Key Points

We found no trials of paliperidone, asenapine, or iloperidone for off-label uses.

Attention deficit-hyperactivity disorder (ADHD). This off-label use was not included in our 2006 evidence report.

We found three placebo controlled trials (PCTs) and one active-control trial for ADHD.

One trial found risperidone superior to placebo in reducing scores on the Children's Aggression Scale–Parent version (CAS-P) in children with no serious co-occurring disorders.

Two trials of aripiprazole showed no effect on SNAP-IV (Swanson, Nolan, and Pelham teacher & parent rating scale) scores compared with placebo in children with bipolar disorder and ADHD.

One trial found risperidone led to greater reduction in SNAP-IV scores than methylphenidate in mentally retarded children with ADHD.

There were no trials of quetiapine, olanzapine, or ziprasidone for ADHD.

Anxiety. This off-label use was not included in our 2006 evidence report.

One recently published systematic review found quetiapine monotherapy superior to placebo for generalized anxiety disorder (GAD), as measured by improvement in the Hamilton Anxiety Scale (HAM-A).

We found 14 PCTs of atypicals for anxiety. Three trials of quetiapine monotherapy for GAD were clinically similar enough to pool; relative risk of responding on the HAM-A favored quetiapine over placebo. There were not enough trials of olanzapine, risperidone, or ziprasidone to pool; these trials had mixed results.

One trial showed no difference between risperidone and paroxetine on HAM-A score improvement. One trial each showed no difference in efficacy between quetiapine and paroxetine or escitalopram.

There were no trials of aripiprazole for anxiety disorders.

Dementia. Our 2006 CER focused on two published meta-analyses on use of atypicals in elderly patients with dementia. They found small but statistically significant effects for treatment with risperidone and aripiprazole, and trends toward efficacy of olanzapine and quetiapine.

The number of new trials published since 2006 justified conducting our own new meta-analyses.

In our pooled analysis of efficacy in treating overall behavioral symptoms such as aggression, motor activity and hostility, aripiprazole, olanzapine, and risperidone were superior to placebo as measured by total scores on BEHAVE-AD, Brief Psychiatric Rating Scale (BPRS), and Neuropsychiatric Inventory Scale (NPI).

Risperidone (six PCTs) was superior to placebo in decreasing psychosis symptoms such as delusions and hallucinations in elderly patients with dementia. Results for aripiprazole (three PCTs) did not meet conventional levels of statistical significance.

In our pooled analysis on agitation outcomes, aripiprazole (two PCTs), olanzapine (four PCTs), and risperidone (six PCTs) were superior to placebo.

There were no trials of ziprasidone in dementia patients.

Three head-to-head trials compared atypicals for dementia; none was found superior.

We pooled five head-to-head trials of atypicals versus haloperidol; there was no statistical difference in effect. There were too few trials to pool by specific atypical. One trial found no difference in effect between risperidone and topiramate.

Depression—major depressive disorder (MDD). Our 2006 CER reported that atypicals were not more effective as augmentation to selective serotonin reuptake inhibitors than placebo at 8 weeks. However, in some trials they led to more rapid improvement (2 to 4 weeks).

Meta-analyses published in 2007 and 2009 found atypicals superior to placebo in increasing response and remission rates, and found no statistical difference between specific atypicals.

By 2011, new trials augmenting selective serotonin reuptake inhibitors/serotonin-norepinephrine reuptake inhibitors (SSRIs/SNRIs) with atypicals had been conducted and published. We conducted new meta-analyses that showed that several atypicals have efficacy in treatment of depression when used as augmentation and that quetiapine is effective as monotherapy.

In our pooled analysis, the relative risk of responding on Hamilton depression (HAM-D) scores for participants taking quetiapine (three PCTs) or risperidone (three PCTs) as augmentation was significantly higher than for those taking placebo.

Other trials reported the Montgomery-Asberg Depression Rating Score (MADRS); the relative risk of responding for participants taking aripiprazole (three PCTs) was significantly higher than for placebo. Risperidone was only included in one PCT that reported MADRS; the drug was statistically superior to placebo. One PCT of ziprasidone reported MADRS outcomes; results were statistically superior to placebo.

The three olanzapine PCTs (included in our original 2006 report) found the drug inefficacious as monotherapy for MDD. Since then, five trials of quetiapine monotherapy have been reported. We conducted a meta-analysis of these trials; the relative risk of remitting on the MADRS was statistically superior for quetiapine compared with placebo.

One trial found quetiapine superior to lithium to HAM-D and MADRS scores.

No head-to-head trials of atypicals for MDD were found.

Eating disorders. This off-label use was not included in our 2006 report.

Five trials of olanzapine were found; three reporting body mass index (BMI) outcomes could be pooled. There was no difference in BMI increase at 1e or 3 months between participants taking olanzapine and those taking placebo. One trial of quetiapine also reported no statistical difference in BMI increase at three months.

There were no trials of aripiprazole, risperidone, or ziprasidone for treatment of eating disorders.

Insomnia. This off-label use was not included in our 2006 report.

We found only one small trial of quetiapine for this use; difference in sleep outcomes was not statistically different from placebo.

Two observational studies of olanzapine and four of quetiapine found promising improvements in sleep quality and sleep onset.

No studies of aripiprazole, risperidone, or ziprasidone for insomnia were found.

Obsessive Compulsive Disorder (OCD). Our 2006 meta-analysis found atypicals had a clinically important benefit when used as augmentation to SSRIs.

Three published meta-analyses reported similar findings.

Our 2011 analysis of PCTs reporting Y-BOCS (Yale-Brown Obsessive Compulsive Scale) outcomes showed significant effects for risperidone (three PCTs) as augmentation in treatment of refractory patients. There were too few trials (two) to permit separate pooling for olanzapine; difference in effect versus placebo was statistically insignificant in both studies.

Two new trials found quetiapine superior to placebo as augmentation to citalopram according to Y-BOCS and Clinical Global Impression Scale - Improvement subscale (CGI-I) scores.

No trials of aripiprazole for OCD were found.

One new trial found quetiapine augmentation of an SSRI superior to augmentation with clomipramine.

One head-to-head trial found no difference in effect between olanzapine and risperidone as SSRI augmentation for OCD. Another head-to-head trial found quetiapine had greater efficacy than ziprasidone for this purpose.

Personality Disorders. Our 2006 CER found three trials of olanzapine and one of aripiprazole for borderline personality disorder (BPD); all reported efficacy of the drug.

Since the original CER was published, PCTs using atypicals for treatment of BPD have shown mixed results. Due to heterogeneity of outcomes, we could not perform a meta-analysis.

Overall, olanzapine had mixed results in seven trials, aripiprazole showed efficacy in two trials, quetiapine had efficacy in one trial, and ziprasidone was found inefficacious in one trial.

Risperidone had mixed results when used to treat schizotypal personality disorder in one small trial.

No head-to-head trials of atypicals for personality disorder were found.

Post-traumatic Stress Disorder (PTSD). Our 2006 CER reported on three PCTs of atypicals as augmentation for PTSD in male veterans and three PCTs as monotherapy in abused women. We had insufficient trials to conduct meta-analysis. The trials for combat-related PTSD had beneficial results, while the other trials had mixed results.

One published meta-analysis of risperidone and olanzapine studies found atypicals superior to placebo as measured by change in CAPS score. Results were not separated by drug.

Another review which included open label trials found small positive effects for risperidone and quetiapine compared with placebo.

In 2011, five PCTs were clinically similar enough to pool using the change in Clinician Administered PTSD Scale (CAPS) as outcome. Risperidone (four trials) was superior to placebo. The other trial found olanzapine superior to placebo.

We also found a trial that reported a 3-fold decline in CAPS scores in patients treated with quetiapine monotherapy compared with patients treated with placebo. (This study did not report exact scores, so could not be pooled.)

In our meta-analysis of risperidone treatment by trial length, pooled results from at least 12 weeks followup were not statistically different from those reported at less than 12 weeks.

