Appendix C.  Review of Research Evidence Supporting Guideline Statements

Statement 1: Assessment of Possible Schizophrenia

Evidence for this statement comes from general principles of assessment and clinical care in psychiatric practice. Expert opinion suggests that conducting such assessments as part of the initial psychiatric evaluation improves diagnostic accuracy, appropriateness of treatment selection, and treatment safety. For additional details, see Guideline I, “Review of Psychiatric Symptoms, Trauma History, and Psychiatric Treatment History,” Guideline II, “Substance Use Assessment,” Guideline III, “Assessment of Suicide Risk,” Guideline IV, “Assessment of Risk for Aggressive Behaviors,” Guideline V, “Assessment of Cultural Factors,” and Guideline VI, “Assessment of Medical Health,” in the APA Practice Guidelines for the Psychiatric Evaluation of Adults, 3rd Edition (American Psychiatric Association 2016a). A detailed systematic review to support this statement is outside the scope of this guideline; however, less comprehensive searches of the literature did not yield any studies related to this recommendation in the context of schizophrenia treatment. Consequently, the strength of research evidence is rated as low.

Statement 2: Use of Quantitative Measures

Evidence for this statement comes from general principles of assessment and clinical care in psychiatric practice. Consequently, the strength of research evidence is rated as low. Expert opinion suggests that conducting such assessments as part of the initial psychiatric evaluation improves diagnostic accuracy, appropriateness of treatment selection, and longitudinal assessment of patient symptoms and treatment effects. This recommendation is also consistent with Guideline VII, “Quantitative Assessment,” as part of the APA Practice Guidelines for the Psychiatric Evaluation of Adults, 3rd Edition (American Psychiatric Association 2016a).

Statement 3: Evidence-Based Treatment Planning

Evidence for this statement comes from general principles of assessment and clinical care in psychiatric practice. A detailed systematic review to support this statement was outside the scope of this guideline; however, less comprehensive searches of the literature did not yield any studies that directly related to this recommendation in the context of schizophrenia treatment. Consequently, the strength of research evidence is rated as low. Nevertheless, in the bulk of the literature reviewed in the Agency for Healthcare Research and Quality (AHRQ) report (McDonagh et al. 2017), pharmacotherapy was included in all treatment arms in the studies of psychosocial interventions. Invariably, in studies of pharmacotherapies, some additional form of clinical intervention is incorporated into treatment and can include elements of patient education, supportive psychotherapy, and other brief interventions.

Statement 4: Antipsychotic Medications

Evidence for this statement comes from the AHRQ review (McDonagh et al. 2017) as well as from other high-quality meta-analyses that examined findings from randomized controlled trials (RCTs) of antipsychotic medications in schizophrenia (Huhn et al. 2019; Leucht et al. 2017). The data from placebo-controlled trials are essential in making an initial determination of whether the benefits of antipsychotic medications outweigh the harms of antipsychotic medications. Placebo-controlled trial data as well as findings from head-to-head comparison studies and network analyses provide additional information on whether the benefits and harms of specific antipsychotic medications suggest preferential use (or nonuse) as compared with other antipsychotic medications. The strength of the research evidence is rated as high in demonstrating that the benefits of treatment with an antipsychotic medication outweigh the harms, although harms are clearly present and must be taken into consideration.

Primary evidence for placebo-controlled antipsychotic trial data came from the systematic review, Bayesian meta-analysis, and meta-regression conducted by Leucht et al. (2017), which included 167 studies (total N  =  28,102) published from 1955 to 2016 that were randomized and double-blinded with placebo control groups. The authors excluded studies of acute treatment with short-acting intramuscular antipsychotic medications and relapse prevention (including studies of long-acting injectable [LAI] antipsychotic agents). Studies of clozapine were excluded because of possible superior efficacy, and studies conducted in China were excluded because of concerns about study quality. Studies were also excluded if subjects had primarily negative symptoms or significant comorbidity in either psychiatric or physical health conditions. The median study duration was 6 weeks, with almost all studies lasting 12 weeks or less in terms of primary study outcomes. None of the studies was focused on first-episode or treatment-resistant samples of subjects, and the mean illness duration was 13.4 years (standard deviation [SD] 4.7), with a mean subject age of 38.7 (SD 5.5). The number of studies available on each drug was highly variable, with chlorpromazine, haloperidol, olanzapine, and risperidone being most often studied, and limited information was available on some antipsychotic medications. Results are provided in Table C–1.

The authors found a moderate benefit of antipsychotic medications, with positive symptoms improving the most but improvements in negative symptoms, depression, quality of life, and social functioning also noted with treatment (Leucht et al. 2017). Side effects were also present but differed substantially among medications. The authors also found, however, that effect sizes for antipsychotic medications have decreased with time over the past 60 years. This seems to result from increasing placebo response rates rather than decreasing medication response, although the benefit of haloperidol as compared with placebo has decreased with time. Not surprisingly, these trends are likely to confound comparisons of newer versus older medications. Although industry sponsorship was associated with a lower effect size as compared with studies funded by other mechanisms, publication bias was observed because of the tendency to avoid publishing studies with no effect of treatment.

In the AHRQ review (McDonagh et al. 2017), few head-to-head comparison studies were available for most of the antipsychotic medications. In terms of functioning, the strength of evidence (SOE) was low. Older second-generation antipsychotics (SGAs; risperidone, olanzapine, quetiapine, ziprasidone) and paliperidone did not differ in terms of global functioning or employment rates, although social functioning with risperidone in an LAI formulation was better than with quetiapine in a single study (Rouillon et al. 2013). Measures of quality of life also showed no difference among older SGAs or between older SGAs and FGAs (specifically, haloperidol and perphenazine) on the basis of a low to moderate SOE.

In terms of response rates (McDonagh et al. 2017), there was no difference between haloperidol and risperidone (16 RCTs, N = 3,452; relative risk [RR] 0.94, 95% confidence interval [CI] 0.87–1.02; moderate SOE), aripiprazole (5 RCTs, N = 2,185; RR 1.01, 95% CI 0.76–1.34; low SOE), quetiapine (6 RCTs, N = 1,421; RR 0.99, 95% CI 0.76–1.30; low SOE), and ziprasidone (6 RCTs, N = 1,283; RR 0.98, 95% CI 0.74–1.30; low SOE). However, response with olanzapine was significantly better than with haloperidol (14 RCTs, N = 4,099; RR 0.86, 95% CI 0.78–0.96; low SOE). In addition, a network meta-analysis of 46 head-to-head RCTs showed a significantly greater likelihood of response with olanzapine (odds ratio [OR] 1.71, 95% CI 1.11–2.68) and risperidone (OR 1.41, 95% CI 1.01–2.00) than quetiapine (low SOE). Olanzapine was also associated with higher remission rates as compared with haloperidol (3 RCTs; pooled RR 0.65, 95% CI 0.45–0.94; I² = 54%; low SOE), but there was no difference in remission rates between haloperidol and ziprasidone (3 RCTs; RR 0.89, 95% CI 0.71–1.12; low SOE).

In terms of core illness symptoms (e.g., delusions, hallucinations, disorganized thinking), all SGAs that were studied were superior to placebo (standardized mean difference [SMD] –0.33 to –0.88; low SOE; McDonagh et al. 2017). Risperidone (21 RCTs, N = 4,020; mean difference [MD] 3.24, 95% CI 1.62–4.86) and olanzapine (15 RCTs, N = 4,209; MD 2.31, 95% CI 0.44–4.18) were associated with greater improvements in total Positive and Negative Syndrome Scale (PANSS) score as compared with haloperidol (moderate SOE), but no differences were noted in other comparisons of FGAs and SGAs (low SOE). With comparisons among SGAs, clozapine improved core illness symptoms more than other SGAs except for olanzapine (network meta-analysis of 212 RCTs; SMDs on PANSS or Brief Psychiatric Rating Scale [BPRS] –0.32 to –0.55; low SOE); olanzapine and risperidone improved core illness symptoms more than the other SGAs except for each other and paliperidone (SMDs –0.13 to –0.26; low SOE); and paliperidone improved core illness symptoms more than lurasidone and iloperidone (SMDs –0.17; low SOE).

For negative symptoms (McDonagh et al. 2017), haloperidol was less effective than olanzapine (5 RCTs, N = 535; MD based on the Scale for the Assessment of Negative Symptoms scores 2.56, 95% CI 0.94–4.18; moderate SOE), aripiprazole (3 RCTs, N = 1,701; MD 0.80, 95% CI 0.14–1.46), olanzapine (14 RCTs, N = 3,742; MD 1.06, 95% CI 0.46–1.67), and risperidone (22 RCTs, N = 4,142; MD 0.80, 95% CI 0.14–1.46), with the latter findings based on negative symptom scores of the PANSS and having a low SOE. Other comparisons of FGAs versus SGAs showed no effects on negative symptoms (low SOE).

In an additional network meta-analysis of 32 antipsychotic medications, Huhn et al. (2019) included 402 placebo-controlled and head-to-head randomized controlled trials that included a total of 53,463 adult participants with acute symptoms and a diagnosis of schizophrenia or a related disorder. Not included were studies that focused on individuals with a first episode of psychosis or treatment resistance and studies in which individuals had concomitant medical illnesses or a predominance of negative or depressive symptoms. For the majority of antipsychotic medications, treatment was associated with a statistically significant reduction in overall symptoms as compared with placebo, and there were few significant differences between individual drugs. With antipsychotic medications that did not differ significantly from placebo, there were numerical differences favoring the antipsychotic medication, and the number of subjects in the network meta-analysis was small, yielding a wide credible interval (CrI). Only clozapine, amisulpride, zotepine, olanzapine, and risperidone exhibited greater efficacy than many other antipsychotic medications for overall symptoms, with the greatest benefit noted with clozapine (SMD –0.89, 95% CrI –1.08 to –0.71). Discontinuation rates for inefficacy paralleled the findings for treatment efficacy (Huhn et al. 2019). In terms of positive symptoms, negative symptoms, and depressive symptoms, the majority of the medications showed a statistically significant difference from placebo, with the exception of several antipsychotic agents for which sample sizes were small and CrIs were wide. Few studies had assessed effects of antipsychotic medications on social functioning. As in the Leucht et al. (2017) meta-analysis, side-effect profiles differed considerably among the antipsychotic medications.

Few studies assessed effects of antipsychotic medications on self-harm, but among patients at high risk, the International Suicide Prevention Trial (InterSePT; Meltzer et al. 2003) found that clozapine was superior to olanzapine in preventing significant suicide attempts or hospitalization to prevent suicide (hazard ratio [HR] 0.76, 95% CI 0.58–0.97; low SOE).

In terms of dose-response effects on antipsychotic medication effectiveness, standard doses of antipsychotic medications are superior to low or very low dose treatment in reducing the risk of relapse (Uchida et al. 2011a). In addition, there is evidence of a dose-response relationship for many antipsychotic medications in short-term trials of acute efficacy (Davis and Chen 2004).

