Psychological Interventions Targeting Auditory Hallucinations in Persons With Psychotic Disorders

Authors: Laura Fässler, Sarah Koop, Felix Opper, Irene Bighelli, Stefan Leucht, Michel Sabé, Malek Bajbouj, Christine Knaevelsrud, Kerem Böge

Abstract

Question

How efficacious are psychological and psychosocial interventions specifically targeting auditory hallucinations (AHs) in schizophrenia spectrum disorders (SSDs), and are there any efficacy differences between therapeutic approaches?

Findings

In this systematic review and meta-analysis including 23 studies and 2016 individuals, psychological and psychosocial interventions targeting AHs in SSDs were associated with significant reductions in AH severity at posttreatment and follow-up compared with controls, with small to moderate effect sizes. Subgroup analyses revealed that avatar therapy was the most effective among the included approaches.

Meaning

Based on the findings in this study, clinicians should consider augmenting pharmacological treatments for persons with AHs and SSDs with psychological and psychosocial interventions specifically targeting AHs, with avatar therapy showing the most promise among the included approaches.

Introduction

Auditory hallucinations (AHs) are considered a hallmark characteristic of schizophrenia spectrum disorders (SSDs), with a lifetime prevalence of 64% to 80% in SSDs. Even though AHs can be considered a fluid rather than categorical phenomenological experience, persons with SSDs commonly report severe distress associated with AHs. According to guidelines, the recommended treatment for SSDs comprises both pharmacological and complementary psychological treatment. Due to limitations in antipsychotic medication, the current literature and clinical practice are shifting toward the ubiquitous recognition of the importance of nonpharmacological approaches to treat symptoms of SSDs, such as AHs. An extensive body of randomized clinical trials (RCTs), systematic reviews, and meta-analyses has examined psychological interventions for SSDs. However, existing studies have predominantly focused on interventions aiming at treating broad symptom clusters, such as positive symptoms, negative symptoms, or functioning. Since psychological approaches are partially lacking in effectively addressing specific symptoms, such as AHs, a wide range of literature suggests targeting interventions for segregated symptoms emerging from established practices. Even though tailored treatments for AHs are promising, the current literature provides no clear overview of available therapies and their overall efficacy.

To date, and to the best of our knowledge, previous meta-analyses on targeted cognitive behavioral therapy (CBT) interventions for AHs and general nonpharmacological interventions for AHs in SSDs fail to provide a comprehensive overview of all available psychological therapies targeting AHs in SSDs. Besides, those studies did not include several additional landmark studies in the field that have been published since. To sum up, the current meta-analytic evidence does not answer the question of how efficacious psychological interventions targeting AHs in SSDs are, nor which underlying method or methods might be superior to others. To overcome these limitations, the present study aims to examine the efficacy of tailored psychological and psychosocial interventions for AHs in SSDs through a systematic review and meta-analysis. An overview of those therapies would provide valuable insights for the development of novel therapeutic strategies in research and for clinicians to implement evidence-based treatment in their practice. Additionally, we compare the effects of different interventions for AHs in SSDs based on their underlying therapeutic approach.

Methods

This systematic review and meta-analysis was preregistered on the PROSPERO database (CRD42023475704). A detailed description of the methodology can be found in the study protocol. Deviations from the preregistration are described in the eAppendix in Supplement 1. We reported results according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guideline.

Search Strategy and Selection Criteria

We included all peer-reviewed RCTs with published data on adults with a diagnosis of SSDs and current AHs. Trials were included if at least 75% of patients in each study arm had a diagnosis of SSDs, regardless of applied diagnostic criteria, illness duration, or symptom severity. We considered any psychological or psychosocial intervention that predominantly targets the treatment of AHs in SSDs without restrictions on setting. Interventions were considered to be tailored for AHs when the implemented treatment was clearly intended to primarily ameliorate or reduce any characteristics of AHs. Comparators comprised active controls, treatment as usual (TAU), waitlist (WL) controls, or a combination of these.