In our meta-analysis by condition, atypicals showed efficacy in treatment of combat-related PTSD but not PTSD in abused women.

No trials of aripiprazole, or ziprasidone for PTSD were found.

No head-to-head trials of atypicals for PTSD were found.

Substance abuse. This off-label use was not included in our 2006 CER.

We found two PCTs of aripiprazole and one of quetiapine that reported the percent of alcohol abusers completely abstinent during followup period. In our pooled analysis, the drugs had insignificant efficacy compared with placebo.

We pooled two PCTs of olanzapine and one of risperidone in cocaine users. There was no difference in efficacy versus placebo as measured by change in Addiction Severity Index (ASI).

One PCT found aripiprazole inefficacious in reducing use of intravenous amphetamine, as measured by urinalysis. Another PCT found aripiprazole inefficacious in reducing craving for methamphetamine.

One PCT of methadone clients found no difference between risperidone and placebo in reduction of cocaine or heroin use.

One trial of aripiprazole versus naltrexone in alcohol abusers found no difference in either mean number of days abstinent or percentage of participants completely abstinent.

One trial augmenting naltrexone with quetiapine found no difference from placebo augmentation in any alcohol use outcomes.

One trial of risperidone versus pergolide found neither more efficacious than placebo in reducing cocaine use.

There were no head-to-head trials of atypicals for substance abuse.

Tourette's syndrome. No new trials of atypicals have been published since our 2006 CER reported that risperidone was superior to placebo in one small PCT, and it was at least as efficacious as pimozide or clonidine for 8 to 12 weeks of therapy in the three other trials. One PCT of ziprasidone showed variable efficacy compared with placebo.

Detailed Analysis

ADHD. This off-label use was not included in our 2006 systematic review. In 2011 we found no prior meta-analyses or systematic reviews on atypical antipsychotics for ADHD. There were four randomized controlled trials (RCTs); two reported on risperidone and two on aripiprazole. The trials lasted either 4 or 6 weeks. Sample sizes were small, ranging from 16 to 45 participants. Trial quality was adequate; the mean Jadad score was 3.5. We were unable to conduct a meta-analysis due to heterogeneity of the outcomes and populations. The studies are displayed in Table 3.

Table 3

Atypical antipsychotics for ADHD.

One risperidone study showed that 100 percent of the patients “responded,” as defined by improving at least 30 percent on CAS-P. This compares WITH 77 percent of the placebo patients.⁷⁷ The other risperidone study⁷⁸ compared risperidone to methylphenidate in children and adolescents with both ADHD and moderate mental retardation. Using SNAP-IV, they found reduced ADHD symptoms with both treatments, with a greater reduction of symptoms with risperidone than methylphenidate. They also found adverse effects of weight gain with risperidone, whereas the other risperidone study had found no weight difference from placebo.⁷⁸

The two studies of aripiprazole involved children with both ADHD and bipolar disorder. Neither showed a difference in ADHD symptoms per the SNAP-IV. One study looked at aripirazole versus placebo and listed adverse events of somnolence and sialorrhea.⁷⁹ The other compared aripiprazole plus placebo versus aripiprazole plus methylphenidate and included the adverse effect of one patient experiencing a severe bipolar mixed episode while on aripiprazole and methylphenidate.⁸⁰

Anxiety. Anxiety is also a new clinical topic not included in our 2006 review. We found two prior meta-analyses on use of atypicals for this condition.⁸¹ One combined OCD trials with trials for GAD; thus, we have excluded it. Another found quetiapine monotherapy significantly better than placebo for for treatment of generalized anxiety disorder.⁸²

Our literature search identified 18 reports of trials that evaluated the use of olanzapine,^83,84 quetiapine,^85-95 risperidone,^96-99 or ziprasidone¹⁰⁰ for the treatment of anxiety. Jadad scores ranged from 2 to 5; mean score was 3.1. Sample sizes varied widely, from 7 to 873. Followup time ranged from same day (for public speaking anxiety) to 1 year. One trial had no placebo comparison group and is discussed under active controlled trials.⁹⁶ Two trials assessed anxiety outcomes in bipolar patients^92,97 so are considered beyond the scope of this report.

Of the remaining 15 PCTs, all but three^83,89,95 reported an outcome measure based on the HAM-A. These three trials studied social anxiety. The first of these trials found olanzapine superior to placebo in the treatment of social anxiety disorder;⁸³ the other two studied quetiapine and did not find it superior to placebo.^89,95

The remaining 12 PCTs ranged from 6 to 18 weeks in duration. One small pilot of quetiapine augmentation of SSRI/venlafaxine versus placebo augmentation was not considered further for analysis due to heterogeneity. This study⁸⁶ included patients with major depression and co-morbid anxiety.

Six remaining PCTs assessed quetiapine or quetiapine augmentation, two evaluated risperidone or risperidone augmentation,^98,99 one assessed olanzapine⁸⁴ and one studied ziprasidone.¹⁰⁰ These trials either reported the mean score on the HAM-A or the percent of participants that responded to treatment as measured by the HAM-A. Since trials did not consistently report the information needed to calculate a weighted mean difference for pooling of the HAM-A total score, we used the number of participants that responded to treatment as the outcome to pool. The trials defined ‘responders’ as participants who decreased their HAM-A score by at least 50 percent.

The one ziprasidone PCT¹⁰⁰ and two PCTs of quetiapine^85,93 did not report the percent or count of participants that responded to treatment and thus could not be pooled. The first of these quetiapine trials used the drug as augmentation of paroxetine for the treatment of refractory generalized anxiety disorder. This study did not find a significant benefit for quetiapine over placebo augmentation.⁸⁵ The second studied quetiapine monotherapy for maintenance treatment of generalized anxiety disorder and found a reduced risk of relapse of anxiety events compared with placebo.⁹³ The ziprasidone PCT¹⁰⁰ reported no difference in HAM-A score at 8 weeks, compared with placebo.

We separated the augmentation studies from studies of monotherapy. One small (N=20) study found quetiapine augmentation of SSRI resulted in more responders on the HAM-A than placebo augmentation (60 percent versus 30 percent) but this difference was not statistically significant.⁹⁰ A similar larger study (N=409) found no statistical difference in HAM-A response rate at eight weeks.⁹¹ The remaining three trials of quetiapine monotherapy versus placebo listed in Table 4, were pooled.^87,88,94 The trials were similar in size, ranging from 710 to 873 participants, and all had a quetiapine 150mg comparison group that was used in the analysis. The results are displayed in Figure 3, along with the olanzapine and risperidone PCTs. The pooled estimate of the relative risk of responding on the HAM-A was 1.26 (95% confidence interval [CI] 1.02, 1.56) in favor of the quetiapine groups. Resulting NNT (number needed to treat) is eight for one responder as measured by HAM-A. The I-squared statistic was 74.4 percent, indicating heterogeneity. Neither Begg's nor Egger's test for publication bias indicated the presence of bias (p=0.462, p=0.239, respectively).

Table 4

Generalized anxiety disorder—PCTs contributing to meta-analysis.

Figure 3

Anxiety % responders on Hamilton Anxiety Scale.

Active Controlled Trials. An 8-week head-to-head trial of risperidone and paroxetine for panic attacks found statistically significant improvements in the HAM-A for both groups and no difference between treatment groups on several other anxiety measures.⁹⁶

Two of the trials in our meta-analysis also had “active” arms. One trial found 50 or 150 mg/day quetiapine as effective at 8 weeks as paroxetine 20 mg/day, but with fewer sexual side effects.⁸⁸ Another trial⁸⁷ found 150 or 300 mg/day quetiapine as effective as 10 mg/day escitalopram at eight weeks.

Dementia. Our 2006 systematic review reported on two published meta-analyses assessing risperidone, quetiapine, and olanzapine for symptoms of dementia in the elderly,^101,102 and one additional meta-analysis solely on risperidone.¹⁰³ In summary, they found small but statistically significant effects for treatment with risperidone and aripiprazole, and trends toward efficacy of olanzapine and quetiapine. Since 2006, one new meta-analysis¹⁰⁴ found no statistically or clinically significant difference between atypicals and placebo. In 2010, we were able to conduct new meta-analyses that included all trials from the previously published analyses plus several newer trials.