Overall discontinuation rates and time to discontinuation reflect whether a treatment is effective but also whether it is tolerable. In this regard, a network meta-analysis of 111 studies (McDonagh et al. 2017) found that rates of discontinuation were lower with the following medications:

• Olanzapine and clozapine as compared with asenapine, cariprazine, iloperidone, lurasidone, olanzapine LAI, quetiapine, risperidone, and ziprasidone (ORs range from 0.42 for clozapine vs. iloperidone to 0.69 for clozapine vs. risperidone)

• Clozapine as compared with monthly paliperidone palmitate LAI (OR 0.56, 95% CI 0.33–0.96)

• Olanzapine as compared with paliperidone (OR 0.67, 95% CI 0.50–0.89)

• Quetiapine extended release (XR) as compared with iloperidone, olanzapine LAI, or quetiapine (ORs 0.26–0.35)

• Risperidone and aripiprazole as compared with iloperidone or quetiapine (ORs 0.61–0.77).

• Risperidone and monthly aripiprazole LAI as compared with iloperidone (ORs 0.52 and 0.62, respectively)

Findings on time to discontinuation are more limited and need replication (low SOE), but they suggest that olanzapine may have a longer time to discontinuation than quetiapine, risperidone, and ziprasidone (4 months on the basis of trial data; 1.5–2.2 months shorter on the basis of observational data); clozapine may have a longer time to discontinuation than olanzapine, risperidone, or quetiapine (7.2–7.8 months in Phase 2E of the Clinical Antipsychotic Trials of Intervention Effectiveness [CATIE] study); and risperidone LAI may have a longer time to discontinuation than aripiprazole, clozapine, olanzapine, quetiapine, or ziprasidone (2.6–4 months).

A network meta-analysis (McDonagh et al. 2017), which used data from 90 head-to-head trials of greater than 6 weeks’ duration, found the risk of withdrawals due to adverse events was less with the following medications:

• Risperidone LAI as compared with clozapine (OR 0.27, 95% CI 0.10–0.71), lurasidone (OR 0.39, 95% CI 0.18–0.84), quetiapine XR (OR 0.43, 95% CI 0.22–0.81), risperidone (OR 0.50, 95% CI 0.25–0.99), and ziprasidone (OR 0.40, 95% CI 0.20–0.82)

• Olanzapine as compared with clozapine (OR 0.39, 95% CI 0.19–0.79), lurasidone (OR 0.57, 95% CI 0.34–0.94), quetiapine (OR 0.62, 95% CI 0.44–0.87), risperidone (OR 0.72, 95% CI 0.55–0.96), and ziprasidone (OR 0.58, 95% CI 0.41–0.82)

• Aripiprazole as compared with clozapine (OR 0.43, 95% CI 0.21–0.88) and ziprasidone (OR 0.64, 95% CI 0.44–0.94)

• Cariprazine as compared with clozapine (OR 0.40, 95% CI 0.17–0.95)

• Iloperidone as compared with clozapine (OR 0.34, 95% CI 0.13–0.91)

These findings had a low SOE, and head-to-head comparison data were not available for all available antipsychotic medications. For haloperidol, withdrawals due to adverse events were significantly higher than with SGAs (moderate SOE), specifically, aripiprazole (8 RCTs, N = 3,232; RR 1.25, 95% CI 1.07–1.47; I² = 0%), olanzapine (24 RCTs, N = 5,708; RR 1.89, 95% CI 1.57–2.27; I² = 0%), risperidone (25 RCTs, N = 4,581; RR 1.32, 95% CI 1.09–1.60; I² = 0%), and ziprasidone (7 RCTs, N = 1,597; RR 1.68, 95% CI 1.26–2.23; I2 = 0%).

Overall adverse event rates also favored SGAs as compared with haloperidol (moderate SOE), specifically aripiprazole (3 RCTs, N = 1,713; RR 1.11, 95% CI 1.06–1.17; I² = 0%), risperidone (8 RCTs, N = 1,313; RR 1.20, 95% CI 1.01–1.42; I² = 84%), and ziprasidone (6 RCTs, N = 1,448; RR 1.13, 95% CI 1.03–1.23; I² = 31%). Among comparisons between SGAs, no differences in overall adverse events were noted (low to moderate SOE).

In terms of mortality, comparisons were difficult because of the short duration of most studies and the small number of reported events in these clinical trials (incidence rates 0%–1.17%). Nevertheless, there were no significant mortality differences found between asenapine versus olanzapine (2 RCTs; RR 2.49, 95% CI 0.54–11.5; low SOE), quetiapine versus risperidone (2 RCTs; RR 3.24, 95% CI 0.72–14.6; low SOE), and paliperidone palmitate LAI (monthly) versus risperidone LAI (2 RCTs; RR 1.26, 95% CI 0.21–7.49; low SOE). Additional findings from retrospective cohort studies found no significant difference in the risk of all-cause (1 study, N = 48,595) or cardiovascular (2 studies, N = 55,582) mortality between risperidone, olanzapine, and quetiapine (low SOE).

For the additional harms data described in the AHRQ report (McDonagh et al. 2017), evidence was relatively limited and did not adjust for known factors that confound risk. Data on cardiac disease are mixed. A large, good-quality retrospective cohort study found no significant differences in the risk of cardiovascular death, acute coronary syndrome, or ischemic stroke between risperidone and olanzapine or quetiapine in patients ages 18–64 years within the first year of starting the drug. However, a large adverse event database study found that clozapine was significantly associated with myocarditis or cardiomyopathy, whereas olanzapine, quetiapine, and risperidone were not. In contrast, other limited evidence suggested an increased risk of cardiac arrest and arrhythmia with risperidone compared with clozapine, and data from CATIE suggested a higher estimated 10-year risk of coronary heart disease with olanzapine compared with risperidone. As compared with FGAs, the SGA aripiprazole showed a lower likelihood of cardiomyopathy or coronary heart disease.

Findings on neurological side effects such as akathisia and parkinsonism also showed significant variability among the head-to-head comparison studies, which makes it difficult to draw overall conclusions about side-effect rates or risk. For new-onset tardive dyskinesia, overall rates were low (3% of subjects treated with risperidone as compared with 1%–2% for other medications). Nevertheless, findings from observational trials suggested a significant increase in risk with risperidone as compared with olanzapine (OR 1.70, 95% CI 1.35–2.14).

Metabolic effects varied with study duration, but clinically important weight gain (defined as a 7% or more increase from baseline) was greater with olanzapine than with aripiprazole (RR 2.31), asenapine (RR 2.59), clozapine (RR 1.71), quetiapine (RR 1.82), risperidone (RR 1.81), and ziprasidone (RR 5.76) across 3.7–24 months. Olanzapine had a significantly greater risk of metabolic syndrome than risperidone (pooled OR 1.60, 95% CI 1.10–2.21; I² = 0%; follow-up of 6 weeks to 3 months) or aripiprazole (pooled OR 2.50, 95% CI 1.32–4.76; I² = 0%; follow-up of 3.5–12 months). In adults, observational evidence indicated an increased risk of new-onset diabetes with olanzapine compared with risperidone (OR 1.16, 95% CI 1.03–1.31). A single study found diabetic ketoacidosis to be increased with olanzapine compared with risperidone (OR 3.5, 95% CI 1.7–7.9), but a second study found no difference in diabetic ketoacidosis, hyperglycemia, or hyperglycemic hyperosmolar state between risperidone and olanzapine, regardless of age group, but a significantly lower risk with quetiapine compared with risperidone in older patients (adjusted HR 0.69, 95% CI 0.53–0.90).

Taken together, the findings of the AHRQ review (McDonagh et al. 2017) complement the meta-analyses of Leucht et al. (2017) and Huhn et al. (2019) in showing efficacy of antipsychotic medications, particularly for core illness symptoms but also for other outcomes. Furthermore, research evidence demonstrates no clear and consistent superiority of one antipsychotic medication as compared with other antipsychotic medications, with the exception of clozapine. In addition, the systematic reviews suggest considerable variability in side-effect profiles among antipsychotic medications, without a clear continuum of risk for individual medications when all side effects are considered.

Statement 5: Continuing Medications

Evidence in support of this statement is based primarily on the evidence for antipsychotic efficacy in improving symptoms and quality of life as well as promoting functioning (see Statement 4 earlier in the appendix). Thus, the strength of research evidence is rated as high.

Additional evidence supporting this statement comes from registry database studies and from discontinuation studies. For example, in a nationwide prospective registry study (N = 6,987) with a 5-year follow-up of individuals with first-onset schizophrenia (Kiviniemi et al. 2013), there was a significant decrease in all-cause mortality in individuals taking SGAs as compared with individuals who were not taking antipsychotic medication (OR 0.69; P = 0.005).

Another nationwide study (N = 8,719) using prospectively collected registry data found that the lowest rates of rehospitalization or death occurred in individuals who received continuing treatment with an antipsychotic medication for up to 16.4 years (Tiihonen et al. 2018). Individuals who discontinued antipsychotic medication had a risk of death that was 174% higher than that in continuous users of antipsychotic medications (HR 2.74, 95% CI 1.09–6.89), whereas the risk of death was 214% higher (HR 3.14, 95% CI 1.29–7.68) in nonusers of antipsychotic medications as compared with continuous users. Rates of treatment failure, which included rehospitalization as well as death, were also lower in individuals who received continuous treatment with an antipsychotic medication. More specifically, 38% of those who discontinued treatment experienced treatment failure as compared with a matched group of continuous users of an antipsychotic medication, in which the rate of treatment failure was 29.3%. For nonusers of antipsychotic medication, treatment failure occurred in 56.5% as compared with 34.3% of a matched group of continuous antipsychotic medication users.

Several meta-analyses have examined mortality-related data with antipsychotic treatment. A meta-analysis of studies with follow-up periods of at least 1 year found that mortality was increased in individuals who did not receive antipsychotic medication as compared with those who were treated with an antipsychotic medication (pooled risk ratio 0.57, 95% CI 0.46–0.76; P < 0.001 based on 22,141 deaths in 715,904 patient years in 4 cohort studies) (Vermeulen et al. 2017). With continuous treatment with clozapine, mortality was found to be lower in long-term follow-up (median 5.4 years) as compared with treatment with other antipsychotic medications (mortality rate ratio 0.56, 95% CI 0.36–0.85, P = 0.007 based on 1,327 deaths in 217,691 patient years in 24 studies) (Vermeulen et al. 2019).

On the basis of 10 RCTs (total N = 776) with mean study duration of 18.6 ± 5.97 months, a meta-analysis of discontinuation studies (Kishi et al. 2019) concluded that relapse rates were lower in individuals with schizophrenia who continued treatment with an antipsychotic medication as compared with those who discontinued treatment (RR 0.47, 95% CI 0.35–0.62; P < 0.00001; I² = 31%; NNT = 3). An additional meta-analysis (Thompson et al. 2018), using somewhat different inclusion and exclusion criteria for studies, also found that relapse rates were lower in individuals who received maintenance treatment (N = 230; 19%; 95% CI 0.05%–37%) as compared with those who stopped the antipsychotic medication (N = 290; 53%; 95% CI 39%–68%). Although caution may be needed in interpreting these results because of methodological considerations (Moncrieff and Steingard 2019), the findings align with expert opinion on the benefits of maintenance treatment with an antipsychotic medication (Goff et al. 2017).

