Association With Hospitalization and All-Cause Discontinuation Among Patients With Schizophrenia on Clozapine vs Other Oral Second-Generation Antipsychotics

Key Points

Question

Meta-analyzing nonrandomized cohort studies, how does clozapine compare with other second-generation antipsychotics (SGAs)?

Findings

In this systematic review and meta-analysis of 63 cohort studies comprising 109 341 participants, although more severely ill patients received clozapine, use of clozapine was associated with better effectiveness outcomes regarding hospitalization, all-cause discontinuation, and overall symptoms but with a higher risk of cardiometabolic-related outcomes vs other SGAs.

Meaning

In a more generalizable sample of patients with schizophrenia than in randomized trials, clozapine was statistically and clinically associated with better effectiveness outcomes compared with other SGAs, but some concerns about safety require careful monitoring and clinical attention.

This systematic review and meta-analysis compares various outcomes of clozapine vs oral nonclozapine second-generation antipsychotics in cohort studies of patients with schizophrenia.

Abstract

Importance

Recent meta-analyses of randomized clinical trials (RCTs) comparing clozapine with nonclozapine second-generation antipsychotics (NC-SGAs) in schizophrenia have challenged clozapine’s superiority in treatment-resistant patients. However, patients in RCTs are not necessarily generalizable to those in clinical practice.

Objective

To conduct a systematic review and meta-analysis to compare various outcomes of clozapine vs oral NC-SGAs in cohort studies.

Data Sources

Systematic literature search in PubMed, PsycINFO, and CINAHL without language restriction from database inception until December 17, 2018.

Study Selection

Nonrandomized cohort studies reporting effectiveness and/or safety outcomes comparing clozapine with NC-SGAs in schizophrenia or schizoaffective disorder.

Data Extraction and Synthesis

Independent investigators assessed studies and extracted data. Using a random-effects model, the study calculated risk ratio (RR) unadjusted for covariates and follow-up duration, number needed to treat/number needed to harm (NNT/NNH) for dichotomous data, and standardized mean difference (SMD) or mean difference (MD) for continuous data.

Main Outcomes and Measures

Coprimary outcomes were hospitalization and all-cause discontinuation. Secondary outcomes included all effectiveness and safety outcomes reported in at least 3 analyzable studies.

Results

Of 8446 hits, 68 articles from 63 individual cohort studies (n = 109 341) (60.3% male; mean [SD] age of 38.8 [6.5] years, illness duration of 11.0 [5.1] years, and study duration of 19.1 [23.3] months) were meta-analyzed. Compared with NC-SGAs, despite greater illness severity (17 studies [n = 38 766]; Hedges g, 0.222; 95% CI, 0.013-0.430; P = .04), clozapine was significantly associated with lower hospitalization risk (19 studies [n = 49 453]; RR, 0.817; 95% CI, 0.725-0.920; P = .001; NNT, 18; 95% CI, 12-40) and all-cause discontinuation (16 studies [n = 56 368]; RR, 0.732; 95% CI, 0.639-0.838; P < .001; NNT, 8; 95% CI, 6-12). Associations were statistically significant for comparisons with quetiapine fumarate and aripiprazole regarding hospitalization and all NC-SGAs, except aripiprazole, for all-cause discontinuation. Clozapine was also significantly associated with better outcomes regarding overall symptoms (SMD, −0.302; 95% CI, −0.572 to −0.032; P = .03) and Clinical Global Impressions scale severity (SMD, −1.182; 95% CI, −2.243 to −0.122; P = .03). Clozapine was significantly associated with increases in body weight (MD, 1.70; 95% CI, 0.31-3.08 kg; P = .02), body mass index (MD, 0.96; 95% CI, 0.24-1.68; P = .009), and type 2 diabetes (RR, 1.777; 95% CI, 1.229-2.570; P = .002; NNH, 27; 95% CI, 13-90).

Conclusions and Relevance

In cohort studies, despite more severely ill patients being treated with clozapine, use of clozapine was associated with better key efficacy outcomes and higher cardiometabolic-related risk outcomes vs NC-SGAs.