We reviewed 38 total trials on dementia. Twenty-seven trials compared an atypical to placebo: five aripiprazole,^105-109 ten olanzapine,^110-119 six quetiapine,^119-124 and eight risperidone.^{115,116,119,125-129} One trial¹²⁰ was later determined to be a duplicate report of a published article¹²⁴ and thus excluded, leaving 37 trials total. Thirteen trials compared an atypical to another active drug.^{112,121,124,125,130-138} Two compared atypical in general to placebo.^139,140 Four trials compared one atypical drug to another;^{116,119,141,142} two are also included in our PCT analyses. Two compared the continuation of an atypical to a cessation group.^143,144 The quality of the trials varied widely, with Jadad scores ranging from zero to five; mean score was 3.0. Mean sample size was 242; range was 16 to 815. Most studies employed flexible dosing, as displayed in Figure 4 to 6. Followup times ranged from same day to 1 year.

Figure 4

Dementia placebo comparisons—total/global scores.

Figure 5

Dementia placebo comparisons—psychosis.

Figure 6

Dementia placebo comparisons—agitation.

Seventeen PCTs reported outcomes between 6 and 12 weeks; this range was considered sufficiently clinically similar to pool. These PCTs are described in detail in Table 5. We grouped study outcomes into three categories: total/global scores, psychosis, and agitation.

Table 5

Dementia—PCTs contributing to meta-analyses.

Total global score includes psychiatric symptoms of delusions, suspiciousness, dysphoria, anxiety, motor agitation, agression, hostility, euphoria, disinhibition, irritability and apathy, as measured by the NPI. Psychosis was measured by subscales of the Behavioral Pathology in Alzheimer's Disease Rating Scale (BEHAVE-AD), BPRS, and NPI, which focus primarily on delusions and hallucinations. Agitation was measured by subscales of the BEHAVE-AD, BPRS, NPI, and Cohen-Mansfield Agitation Inventory, and included the symptoms physical aggression, verbal aggression, excitability, oppositional behaviors, and excessive motor ability.

Several PCTs contained more than one treatment arm; these studies compared different doses of atypicals. For our main efficacy analyses, we pooled these arms together and present one resulting intervention outcome for each trial. This was most often done for aripiprazole trials that included a 2, 5, and 10 mg arm. We present the results by dosage later in the relevant section (Key Question 5).

There was a positive, significant difference between the atypicals as a class and placebo for all three outcome measures: total/global scores (standardized mean difference [SMD] 0.17 [95% CI 0.08, 0.25]), psychosis (SMD 0.12 [95% CI 0.04, 0.19]), and agitation (SMD 0.20 [95% CI 0.12, 0.27]). While the minimum clinically important difference is not known, these effect sizes are generally considered “small” in magnitude. The I-squared values indicated moderate heterogeneity (range 30.3 percent–53.1 percent). Results are displayed in Figures 4 to 6.

For aripiprazole, olanzapine, and risperidone, the pooled estimate of effect on the total/global score was statistically significant, with an effect size of between 0.12 and 0.20. The pooled estimate of effect for quetiapine was similar (0.13) but this was not statistically different from zero. This effect size is “small.” Corroborating this conclusion is the observation that the mean difference in the pooled NPI total score between treatment and placebo was 3.41 points, which is close to the minimum clinically observable change of 4 points.¹⁴⁵ Individual studies suggested that higher dose of aripiprazole (10mg/day)¹⁰⁷ or risperidone (2mg/day)¹²⁷ were possibly more effective than lower doses, although these findings have not been replicated, dose effects are not addressed in many trials, and dose-response trends across studies are inconsistent. Only the pooled analysis for risperidone had substantial heterogeneity (I-squared = 74.6 percent). There was no evidence of publication bias. Only risperidone had enough studies to conduct a sensitivity analysis based on quality; no difference was found. For treatment of psychosis, results favored risperidone when compared with placebo. As measured by the psychosis subscale of the NPI, pooled estimate of SMD in effect size was 0.20 (95% CI 0.05, 0.36) for risperidone (five trials). Results for aripiprazole (three trials), olanzapine (five trials), and quetiapine (three trials) did not meet conventional levels of statistical significance. Standardized mean difference for aripiprazole was 0.14 (95% CI -0.02, 0.29), for olanzapine was 0.05 (95% CI -0.07, 0.17) and for quetiapine 0.04 (95% CI -0.11, 0.18).

Pooled estimates of SMD in effect size for agitation were 0.19 (95% CI 0.07, 0.31) for olanzapine (four trials), and 0.22 (95% CI 0.09, 0.35) for risperidone (six trials); once again these trials are generally considered “small” effects. Two trials of aripripazole reported positive results. Results for quetiapine (five studies) were not significant.

Active Controlled Trials. We conducted a meta-analysis by pooling five trials that compared atypicals to haloperidol on total score.^{124,125,132,133,136} Information from these trials is displayed in Table 6. Difference between atypicals and haloperidol was not significant. There were too few trials to pool results separately by drug. Regarding psychosis symptoms, we found one trial which showed no difference in efficacy between olanzapine and haloperidol. Results are displayed in Figures 7 and 8. We also found one trial of risperidone versus olanzapine¹³⁸ for dementia. Differences in total/global score and agitation score were not statistically significant.

Table 6

Dementia atypical versus haloperidol—PCTs contributing to analysis.

Figure 7

Dementia: atypical versus haloperidol—total/global scores.

Figure 8

Dementia: atypical versus haloperidol—agitation.

Head-to-head Trials. Three head-to-head trials, described in Table 7, compared atypicals on total/global scores and psychosis outcomes.^116,119,142 None was found superior. Results are displayed in Figures 9 to 11.

Table 7

Dementia head-to-head studies contributing to analysis.

Figure 9

Dementia head-to-head studies olanzapine or quetapine versus risperidone—total/global scores.

Figure 10

Head-to-head studies: olanzapine or quetapine versus risperidone—psychosis.

Figure 11

Dementia head-to-head studies: olanzapine or quetapine versus risperidone—agitation.

Depression. This section focuses on MDD; we excluded other types of depression, including bipolar depression or depression with psychotic features. For MDD, our 2006 CER reported that atypicals were not superior to placebo as augmentation to SSRIs at 8 weeks. However, in some trials they led to more rapid improvement (2 to 4 weeks). Since then, Papakostas published a meta-analysis on MDD in 2007¹⁵¹ and updated it in 2009.¹⁵² Both versions found atypicals superior to placebo in increasing response and remission rates, and found no statistical difference between the specific atypicals. Both versions included olanzapine, risperidone, and quetiapine; the most recent version added aripiprazole.

Our literature search identified 26 new studies of atypical antipsychotics as monotherapy or augmentation for MDD published since our original CER, 18 of which were not in the prior systematic reviews. Quality of trials ranged from 1 to 5 on the Jadad scale; mean score was 2.7. Sample sizes were usually large, with the mean close to 200. Followup times ranged from 4 weeks to 1 year.

The majority of the trials studied augmentaition of SSRIs in treatment refractory patients: four of these were PCTs of aripiprazole,^153-156 seven were PCTs of quetiapine,^86,157-162 and five were PCTs of risperidone.^163-167 One quetiapine PCT augmented treatment with cognitive behavioral therapy (CBT).¹⁵⁸ The results of this trial were suggestive of an added benefit of quetiapine over placebo. However, it was not considered for further analysis as it was deemed to be insufficiently clinically similar to the other studies. There were also six PCTs of quetiapine extended release (ER) as monotherapy.^168-173 Four other trials were not placebo controlled^174-177 and thus could not be included in our pooled analyses. They will be discussed later under “active controlled trials.”