Statement 6: Continuing the Same Medications

Evidence in support of this statement includes the evidence described for antipsychotic efficacy (see Statement 4 earlier in the appendix) and the evidence for continuing with antipsychotic treatment (see Statement 5 earlier in the appendix). Additional evidence that specifically addresses this guideline statement comes from randomized trials of a change in antipsychotic medication. On the basis of these studies, the strength of research evidence is rated as moderate.

The CATIE study provided important findings on medication changes (Essock et al. 2006). At the time of randomization, some individuals happened to be randomly assigned to a medication that they were already taking, whereas other individuals were assigned to a different antipsychotic medication. Individuals who were assigned to change to a different antipsychotic medication (N = 269) had an earlier time to all-cause treatment discontinuation than those assigned to continue taking the same antipsychotic medication (N = 129; Cox proportional HR 0.69; P = 0.007). Although a change from olanzapine to a different antipsychotic medication was beneficial in terms of weight gain, there were no other differences in outcome measures for individuals who switched medications as compared with those who continued with the same treatment (Rosenheck et al. 2009).

Additional evidence comes from an RCT aimed at reducing the metabolic risk of antipsychotic treatment by changing medication from olanzapine, quetiapine, or risperidone to aripiprazole (Stroup et al. 2011). Individuals were followed for 24 weeks after being assigned to continue taking their current medication (N = 106) or to switch to aripiprazole (N = 109). Although the two groups did not differ in the proportion of individuals with medication efficacy (as measured by the PANSS total score or change in Clinical Global Impression [CGI] severity score), individuals who switched medication were more likely to stop medication (43.9% vs. 24.5%; P = 0.0019), and treatment discontinuation occurred earlier in those who switched medication as compared with those who did not (HR 0.456, 95% CI 0.285–0.728; P = 0.0010). However, modest but statistically significant changes did occur in weight, serum non-high-density lipoprotein cholesterol, and serum triglycerides in individuals who switched to aripiprazole as compared with those who continued with olanzapine, quetiapine, or risperidone.

Together, these findings suggest that changes in antipsychotic medications may be appropriate for addressing significant side effects such as weight or metabolic considerations, but switching medications may also confer an increased risk of medication discontinuation, with associated risks of increased relapse and increased mortality.

Statement 7: Clozapine in Treatment-Resistant Schizophrenia

Evidence on clozapine comes from multiple RCTs, observational studies (including clinical trials and studies using administrative databases), and meta-analyses. In some instances, the studies were limited to individuals with treatment-resistant schizophrenia, whereas in other studies a formal determination of treatment resistance was not reported or possible. Nevertheless, most information about clozapine will be of relevance to patients with treatment-resistant schizophrenia because, in current practice, most individuals receive clozapine only after a lack of response to other treatments.

In comparisons of SGAs, the AHRQ report (McDonagh et al. 2017) found that, independent of prior treatment history, clozapine improved core illness symptoms more than other SGAs (except for olanzapine) and was associated with a lower risk of suicide or suicide attempts than olanzapine, quetiapine, and ziprasidone (low SOE). In addition, in treatment-resistant patients, clozapine treatment was associated with a lower rate of treatment discontinuation due to lack of efficacy than the other SGAs that were studied. It is not clear whether rates of overall treatment discontinuation with clozapine may be influenced by the increased frequency of clinical interactions related to the more intensive monitoring with clozapine as compared with other antipsychotic medications.

The AHRQ review drew on several meta-analyses related to treatment-resistant schizophrenia (Ranasinghe and Sin 2014; Samara et al. 2016; Souza et al. 2013); however, some additional studies are also relevant to this guideline statement. A meta-analysis by Siskind et al. (2016b) had considerable overlap with the meta-analysis of Samara et al. (2016) in terms of the included studies. Despite this, the findings of the two meta-analyses were somewhat different, likely due to differences in the inclusion criteria and analytic approach (Samara and Leucht 2017). Samara et al. (2016) found few significant differences in outcomes and did not find clozapine to be significantly better than most other drugs in treatment-resistant schizophrenia. Siskind et al. (2016b) found no difference for clozapine compared with other antipsychotic medications in long-term studies but did find clozapine to be superior to other medications in short-term studies and across all studies in reducing total psychotic symptoms (24 studies, N = 1,858; P < 0.005). Similarly, in terms of response to treatment (as reflected by a 20%–30% reduction in symptoms), clozapine showed higher rates of response than comparators in short-term studies of treatment-resistant schizophrenia (8 studies, total N = 598 for clozapine, 620 for comparators; RR 1.17, 95% CI 1.07–2.7; P = 0.03; absolute risk reduction 12.48%, 95% CI 7.52–17.43; NNT = 9). Again, however, studies that assessed long-term response showed no difference between clozapine and comparators. A greater benefit of clozapine than comparators was also seen when the analysis was limited to non-industry-funded trials (6 studies, N =  208; RR 1.68, 95% CI 1.20–2.35; P = 0.002). In a subsequent meta-analysis using data from the same studies, Siskind et al. (2017b) found that 40.1% of treatment-resistant individuals who received clozapine had a response, with a reduction of PANSS scores of 25.8% (22 points on the PANSS) from baseline.

In an additional network meta-analysis of 32 antipsychotic medications, Huhn et al. (2019) analyzed 402 placebo-controlled and head-to-head RCTs that included a total of 53,463 adult participants with acute symptoms and a diagnosis of schizophrenia or a related disorder. Studies that focused on individuals with a first episode of psychosis or treatment resistance were excluded, as were studies in which individuals had concomitant medical illnesses or a predominance of negative or depressive symptoms. Only clozapine, amisulpride, zotepine, olanzapine, and risperidone exhibited greater efficacy than many other antipsychotic medications for overall symptoms, with the greatest benefit noted with clozapine (SMD –0.89, 95% CrI –1.08 to –0.71). Clozapine also was statistically better than placebo and the majority of the other antipsychotic medications in terms of all-cause discontinuation (SMD 0.76, 95% CrI 0.59–0.92) and its effects on positive symptoms (SMD –0.64, 95% CrI –1.09 to –0.19), negative symptoms (SMD 0.62, 95% CrI –0.84 to –0.39), and depressive symptoms (SMD –0.52, 95% CrI –0.82 to –0.23).

Findings from studies using administrative databases also suggest benefits of treatment with clozapine. For example, a prospective nationwide study conducted over a 7.5-year period in Sweden (Tiihonen et al. 2017) found significantly reduced rates of rehospitalization with the use of clozapine as compared with no antipsychotic treatment (HR 0.53, 95% CI 0.48–0.58). In addition, the reduction in rehospitalization with clozapine was comparable to reductions in rehospitalization with LAI antipsychotic medications, whereas other oral formulations of antipsychotic medications had higher risks of rehospitalization. In comparison with oral olanzapine, clozapine had a lower rate of treatment failure (HR 0.58, 95% CI 0.53–0.63) that was comparable to the rate of treatment failure with LAI antipsychotic medications (range of HRs 0.65–0.80).

Similar benefits of clozapine were found in analysis of prospective registry data from Finland obtained for all persons with schizophrenia who received inpatient care from 1972 to 2014 (Taipale et al. 2018a). Of the 62,250 individuals in the prevalent cohort, 59% were readmitted during follow-up time of up to 20 years (median follow-up duration 14.1 years). Among oral antipsychotic medications, clozapine was associated with the lowest risk for psychiatric readmission compared with no antipsychotic use (HR 0.51, 95% CI 0.49–0.53) and for all-cause readmission (HR 0.60, 95% CI 0.58–0.61). For the 8,719 individuals with a first episode of schizophrenia, risks of psychiatric readmission and all-cause readmission were also reduced (HR 0.45, 95% CI 0.40–0.50 and HR 0.51, 95% CI 0.47–0.56, respectively).

A meta-analysis that examined effects of clozapine on hospital use also found benefits for clozapine (Land et al. 2017). Although the vast majority of studies in the meta-analysis were observational studies, use of clozapine as compared with other antipsychotic medications was associated with a significant decrease in the proportion of individuals who were hospitalized (22 studies, N = 44,718; RR 0.74, 95% CI 0.69–0.80; P < 0.001), although the time to rehospitalization did not differ.

In terms of suicide risk, subjects in the InterSePT study (Meltzer et al. 2003) who met criteria for treatment-resistant schizophrenia showed benefits of clozapine that were comparable to the benefits seen in the overall sample. For the sample as a whole, clozapine was superior to olanzapine in preventing significant suicide attempts or hospitalization to prevent suicide in high-risk patients (HR 0.76, 95% CI 0.58–0.97). Fewer clozapine-treated patients in the InterSePT study attempted suicide (P = 0.03); required hospitalizations (P = 0.05) or rescue interventions (P = 0.01) to prevent suicide; or required concomitant treatment with antidepressants (P = 0.01) or with anxiolytics or soporifics (P = 0.03). Subjects treated with clozapine were also less likely to have CGI severity of suicidality scale ratings of “much worse” or “very much worse” (HR 0.78, 95% CI 0.61–0.99) than subjects treated with olanzapine.

In terms of mortality risk, a population-based cohort study of 2,370 patients with treatment-resistant schizophrenia found a higher rate of self-harm in individuals treated with non-clozapine antipsychotic medications than in those treated with clozapine (HR 1.36, 95% CI 1.04–1.78) (Wimberley et al. 2017). There was also a higher rate of all-cause mortality in patients not receiving clozapine than in those treated with clozapine (HR 1.88, 95% CI 1.16–3.05); however, the comparator group included individuals who were not taking any antipsychotic medication. When the study subjects were limited to those who were adhering to treatment, the higher mortality during treatment with other antipsychotic medications did not reach statistical significance. In the year after clozapine discontinuation, an increase in mortality was observed (HR 2.65, 95% CI 1.47–4.78), consistent with benefits of clozapine treatment in reducing overall mortality. Another cohort study also found significant benefits of clozapine on all-cause mortality in individuals with treatment-resistant schizophrenia (adjusted HR 0.61, 95% CI 0.38–0.97; P = 0.04) (Cho et al. 2019). These findings are also consistent with results of a meta-analysis that showed significantly lower rates of long-term crude mortality in patients who received continuous treatment with clozapine as compared with patients treated with other antipsychotic medications (mortality rate ratio 0.56, 95% CI 0.36–0.85; P = 0.007) (Vermeulen et al. 2019).

