within the Vivli secure computing environment. ²¹

People with lived experience of schizophrenia and/or current antipsychotic treatment, both women and men, their relatives, and representatives from the patient and family organizations Bündnis für psychisch erkrankte Menschen (BASTA) and Aktionsgemeinschaft der Angehörigen psychisch Kranker (ApK e.V. Bavaria) are actively involved in all stages of the project, including reactor, advisory, and leadership roles. Patient and public involvement will be reported in accordance with the GRIPP2 short-form checklist. ⁶¹

During the conceptualization phase and preparation of the funding application (2023–2024), we held joint discussions with individuals with lived experience who identified the investigation of sex differences in antipsychotic effects as a high-priority topic. They contributed to key elements of the study design, including the definition of inclusion criteria and selection of outcomes, with particular emphasis on examining dose-effects and adverse events in addition to efficacy.

To ensure a multidisciplinary perspective, we established a steering committee composed of clinicians, methodologists, statisticians, and individuals with lived experience (listed as co-authors or acknowledged in the protocol). The committee held its kick-off meeting at the start of the project, where the study protocol was discussed, including the selection of outcomes and covariates potentially influencing sex differences (see Table 1), as well as the rationale for a stepwise synthesis strategy (see “Data synthesis”). The committee meets biannually to support progress across all stages of the review, including the interpretation of findings and the joint dissemination of results in accessible language together with individuals with lived experience.

We plan to publish the systematic review and meta-analysis as open access in peer-reviewed journals, potentially resulting in multiple publications, and to present the findings at scientific conferences. The results will be reported in accordance with the PRISMA statement ²⁹ and its extensions for network meta-analyses ⁶² and IPD, ³⁰ as well as the SAGER guidelines. ²² Lay summaries will be prepared and disseminated with the support of patient and caregiver groups (see “Patient and Public Involvement”). The findings are expected to inform clinical treatment guidelines and digital tools for shared-decision making. ^{63, 64} To support future research, we will make the aggregated data generated during the analysis freely available, including sex-disaggregated summary data for all outcomes in each study. This will be done in accordance with the FAIR principles, ⁶⁵ and we will comply with any terms outlined in the data use agreements with Vivli and YODA.

As of the date of the first submission of this protocol on 27.08.2025, we have completed the formal study identification in Vivli and initiated, but not yet completed, the screening of search results against the inclusion criteria, the receipt of IPD, and data harmonization. We have not yet commenced data synthesis.

As noted in the PROSPERO registration, a preliminary search in Vivli identified 44 trials (18,568 participants) on eight antipsychotics (aripiprazole, haloperidol, lurasidone, olanzapine, paliperidone, quetiapine, risperidone, and ziprasidone) and placebo. Following project initiation and ongoing screening, no new eligible studies have been found, and IPD from a few trials are no longer available, as anonymized data including information on sex are not accessible. The current estimate includes 37 studies (15,750 participants) for the above-mentioned interventions except for lurasidone, with additional data potentially identified after screening is complete.

There were no changes from the PROSPERO registration, except for further specification and clarification of the methods, such as a clearer presentation of the data synthesis approach and the assessment of confidence in the evidence.

There were some changes from earlier versions of the protocol registered on Vivli and YODA, which were prepared during the funding application process. Specifically, we will not apply any restrictions in the eligibility criteria based on the dose of antipsychotics, as examining dose effects was deemed a crucial component of the data synthesis. In addition, we revised the tentative list of variables ( Table 1) to focus on core factors, whereas earlier versions included an extended list containing variables that may not be considered in the current analysis due to lower relevance and/or limited expected availability across studies (e.g., current living situation, employment status).

Any deviations or additional specifications will be reported and explained in the manuscript and reflected in updates to the PROSPERO registration.

The project received approval from the Technical University of Munich ethics committee (2025-270-S-NP). This research project involves an independent secondary analysis of an existing anonymized dataset from already completed clinical trials, accessed through the Vivli platform. Vivli’s policy is to only accept anonymized data. It is the responsibility of the data contributors to ensure that they can share the data being requested and meet all global regulations and policy before sharing the data. The data contributors have conducted all necessary checks during the data contributor review.