Outcomes consistently reported in the PCTs included the HAM-D total score, percent responders and percent remitted, and the MADRS total score, percent responders and percent remitted. Several trials reported both HAM-D and MADRS outcomes.^{156,161,162,165,168,169,171,172}. Since the information needed to calculate an effect size for the mean MADRS and HAM-D total scores was not consistently reported, we pooled the percent responded and remitted on each scale. The patient populations were reviewed by a psychiatrist to determine level of severity, age, comorbid illness and other factors to verify that these populations were similar enough to pool. The outcomes were measured between 4 and 8 weeks, considered sufficiently clinically similar to pool. Several PCTs contained more than one treatment arm; these studies compared the effects of different doses of atypicals. For our main efficacy analyses, we pooled these arms together and present one resulting intervention outcome per trial. We present the results by dosage later in the relevant section (Key Question 5).

Three trials only reported continuous outcomes, thus they were not included in pooled analyses, which used binary outcomes (e.g., percent responding or percent remitted).^153,164,173 The first of these studied risperidone augmentation of antidepressant medication and found a significant decrease in suicidal ideation with risperidone versus placebo.¹⁶⁴ The second compared quetiapine monotherapy with placebo and found that quetiapine significantly increased the time to a depressed event, compared with placebo.¹⁷³ The third compared aripiprazole augmentation of an antidepressant to placebo augmentation. They reported a significantly greater change in MADRS total score in those receiving aripiprazole.¹⁵³ Additionally, one study did not report outcome data by arm, only overall, so was not included in pooled analysis.¹⁶³ In that study, risperidone augmentation of antidepressant therapy was reported to result in symptomatic remission in a substantial number of patients with chronic resistant depression, compared with placebo.

We conducted six meta-analyses with data from the remaining PCTs:

• Percent remitted on the HAM-D, augmentation.
• Percent responded on the HAM-D, augmentation.
• Percent remitted on the MADRS, augmentation.
• Percent responded on the MADRS, augmentation.
• Percent remitted on the MADRS, monotherapy.
• Percent responded on the MADRS, monotherapy.

HAM-D meta-analyses, augmentation trials. A person was considered remitted if their HAM-D score was less than or equal to 7 (on the HAM-D 17) or a less than or equal to 8 (on the HAM-D 24) for two consecutive visits. Two trials (from one article¹⁷⁸) from our 2006 systematic review reported percent responded and percent remitted on the HAM-D; we include them in the current meta-analyses. The eight total trials that reported the number of participants classified as remitters using the HAM-D ranged in duration from 4 to 8 weeks.^{86,157,161,165-167,178} As displayed in Table 8, the size of these trials ranged from 34 to 274 patients. Only quetiapine and risperidone had a sufficient number of studies to pool estimate of effect by drug. As displayed in Figure 12, the random effects pooled estimate of the relative risk of remitting on the HAM-D for those treated with quetiapine versus placebo was 2.76 (95% CI 1.21, 6.28), and for those taking risperidone was 2.10 (95% CI 1.43, 3.09). This is equivalent to a NNT (number needed to treat) of five for quetiapine and eight for risperidone.

Table 8

Depression—placebo-controlled augmentation trials contributing to HAM-D meta-analysis.

Figure 12

Depression—HAM-D % remitted, augmention.

Responders on the HAM-D were identified in the same eight trials. A responder was defined as someone who had at least a 50 percent reduction in HAM-D score from randomization to followup. We were only able to calculate a pooled estimate of effect for quetiapine and risperidone, as olanzapine had only two trials. As displayed in Figure 13, the random effects pooled estimate of the relative risk of responding on the HAM-D for participants taking quetiapine compared with placebo was 2.30 (95% CI 1.35, 3.92), while for risperidone it was 1.50 (95% CI 1.20, 1.87). This is equivalent to an NNT of three for quetiapine and seven for risperidone. The overall I-squared statistic for these eight trials indicated no heterogeneity (0.0 percent). Neither Begg's nor Egger's test were statistically significant (p=0.711,p=0.245, respectively).

Figure 13

Depression—HAM-D % responded, augmentation.

MADRS meta-analyses, augmentation trials. On the MADRS scale, the definition of a remitted participant differed slightly between trials. A person was considered remitted if their MADRS score was from 8 to 10, depending on the study. The seven trials that reported the number of participants classified as remitters ranged in duration from 4 to 8 weeks.^{154-156,159,160,162,165} As displayed in Table 9, the size of these trials ranged from 97 to 493. Only aripiprazole and quetiapine had a sufficient number of studies to report the pooled estimate of effect per drug. As displayed in Figure 14, the random effects pooled estimate of the relative risk of remitting on the MADRS for those treated with aripiprazole versus placebo was 1.57 (95% CI 1.24, 2.00); for those taking quetiapine it was 1.24 (95% CI 0.82, 1.88). The I-squared statistics for these two analyses were 0 and 82.8, respectively. Begg's test approached significance (p = .072) and Egger's test was significant (p = .018) indicating possible publication bias.

Table 9

Depression—placebo-controlled augmentation trials contributing to MADRS meta-analysis.

Figure 14

Depression—MADRS % remitted, augmentation.

Responders on the MADRS were identified in all but one trial¹⁶⁰ that reported remitters. A responder was defined as someone who had at least a 50 percent reduction in MADRS score from randomization to followup. We were able to calculate a pooled estimate of effect for aripiprazole, which had three trials. Quetiapine was included in two trials, while risperidone was included in only one. As displayed in Figure 15, the random effects pooled estimate of the relative risk of responding on the MADRS for those participants taking aripiprazole compared with placebo was 1.66 (95% CI 1.37, 2.01); for an NNT of seven. The I-squared statistic for this analysis was 0.0. Begg's test was not significant (p = 0.260), while Egger's test approached significance (p = .069).

Figure 15

Depression—MADRS % responded, augmentation.

MADRS meta-analyses, monotherapy trials. The five monotherapy trials for MDD ranged in length from 6 to 9 weeks.^168-172 The number of enrollees ranged from 310 to 723; all studied quetiapine and reported both on both remitters and responders. Details of the studies are displayed in Table 10. As displayed in Figure 16, the random effects pooled estimate of remitting on the MADRS for those treated with quetiapine versus placebo was 1.43 (95% CI 1.07, 191). Begg's and Egger's tests were not statistically significant (p = 0.86, p = .142, respectively). Figure 17 presents the results using percent of patients responding. Quetiapine patients were significantly more likely to respond (OR 1.49, 95 percent CI 1.23, 1.81) than placebo patients. Begg's and Egger's test were both statistically significant (p = .027 each) indicating the possibility of publication bias. The I-squared statistic for each analysis was 70.7 percent and 72 percent respectively.

Table 10

Placebo-controlled monotherapy trials contributing to MADRS meta-analyses.

Figure 16

Depression—MADRS % remitted, monotherapy.

Figure 17

Depression—MADRS % responded, monotherapy.

Active Controlled Trials. There were four active controlled trials of atypicals for the treatment of MDD. One study included two parallel 8-week double-blind trials comparing treatment with a combination of olanzapine and fluoxetine versus olanzapine alone versus fluoxetine alone.¹⁷⁴ The authors report that the pooled results of the two studies found significant differences in mean change of MADRS scores for the olanzapine/fluoxetine combination, compared with either fluoxetine or olanzapine alone. Another trial evaluated quetiapine versus lithium for 56 days and found greater improvement with quetiapine, according to HAM-D, MADRS, and Wildlocher Psychomotor Retardation Scales scores.¹⁷⁵ An 8-week trial compared zisprasidone at differing levels augmenting sertraline to sertraline alone.¹⁷⁶ This trial found a greater improvement in CGI-S and MADRS scores augmenting with ziprasidone at 160mg than either augmentation with ziprasidone at 80mg or sertraline alone. However, there was no significant difference in HAM-D 17, CGI-I, or HAM-A scores. The final non –placebo-controlled trial compared quetiapine as augmentation of paroxetine or venlafaxine to venlafaxine or paroxetine alone.¹⁷⁷ This 12-week trial found an improvement in HAMD-17 scores for all groups, with the quetiapine- paroxetine combination showing the greatest improvement, followed by the quetiapine-venlafaxine combination, then paroxetine only and finally venlafaxine only.