In terms of side effects with clozapine, a network meta-analysis conducted as part of the AHRQ report (McDonagh et al. 2017) showed that clozapine had a higher risk of study withdrawal due to adverse events than some other SGAs (low SOE) but did not show differences in overall rates of adverse events as compared with risperidone (low SOE). In the meta-analysis by Siskind et al. (2016b), individuals treated with clozapine had a higher likelihood of experiencing sialorrhea (P < 0.001; number needed to harm [NNH] = 4), seizures (P < 0.05; NNH = 17), tachycardia (P < 0.01; NNH = 7), fever (P < 0.01; NNH = 19), dizziness (P < 0.01; NNH = 11), sedation (P < 0.001; NNH = 7), constipation (P < 0.05; NNH = 12), and nausea or vomiting (P < 0.05; NNH = 19) than individuals treated with comparator antipsychotic medications. In the meta-analysis by Leucht et al. (2013), all-cause treatment discontinuation was less likely with clozapine than placebo (OR 0.46, 95% CI 0.32–0.65), as were extrapyramidal side effects (OR 0.3, 95% CI 0.17–0.62). In contrast, weight gain (SMD 0.65, 95% CI 0.31–0.99) and sedation (OR 8.82, 95% CI 4.72–15.06) were more likely with clozapine than placebo.

In an Australian national survey of 1,049 people with a diagnosis of schizophrenia or schizoaffective disorder who reported taking any antipsychotic medication (Siskind et al. 2017a), the proportion of individuals with diabetes, obesity, and metabolic syndrome was higher in individuals taking clozapine as compared with other antipsychotic medications (adjusted ORs 1.744, 1.899, and 2.300, respectively; P < 0.001). In addition, clozapine was associated with a greater proportion of individuals with dry or watery mouth (adjusted OR 2.721; P < 0.001), difficulty swallowing (adjusted OR 1.754; P < 0.01), constipation (adjusted OR 1.996; P < 0.001), dizziness/vertigo (adjusted OR 1.571; P < 0.01), and palpitations (adjusted OR 1.543; P < 0.05). The proportion of individuals who reported trembling or shaking was significantly less in those treated with clozapine as compared with other antipsychotic agents (adjusted OR 0.581; P < 0.01).

In the network meta-analysis by Huhn et al. (2019), individuals treated with clozapine were less likely to require use of an antiparkinsonian medication (SMD 0.46, 95% CrI 0.19–0.88) than those treated with other antipsychotic agents or placebo. However, clozapine was associated with a greater degree of weight gain (SMD 2.37, 95% CrI 1.43–3.32), sedation (SMD 3.02, 95% CrI 2.52–3.37), and experiencing at least one anticholinergic side effect (SMD 2.21, 95% CrI 1.26–3.47) than placebo.

Statement 8: Clozapine in Suicide Risk

For individuals with schizophrenia who are at substantial risk for suicide, evidence on the use of clozapine comes from retrospective cohort studies and a large pragmatic, open-label RCT (N = 980). Consequently, the strength of research evidence is rated as moderate.

On the basis of findings from the InterSePT study (Meltzer et al. 2003), the AHRQ report (McDonagh et al. 2017) concluded that clozapine was superior to olanzapine in preventing significant suicide attempts or hospitalization to prevent suicide in high-risk patients (HR 0.76, 95% CI 0.58–0.97; moderate SOE). Fewer clozapine-treated patients in the InterSePT study attempted suicide (P = 0.03); required hospitalizations (P = 0.05) or rescue interventions (P = 0.01) to prevent suicide; or required concomitant treatment with antidepressants (P =  0.01) or with anxiolytics or soporifics (P = 0.03). Although there was not a significant difference in suicide deaths (5 for clozapine and 3 for olanzapine), Kaplan-Meier life table estimates indicated a significant reduction in the 2-year event rate in the clozapine group (P = 0.02), with an NNT of 12. Data from other RCTs, in which suicide-related outcomes were reported as adverse events, showed very low event rates and no differences among antipsychotic medications.

One large retrospective study (Kiviniemi et al. 2013) used a nationwide registry to follow up with patients presenting with a first episode of schizophrenia (N = 6,987). At 5 years, the risk of suicide in patients treated with clozapine was significantly reduced (OR 0.29, 95% CI 0.14–0.63), whereas suicide risk in those treated with risperidone, olanzapine, or quetiapine was comparable to the risk with no antipsychotic treatment. Another large nationwide study (N = 9,567) of patients newly starting treatment with SGAs found lower rates of suicide attempts in those beginning with clozapine as compared with other drugs studied (Bitter et al. 2013). Suicide attempt rates were 1.1% at 1 year in those treated with clozapine in contrast to suicide attempt rates that ranged from 2.1% to 3.7% for other SGAs at 1 year. The suicide attempt rate with clozapine treatment was also reduced as compared with the 6 months prior to clozapine initiation (2.2% prior to clozapine as compared with 1.1% after clozapine initiation).

For a discussion of the evidence related to the side effects of clozapine, see Statement 7 earlier in the appendix.

Statement 9: Clozapine in Aggressive Behavior

Evidence for the use of clozapine for individuals with substantial aggressive behavior is limited, and the strength of research evidence is rated as low. A systematic review on pharmacological management of persistent hostility and aggression in persons with schizophrenia spectrum disorders found 92 articles with sufficient methodological information to evaluate, although none were at low risk of bias (Victoroff et al. 2014). The authors found two studies (1 RCT, N = 157; 1 open-label, N = 44) showing that in inpatients with schizophrenia spectrum disorders, clozapine was superior to haloperidol in reducing scores on the Overt Aggression Scale (Conley et al. 2003; Ratey et al. 1993). Another RCT conducted in physically assaultive inpatients (N = 100) also found clozapine to be superior to haloperidol or olanzapine in reducing scores on the Overt Aggression Scale (Krakowski et al. 2006, 2008). In reducing hostility (as measured by PANSS or BPRS hostility items), 4 RCTs (3 in inpatients, 1 in outpatients) reported superiority of clozapine as compared with FGAs. Two of these studies (N = 48 and N = 151) compared clozapine with chlorpromazine (Claghorn et al. 1987; Niskanen et al. 1974), and the other 2 studies (N = 167 and N = 71) compared clozapine with haloperidol (Citrome et al. 2001; Kane et al. 2001).

These findings support the opinions of many experts in viewing clozapine as beneficial in patients at substantial risk of aggressive behaviors. Nevertheless, additional evidence from well-designed clinical trials is needed. For a discussion of the evidence related to the side effects of clozapine, see Statement 7 earlier in the appendix.

Statement 10: Long-Acting Injectable Antipsychotic Medications

Evidence for this guideline statement comes from the AHRQ review (McDonagh et al. 2017) as well as from other RCTs, registry database studies, cohort studies, “mirror image” studies, and meta-analyses of such trials. The findings from these studies are mixed because RCTs show few differences in outcomes between LAI antipsychotic medications and oral antipsychotic agents, whereas observational trials show consistent benefits of LAI formulations. There are a number of possible explanations for these apparent disparities related to the design of the studies and differences in study populations (Correll et al. 2016; Fagiolini et al. 2017). Individuals who are agreeable to participating in a randomized clinical trial are more likely to be adherent to treatment than a broader population of individuals with a particular diagnosis. A greater focus on adherence-related questions and a greater frequency of visits may occur in an RCT as compared with treatment as usual, which may also influence adherence or outcomes. Consequently, the possible advantages of LAIs over oral formulations in promoting or assuring adherence may be less salient in RCTs as compared with observational trials. Although each type of study has advantages and disadvantages, observational trials that use registry databases are also able to examine outcomes among large numbers of individuals over many years of follow-up in contrast to the smaller numbers and shorter follow-up periods of RCTs.

In the AHRQ review (McDonagh et al. 2017), the ability to draw conclusions about the comparative effectiveness of LAI antipsychotic medications is limited by the relatively small number of head-to-head comparison studies among LAI antipsychotic medications or for LAI formulations as compared with oral agents. Few studies assessed differences in symptoms with treatment, but clozapine was noted to be superior to aripiprazole LAI (monthly or every 6 weeks), olanzapine LAI, paliperidone LAI (monthly and every 3 months), and risperidone LAI, whereas paliperidone LAI every 3 months was superior to oral lurasidone (low SOE). For the few comparisons where data on response were available, no differences were found (low SOE). Risperidone LAI was significantly better than quetiapine in social function over 24 months but did not differ from quetiapine on measures of quality of life (low SOE). No difference in social function was found between monthly paliperidone palmitate LAI and biweekly risperidone LAI (low SOE). In terms of quality of life, oral aripiprazole and monthly aripiprazole LAI did not differ from one another with up to 2 years of follow-up (low SOE).

In terms of findings on harms, a network meta-analysis of 90 head-to-head RCTs showed that risperidone LAI had a significantly lower risk of withdrawal due to adverse events than clozapine (OR 0.27, 95% CI 0.10–0.71), lurasidone (OR 0.39, 95% CI 0.18–0.84), quetiapine XR (OR 0.43, 95% CI 0.22–0.81), risperidone (OR 0.50, 95% CI 0.25–0.99), or ziprasidone (OR 0.40, 95% CI 0.20–0.82). No differences in overall adverse events were found between aripiprazole as compared with monthly aripiprazole LAI or between paliperidone or monthly paliperidone palmitate LAI as compared with risperidone LAI (low SOE). In addition, no differences in extrapyramidal side effects were seen in a 28-week trial of aripiprazole and monthly paliperidone LAI or a network meta-analysis comparing monthly and 4- to 6-week aripiprazole LAI. However, monthly aripiprazole LAI had a greater incidence of extrapyramidal side effects (RR 1.88) and worse akathisia than oral aripiprazole in the short term but not at 1 year. For mortality, no significant difference was found between monthly paliperidone palmitate LAI versus risperidone LAI (RR 1.26, 95% CI 0.21–7.49) on the basis of 2 RCTs of 4–24 months’ duration (low SOE).

A number of other meta-analyses of RCTs are also available and provide complementary information to the findings in the AHRQ review. Ostuzzi et al. (2017) conducted a systematic review and meta-analysis of RCTs that compared the oral and LAI formulations of the same antipsychotic medication and included risperidone (6 studies), olanzapine (2 studies), aripiprazole (3 studies), zuclopenthixol (1 study), fluphenazine (7 studies), and haloperidol (2 studies). There was a small benefit for aripiprazole LAI (2 studies, N = 986; RR 0.78, 95% CI 0.64–0.95; high SOE) as compared with oral aripiprazole for all-cause discontinuation, a small benefit for oral olanzapine as compared with olanzapine LAI for discontinuation due to inefficacy (2 studies, N = 1,445; RR 1.52, 95% CI 1.12–2.07; low SOE), and a small benefit for risperidone LAI as compared with oral risperidone for hyperprolactinemia (5 studies, N = 891; RR 0.81, 95% CI 0.68–0.98; moderate SOE). The other comparisons showed no differences for these outcomes, and there were also no differences noted for nonresponse rate, relapse rate, dropouts for adverse events, extrapyramidal symptoms, or weight gain.