The primary outcome of our study was the between-group change in AH severity from baseline to posttreatment. Secondary outcomes comprised follow-up assessments of AH severity, changes in AH characteristics, and other clinical outcomes at posttreatment and follow-up (time frame from posttreatment to follow-up without additional treatment). We only extracted data measured with validated rating scales. Besides, we analyzed study and treatment dropouts as an acceptability measure and extracted adverse events (AEs) as reported in each study.

We searched the databases Embase, MEDLINE, PsycArticles, PsycInfo, PSYNDEX, and the Cochrane library for eligible RCTs that were published until August 18, 2025. In addition, we searched for relevant studies in previous reviews and meta-analyses. The initial search was conducted on November 18, 2024, with an update search on all databases on August 18, 2025 (eFigure 1 in Supplement 1). Search terms were related to relevant characteristics of inclusion criteria (eMethods 2 in Supplement 1).

Two reviewers (L.F. and S.K.) independently screened references in a 2-stage procedure. First, abstracts and titles were reviewed using the Systematic Review Facility program. Disagreements were resolved through discussion, and unresolved disagreements were included in the next screening stage. The second stage comprised full-text screenings, again independently assessed by L.F. and S.K. using a structured spreadsheet. Disagreements were resolved through discussion and, in case of doubt, rated by another independent reviewer. L.F. contacted authors to clarify missing or unclear information (eAppendix 3 in Supplement 1).

Two researchers (L.F. and S.K.) independently extracted data from selected studies (cross-checked by F.O.), using Microsoft Excel. Disagreements were discussed and study authors were contacted in case of uncertainty. Risk of bias was independently evaluated by L.F. and F.O. for the primary outcome, using the Cochrane Risk of Bias 2 tool operationalized by the Metapsy assistant. For 1 trial, S.K. served as second evaluator due to involvement of L.F. Disagreements were discussed and, in case of doubts, rated by K.B. The risk domains randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, selection of reported result, and overall risk of bias were assessed for possible biases and categorized into low risk of bias, some concerns, and high risk of bias. In addition, risk of publication bias was examined using funnel plots.

Data Analysis

We performed a meta-analysis using random-effect models due to anticipated heterogeneity. All analyses were performed in R version 4.5.1 (R Foundation) using RStudio with the metafor and tidyverse packages, and a significance level of α = .05. We calculated standardized mean differences (SMDs) for continuous outcomes, using Hedges* g* as a metric for effect sizes. Additionally, we conducted analyses on the number needed to treat (NNT) (eTable 8 in Supplement 1) and assessed confidence in the estimate with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach (eTable 2 in Supplement 1). Treatment and study dropouts were analyzed with risk ratios and dropout proportions, using novel generalized linear mixed models to implement a 1-step approach that reduces biases. To avoid carryover effects in crossover trials, we only used the first phase (before) crossover. We preferred to extract data based on intention to treat and imputation methods if reported.

Heterogeneity across studies was tested with the Cochran Q test and reported with the I^2^, following the Higgins interpretation of heterogeneity classification (0% = no, 25% = low, 50% = moderate, and 75% = high heterogeneity). Heterogeneity across trials was illustrated by funnel plots. When Q tests were significant, we additionally conducted sensitivity jackknife analyses to assess the effect of singular studies on heterogeneity. Publication bias was tested with the Egger test and funnel plots for visual assessment.

We performed sensitivity analyses according to the risk of bias judgment and control groups (active vs TAU vs WL + TAU) and analyzed potential moderations again by controls, risk of bias, and treatment duration. As sensitivity measures, we excluded studies with WL control groups, which have been shown to potentially affect treatment outcomes by nocebo effects, and studies from China due to concerns about the trustworthiness of data. We performed subgroup analyses according to the different interventional approaches used in included studies when at least 2 coherent observations could be aggregated. We categorized interventions as (1) CBT interventions; (2) avatar therapy; (3) relating and dialogue therapy; (4) self-help, online based, and computerized programs (digital), and (5) integrative approaches combining different therapeutic approaches. For subgroup and sensitivity analyses, we used mixed-effects meta-regression. All outcomes are presented with 95% CIs and were visualized with forest plots.