Head-to-Head Trials. No trials comparing specific atypical antipsychotics for MDD were found.

Eating Disorder. This off-label use was not included in our 2006 systematic review. We found one systematic review on this topic; it included RCTs, observational studies, and case reports.¹⁷⁹ The review found evidence of improvement in psychological symptoms, but not in weight gain. Our literature search identified five trials that assessed olanzapine for this use^180-184 and one of quetiapine.¹⁸⁵ Mean quality score was 2.0 on the Jadad scale. Trials ranged in length from 2 to 3 months. Sample sizes were small, with 15 to 45 participants, per trial. All of the RCTs were placebo controlled except for one small head-to-head trial that compared olanzapine to a conventional antipsychotic, chlorpromazine,¹⁸¹ in which the olanzapine group had a significant reduction in anorexic rumination. This trial was excluded from quantitative analysis, which included only placebo comparisons.

Four of five remaining studies reported BMI at times between 1 and 13 weeks. One study that only reported weight gain per week was excluded from further analysis.¹⁸⁴ In that study, there were no differences in weight gain by whether they were treated with olanzapine.

The sample size of the four remaining trials ranged from 20 to 34. These trials were deemed clinically similar to justify meta-analysis at 1 and 3 months; their results are displayed in Table 11.^{180,182,183,185} (BMI is measured such that the desired effect is an increase.)

Table 11

Eating disorder—PCTs contributing to meta-analysis.

The random effects pooled weighted mean difference in BMI from baseline to 1 month of treatment with olanzapine was .004 (95% CI -0.56, 0.57). At 3 months the random effects pooled estimate was 0.25 (95% CI -0.34, 0.84) (Figure 18). The I-squared statistic for each time point indicated low heterogeneity. Neither Begg's or Egger's test for publication bias were statistically significant at either time point (1 month p=0.30, p=0.21 respectively; 3 months p=0.73, p=0.68 respectively).

Figure 18

Eating disorders—BMI.

Active Controlled Trials. There were no active controlled trials of atypicals for eating disorders.

Head-to-Head Trials. There were no head-to-head trials of atypicals for eating disorder.

Insomnia. This off-label use was not included in our 2006 CER. We found no meta-analyses or systematic reviews on the use of atypical antipsychotics for insomnia treatment. We found only one small RCT conducted in Thailand. Although the quetiapine group increased total sleep time by 125 minutes, compared with an increase of 72 minutes in the placebo group, the difference was not statistically significant, due to small sample size (N=13). Because of the paucity of information on this use, we describe six observational studies identified in our literature search; two utilized olanzapine while four utilized quetiapine. Study characteristics are displayed in Table 12 and 13.

Table 12

Atypical antipsychotics for insomnia, observational studies—olanzapine.

Table 13

Atypical antipsychotics for insomnia, observational studies—quetiapine.

One olanzapine study treated 12 patients with insomnia related to major depressive disorder for three weeks. These patients experienced improvements in sleep efficiency, subjective sleep quality and slow wave sleep.¹⁸⁶ The other olanzapine study included case reports of nine patients with different sleep disorders followed for up to 3 years. Eight patients experienced improvements in sleep including sleep latency, total sleep time, decreased nightmares and unspecified improvements.¹⁸⁷ In both studies, the dosages ranged from 2.5mg to 10mg each night and measurements were done both subjectively and per polysomnogram.

Quetiapine was used to treat insomnia of various causes including: primary insomnia,¹⁸⁸ insomnia of drug withdrawal,¹⁸⁹ tamoxifen- related insomnia¹⁹⁰ and insomnia of Parkinson's disease.¹⁹¹ The dosages ranged from 12.5mg to 225mg each night, and the patients were treated from 6 weeks to 3 months. Sleep was measured both objectively using a polysomnogram¹⁸⁸ and subjectively using the Pittsburgh Sleep Quality Inventory,^188,191 the Italian version of the Insomnia Severity Index Scale,¹⁹⁰ Speigal Sleep Questionnaire,¹⁸⁹ and the Epworth Sleep Scale.¹⁹¹ All studies showed improvements in sleep including total sleep time, sleep efficiency,¹⁸⁸ overall quality of sleep,¹⁸⁹ all aspects of sleep,¹⁹¹ and unspecified improvements.¹⁹⁰ Of note, one study did not show an improvement in sleep latency¹⁸⁸ while two others did.^189,191

Obsessive-Compulsive Disorder. Our 2006 CER concluded that atypicals have a clinically meaningful benefit when used as augmentation therapy in patients with OCD. That report included a meta-analysis we conducted using the Yale-Brown Obsessive Compulsive Scale (Y-BOCS) as outcome; both quetiapine and risperidone augmentation increased the odds of response, when compared with augmentation with placebo. (At that time, there were too few trials of olanzapine to permit pooling.) Three other meta-analyses assessing atypical antipsychotics as augmentation for treatment-resistant OCD patients were published around the time of our first evidence report. Two^192,193 included trials of risperidone, quetiapine, and olanzapine. They both found the atypicals have efficacy in increasing the number of responders on the Y-BOCS. Risperidone was statistically significant, while quetiapine and olanzapine had a trend toward efficacy that was not statistically significant. The other meta-analysis included only quetiapine; the authors pooled two trials and found the drug superior to placebo, as measured by changed in total Y-BOCS score.¹⁹⁴

Our literature search identified eight reports of trials published after our 2006 CER.^195-202 Trials were relatively small compared with trials for dementia, anxiety, and depression; sample sizes ranged from 18 to 66. Four were controlled trials of an atypical antipsychotic versus another drug, with no placebo group. These will be discussed below in the section on active controlled and head to head trials.

The other five trials reported on PCTs of augmentation. These trials ranged in duration from 8 to 12 weeks and measured the change in Y-BOCS as the primary outcome measure. Three evaluated the treatment of OCD with quetiapine plus citalopram or placebo plus citalopram,^199,201,202 in patients with OCD who were currently not taking any pharmacotherapy. All three of these studies are related and may in fact be from one trial: they are from the same group of authors using nearly identical protocols, and two studies report on 76 patients while one study reports on 82 patients. These papers report that in treatment-naïve patients quetiapine augmentation was superior to placebo according to improvement in both Y-BOCS and CGI-I scores. The final two studies evaluated quetiapine plus serotonin reuptake inhibitor (SRI) or placebo plus SRI treatment.^196,197 One of these reported duplicate data to an already-included study²⁰³ and was therefore excluded from our pooled analysis.

These final two new RCTs that evaluated quetiapine augmentation versus placebo^196,197 along with four RCTs identified in our original report that evaluated the same treatment,^204-206 two from the original report that evaluated olanzapine augmentation versus placebo,^207,208 and three from the original report that assessed risperidone augmentation versus placebo^209-211 were deemed sufficiently clinically similar to justify meta-analysis. These trials are displayed in Table 14.

Table 14

OCD—PCTs contributing to meta-analysis.

These 10 trials used the Y-BOCS as the primary outcome, classifying “responders” as those achieving a 25 to 35 percent improvement on the Y-BOCS total score. The sample sizes ranged from 16 to 45. The outcome “responders” on the Y-BOCS was measured at 6 to 12 weeks. A few PCTs^204,207,210 reported very wide confidence intervals; these trials were published earlier (2002 to 2004) than the rest.

The meta-analysis results are displayed in Figure 19. There were enough studies to calculate a pooled estimate of relative risk for risperidone and quetiapine. The relative risk of “responding” on the Y-BOCS for those in the quetiapine augmentation arm versus those in the placebo arm was 2.36 (95% CI 0.85, 6.57). The relative risk of “responding” on the Y-BOCS for those in the risperidone augmentation arm was 3.92 (95% CI 1.27, 12.13). This results in an NNT (number needed to treat) of four for quetiapine and five for risperidone. The I-squared statistic was 56.1 percent, indicating some heterogeneity. Both Begg's and Eggar's test indicated the possibility of publication bias (p=0.002,p=0.002 respectively).