Kishimoto et al. (2014) examined the relative efficacy of LAI antipsychotic medications as compared with oral antipsychotic medications in relapse prevention and found that LAIs were similar to oral agents in outpatient studies lasting ≥ 1 year (12 studies; RR 0.93, 95% CI 0.71–1.07; P = 0.31) and at the longest study time point across all settings (21 studies, N = 4,950; RR 0.93, 95% CI 0.80–1.08; P = 0.35). When analyzed by drug, fluphenazine LAI showed greater benefit than oral antipsychotic agents in preventing relapse (RR 0.79, 95% CI 0.65–0.96; P = 0.02); however, the authors note that this may be mediated by a cohort effect rather than a drug-specific effect because all of the fluphenazine studies predated 1992. For drug inefficacy, calculated as the sum of relapses plus discontinuations due to inefficacy, fluphenazine LAI was again superior to oral antipsychotic agents (8 studies, N = 826; RR 0.78, 95% CI 0.66–0.91; P = 0.002), whereas olanzapine LAI was inferior to oral antipsychotic agents (N = 1,445; RR 1.52, 95% CI 1.12–2.07; P = 0.007). In preventing hospitalization, fluphenazine LAI was also superior to oral antipsychotic agents (4 studies, N = 197; RR 0.82, 95% CI 0.67–0.99, P = 0.04); however, pooled data for LAIs as compared with oral antipsychotic agents showed no statistically significant effects (10 studies, N = 2,296; RR 0.88, 95% CI 0.75–1.03; P = 0.09). No differences in all-cause discontinuation were noted in pooled analyses or for individual LAIs (fluphenazine, haloperidol, zuclopenthixol, risperidone, olanzapine).

A separate meta-analysis (Kishi et al. 2016b) examined the efficacy of paliperidone LAI or risperidone LAI as compared with oral antipsychotic agents in patients with a recent-onset psychotic disorder. Although there was significant heterogeneity in the study findings, LAIs and oral agents were comparable overall in relapse prevention (3 studies; N = 875). The LAIs did have fewer study discontinuations because of inefficacy (RR 0.34, NNT = –50) or nonadherence (RR 0.30, NNT = –33), but LAIs also had a higher incidence of tremor (RR 2.38) or at least one adverse effect (RR 1.13). In terms of mortality with LAIs as compared with oral antipsychotic agents, another meta-analysis of 52 RCTs (Kishi et al. 2016a) found no difference between LAIs and placebo or oral antipsychotics in all-cause death or death due to suicide (total N = 17,416; LAI antipsychotics = 11,360, oral antipsychotics = 3,910, and placebo = 2,146; LAI antipsychotics vs. placebo = 28.9, LAI antipsychotics vs. oral antipsychotics = 64.5).

Another RCT, the Preventing Relapse Oral Antipsychotics Compared to Injectables Evaluating Efficacy (PROACTIVE) study (P. F. Buckley et al. 2015), was a multisite trial conducted at 8 academic centers in the United States. The researchers randomly assigned patients with schizophrenia or schizoaffective disorder to risperidone LAI or the physician’s choice of an oral SGA for up to 30 months. Subjects were outpatients who were neither resistant to treatment nor experiencing a first episode of psychosis. Approximately half of the subjects (161 of 305) discontinued treatment before the end of the trial. There was no significant difference noted in the proportion with a relapse (42% for risperidone LAI vs. 32% for oral SGA; P = 0.08), time to first relapse (P = 0.13), or time to first hospitalization (P = 0.30). In addition, no significant differences between risperidone LAI and oral SGAs were noted for the bulk of symptom ratings (anxiety-depression, negative symptoms, excitement, affective flattening, avolition, asociality-anhedonia, CGI severity and CGI improvement). However, changes in symptom scores did differ between the treatment arms, with lower Scale for the Assessment of Negative Symptoms (SANS) alogia scores with oral SGAs and greater improvement in psychotic symptoms and BRPS total scores with risperidone LAI. In patients followed after an initial relapse, 32 (11%) had two relapses, and 13 (4%) had three relapses, with no significant differences in the rate or time to successive relapse between those treated with risperidone LAI and those treated with oral SGAs (Buckley et al. 2016).

In contrast to the findings from RCTs, observational studies often find benefits of LAI antipsychotic formulations as compared with oral antipsychotic formulations. Tiihonen et al. (2017) used a prospective national database in Sweden (with individuals as their own controls) to examine the risk of treatment failure, which was defined as psychiatric rehospitalization, admissions due to a suicide attempt, discontinuation of antipsychotic medication or switch to another antipsychotic medication, or death. Of the 29,823 patients, 43.7% were rehospitalized, and 71.7% met criteria for treatment failure. The LAI formulations of antipsychotic medications were associated with a 20%–30% lower risk of rehospitalization as compared with oral formulations of an antipsychotic (HR 0.78, 95% CI 0.72–0.84 for the total cohort; HR 0.68, 95% CI 0.53–0.86 for the incident cohort). For specific LAI antipsychotic medications as compared with no use of antipsychotic medication, rehospitalization risk was lowest with once-monthly paliperidone LAI (HR 0.51, 95% CI 0.41–0.64), zuclopenthixol LAI (HR 0.53, 95% CI 0.48–0.57), perphenazine LAI (HR 0.58, 95% CI 0.52–0.65), and olanzapine LAI (HR 0.58, 95% CI 0.44–0.77). Of the oral medications, rehospitalization rates were lowest with clozapine (HR 0.53, 95% CI 0.48–0.58). Rates of treatment failure were also lowest with clozapine (HR 0.58, 95% CI 0.53–0.63) and with LAI antipsychotic formulations as compared with other oral formulations (HR values for LAI formulations: perphenazine LAI 0.65, haloperidol LAI 0.67, zuclopenthixol LAI 0.69, paliperidone LAI 0.72, flupentixol LAI 0.75, olanzapine LAI 0.77, and risperidone LAI 0.80).

Tiihonen et al. (2011) also compared LAI antipsychotics with their equivalent oral formulation in a nationwide cohort of 2,588 consecutive patients in Finland who had an initial admission with a diagnosis of schizophrenia. Of those individuals, only 58.2% used an antipsychotic medication after discharge, and 45.7% of the cohort continued to take an antipsychotic for at least 30 days. For rehospitalization as well as for all-cause discontinuation, LAI antipsychotic had a lower adjusted hazard ratio (aHR) than the equivalent oral formulation (aHR 0.36, 95% CI 0.17–0.75; P = 0.007 and aHR 0.41, 95% CI 0.27–0.61; P = <0.0001, respectively). For each LAI antipsychotic as compared with its oral equivalent, rehospitalization was lower with haloperidol LAI (aHR 0.12, 95% CI 0.01–1.13; P = 0.06) but not perphenazine LAI (aHR 0.53, 95% CI 0.22–1.28; P = 0.16) or risperidone LAI (aHR 0.57, 95% CI 0.30–1.08; P = 0.09). Use of an LAI antipsychotic was also associated with lower rates of all-cause discontinuation for haloperidol LAI, perphenazine LAI, and risperidone LAI as compared with their oral equivalents (aHR 0.27, 95% CI 0.08–0.88; P = 0.03; aHR 0.32, 95% CI 0.19–0.53; P = <0.0001; and aHR 0.44, 95% CI 0.31–0.62; P = <0.0001, respectively). Zuclopenthixol LAI showed no difference from its equivalent oral formulation in either rehospitalization or all-cause discontinuation.

Taipale et al. (2018a) used the same nationwide Finnish health care registry to assess the long-term effectiveness of antipsychotic medications on the risk of psychiatry rehospitalization over follow-up periods of up to 20 years (median of 14.1 years). The sample included a prevalence cohort of 62,250 individuals as well as 8,719 individuals who were followed prospectively after a first episode of psychosis. The risk of psychiatric rehospitalization was lower with LAI antipsychotic medications than with oral antipsychotic formulations (LAI FGAs HR 0.46, 95% CI 0.40–0.54; LAI SGAs HR 0.45, 95% CI 0.39–0.52; oral FGAs HR 0.67, 95% CI 0.60–0.74; oral SGAs HR 0.57, 95% CI 0.53–0.61) in first-episode patients, as was the risk of all-cause hospitalization (LAI FGAs HR 0.58, 95% CI 0.51–0.66; LAI SGAs HR 0.56, 95% CI 0.50–0.63; oral FGAs HR 0.80, 95% CI 0.74–0.87; oral SGAs HR 0.69, 95% CI 0.66–0.73), with similar patterns noted in the prevalence cohort.

A nationwide registry was also used by Taipale et al. (2018b) to examine all-cause mortality and its relationship to medication treatment among patients with schizophrenia in Sweden. Information was available on 29,823 individuals between 2006 and 2013, of which 4,603 patients were in the incident cohort. For LAI SGAs, the cumulative mortality rate was about one-third lower than the mortality rate for equivalent oral antipsychotics in pairwise analyses (aHRs 0.67, 95% CI 0.56–0.80). Those taking an LAI formulation of an SGA had the lowest cumulative mortality (7.5%), with median follow-up of 6.9 years. Corresponding rates of cumulative mortality were 8.5% for oral SGAs, 12.2% for oral FGAs, 12.3% for LAI FGAs, and 15.2% in those who were not taking an antipsychotic medication. As compared with LAI SGAs, corresponding aHRs were 1.52 (95% CI 1.13–2.05) for oral SGAs, 1.37 (95% CI 1.01–1.86) for LAI FGAs, 1.83 (95% CI 1.33–2.50) for oral FGAs, and 3.39 (95% CI 2.53–4.56) in those who were not taking an antipsychotic medication.

MacEwan et al. (2016b) used a multistate database of U.S. Medicaid patients to examine the probability of rehospitalization after an index admission with LAI antipsychotic treatment as compared with oral antipsychotic treatment. Using multivariate logistic regression with propensity score matching for 1,450 patients with a diagnosis of schizophrenia, an LAI antipsychotic medication was associated with a lower probability of readmission at 60 days postdischarge (adjusted OR 0.60, 95% CI 0.41–0.90) but not at 30 days postdischarge.

Kishimoto and colleagues conducted meta-analyses of cohort studies and mirror image studies to compare the effectiveness of LAIs versus oral antipsychotic agents in terms of hospitalization and treatment discontinuation (Kishimoto et al. 2013, 2018). On the basis of 42 prospective and retrospective cohort studies (total N = 101,624; mean follow-up 18.6 ± 10.0 months), LAIs were found to be superior to oral antipsychotics in terms of all-cause discontinuations (10 studies, N =  37,293; risk ratio 0.78, 95% CI 0.67–0.91; P = 0.001) and hospitalization rates (15 studies, 68,009 person-years; rate ratio 0.85, 95% CI 0.78–0.93; P < 0.001) but not hospitalization risk or days of hospitalization (Kishimoto et al. 2018). However, the patients treated with an LAI antipsychotic medication had longer illness durations than those treated with oral formulations of antipsychotic medication, which may have influenced the findings. In 25 mirror image studies that followed patients for at least 6 months before and after a transition between medication formulations, LAI antipsychotic medications were superior to oral antipsychotic medications in preventing hospitalization (16 studies, N = 4,066; risk ratio 0.43, 95% CI 0.35–0.53; P < 0.001) and in decreasing the number of hospitalizations (15 studies, 6,342 person-years; rate ratio 0.38, 95% CI 0.28–0.51; P < 0.001) (Kishimoto et al. 2013).

