Effectiveness of Suicide-Focused Psychosocial Interventions in Psychosis: A Systematic Review and Meta-Analysis

Suicide is a leading cause of preventable death in the United States and worldwide (1, 2). Pertaining to this rising public health concern, psychotic disorders, including schizophrenia, are associated with a significantly higher risk of suicide compared with the risk in the general population (3–5). Literature suggests that the risk of suicide death in this population has been found to be 10 to 20 times higher than that in the general population (6, 7). Pertaining to ideation, plan, and attempt, a systematic review and meta-analysis by Yates and colleagues (8) of 10 studies identified an increased risk of suicide ideation and attempt among individuals with psychotic experiences compared with the general population. Highlighting the clinical importance of this finding, a growing body of literature supports the strong associations between psychosis, suicide ideation, attempt, and suicide death. Data indicate that 40%−50% of individuals diagnosed as having schizophrenia experience suicidal thoughts (9), 20%−50% attempt suicide (10), and 4%−13% end life by suicide (9, 11). Given that symptoms of psychosis are a clinical marker of risk of future suicidal behavior (8), suicide ideation, plan, and attempt are important factors and treatment targets within this at-risk population.

Despite the impact of suicidality on public health, knowledge of risk and protective factors for suicide ideation, plan, attempt, and death among individuals with psychosis is limited; in particular, effective evidence-based interventions aimed at suicide prevention in this population are lacking (3, 10, 12). Although a number of studies have evaluated risk factors related to both suicide ideation and attempt among individuals with schizophrenia and psychosis (12–20), few controlled trials have been conducted investigating the effectiveness of psychosocial (i.e., nonpharmacological therapeutic interventions that address the psychological, social, personal, relational, or vocational problems associated with mental disorders) interventions for addressing suicide ideation, plan, attempt, and death in this population. Tarrier and colleagues (21) conducted a systematic review and meta-analysis to examine cognitive-behavioral interventions to reduce suicidal behavior (suicide ideation, plan, attempt, or suicide death) among adolescents and adults in different diagnostic groups. A significant effect was found in 28 studies for cognitive-behavioral interventions in reducing suicidal behavior; however, the samples primarily included individuals without psychosis.

To date, little is known about the efficacy of existing interventions specifically designed to address risk of suicide among individuals with psychosis, and no prior meta-analyses of psychosocial interventions in this population have been published in this important area. Donker and colleagues (22) published a systematic review of suicide prevention among those with schizophrenia spectrum disorders and psychosis; however, these authors did not produce a meta-analysis because of the small number of analyzable studies (N=6). Therefore, because suicide is a leading cause of death among individuals with schizophrenia spectrum and other psychotic disorders (3–7, 23–25), existing empirically supported therapeutic interventions for individuals experiencing psychosis may benefit markedly from an evaluation of effective treatments targeting risk of ideation, attempt, and suicide death. Furthermore, identifying efficacious therapeutic interventions for suicide prevention in this population could improve clinical care and highlight strategies for use in treatment efforts in early phases of illness, guide ensuing public health and psychiatric research, and lead to an overall reduction of the high prevalence of suicide in this population. Accordingly, given these objectives, the study reported here was a systematic review and meta-analysis that included controlled trials of suicide-focused psychosocial interventions for individuals with psychosis. We hypothesized that psychosocial interventions with a focus on suicide prevention would be related to reductions in suicide ideation, plan, attempt, and death among individuals with psychosis.

Methods

Search Strategy

The literature search was led by an experienced meta-analyst (A.Z.) in consultation with the university health sciences librarian to identify all relevant literature reporting on psychosocial interventions for suicide ideation, plan, attempt, and death among individuals with psychotic symptoms. The search period covered the years from inception to June 30, 2019. Following the Cochrane guideline, the team conducted an extensive search of literature across eight electronic databases and a manual search of four major psychiatry journals, two professional Web sites (gray literature), and reference lists in included studies (see online supplement for a detailed search protocol). Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines for reporting of systematic reviews were followed and fulfilled. Covidence, a Cochrane-recommended software, was used to manage the screening process (see online supplement). Our study was exempt from institutional review board review because it examined the report of published data and effect sizes.