Figure 19

OCD—responders improving 25–35% on Y-BOCS.

Active Controlled Trials. One trial compared an atypical antipsychotic plus SSRI plus CBT to SSRI alone plus CBT for the treatment of OCD.¹⁹⁸ These receiving the atypical were treatment resistant and therefore sicker than the other group, but did have a mean reduction in Y-BOCS of 10 points. Another trial evaluated quetiapine plus an SSRI compared with clomipramine plus an SSRI.²⁰⁰ Quetiapine augmentation produced a significant reduction in the Y-BOCS score, while clomipramine augmentation did not.

Head-to-Head Trials. One trial evaluated the treatment of OCD with olanzapine versus risperidone;¹⁹⁵ it found no statistically significant differences between treatment groups. Another trial evaluated the treatment of OCD with quetiapine versus ziprasidone both adjunctive with an SRI.²¹² This trial reported a 80 percent improvement in Y-BOCS score for the quetiapine group and a 44.4 percent improvement for the ziprasidone group.

Personality Disorder. Our 2006 CER found promising results for this off-label use. Three PCTs of olanzapine and one of aripiprazole reported efficacy compared with placebo. Since 2006, several additional trials have been published showing mixed results. These studies are displayed in Table 15.

Table 15

PCTs for personality disorder.

In 2010, a meta-analysis on the efficacy of antipsychotics in treatment of personality disorders²¹³ was published. It included three studies of olanzapine^214-216 and one each of risperidone²¹⁷ and aripiprazole.²¹⁸ It also included several studies of conventional antipsychotics, and pooled all the antipsychotics together, without separating out the effects of atypicals. Therefore, we will not report the results of this analysis.

Borderline Personality Disorder (BPD). Since our initial CER, eight placebo-controlled trials of personality disorders have been published; seven were on BPD. Four of these studies showed an improvement with treatment. Two of these studies involved the same population of patients, first reporting after 8 weeks of treatment and then again after 18 months. In those studies, aripiprazole was the treatment used.^218,219

Another study showed improvement when 5–10mg olanzapine was used each day but no change from placebo when 2.5mg was used.²²⁰ A study reporting only the psychotic symptoms associated with BPD found an improvement with quetiapine.²²¹

Of note, the three studies that did not show an improvement used ziprasidone or olanzapine and, though there was no difference in response from placebo, both groups of patients in each study showed improvement overall with the treated patients showing a faster time to response.^222-224 Studies were too heterogeneous to perform meta-analyses.

Active Controlled Trials. One small RCT of olanzapine versus haloperidol for borderline personality disorders in female inpatients,²²⁵ reported patients in both groups improved considerably regarding hostility, depressive mood, and anxiety. However, differences between groups were not statistically significant.

Observational Study. Our search found only one small pilot study of paliperidone for off-label conditions.²²⁶ Although it is a very small observational study, we include here due to lack of any other relevant data on this drug.

There were eight patients with borderline personality disorder and no other current interfering psychiatric disorder such as psychotic disorders, bipolar disorder, cognitive disorder, major depression or substance abuse. The patients were given paliperidone ER for 12 weeks in a dose of 3–6mg per day. Of the six patients who completed the study, the publication lists that paliperidone was efficacious in reducing global symptoms and “a few core symptoms” of borderline personality disorder. However, there is no specific data regarding these results. There were reports of adverse effects including extrapyramidal symptoms (EPS), insomnia, and agitation, and two patients dropped out of the study, one for noncompliance and the other for gastrointestinal adverse effects.

Schizotypal Personality Disorder. One study measured cognitive symptoms in schizotypal personality disorder and found no significant difference from placebo in those treated with risperidone.²²⁷

Post Traumatic Stress Disorder(PTSD). Our 2006 CER found six PCTs of atypicals for PTSD. Due to heterogeneity, we could not conduct meta-analysis. Trials for combat-related PTSD reported benefits, while trials in abused woman reported mixed results.

Two systematic reviews on use of atypical antipsychotics as monotherapy or medication augmentation for patients with PTSD^228,229 were published after our original evidence report. One included five studies of risperidone^230-234 and two studies of olanzapine.^235,236 It found that atypicals have benefit (compared with placebo) as measured by the change in CAPS score. The publication did not report separate results by drug or clinical subtype. Another review²²⁹ included 10 double-blind RCTs and eight open-label trials. Small positive effects were found for risperidone and quetiapine. Results for olanzapine were mixed.

Our literature search identified three new placebo-controlled trials that assessed risperidone for the treatment of PTSD^233,237,238 and two of quetiapine.^239,240 Combined with the studies included in the original CER, there were eight studies of risperidone^{230-234,237,238,241} two studies of olanzapine^235,236 and two for quetiapine^239,240 for PTSD. Trials were small, ranging from 15 to 94 participants. Quality scores ranged from two to four on the Jadad scale.

Two trials identified in the new literature search reported on the same trials we included in our previous report.^233,238 We selected the most current article to include in our new CER.^233,234 This left 10 trials varying in duration from 5 to 16 weeks. All but two of these trials measured the CAPS. One that did not utilize the CAPS studied showed no difference in improvement between the olanzapine and placebo group.²³⁶ The other found risperidone superior to placebo in reducing irritability and intrusive thought symptoms of PTSD.²³² One risperidone PCT²³³ reported that the treated population showed a significant difference in the CAPS score at endpoint, compared with placebo, but did not report the exact numbers. Similarly, one study reported a 3-fold decline in CAPS scores in patients treated with quetiapine monotherapy compared with placebo.²³⁹ This study did not report exact scores, so could not be pooled in our meta-analysis. Another study found quetiapine/setraline combination superior to setraline plus placebo according to decrease in CAPS score from baseline to 8 weeks.²⁴⁰

Thus, the five remaining PCTs, which were clinically comparable, were pooled in our meta-analyses.^{230,231,234,235,237} The trials are displayed on Table 16. The sample sizes ranged from 19 to 65, while the duration ranged from 5 to 16 weeks, with three trials reporting results at eight weeks. Risperidone dose ranged from 0.5mg to 3mg daily. The one trial of olanzapine used 15mg daily dose.

Table 16

PTSD—PCTs contributing to meta-analyses.

The similar outcome across these five trials was the CAPS total score. A lower CAPS score indicates fewer PTSD symptoms. Thus, we calculated a weighted mean difference for each study and a positive weighted mean difference means an improvement in CAPS total score from baseline to follow up.

We stratified our analyses first by drug (Figure 20). Only risperidone had enough eligible studies to calculate a pooled estimate.^{230,231,234,237} The random effects pooled estimate for risperidone was 6.47 (95% CI 0.32, 12.61). The one olanzapine study²³⁵ had an effect size of 12.13 (95% CI 0.97, 23.29). The overall random effects pooled estimate for risperidone and the one olanzapine study was 7.79 (95% CI 2.40, 13.17), with an I-squared = 0.0 percent. The clinical importance of this 6- or 12-point weighted mean difference needs to be considered in the context that the range of this instrument is 0–136 points, and the standardized mean differences are 0.40, which is normally considered “moderate” in size.

Figure 20

PTSD—by drug–difference in CAPS score.

We also performed a meta-analysis stratified on combat status (Figure 21). We included three studies that included patients with PTSD from combat situations,^230,234,235 and two studies that primarily included abused women with PTSD.^231,237 We only had a sufficient number of studies to perform pooled analysis on the combat studies. We found a random effects pooled estimate of 7.95 (95% CI 1.06, 14.84).

Figure 21

PTSD—by combat status–difference in CAPS score.

Finally, we performed a meta-analysis of the risperidone studies stratified by followup time (Figure 22) divided the studies in to durations greater than or equal to 12 weeks or less than 12 weeks. One study reported outcomes in both time frames,²³⁷ one additional study reported outcomes at 12 weeks or more²³⁰ and two additional studies reported outcomes at less than 12 weeks.^231,234 We only had a sufficient number of studies to report a pooled effect for the less than 12 week outcomes. The random effects pooled estimate for these three studies was not statistically significant (3.23, 95% CI -5.47, 11.93), indicating that we did not find an improvement in CAPS scores for risperidone treatment over placebo at less than 12 weeks.