Statement 11: Anticholinergic Medications for Acute Dystonia

This recommendation is based on expert opinion and is supported by studies of the prophylactic use of anticholinergic medications to reduce the risk of acute dystonia in the initial phases of antipsychotic therapy. The strength of research evidence for this guideline statement is rated as low. No studies were found that specifically examined the treatment of acute dystonia with anticholinergic medications in a randomized or controlled manner, although intramuscular administration of an anticholinergic agent is widely viewed as the treatment of choice for acute dystonia associated with antipsychotic therapy (Stanilla and Simpson 2017).

Information on the use of anticholinergic medications to prevent acute dystonia associated with antipsychotic therapy comes from a review of 9 studies (Arana et al. 1988), of which 4 were randomized, blinded trials (total N = 232); 2 were open trials (total N = 856); and 3 were retrospective studies (total N = 278). On the basis of data from all of these studies, prophylactic use of an anticholinergic medication was associated with 1.9-fold reduction in risk of acute dystonia (14.8% without prophylaxis vs. 7.7% with prophylaxis). In patients who received a high-potency antipsychotic agent (e.g., haloperidol), the benefits of prophylactic anticholinergic medication were even more pronounced (5.4-fold reduction in risk; 46.8% without prophylaxis vs. 8.7% with prophylaxis).

A subsequent study of consecutive psychiatric admissions (N = 646) showed a lower rate of acute dystonia in patients who received anticholinergic prophylaxis (8.5% without anticholinergic prophylaxis vs. 2.8% with anticholinergic prophylaxis), and rates of acute dystonia were greater in individuals treated with a high-potency antipsychotic agent (Spina et al. 1993). A small double-blind RCT (N = 29) showed a decrease in acute dystonia associated with antipsychotic therapy in patients who received benztropine as compared with placebo, but the results did not reach statistical significance (Goff et al. 1991). These studies suggest therapeutic effects of anticholinergic medications in acute dystonia associated with antipsychotic therapy, and although the studies were conducted in patients who received FGAs, they likely would also apply to acute dystonia when it occurs with use of SGAs.

Statement 12: Treatments for Parkinsonism

This statement is based on expert opinion, and, consequently, the strength of research evidence is rated as low. Knowledge of pharmacology and pharmacokinetics suggests that side effects such as parkinsonism may be diminished by reducing the dose of a medication or changing to a medication with a different side-effect profile and a lesser propensity for treatment-related parkinsonism. Clinical experience also suggests that an anticholinergic medication can be used to treat antipsychotic-associated parkinsonism (Stanilla and Simpson 2017). A good-quality systematic review assessed the use of anticholinergic medication compared with placebo for parkinsonism associated with antipsychotic therapy (Dickenson et al. 2017). Although many studies of anticholinergic treatment for parkinsonism were conducted decades ago and suggested benefits of anticholinergics, few of these studies met the systematic review’s inclusion criteria. In addition, sample sizes in the two included studies were small, and no definitive conclusions could be drawn from the systematic review.

Statement 13: Treatments for Akathisia

This statement is based on expert opinion, and, consequently, the strength of research evidence is rated as low. Knowledge of pharmacology and pharmacokinetics suggests that side effects such as akathisia may be diminished by reducing the dose of a medication or changing to a medication with a different side-effect profile and a lesser propensity for treatment-related akathisia. A good-quality systematic review identified some benefits of benzodiazepines for akathisia associated with antipsychotic therapy (Lima et al. 2002), but only 2 studies (total N = 27) met the inclusion criteria. Another good-quality systematic review assessed the use of β-adrenergic blocking agents in akathisia and also found insufficient evidence to draw conclusions about therapeutic benefits (3 RCTs, total N = 51; Lima et al. 2004). In addition, no reliable evidence was found to support or refute the use of anticholinergic agents as compared with placebo for akathisia associated with antipsychotic therapy (Rathbone and Soares-Weiser 2006). The literature search did not identify well-designed trials published after these systematic reviews that shed additional light on any of these treatment approaches.

Statement 14: VMAT2 Medications for Tardive Dyskinesia

This statement is based on information from a good-quality systematic review (Solmi et al. 2018b) on deutetrabenazine and valbenazine treatment; information on tetrabenazine comes from less robust clinical trials. The strength of research evidence for this guideline statement is rated as moderate.

For deutetrabenazine, data were available from 2 double-blind, placebo-controlled RCTs (Anderson et al. 2017; Fernandez et al. 2017) that enrolled subjects with moderate to severe tardive dyskinesia. Each trial lasted 12 weeks, and the dosage of deutetrabenazine was 12–48 mg/day. Treatment with deutetrabenazine was associated with a significant decrease in total Abnormal Involuntary Movement Scale (AIMS) scores (N = 413; SMD –0.40, 95% CI –0.19 to –0.62, P < 0.001; weighted mean difference [WMD] –1.44, 95% CI –0.67 to –2.19, P < 0.001) and significantly greater rates of response (defined as an AIMS score reduction of at least 50%; RR 2.13, 95% CI 1.10–4.12, P = 0.024; NNT = 7, 95% CI 3–333, P = 0.046) (Solmi et al. 2018b). The rate of treatment response increased with treatment duration during the open-label extension phase of the study (Hauser et al. 2019). Deutetrabenazine was well tolerated, with trial completion rates and rates of adverse effects similar to rates with placebo (Solmi et al. 2018b).

For valbenazine, data were available from 4 double-blind, placebo-controlled trials (total N = 488) of 4–6 weeks each using a valbenazine dosage of 12.5–100 mg/day in individuals with moderate to severe tardive dyskinesia (Citrome 2017b; Correll et al. 2017a; Factor et al. 2017; Hauser et al. 2017; Josiassen et al. 2017; Kane et al. 2017; O’Brien et al. 2015). Treatment with valbenazine was associated with a significant decrease in total AIMS scores (N = 421; SMD –0.58, 95% CI –0.26 to –0.91, P < 0.001; WMD –2.07, 95% CI –1.08 to –3.05, P < 0.001) and significantly greater rates of response (RR 3.05, 95% CI 1.81–5.11, P < 0.001; NNT = 4, 95% CI 3–6, P < 0.001; Solmi et al. 2018b). With valbenazine, as with deutetrabenazine, the rate of treatment response increased with treatment duration during the open-label extension phase of the study (Factor et al. 2017). Furthermore, in the randomized KINECT 3 study, a dose-response relationship was observed, with greater benefit at dosages of 80 mg/day as compared with 40 mg/day (Hauser et al. 2017). Valbenazine was well tolerated, with trial completion rates and rates of adverse effects that were similar to rates with placebo (Solmi et al. 2018b).

For tetrabenazine, prospective placebo-controlled data are more limited and include a single-blind trial of 20 subjects (Ondo et al. 1999), a double-blind crossover trial of 6 subjects (Godwin-Austen and Clark 1971), and another double-blind crossover trial of 24 subjects (Kazamatsuri et al. 1972). Although benefits of tetrabenazine were seen at dosages of up to 150 mg/day, the quality of evidence is not sufficient to draw robust conclusions or conduct meta-analyses (Leung and Breden 2011; Solmi et al. 2018b). Adverse effects that were more frequent with tetrabenazine than placebo included drowsiness, sedation or somnolence, parkinsonism, insomnia, anxiety, depression, and akathisia.

Although this statement specifically relates to the use of a reversible inhibitor of the vesicular monoamine transporter 2 (VMAT2) (e.g., deutetrabenazine, tetrabenazine, valbenazine), the guideline writing group also reviewed evidence for other possible treatments for tardive dyskinesia. Amantadine has been mentioned in the literature as a treatment for tardive dyskinesia, but evidence for its use is extremely limited. One randomized double-blind crossover trial (Angus et al. 1997) included only 16 patients and had significant attrition. Another randomized double-blind crossover trial (Pappa et al. 2010) also had a small sample (N = 22), and the period of treatment was only 2 weeks. Thus, data from these trials are insufficient to support use of amantadine for treatment of tardive dyskinesia. Other studies of treatments for tardive dyskinesia have been discussed in systematic reviews, as summarized in Table C–2. On the basis of these findings, there is insufficient evidence to support a guideline statement on use of these treatments in individuals with tardive dyskinesia.

Statement 15: Coordinated Specialty Care Programs

This recommendation is based on evidence from four clinical trials as presented in the AHRQ review (McDonagh et al. 2017) as well as an additional study (Anderson et al. 2018) that showed reduced mortality at 2 years for those who had participated in an early intervention program. The strength of research evidence for this statement is rated as moderate on the basis of the moderate SOE found for multiple key outcomes.

Many studies of coordinated specialty care programs were not included in the AHRQ review, although there is a long history of using such programs worldwide for early identification and treatment of attenuated psychosis syndrome, related syndromes of high psychosis risk, or first-episode psychosis (J. Addington et al. 2017; Cotton et al. 2016; Malla and McGorry 2019); however, these programs were not within the scope of this guideline recommendation because they include individuals who do not have a psychiatric diagnosis or who have diagnoses other than schizophrenia at later follow-up times (Fusar-Poli et al. 2016; Iorfino et al. 2019). Only one of the studies in a meta-analysis of early intervention programs (Marshall and Rathbone 2011) was included in the AHRQ analysis; all other studies were excluded because of such factors as small sample size, enrollment of inpatients, lack of applicability to the U.S. population, and/or use of a one-off intervention.

Pooled results from studies of individuals with a first episode of psychosis (McDonagh et al. 2017) found that up to 2 years of treatment with a coordinated specialty care program was associated with higher global functioning based on Global Assessment of Functioning (GAF) and Global Assessment Scale scores (3 RCTs; WMD 3.88, 95% CI 0.91–6.85; moderate SOE), significantly more people working or in school (3 RCTs; RR 1.22, 95% CI 1.01–1.47; moderate SOE), significantly higher ratings of quality of life (2 RCTs; effect size 0.84, 95% CI 0.14–1.55; moderate SOE), and a greater rate of retention in treatment (RR 1.27, 95% CI 1.16–1.38; Cochran’s Q = 0.03, degrees of freedom = 1; high SOE) as compared with usual care. Coordinated specialty care program participants were also less likely to relapse compared with those in usual care (2 RCTs; RR 0.64, 95% CI 0.52–0.79; moderate SOE). These treatment effects were not sustained and had generally dissipated by 5 years after treatment discontinuation. In addition, as compared with usual care, there were no significant effects of coordinated specialty care programs on housing status (2 RCTs; low SOE), self-harm (N = 506; RR 0.93, 95% CI 0.06–14.81), or total PANSS scores (3 RCTs; WMD –2.53, 95% CI –5.45 to 0.39; low SOE).