Study Selection

Two authors (J.L. and M.H.) independently assessed all titles and abstracts along with the full text of potentially relevant articles. To be eligible for inclusion, a study needed to be a controlled trial; evaluate psychosocial interventions (nonmedical, nonpharmacological behavioral treatments); include suicide ideation, plan, attempt, and death variables, with intent to die; include individuals with psychotic symptoms; and be written in English. If a study included individuals with a history of psychotic symptoms but no current symptoms, it was excluded. The presence of psychotic symptoms was determined by the report of inclusion criteria involving diagnosis or use of a screening process to identify psychosis (e.g., Mini Neuropsychiatric Interview). We decided before conducting the search that if a study included both individuals with and without psychotic symptoms, it would be included only if over 70% of the participants reported the presence of psychotic symptoms. Also, studies were excluded if outcomes involved nonsuicidal self-injurious behavior.

Data Extraction

A predeveloped coding sheet was used to extract the following data (see online supplement). First, bibliographical information included authors, title, geographic area of the study, and type of report. Second, participant characteristics included age, gender, race-ethnicity, marital status, duration of psychosis, and socioeconomic status. Third, study design included the nature of the intervention (i.e., psychotherapeutic intervention versus case management), intervention integrity (i.e., cognitive-behavioral therapy [CBT] only versus CBT plus motivational interviewing), randomization method, comparison group type (i.e., treatment as usual or waitlist attention control), intervention format, and delivery setting. Fourth, intervention fidelity included provider discipline, educational background, clinical experience, whether the provider received training, and whether the provider received ongoing supervision during the study. Finally, outcome variables and their respective measurement were identified. Interscreener and interrater reliability were satisfactory, with consensus rates of 95% and 89%, respectively. All disagreements were resolved either through discussion between the two coders or by an independent third coder (L.A.B.).

Quality Assessment

The risk of bias of the included trials was assessed by using both the Cochrane Collaboration’s tool for assessing risk of bias (26) and the Jadad scale for reporting randomized controlled trials (i.e., the Oxford quality scoring system) (27). Although the two assessment tools are similar overall, the two tools provided complementary evaluation information and were both included in our quality assessment.

Statistical Analysis

Trials eligible for this meta-analysis used continuous and categorical variables to assess suicide ideation, plan, attempt, and death. For continuous outcomes, we calculated standardized mean differences (i.e., the Hedges’s g statistics) as treatment effect size estimates. For categorical outcomes, we calculated odds ratios (ORs) as treatment effect size estimates. In order to pool Hedges’s g effect size estimates and OR effect size estimates together, we converted the Hedges’s g effect size estimates into the OR metric so that all effect size estimates were comparable to each other (28). We then estimated between-study heterogeneity by using Cochran’s Q test and I² to aid the meta-analysis modeling strategy between using a fixed- versus a random-effects model. In cases when a study presented more than one effect size for the same construct, we selected one effect size that was most comparable to other effect sizes in terms of time of assessment (e.g., postintervention assessment). In cases when a study presented more than one construct (e.g., both suicide ideation and suicide attempt outcomes were presented), we included all outcomes in our final analysis. Sensitivity analysis was conducted to evaluate whether such analytical decisions changed the results.

Moderator Analysis via Meta-Regression

Given the heterogeneity statistic (Q=12.46, df=13, p=0.41, I²=84%) and the variation in conceptual understandings of psychosocial interventions across different populations, outcomes, and interventions, we used a random-effects meta-regression model to explore whether outcome type, treatment type (i.e.., therapeutic versus supportive intervention), illness phase (i.e., first-episode psychosis versus midrange psychosis versus chronic psychosis), and treatment modality (i.e., individual versus family-based interventions) moderated treatment effect. The analysis was carried out using the rma and metagen function from the metafor package in the R software program.

Results

Search Results, Study Characteristics, and Quality of Studies

Of an initial pool of 1,189 records (after duplications were removed), 18 studies were eligible for full-text review. A total of 16 studies were included for systematic review, and 11 studies that included 14 effect sizes were included in the meta-analysis (N=4,829 participants). Eight studies reported OR statistics, and three studies reported standardized mean difference statistics. Study characteristics are presented in Table 1 (3, 29–43). Overall, the included studies reported satisfactory quality of rating and relatively low risk of bias (for details, see the online supplement).

Overall Pooled and Subgroup Meta-Analysis Result

A total of 11 studies (14 effect sizes) met the inclusion criteria for meta-analysis: seven suicide ideation outcomes, three suicide attempt outcomes, three suicide death outcomes, and one suicide plan outcome. Because of the small sample size, we first pooled together all suicide ideation, plan, attempt, and death variables for meta-analysis. Using a random-effects model (Q=79.63, df=13, p<0.01, I²=84%), the pooled OR of psychosocial interventions for all outcomes was 0.57 (95% confidence interval [CI]=0.41–0.78). Specifically, for suicide ideation outcomes, a fixed-effects model (Q=7.68, df=6, p=0.26) reported a pooled OR of 0.73 (95% CI=0.55–0.97). For suicide death, a fixed-effects model (Q=3.10, df=2, p=0.21) reported a pooled OR of 0.45 (95% CI=0.30–0.68). Finally, a fixed-effects model (Q=1.81, df=2, p=0.40) of pooled suicide attempt outcomes revealed a nonsignificant OR of 0.78 (95% CI=0.57–1.07). Findings are presented in Figure 1.