Figure 22

PTSD—by time–difference in CAPS score.

Active Controlled Trials. We found no active controlled trials of atypicals for PTSD.

Head-to-Head Trials. We found no head-to-head trials of atypicals for PTSD.

Substance Abuse. This off-label use was not included in our 2006 CER. Our literature search identified 33 studies that evaluated the use of atypical antipsychotics for use in alcohol or drug abuse. Nine trials were excluded from further analysis because they included patients with schizophrenia or schizophrenia-related psychosis ^242-249 or bipolar disorder.²⁵⁰

Of the remaining 24 trials, ten evaluated aripiprazole,^251-254 olanzapine,^255-258 and quetiapine^259,260 treatment for alcohol abuse and dependence. Ten trials assessed aripiprazole,^261,262 olanzapine,^263-265 and risperidone^266-270 treatment for cocaine abuse and dependence. The quality and size of trials varied widely; Jadad scores ranged from 0 to 5, and sample size ranged from 3 to 262 participants.

Two trials assessed aripiprazole versus placebo for amphetamine/methamphetamine abuse^271,272 and one trial evaluated olanzapine versus an SSRI and benzodiazepine for the treatment of heroin abuse and dependence.²⁷³ The two amphetamine/methamphetamine treatment trials found aripiprazole not likely to be an efficacious treatment.^271,272 The heroin treatment trial found olanzapine did not improve addictive behavior or relapse.²⁷³ Another trial assessed treatment of concurrent cocaine and heroin dependence with the combination of methadone with risperidone at 2 or 4mg or placebo. This trial found no difference in reduction of cocaine or opiate use, between the three groups.²⁷⁴

Alcohol. The 10 trials that evaluated treatment of alcohol abuse ranged in duration from a few hours to 16 weeks. Two trials reporting on outcomes only after a specified number of drinks or several hours were not included.^255,258

The most commonly reported outcome was drinking abstinence, which was reported in seven of the remaining eight trials. The one trial²⁵⁷ that did not report on abstinence compared olanzapine and placebo's effects on alcohol craving. After 2 weeks of treatment, they found that those participants with the longer repeat allele of the DRD4 VNTR gene responded to olanzapine with reduced craving and alcohol use whereas the participants with the shorter alleles did not.

Of the seven trials that reported on abstinence, two did not report sufficient data to calculate an effect size to use in further analyses.^254,260 In one study,²⁵⁴ aripiprazole was found no better than placebo on the main outcome of percentage of days abstinent. In another study²⁶⁰ there were statistically significant differences for any primary drinking outcomes between patients treated with quetiapine plus naltrexone versus placebo plus naltrexone. Of the remaining trials, two reported only the number or percentage of patients that were completely abstinent at 12 weeks,^252,259 one reported the number of patients that were completely abstinent and the number or percentage of days abstinent,²⁵² and one trial reported the number of days abstinent.²⁵⁶

We performed meta-analysis for percentage of patients completely abstinent. Results are displayed in Figure 23. Three trials reported the number or percentage of patients completely abstinent during the followup period, which ranged from 8 days to 16 weeks. Two evaluated aripiprazole^252,253 and one assessed quetiapine.²⁵⁹ The trials are displayed in Table 17. The size of these trials ranged from 30 to 288 participants. The overall random effects pooled estimate of the relative risk of remaining completely abstinent was 1.42 (95% CI 0.36, 5.67). The overall I-squared statistic was 80.4 percent. Neither Begg's nor Eggar's test indicated publication bias (p = .296, p = .308, respectively).

Figure 23

Substance abuse—alcohol complete abstinence.

Table 17

Alcohol abuse—PCTs contributing to meta-analyses.

Active Controlled Trials. One study compared naltrexone with aripiprazole; there was no difference in either mean number of days abstinent or percentage of group completely abstinent.²⁵¹

Cocaine. One published meta-analysis²⁷⁵ assessed use of atypicals in treatment of cocaine dependence. It included three trials of risperidone^268,270,274 and three of olanzapine.^245,264,265 Outcome was rate of dropout from residential and outpatient substance abuse treatment programs. The analysis found no significant difference bteween atypicals and placebo; effect was not reported separately by medication.

Ten trials with placebo comparisons reported on the treatment of cocaine abuse or dependence with aripiprazole,^261,262 olanzapine,^263-265 and risperidone.^266-270 These trials ranged from several days to 20 weeks. Outcomes reported varied greatly; most consistently reported outcomes were the Cocaine Craving Questionnaire (CCQ) and the ASI-drug. Two trials reported neither the CCQ or the ASI;^267,268 they were not considered for further analysis. The first reported that risperidone improved neuropsychological impairment in cocaine-withdrawn patients. ²⁶⁷ One was an active-control trial that will be discussed below.

Five of the remaining eight cocaine abuse PCTs reported the CCQ.^{261-263,265,270} None reported usable CCQ data that would allow us to calculate an effect size estimate; thus, we could not use the CCQ as a poolable outcome. Five of the eight trials reported the ASI-drug composite score, two of which had no usable data.^263,266 The first of these compared olanzapine to placebo for 16 weeks. They found that olanzapine was not superior to placebo in decreasing use, cravings, or addiction severity.²⁶³ The second of these compared risperidone to placebo in the treatment of cocaine dependence. There was no reduction in cocaine use after 12 weeks of treatment.²⁶⁶ The remaining three trials were considered comparable enough to justify meta-analysis, pooling on the continuous ASI-drug composite score outcome.^264,265,269 The trials are listed in Table 18; meta-analyses results are displayed in Figure 24.

Table 18

Cocaine—PCTs contributing to meta-analysis.

Figure 24

Cocaine—ASI drug composite.

These trials of olanzapine^264,265 and risperidone²⁶⁹ treatment for cocaine abuse ranged in size from 30 to 68 participants and lasted from 8 to 12 weeks. We calculated a weighted mean difference for the effect size estimate in which a positive weighted mean difference favors the treatment arm. The overall random effects pooled estimate for the difference in ASI-drug composite score was 0.001 (95% CI -0.41, 0.043). The I-squared statistic indicated no heterogeneity. Neither Begg's nor Eggar's test indicated publication bias (p=1.00,p=0.928 respectively).

Active Controlled Trials. We found one trial that compared risperidone with pergolide.²⁶⁸ There was no statistical difference from placebo in reducing cocaine use.

Tourette's. We found no new clinical trials that studied atypical antipsychotics for Tourrette's syndrome published after our original CER. That CER reported risperidone was superior to placebo in one small trial, and it was at least as effective as pimozide or clonidine for 8 to 12 weeks of therapy in the three other trials. One trial of ziprasidone showed variable efficacy compared with placebo.

There were two observational studies^276,277 of aripiprazole that reported effectiveness data after our 2006 CER; information is displayed in Table 19. One was a retrospective observational study for the treatment of tics with or without comorbid explosive disorder. Thirty-seven patients aged 8–18 years old were treated with aripiprazole 2.5-40 mg for 12 weeks. All of the 29 subjects who completed the trial experienced a reduction in their tic severity. However, eight subjects discontinued early due to inability to tolerate the medication.²⁷⁶

Table 19

Atypical antipsychotics for Tourette's syndrome.

The other study treated 24 patients, aged 7–18 years old with a mean aripiprazole dose of 9.8mg for eight weeks. Overall, there was a 52.8 percent reduction in Yale Global Tic Severity Scale scores, and 19 of the 24 were described as “much” or “very much” improved per CGI-I. Six of the patients discontinued due to adverse effects.²⁷⁷

Discussion

We conducted an extensive literature search, data abstraction, and meta-analyses whenever possible to assess the efficacy and comparative effectiveness of atypical antipsychotics for off-label use. Since the publication of our original CER in 2006, many new high-quality controlled trials have been published; we were able to add many to our prior quantitative analyses. Our results are summarized in Table 20. It is important to note that we found no trials of the three newest atypicals—asenapine, iloperidone, and paliperidone—for off-label uses.