One study (Bertelsen et al. 2007; Nordentoft et al. 2002; Secher et al. 2015) found no differences in rates of accidental death (RR 0.31, 95% CI 0.01–7.59) or unexplained death (RR 0.31, 95% CI 0.01–7.56) at 2 years and no difference in mortality at 10-year follow-up (RR 0.92, 95% CI 0.45–1.88) between individuals who received a coordinated specialty care intervention as compared with usual care. In contrast, in an early intervention program for psychosis in Ontario, Canada (Anderson et al. 2018), rates of self-harm behavior (HR 0.86, 95% CI 0.18–4.24) and suicide (HR 0.73, 95% CI 0.29–1.80) did not differ during the initial 2 years after enrollment as compared with usual care, but rates of all-cause mortality were lower in the coordinated specialty care intervention group (HR 0.24, 95% CI 0.11–0.53; absolute risk reduction 2.5%; NNT = 40). As compared with those who received usual care, individuals in the coordinated specialty care program also saw a psychiatrist more rapidly (user median days = 13 compared with nonuser median days = 78), were more likely to have contact with a psychiatrist (HR 6.05, 95% CI 5.30–6.91), and were more likely to be hospitalized (HR 1.42, 95% CI 1.18–1.71) but were less likely to have emergency department visits (HR 0.71, 95% CI 0.60–0.83) or primary care contacts (HR 0.46, 95% CI 0.41–0.52).

Statement 16: Cognitive-Behavioral Therapy

Evidence in support of this statement comes from multiple RCTs and meta-analyses as described in the AHRQ review (McDonagh et al. 2017). The strength of the research evidence is rated as moderate on the basis of the evidence of CBTp benefits for core illness symptoms and short-term functioning.

In terms of overall symptoms, the AHRQ report relied primarily on a systematic review of 34 RCTs (Jauhar et al. 2014) that found CBTp to be more effective than usual care at improving overall symptoms based on symptom-based rating scales such as the PANSS and BPRS (SMD –0.33, 95% CI –0.47 to –0.19; I² = 68%; moderate SOE). The effect was less pronounced but remained significant (95% CI –0.27 to –0.03) when the analysis was restricted to trials with blinded outcome assessments. Because this review did not conduct stratified analysis by format, it is not possible to tell whether distinctions in outcome exist for individual as compared with group CBTp. Although the included studies ranged in duration from 8 weeks to 5 years, analysis of shorter as compared with longer durations of treatment was not conducted, limiting the ability to determine whether more prolonged treatment is able to maintain shorter-term treatment gains. For negative symptoms, there was no meaningful difference noted between CBTp and usual care based on data from two systematic reviews (Jauhar et al. 2014; Velthorst et al. 2015) (low SOE).

The AHRQ report (McDonagh et al. 2017) also found CBTp to be associated with improvements in global function in the short term (≤ 6 months since CBTp initiation) as measured using the GAF scale (6 trials: MD 5.35, 95% CI 1.05–9.65; I² = 77%). Removing the one study that used group CBTp from the analysis strengthened the effect and eliminated statistical heterogeneity. In one study that focused on global function (van der Gaag et al. 2011), a higher proportion of CBTp patients had normal functioning after 6 months of treatment as compared with patients who received usual care (28% vs. 14%; RR 2.21, 95% CI 1.25–3.93). Another study (Zimmer et al. 2007) found short-term benefits of an integrated cognitive-behavioral intervention on social and/or occupational function (as measured by the Social and Occupational Functioning Assessment Scale and overall functioning (GAF mean difference 5.69, 95% CI 2.05–10.97; p = 0.000). Benefits of CBTp on global, social, and occupational function were not maintained for more than 6 months after treatment cessation (low SOE). CBTp also improved quality of life more than usual care in the short term but not with longer periods of follow-up (low SOE).

Statement 17: Psychoeducation

Evidence in support of this statement comes from a good-quality systematic review as described in the AHRQ report (McDonagh et al. 2017). The strength of the research evidence is rated as moderate on the basis of the evidence of psychoeducation benefits on relapse rates.

The 10 RCTs (total N = 1,125) of psychoeducation included in the systematic review (Pekkala and Merinder 2002) varied in length (with duration of follow-up as long as 5 years), included diverse interventions, and used individual and group techniques. Although most of the studies included stabilized outpatients and were conducted in North America and northern Europe, many studies contained some individuals with a diagnosis other than schizophrenia. On the basis of the data from 6 trials, psychoeducation had a greater effect than usual care on relapse rates (with or without readmission) at 9–18 months of follow-up (RR 0.80, 95% CI 0.70–0.92; moderate SOE). Psychoeducation was also superior to usual care in terms of global functional outcomes at 1 year of follow-up (3 RCTs; MD –5.23, 95% CI –8.76 to –1.71; low SOE).

In terms of potential harms, few studies reported adverse outcomes. Nevertheless, for psychoeducation as compared with usual care, no differences were observed in the number of deaths, which were small in both groups, and rates of all-cause study dropout were also comparable between study groups (McDonagh et al. 2017).

Statement 18: Supported Employment Services

Evidence in support of this statement comes from one study comparing supported employment with usual care and an RCT and meta-analysis comparing supported employment with other vocational interventions as described in the AHRQ report (McDonagh et al. 2017). The strength of the research evidence is rated as moderate on the basis of the evidence of benefits for supported employment on obtaining competitive work.

The AHRQ review (McDonagh et al. 2017) found that supported employment, using the individual placement and support (IPS) model, results in better employment outcomes than usual care with 2 years of follow-up. Patients receiving IPS in one fair-quality trial (N = 204) were significantly more likely to obtain competitive work than those receiving usual care (75% vs. 27.5%, P = 0.001; Mueser et al. 2004). They were also more likely to obtain any form of paid work than those receiving usual care (73.9% vs. 53.6%). A large RCT (N = 1,273), with both usual care and vocational training comparisons, showed similar benefits of IPS (55% vs. 34%, P < 0.001 overall and 22% vs. 12%, P < 0.001 in the subgroup of study subjects with schizophrenia) (Cook et al. 2005). These findings are consistent with findings of a good-quality systematic review of 14 RCTs (N = 2,265) in which other vocational training interventions were used as controls (Kinoshita et al. 2013). Together, these studies provide a moderate SOE for benefits of supported employment using IPS. Although associated with a lower SOE, supported employment also showed benefits in terms of working more than 20 hours per week (13% vs. 34%, P = 0.00), having more weeks of employment overall (24 more weeks competitive and 11 more weeks any employment, P < 0.001), and longer tenure per individual job (4 weeks, P = 0.048) than those in either usual care, other vocational interventions, or both. Patients receiving IPS also reported earning more money than those in usual care ($2,078/month vs. $617.59/month, P < 0.001).

Several other meta-analyses of supported employment using somewhat different analytic methods and different inclusion and exclusion criteria than the AHRQ review found similar benefits of supported employment using the IPS approach (Frederick and VanderWeele 2019; Suijkerbuijk et al. 2017). In terms of outcomes unrelated to work, findings are less consistent but suggest potential advantages to supported employment in reducing symptoms and hospitalization risk (Burns et al. 2007; Hoffmann et al. 2014; Luciano et al. 2014).

Statement 19: Assertive Community Treatment

This recommendation is based on information from the AHRQ review (McDonagh et al. 2017), which used a good-quality systematic review (14 RCTs; N = 2,281) as a primary source (Marshall and Lockwood 2000) and also considered one additional RCT (N = 118; Sytema et al. 2007). The strength of research evidence for this guideline statement is rated as moderate on the basis of the moderate SOE found for multiple key outcomes.

The AHRQ review (McDonagh et al. 2017) focused on assertive community treatment (ACT) alone as compared with usual care and did not include a recent review in which evidence for ACT was combined with evidence for intensive case management. Significant variability was noted in study populations, with a range of ages, demographic characteristics, diagnoses, and eligibility criteria (e.g., frequent or recent hospitalization, other risk factors for reduced continuity of care). In addition, the degree of fidelity to ACT principles was often unclear, which may influence effectiveness.

Individuals who received ACT were less likely to discontinue treatment and be lost to follow-up than individuals who received usual care (12 trials; OR 0.51, 95% CI 0.41–0.63; moderate SOE). They were also less likely to be admitted to a hospital compared with individuals who received usual care (6 RCTs; OR 0.59, 95% CI 0.41–0.85; I² = 73%), and many of the reported studies also showed a decrease in the number of days in the hospital. Furthermore, individuals who received ACT were less likely to be unemployed (3 trials; OR 0.46, 95% CI 0.21–0.99; I² = 34%), homeless (4 trials; OR 0.20, 95% CI 0.09–0.47; I² = 52%), or living nonindependently (4 trials; OR 0.52, 95% CI 0.35–0.79; I² = 0%) compared with usual care (moderate SOE). Core illness symptoms also improved with ACT, but the degree of improvement was comparable to that in the usual care group (3 trials, N = 255; MD –0.14, 95% CI –0.36 to 0.08; moderate SOE). As compared with usual care, there was no significant difference in social function (pooled analysis of 3 studies; MD 0.03, 95% CI –0.28 to 0.34; low SOE) and no significant differences in arrests (2 trials, total N = 604; OR 1.17, 95% CI 0.60–2.29; I² = 0%), imprisonment (4 trials, total N = 471; OR 1.19, 95% CI 0.70–2.01; I² = 27%), or police contacts (2 trials, total N = 149; OR 0.76, 95% CI 0.32–1.79; I² = 84%) with ACT (low SOE). Findings from the additional RCT were generally consistent with the meta-analytic results. Only 2 trials reported information on quality of life, with one finding a small but statistically significant difference and the other showing no difference.

In individuals with co-occurring schizophrenia and a substance use disorder, one good-quality systematic review of 32 trials (N = 3,165) examined differences between integrated ACT and usual care (Hunt et al. 2013). For most outcomes of interest, only one or two of the studies from the systematic review contributed relevant data; however, these limited data showed no differences between integrated ACT and usual care for substance use, treatment discontinuation, function, or mortality through follow-up durations of up to 36 months.

Statement 20: Family Interventions

This guideline statement is based on 1 fair-quality systematic review (27 non-Chinese studies, total N = 2,297; Pharoah et al. 2010) and 6 additional studies (total N = 562; Barrowclough et al. 1999; Dyck et al. 2000; Garety et al. 2008; Kopelowicz et al. 2012; Mayoral et al. 2015; Sellwood et al. 2001, 2007; Valencia et al. 2007) as described in the AHRQ review (McDonagh et al. 2017). Because most family interventions are aimed at reducing relapse, the strength of research evidence for this statement is rated as moderate on the basis of the moderate SOE found for relapse in the AHRQ review with medium-term follow-up, although other outcomes had a low SOE.

The studies of family intervention described in the AHRQ review (McDonagh et al. 2017) had significant variation in content and methods of the delivered intervention (e.g., psychoeducation, motivational interviewing, behavioral family therapy, support groups, 24-hour support, communication training, stress management, role-play, homework, goal setting, development of social networks), measured outcomes, study design (e.g., setting, treatment duration, treatment frequency, follow-up duration, single-family versus multiple-family format, family members only vs. family members plus patient), and demographics of the study population (e.g., age, illness duration, symptom severity at baseline). Most studies had small sample sizes, and most had some risk of bias due to lack of reporting of randomization procedures or outcome assessment blinding. Studies conducted in China were excluded because of concerns about their applicability to Canadian and U.S. populations.