Sensitivity analysis suggested that all results remained the same regardless of the analytical methods used, which strengthened the robustness of our findings (see online supplement for forest plot of the sensitivity analyses).

Exploratory Moderator and Subgroup Analysis

Given the significant heterogeneity across effect size estimates in the main model, we conducted exploratory analyses to determine whether the type of suicide outcome moderated treatment effect. Compared with the difference in treatment effect for suicide ideation outcomes, the difference in treatment effect for suicide plan, suicide attempt, and suicide death outcomes was not statistically significant (b=–0.38, b=0.06, and b=–0.48, respectively). Other exploratory moderator analyses investigating treatment type, illness phase, and delivery modality as potential moderators were all statistically nonsignificant (results not presented). In addition, exploratory subgroup analysis was performed that included only first-episode psychosis studies (N=6). The six studies reported an overall statistically significant treatment effect of psychosocial intervention for suicide outcomes (OR=0.72, p<0.001, and OR=0.61, p<0.001 using fixed- and random-effects models, respectively).

Discussion

Findings of this meta-analysis showed a statistically significant treatment effect that pooled across three (ideation, attempt, and suicide death) of the four outcomes of focus (ideation, plan, attempt, and suicide death) for controlled trials of suicide-focused psychosocial interventions for individuals with psychosis. Even though it is known that risk of suicide is significantly higher among individuals with psychosis compared with the general population (3–5), research on evidence-based interventions for suicide prevention in this vulnerable population is limited. Tarrier and colleagues’ (21) systematic review and meta-analysis found a significant treatment effect for cognitive-behavioral interventions in reducing suicidal behavior among different diagnostic groups. However, psychosocial interventions other than cognitive-behavioral approaches were not examined, the sample included adolescents in addition to adults, and only three of 28 studies included participants with psychosis.

Our finding that psychosocial interventions were related to a significant decrease in the suicide ideation, plan, attempt, and death outcomes among individuals with psychosis is promising. A significant treatment effect for psychosocial interventions in relation to ideation and suicide death was found, but no effect was found for attempt. The limited number of empirical investigations of suicide-focused psychosocial interventions for individuals with psychosis makes it challenging to speculate about potential clinical reasons for the nonsignificant finding regarding suicide attempt. First, although some studies did not measure or report on attempt as an outcome, it is possible that some interventions reduced attempts and that the reduction was either not captured empirically or not reported. Second, because suicide prevention interventions inherently focus on suicide ideation, plan, attempt, and death (i.e., an intervention cannot solely target attempt without addressing or touching upon ideation), it is possible that the main effect of treatment resided in a different or more prevalent outcome (e.g., ideation) compared with attempt. Finally, a third possible explanation is the potential role of impulsivity. Impulsivity is known to be related to suicide attempt, and data have shown that individuals with psychotic disorders display greater impulsivity compared with the general population (44). Although impulsive suicide attempts have not been examined in a population with psychosis, such attempts are related in the literature to lower levels of depression and lethality compared with planned attempts (45). As a result, it is possible that our finding of a nonsignificant effect for treatment of suicide attempt is related to the presence of impulsive versus planned attempt in this population. Future research is needed to examine this theory. Given these possibilities, however, it is important to note that the finding of a nonsignificant effect for treatment of suicide attempt is likely the result of only three studies meeting inclusion criteria with statistically extractable suicide attempt data, which limited the available data and statistical power. Furthermore, moderation analyses indicated that the pooled treatment effect did not differ significantly by suicide outcome; thus, the psychosocial interventions improved the suicide ideation, plan, attempt, and death outcomes overall, without variation by ideation, attempt, or suicide death.