Results

Study Selection Process

A total of 7408 references were identified by the systematic search in 6 databases, and 5964 references were retrieved for the abstract and title screening. Of these, 109 full-text articles were examined (Figure 1). In total, we included 23 relevant studies (published between 2004 and 2025), 22 of which had usable data. For 3 studies, we were able to include unpublished data and additional information provided by the authors (eAppendix 4 in Supplement 1).

Characteristics of Studies and Samples

The studies comprised a total of 2016 participants, involving 1081 participants in the intervention and 935 in the control groups. The mean (range) sample size was 87.61 (20-345) participants, with a mean (range) age of 38.61 (25.91-45.95) years. Of the 2016 participants, 1074 identified as male (53.27%), 888 as female (44.05%), 8 (0.40%) as another gender, and 46 (2.28%) as having no gender reported.

The included trials reported results from different treatment approaches (Table 1; eAppendix 5 in Supplement 1). All studies were RCTs, with 3 studies using a partial crossover design. Furthermore, most interventions were delivered within a face-to-face setting, except for 5 studies implementing group therapy and 3 studies examining digital and unblended interventions. Control conditions predominantly comprised TAU alone (n = 14), with 3 studies using enhanced supportive counseling, 1 study using the befriending intervention, 2 studies using CBT, and 3 studies using WL conditions with additional TAU as comparators. Trial durations and number of sessions varied, with a median (range) trial duration of 12 (6-39) weeks.

Risk of Bias Assessment

Risk of bias assessment revealed a total of 6 studies with an overall low risk, another 5 studies raised some concerns, and the remaining 11 were judged as having a high risk of bias (eFigure 2 in Supplement 1).

Primary and Secondary Outcomes on AH Measures

At posttreatment, targeted interventions reduced the severity of AHs compared to controls with a small effect size (SMD, −0.18; 95% CI, −0.31 to −0.05; P = .007; NNT, 16; GRADE, low evidence) and low heterogeneity (I^2^, 31.81%; P = .06) (Figure 2). This effect was moderated by control groups (Q, 10.23; P = .02), risk of bias (Q, 9.59; P = .02), and the therapeutic approach (Q, 26.07; P < .001), while treatment duration did not moderate results (Q, 2.55; P = .11) (eTable 7 in Supplement 1). Visual inspection of the funnel plot and Egger test (z, −0.94; P = .35) did not indicate publication bias. See eTable 2 in Supplement 1 for GRADE ratings and eTable 8 in Supplement 1 for NNT results.

At follow-up (mean [range], 27.24 [12-78] weeks), the effect size for AHs severity remained small, favoring targeted interventions for AHs compared to controls (SMD, −0.15; 95% CI, −0.27 to −0.03; P = .01) with nonapparent heterogeneity (I^2^, 5.28%; P = .42). The effect size for posttreatment frequency (SMD, −0.28; 95% CI, −0.45 to −0.10; P = .002) was small to moderate and again favored interventions with moderate heterogeneity (I^2^, 47.37%; P = .006). A jackknife analysis showed that no single study was associated with heterogeneity. Effects on AH distress at posttreatment were small and favored targeted interventions (SMD, −0.19; 95% CI, −0.34 to −0.04; P = .01) with low heterogeneity (I^2^, 33.05%; P = .07). Effect sizes for beliefs about voices (malevolence) at posttreatment were small and did not significantly favor targeted interventions for AHs (SMD, −0.09; 95% CI, −0.24 to 0.07; P = .26).