In the AHRQ review (McDonagh et al. 2017), family interventions resulted in significantly lower relapse rates than usual care when measured at 0–6 months (3 RCTs; N = 244; 23% vs. 37%; RR 0.62, 95% CI 0.41–0.92; low SOE), 7–12 months (19 RCTs; N = 1,118; 30% vs. 44%; RR 0.67, 95% CI 0.54–0.83; moderate SOE), 13–24 months (9 RCTs; N = 517; 49% vs. 61%; RR 0.75, 95% CI 0.58–0.99; low SOE), and 5 years posttreatment (2 RCTs; N = 140; 78% vs. 94%; RR 0.82, 95% CI 0.72–0.94; low SOE) but not at 25–36 months. The strongest evidence for effects of family interventions on relapse occurred in studies that included at least 10 treatment sessions over 7–12 months.

Improvements in core illness symptoms (4 RCTs; N = 223; SMD –0.46, 95% CI –0.73 to –0.20; low SOE) and negative symptoms (3 trials; N = 163; SMD –0.38, 95% CI –0.69 to –0.07; low SOE) were also found with family intervention compared with usual care. Unemployment (4 trials; N = 230; 75% vs. 66% after 6–12 months of follow-up; RR 1.09, 95% CI 0.91–1.29; I² = 0%; low SOE), independent living (3 RCTs; N = 164; 57% vs. 63% at 1 year; RR 0.83, 95% CI 0.66–1.03; low SOE), and reduction in self-harm (6 trials; N = 314; 4% vs. 6%; RR 0.85, 95% CI 0.24–3.02; I² = 23%; low SOE) were not found to be different between family intervention and usual care groups (low SOE). For social functioning, quality of life, family burden, and nonsuicide mortality, there was insufficient evidence to draw any conclusions from the available studies. Rates of treatment discontinuation varied with time in the study, but family interventions either did not differ from usual care or had fewer treatment dropouts than usual care (McDonagh et al. 2017; low SOE).

Statement 21: Self-Management Skills and Recovery-Focused Interventions

This guideline statement is based on evidence provided by a fair-quality systematic review (13 studies; total N = 1,404; Zou et al. 2013) and 1 additional fair-quality study (N = 210; Hasson-Ohayon et al. 2007) as described in the AHRQ review (McDonagh et al. 2017), as well as a meta-analysis of person-oriented recovery approaches (7 RCTs, N = 1,739; Thomas et al. 2018). The strength of research evidence for this statement is rated as low on the basis of the low SOE found for the majority of outcomes in the AHRQ review and a significant risk of bias (consistent with a low SOE) for most of the studies in the meta-analysis of person-oriented recovery approaches.

For illness self-management training and for recovery-focused interventions, interpretation of the evidence can be challenging because of the degree of heterogeneity in the content and format of the interventions. For example, illness self-management training programs are designed to improve knowledge, management of symptoms, and social and occupational functioning, with a primary goal of reducing the risk of relapse by focusing on medication management, recognizing signs of relapse, and developing a relapse prevention plan and coping skills for persistent symptoms (McDonagh et al. 2017; Substance Abuse and Mental Health Services Administration 2010a). Recovery-focused interventions can include similar approaches but are focused primarily on supporting a recovery-oriented vision that strives for community integration in the context of individual goals, needs, and strengths (Le Boutillier et al. 2011; Thomas et al. 2018). Activities of recovery-focused interventions incorporate opportunities for participants to share experiences and receive support as well as practicing strategies for success in illness self-management. With illness self-management, the interventions were typically administered in a group format, whereas recovery-focused interventions included a mix of group and individual formats as well as a mix of peer- and professional-led activities. Both illness self-management and recovery-focused interventions had significant variations in session content, duration, and number.

In terms of outcomes with illness self-management, the AHRQ review (McDonagh et al. 2017) noted a reduction in core illness symptom severity based on the BPRS (5 RCTs; pooled WMD –4.19, 95% CI –5.84 to –2.54; moderate SOE) and a reduced likelihood of relapse with completion of 10 or more self-management sessions (OR 0.41, 95% CI 0.21–0.79; low SOE). Effects of intervention were reduced if low-fidelity treatment was given or if fewer self-management sessions were completed. No significant effect of illness self-management was noted for negative symptoms (low SOE).

With recovery-focused interventions (Thomas et al. 2018), individuals in the intervention group showed a modest improvement in person-oriented recovery, empowerment, and hope immediately after the intervention (effect size 0.24, 95% CI 0.04–0.44) and at follow-up (effect size 0.21, 95% CI 0.06–0.35). Moderator analysis suggested that the greatest improvement was seen when mental health professionals and peer providers collaborated in treatment delivery.

Statement 22: Cognitive Remediation

This guideline statement is based on evidence provided by 2 good-quality systematic reviews (57 studies, total N = 2,885; Cella et al. 2017; Wykes et al. 2011), 1 good-quality trial (N = 90; Deste et al. 2015; Vita et al. 2011), and 3 fair-quality trials (N = 56–156; Farreny et al. 2012; Mueller et al. 2015; Twamley et al. 2012). The strength of research evidence for this statement is rated as low on the basis of the low SOE found for the majority of outcomes.

Studies included in the AHRQ review (McDonagh et al. 2017) used standard cognitive remediation principles (Saperstein and Medalia 2012) and usual care control comparisons, but other population and study characteristics varied (e.g., population demographics, treatment setting, individual vs. group format, drill and practice vs. drill plus strategy methodology, sessions per week, treatment duration, follow-up duration, extent of treatment fidelity, baseline symptom severity, computerized vs. noncomputerized content delivery, presence of active comparator condition).

Overall, as compared with usual care, use of cognitive remediation for 15–16 weeks of treatment was associated with small positive effects on core illness symptoms (2 trials, N = 153; SMD –0.62, 95% CI –1.01 to –0.24; low SOE), but effects were inconsistent among the studies, and symptom improvement was not sustained following treatment removal (8 RCTs; effect size 0.17, 95% CI –0.03 to 0.48). Cognitive remediation as compared with usual care was also associated with improvements in negative symptoms (1 systematic review of 18 RCTs; effect size –0.36, 95% CI –0.52 to –0.20; moderate SOE) as well as small positive effects on social, occupational, and global function (6 RCTs; effect sizes of 0.16–0.40; low SOE). Effects of intervention on cognitive functioning were outside the scope of the AHRQ review, but some evidence suggests that improved cognitive function can result from treatment with cognitive remediation, with indirect benefits for global function (Harvey et al. 2018). Only 1 study reported on health-related quality of life, and study limitations preclude drawing conclusions on this outcome. Treatment with cognitive remediation did not differ from usual care in terms of rates of treatment discontinuation (McDonagh et al. 2017). Cognitive remediation also seems to be acceptable to individuals who receive treatment in clinical settings as compared with research settings (Medalia et al. 2019).

Although the AHRQ review did not specifically assess cognitive outcomes with cognitive remediation, this has been a major focus of study and the primary target of cognitive remediation as an intervention. Results from meta-analyses (Revell et al. 2015; Wykes et al. 2011) and more recent randomized trials (D’Amato et al. 2011; Donohoe et al. 2018; Gomar et al. 2015; Keefe et al. 2012; McGurk et al. 2016; Mueller et al. 2015; Reeder et al. 2017) have been mixed, with significant heterogeneity in the degree of cognitive improvement (if any), the domains of cognitive improvement, and the persistence and generalizability of improvements. This may not be surprising given the wide variety of cognitive remediation approaches and formats that have been used in an effort to enhance cognitive processes such as attention, memory, executive function, social cognition, or meta-cognition (Delahunty and Morice 1996; Medalia et al. 2018; Pentaraki et al. 2017; Reeder et al. 2016; Wykes et al. 2011). There are also no clear-cut factors that are predictive of whether cognitive improvement will occur (Reser et al. 2019), which makes it difficult to determine how to target the intervention to individuals who are most likely to respond. Nevertheless, cognitive remediation does seem to result in improvements in cognition in individuals with schizophrenia, at least on a short-term basis (Harvey et al. 2018; Revell et al. 2015).

Statement 23: Social Skills Training

This guideline statement is based on 3 fair-quality RCTs (total N = 384; Bartels et al. 2014; Mueser et al. 2010; Valencia et al. 2007, 2013) as described in the AHRQ review (McDonagh et al. 2017). The strength of research evidence for this statement is rated as low on the basis of the low SOE found for outcomes in the AHRQ review with social skills training.

In the trials of social skills training that were included in the AHRQ review (McDonagh et al. 2017), sessions were held weekly for 24–52 weeks and included specific, progressive intervention modules on such topics as management of symptoms and medication, improving social and family relationships, and increasing functional skills such as money management. Goals of social skills training included enhanced psychosocial function and reductions in relapse and need for hospitalization. Demographic parameters, diagnoses of participants, and outcome measures varied among the trials.

Social function was noted to be significantly improved with social skills training as compared with usual care (SMD on GAF at 6 months 1.60, 95% CI 1.19–2.02; SMD on GAF at 1 year 2.02, 95% CI 1.53–2.52; SMD on Multnomah Community Ability Scale at 2 years 0.65, 95% CI 0.36–0.95; low SOE), but it was not clear whether gains were maintained after treatment discontinuation. Core illness symptoms, as measured by the PANSS, also showed more improvement with social skills training as compared with usual care (SMD at 6 months –1.50, 95% CI –1.92 to –1.09; SMD at 2 years –0.81, 95% CI –1.22 to –0.40; low SOE). Negative symptoms also improved with social skills training as compared with usual care (SMD at 6 months –1.30, 95% CI –1.70 to –0.90; SMD at 1 year –0.82, 95% CI –1.23 to –1.40; SMD at 2 years –0.45, 95% CI –0.74 to –0.15; low SOE), and in one study gains were maintained 1 year after treatment had ended. It was unclear whether relapse rates were affected by social skills training because of a small number of studies, small sample sizes, and small numbers of individuals who experienced relapse. In terms of treatment discontinuation, individuals who received social skills training did not differ from those in the usual care group (RR 1.10 at 1 year, 95% CI 0.92–1.31; RR 1.01 at 2 years, 95% CI 0.88–1.16; low SOE), with high rates of treatment retention in both groups.

Statement 24: Supportive Psychotherapy

This guideline statement is based on studies that compared supportive psychotherapy with usual care (total N = 822) in 1 good-quality systematic review (L. A. Buckley et al. 2015) as described in the AHRQ review (McDonagh et al. 2017). The strength of research evidence for this statement is rated as low on the basis of the low SOE found for study outcomes in the AHRQ review.

The studies in the systematic review (L. A. Buckley et al. 2015) were aimed primarily at helping patients with coping abilities and maintaining levels of functioning. In other respects, there was significant variation in measured outcomes, study design (e.g., setting, treatment duration, treatment frequency, follow-up duration), and demographics of the study population (e.g., age, illness duration, symptom severity at baseline). In addition, most of the included studies had some risk of bias.

The AHRQ review (McDonagh et al. 2017) found no difference in global or social function on the basis of 2 studies, but study results were not able to be pooled for analysis. Four RCTs reported information on study attrition, and no significant difference was noted between supportive psychotherapy and usual care (N = 354; RR 0.86, 95% CI 0.53–1.40; low SOE). For other outcomes, evidence was available from only a single study and sample sizes were small, making it difficult to draw reliable conclusions.