In this systematic review and meta-analysis, males outnumbered females in the included studies, and in the studies that reported race, most participants identified as non-Hispanic Caucasian or Chinese. Studies took place in various psychiatric settings and among participants whose average age ranged from 21 to 51. These descriptive details of the included studies are important because they highlight gaps on which future research can focus. Six studies did not report participants’ stage of illness, one study included participants with chronic illness (of more than 14 years), one included participants in a second episode of psychosis or later, one included participants with a median 6 years of illness, and seven included participants who were in an early or first-episode phase of illness. Studies on risk and protective factors for suicide tend to include participants with longer illness duration and chronic psychotic symptoms (46–49), and it is apparent that the psychosocial intervention literature is increasingly investigating the population experiencing a first episode in relation to suicide, which is particularly important given a higher risk of suicide among these individuals compared with individuals with longer duration of illness (50, 51). It was encouraging to observe that subgroup analysis of studies focusing only on participants in an early or first-episode stage of illness reported an overall statistically significant treatment effect, which provides preliminary support for the use of psychosocial interventions for suicidal intervention among these patients.

We observed that most participants randomly assigned to treatment conditions received either a mixture of therapeutic (e.g., CBT) and supportive (e.g., case management) interventions (N=8 studies) or therapeutic interventions (N=7 studies); cognitive (N=2 studies) and cognitive-behavioral (N=4 studies) interventions were most often delivered, which further supports the understanding that cognitive-behavioral interventions are among the most researched psychosocial treatments (52–54). Not only did we note a variety of treatment modalities for suicidal intervention among participants with psychosis, we also observed a wide range in the number of sessions delivered to participants—6 weeks to 5 years—suggesting that the field is testing a variety of lengths of treatment and pointing toward the need for dosage testing. As for providers, most reported being a trained psychiatric provider (e.g., psychiatrist, psychologist, or social worker), and others utilized an interdisciplinary team-based treatment structure (e.g., psychiatrist, psychologist, nurse, social worker, and vocational specialist). With increasing enthusiasm for multi- and interdisciplinary treatment teams in behavioral health care (2, 55–58), it is promising to observe that almost 40% of the studies reviewed here used a team-based delivery approach.

This study points toward the need for additional research on the relationship between psychosocial interventions and suicide ideation, plan, attempt, and death among individuals with schizophrenia and other psychotic disorders, particularly given the small number of studies that met criteria for meta-analysis inclusion. Further investigations should examine the specific effects of interventions on the differing outcomes of suicide ideation, attempt, and suicide death. In particular, more research is needed regarding the effects of intervention on suicide attempts. Given the well-documented findings that suicide attempt and death vary by demographic factors, including age, gender, race, and education (59), more research is needed to understand how interventions affect differing experiences of suicide ideation, plan, attempt, and death. Findings of such research could inform the adaptation and delivery of suicide prevention efforts based on a client’s clinical presentation and demographic risk factors.

This study must be considered in light of several potential limitations, many of which are inherent to systematic review and meta-analysis methodologies. First, despite a rigorous process for searching and reviewing the literature, it is always a possibility that not all available studies that met the inclusion criteria were included, which may have changed the overall findings, although this is unlikely. Second, although two independent reviewers engaged in a process of blind review and coding, results of this study remain susceptible to human error. Third, we included only controlled trials, which had an impact on the external validity of our findings because results from studies with quasi-experimental and nonexperimental designs were not considered. Fourth, generalizability of findings must be considered because most study participants identified as Chinese and as experiencing a first episode of psychosis.

Fifth, this study focused on psychosocial interventions targeting suicide ideation, plan, attempt, and death. As a result, some empirically supported interventions for schizophrenia specifically were not included (e.g., social skills training [60] and supported employment [61]), because these interventions focus on outcomes other than suicide ideation, plan, attempt, and death. Sixth, measurement of suicide ideation, plan, attempt, and death outcomes varied across studies, with some using multi-item measures designed to assess for suicide outcomes and others using single items of a depression or general psychopathology scale. These differences in measurement may have influenced results, and this issue highlights the importance of future research using standardized multi-item scales validated to measure a given construct. Finally, findings of this study were likely to be affected by a small sample size, resulting in low statistical power. This further highlights the importance of conducting more high-quality trials focused on interventions for suicide ideation, plan, attempt, and death.

Conclusions

Although individuals with schizophrenia and other psychotic disorders are at high risk of suicide, the literature on efficacy of psychosocial treatment approaches remains limited. This meta-analysis found that psychosocial interventions examined in 11 studies had a significant impact on reducing suicide ideation, plan, attempt, and death in this population. More research is needed, however, to determine how interventions differentially affect the specific outcomes of suicide ideation, attempt, and death. Further research in these areas can inform individualized adaptation and delivery of suicide prevention interventions based on clinical presentation and risk factors, offering tools to address an urgent and understudied issue in mental health care.