Secondary Outcomes on Other Clinical Measures

At posttreatment, findings suggest that targeted psychological and psychosocial interventions for AHs reduced total symptoms (SMD, −0.29; 95% CI, −0.48 to −0.11; P = .002) with no heterogeneity (I^2^, 0%; P = .50), overall positive symptoms (SMD, −0.24; 95% CI, −0.38 to −0.11; P < .001) with low heterogeneity (I^2^, 33.80%; P = .08), delusions (SMD, −0.27; 95% CI, −0.41 to −0.12; P < .001) with no heterogeneity (I^2^, 0%; P = .95), depression (SMD, −0.19; 95% CI, −0.30 to −0.07; P = .001) with nonapparent heterogeneity (I^2^, 17.64%; P = .26), and anxiety (SMD, −0.28; 95% CI, −0.48 to −0.08; P = .006) with low heterogeneity (I^2^, 41.30%; P = .08). However, negative symptoms were not decreased by targeted interventions at posttreatment (SMD, 0.01; 95% CI, −0.13 to 0.14; P = .90) and showed no heterogeneity (I^2^, 0%; P = .97). Outcomes are summarized in Table 2. Overall, the investigation of funnel plots and Egger tests did not reveal publication bias for any secondary outcome (eTable 5 in Supplement 1). Reports on all analyses and funnel plots can be found in Supplement 1.

Subgroup and Sensitivity Analyses

According to their therapeutic methods and usable data, 6 studies examined avatar therapy, 3 studies CBT, 2 studies dialogue and relating therapy, 4 studies digital approaches, and 3 studies integrative interventions. Subgroup analyses showed that avatar therapy was the only therapeutic method that significantly reduced the severity (SMD, −0.37; 95% CI, −0.51 to −0.22; P < .001), frequency (SMD, −0.34; 95% CI, −0.48 to −0.19; P < .001), and distress (SMD, −0.29; 95% CI, −0.44 to −0.15; P < .001) of AHs at posttreatment compared to control conditions (Table 2; eTable 3 in Supplement 1).

Further sensitivity analyses showed that only targeted interventions of studies with an overall low risk of bias (k = 7) significantly reduced the severity of AHs at posttreatment compared to controls (SMD, −0.27; 95% CI, −0.44 to −0.09; P = .004). Besides, only studies with an active control condition (k = 5; SMD, −0.32; 95% CI, −0.50 to −0.15; P < .001) significantly favored the intervention groups in reducing AH severity at posttreatment. Sensitivity analyses excluding studies with WL control groups (k = 3) and studies conducted in China (k = 1) revealed findings on AH severity and secondary outcomes comparable to primary analyses, indicating robustness of results (eTable 4 in Supplement 1).

Acceptability and AEs

Reports on AEs or adverse treatment reactions were sparse, with only 6 studies assessing and reporting AEs in detail and defining their possible relation to the implemented treatments. Some studies assessed and reported additional AEs such as rehospitalizations, deaths, deterioration of symptoms, or unspecified serious AEs. Eight studies did not report on AEs (eResults 1 in Supplement 1).

The aggregated proportion of treatment dropout in the intervention groups was 17.31% with moderate heterogeneity (Table 3). In control groups, analyses revealed an aggregated treatment dropout proportion of 12.18% with high heterogeneity. Regarding study dropout, the aggregated proportion of the intervention groups was 15.01% with no heterogeneity. For control groups, the aggregated proportion was 16.21% with moderate heterogeneity. There were no differences in treatment (risk ratio, 0.97; 95% CI, 0.66 to 1.43, P = .87) or study (risk ratio, 0.91; 95% CI, 0.74 to 1.11, P = .36) dropout between the intervention and control groups.

Discussion

To the best of our knowledge, this is the first systematic review and meta-analysis of psychological and psychosocial interventions specifically targeting AHs in SSDs, synthesizing evidence from 23 RCTs (N = 2016). Overall, interventions were superior to control conditions in reducing the severity of AHs at posttreatment and follow-up with small to medium effect sizes (GRADE = low evidence). Subgroup analyses revealed that avatar therapy was consistently the most effective approach (GRADE = low evidence), with medium effect sizes at posttreatment and follow-up. This may be attributable to the AH-specific, exposure-based nature of avatar interventions, which could inform researchers and clinicians to further develop targeted interventions. In contrast, CBT, integrative, relating, and digital approaches also indicated moderate but nonsignificant effects on different characteristics of AHs, possibly due to underpowered subgroups. As shown in an extensive body of studies, overall CBT approaches are effective and robust in reducing positive symptoms in general and, among them, AHs in SSDs. However, those more generic interventions do not target specific symptoms alone but rather broader symptom clusters which may explain our contrary results to other studies on CBT. Besides, some studies did not aim to reduce the severity of AHs as the primary outcome and were powered for pilot results, emphasizing that conclusions are limited by original study aims and statistical power. Included studies reported posttreatment and follow-up periods inconsistently, with substantial variation across trials. Only a minority of studies assessed long-term follow-up (≥12 months), resulting in limited evidence regarding the long-term efficacy of psychological interventions targeting AHs. Future trials should address this limitation, particularly given the often chronic course of SSDs.