Table 20

Summary of strength of evidence of efficacy, by drug and condition.

We found that aripiprazole, olanzapine, and risperidone had small but statistically significant effects in treating agitation, psychosis, and behavioral symptoms of dementia. Because of the plethora of trials, the large sample sizes enrolled in each trial (usually 300 or more), the quality of trials (mean Jadad score 3.2) and the consistency of the results, the strength of the evidence is high. However, the clinical benefits must be balanced against significant side effects and potential harms. (See results of Key Question 4, later in this report.) In addition, most trials used flexible dosing, so we were unable to determine the most appropriate dosage.

Moderate strength evidence suggests efficacy of aripiprazole, quetiapine, and risperidone as augmentation in treatment of MDD in patients who respond inadequately to SSRIs/SNRIs. Effect sizes are moderate to large, with patients one and a half to two times as likely to respond as with SSRIs alone. Also, a few trials found efficacy for ziprasidone and olanzapine; the strength of evidence is low for these two drugs, but this rating could change with the publication of additional successful trials. Quetiapine is also effective as monotherapy for MDD; strength of evidence is moderate. Strength could increase to high if non–industry-funded studies are published with similar results.

We found moderate strength evidence of efficacy of risperidone as augmentation therapy for OCD. Trials for OCD tend to be much smaller than those for dementia and depression; sample sizes ranged from 15 to 45 for the 10 trials contributing to our efficacy meta-analysis. The mean quality of trials is lower (2.2 on Jadad scale). Results are also less consistent. For example, in the only two PCTs of olanzapine, percentage of participants responding as measured by Y-BOCS did not differ from placebo. In contrast, a head-to-head trial (with no placebo) found no difference in efficacy between olanzapine and risperidone, a drug with moderate evidence of efficacy.

There is also moderate strength evidence of efficacy in reducing symptoms of combat-related PTSD from several small trials of risperidone. We also found two studies of olanzapine for PTSD; they reported conflicting results. There is low strength of evidence based on two positive trials of quetiapine. Trials of PTSD tend to be of lower quality and smaller size than the depression augmentation and dementia trials. Mean Jadad score was 2.7; only two PTSD trials had over 40 participants. New, preferably larger trials must be conducted before the strength of evidence can be increased.

Regarding borderline personality disorder, strength of evidence of efficacy is low or very low for all atypicals other than risperidone, where we found no trials. Olanzapine had the most trials (seven) but results were inconsistent. Of note, however, in the olanzapine studies that showed no difference between drug and placebo groups at 12-week followup, both groups of patients showed improvement overall, with the treated patients showing a faster time to response.

We added eating disorders, anxiety, insomnia, and substance abuse to our 2011 report. With the exception of generalized anxiety disorder, there is little scientific evidence that atypicals are useful in addressing symptoms of these conditions. Moderate evidence suggests that olanzapine, risperidone, and aripiprazole have no efficacy in substance abuse treatment, and that olanzapine treatment does not lead to weight increase in eating disorder patients, compared with placebo. We did find moderate evidence of efficacy of quetiapine in treating generalized anxiety disorder. There were too few trials of olanzapine, risperidone or ziprasidone for anxiety to pool; these trials had mixed results. Importantly, anxiety trials had larger samples (mean N = 122) and higher quality (mean Jadad score = 3.1) than most trials for OCD, PTSD, substance abuse, and eating disorders.

Finally, we reviewed trials of children and adolescents with Tourette's syndrome or ADHD; evidence of efficacy was low for use of atypicals for these conditions. No Tourette's trials have been published since our 2006 CER which reported that risperidone is at least as effective as pimozide or clonidine. Only one small trial has studied atypicals for ADHD in children with no major co-occurring disorders; risperidone users were more likely to respond than placebo patients.

These findings are valuable and can help psychiatrists make better clinical decisions based on the latest evidence. Findings are summarized in Table 20. The symbol “O” below indicates areas where we found no clinical trials of a particular atypical for that condition, while “-” indicates evidence of inefficacy for a condition, according to the psychometric measures our team considered most important. In summary, ziprasidone has no evidence of efficacy for any off-label use other than depression. The four other atypicals have shown efficacy in treating dementia, depression, and a few other conditions, depending on drug.

Our literature search procedures were extensive and included canvassing experts from academia and industry regarding studies we may have missed. However, the possibility of publication bias still exists. Table 21 below displays our assessment of heterogeneity by condition and outcome. For the most part, our assessment did not yield evidence of unexplained heterogeneity. Two exceptions include the MADRS for depression and the Y-BOCS (percent of participants responded) outcome used in our OCD meta-analysis. In our analysis of atypicals as augmentation and monotherapy in treatment of major depressive disorder (MDD), possible publication bias appeared in studies reporting the MADRS (Begg's p=.072, Egger's p=.019 for augmentation, percent remitted; Begg's p=.027, Egger's p=.027, for monotherapy, percent responded). We conducted additional augmentation meta-analyses using HAM-D outcomes; efficacy results were similar, but no heterogeneity was detected. Thus, our confidence that some atypicals have efficacy in treating depression remains. Heterogeneity was also evident in studies assessing the efficacy of atypicals for OCD (Begg's p=0.002, Egger's p=0.001). This heterogeneity was likely due to patient enrollment criteria; studies used different definitions of “refractory” and “treatment resistant.” Another published meta-analysis of atypicals for OCD¹⁹² found similar efficacy results but no heterogeneity according to statistical tests.

Table 21

Analysis of publication bias.

An important limitation common to systematic reviews is the quality of the original studies included. In order to measure the quality of clinical trials we used the Jadad scale.¹⁷ As empirical evidence regarding other study characteristics and their relationship to bias is lacking, we did not attempt to use other criteria. However, other aspects of the design and execution of a trial may be related to bias, but we do not yet have good measures of these elements. In our 2006 CER on off-label use of atypicals, we conducted a sensitivity analysis on the relationship between trial quality and effect size; the better quality trials reported an effect size 25 percent smaller than did lower quality trials. This finding increases the likelihood that a synthesis of results of all studies—whether narrative or quantitative—may produce inflated estimates of efficacy. As stated above, the higher general quality of the dementia and depression augmentation studies led to a greater strength of evidence rating for those uses.

Applicability of research to the larger treatment population is important in interpreting the results of the included studies. The participation rate, the intended target population, representativeness of the setting, and representativeness of the individuals must be known to assess applicability. Such data were reported unevenly in the studies we reviewed. The dementia trials were most often conducted in nursing homes, hospitals, or assisted living facilities. According to our review on utilization patterns, these settings represent where atypicals are most often used in the elderly. Studies for other conditions were not particularly representative. For example, three of the four trials for ADHD were conducted in children with severe co-occurring conditions, such as bipolar disorder or mental retardation. Subjects in substance abuse trials were usually enrolled in outpatient or residential treatment programs. However, there was one trial of non–treatment-seeking subjects;²⁵³ it is unlikely that atypicals would be used in the real world without some initial detoxification or simultaneous treatment program.

In the studies of atypicals as augmentation for SSRI or SNRI patients with MDD, it was often unclear whether patients were simultaneously undergoing psychotherapy. One article¹⁶⁵ specifically stated that subjects were prohibited from initiating such therapy during the trial, but other articles were unclear on the issue. Thus, we don't know whether treatment over and above the medication influenced the study results. It is important to note that subjects in depression trials were recruited from both primary care and mental health centers, as depression patients have been increasingly treated in primary care settings.

We found only one small trial (N=13) of atypicals for treatment of insomnia. Observational studies of Insomnia included patients with Parkinson's disease, MDD, polysubstance abuse withdrawal symptoms, and tamoxifen-induced insomnia. Thus, the results of these studies should not be applied to the general population.