Compared with TAU and WL, only studies using active control conditions showed significant reductions in AH severity, contradicting previous evidence suggesting larger effects with less active controls. Moreover, only studies rated as having a low risk of bias demonstrated significant reductions in AH severity, underscoring the influence of bias domains on therapeutic outcomes. However, both patterns likely reflect structural dependencies, as active controls were almost exclusively implemented in avatar studies with predominantly low risk of bias, which was the most effective intervention overall. Nevertheless, these results also suggest superior effects of avatar treatments relative to more generic control approaches and the positive influences of advanced treatments. Future research should carefully consider potential moderating study characteristics, such as methodological quality, to ensure robustness of results and comparability across trials.

Our results also suggest that targeted interventions seem to be more efficacious in addressing the frequency of AHs than in reducing the overall severity and other characteristics of AHs. Despite this, no treatment effect could be observed for beliefs about AHs (malevolence), which seem to be less influential than other AH characteristics. Overall positive symptoms, total symptoms, delusions, anxiety, and depression were also reduced at posttreatment and, primarily, at follow-up, with avatar therapy being again the most effective approach. Contrary to other meta-analyses on general psychological interventions for SSDs, negative symptoms could not be decreased by tailored approaches in our analysis. We assume that studies examining general interventions that target wider symptom clusters tend to yield more general effects on both positive and negative symptoms while symptom-specific interventions appear to account for improvements in the targeted and closely associated symptoms, as supported by our findings.

The comparable aggregated study dropout in the intervention (15.01%) and control (16.21%) groups aligns with dropout rates in comparable studies on psychological interventions for SSDs, indicating high acceptability of treatments. Reports on AEs and side effects were limited, often insufficiently reported, and mostly focused on serious AEs, thereby neglecting milder but clinically relevant negative effects of interventions. Future studies should use systematic and standardized assessments of AEs to enable a more nuanced evaluation of treatment acceptability and to improve adherence, tolerability, and associated efficacy. Reports on negative effects should align with international consensus recommendations to ensure reporting on both harms and benefits of psychological interventions. In line with best-practice guidelines, researchers should clearly differentiate between adverse treatment reactions, side effects, and serious AEs, and specify their expectedness and their potential relation to treatments.

Limitations

The findings of the current study must be considered alongside some limitations. First, even though statistical heterogeneity was often low and controlled by suitable statistics, the underlying methods such as scaling of measures and therapeutic approaches varied across trials, with some that were only examined by a very small number of studies, limiting statistical power. Second, superiority of subgroups should be interpreted with caution due to the absence of head-to-head trials that need further examination. Third, most studies were rated as having a high risk of bias or some concerns, highlighting the need for future rigorously designed trials to strengthen the reliability and validity of reported effect sizes. Fourth, detailed descriptions of pharmacological treatments were sparse, restricting disentanglement of results, although randomization likely balanced medication effects. Fifth, secondary outcomes were measured with heterogeneous measures, which might have contributed to statistical heterogeneity. Sixth, we only aggregated effect sizes for the first follow-up with substantial differences across trials (3-18 months), limiting overall evidence on long-term efficacy.

Conclusions

This systematic review and meta-analysis provides evidence that psychological and psychosocial interventions targeting AHs are effective in reducing AH characteristics and SSD symptoms, with avatar therapy showing the most consistent benefits. Effect sizes ranged from low to medium, depending on approaches, control conditions, or study risk of bias. These findings could influence future studies and clinical guidelines by advancing the development of targeted, novel approaches for AHs that address relevant treatment characteristics. The use of prospectively specified inclusion criteria and a prespecified and published methodological plan strengthens confidence in the validity and reproducibility of these findings.