Introduction

Although antipsychotics are the main treatment for schizophrenia, approximately 30% patients do not respond to 2 or more antipsychotic trials of adequate dosage and duration.¹ These patients are considered to have treatment-resistant schizophrenia (TRS).

Clozapine is the criterion standard medication for TRS,^2,3,4,5 whose superior effectiveness is supported by a network meta-analysis of acute treatment associations, although the population was not specifically treatment resistant.⁶ However, recent meta-analyses of masked randomized clinical trials (RCTs) comparing clozapine with other oral nonclozapine second-generation antipsychotics (NC-SGAs) reported inconsistent results.^7,8 One network meta-analysis found superiority of clozapine only in TRS compared with first-generation antipsychotics but not NC-SGAs, either individually (ie, olanzapine, risperidone, or ziprasidone hydrochloride) or pooled.⁷ These surprising results have been challenged as being related to methodological issues, including lower clozapine dosages than in clinical care, and, especially, the possibility of sampling bias in that patients recruited in RCTs are less representative of real-world patients, being less severely ill and having greater illness insight, and thus more likely to respond to NC-SGAs.⁹ Studies on long-acting injectable (LAI) antipsychotics yielded similarly conflicting findings¹⁰ in that LAI antipsychotics were not superior to oral antipsychotics in a meta-analysis of RCTs,¹¹ whereas meta-analyses of mirror-image studies¹² and cohort studies,¹³ expected to better reflect real-world patients and practice than RCTs, exhibited consistent superiority of LAI antipsychotics.

Therefore, real-world effectiveness of clozapine should also be explored beyond RCTs, as in cohort studies. However, cohort studies also have a potential sampling bias in that more severely ill and treatment-resistant patients are more likely treated with clozapine than with other oral NC-SGAs. Nevertheless, this would be a conservative bias, which would strengthen the results if clozapine was superior to NC-SGAs.

Therefore, to complement the meta-analyses of clozapine RCTs^7,8 we conducted a meta-analysis of cohort studies comparing clozapine with oral NC-SGAs. We hypothesized that clozapine would be associated with better effectiveness outcomes compared with NC-SGAs, although clozapine-treated patients would have greater illness severity and/or chronicity.

Methods

We conducted a systematic review and meta-analysis of 68 articles from 63 cohort studies. The meta-analysis was performed in accord with the Meta-analysis of Observational Studies in Epidemiology (MOOSE) reporting guideline.¹⁴

Literature Search

At least 2 independent authors (T.M., F.M., and M.T.) conducted a systematic literature search in PubMed, PsycINFO, and CINAHL without language restriction from database inception until December 17, 2018, using the following search terms: (clozapin* or clozaril* or leponex* or denzapin* or zaponex*) AND (schizophrenia OR schizoaffective OR schizophreniform). The electronic search was supplemented by a manual review of reference lists from eligible publications and relevant reviews.

Inclusion and Exclusion Criteria

At least 2 authors (T.M., F.M., and M.T.) independently screened selected full-text articles. Any disagreements were resolved by consensus or involvement of the senior author (C.U.C). Eligible study characteristics were (1) nonrandomized, prospective, or retrospective cohort studies; (2) at least 4-week duration (even if study duration differed between groups); (3) effectiveness and/or safety outcomes comparing clozapine with oral NC-SGAs (<30% SGA-LAI antipsychotics); (4) at least 15 patients per study; (5) mean age at least 18 years; and (6) at least 70% of participants diagnosed as having schizophrenia, schizoaffective, or schizophreniform disorder.

To clarify the specific association of clozapine, there were some exclusions. We excluded (1) treatment episode–based studies in which 1 patient could be counted in more than 1 medication group and (2) studies starting the observation period after clozapine had been initiated (except for hospitalization-related outcomes where index discharge was the starting point, which can be after clozapine or NC-SGA initiation).

Outcomes

Coprimary outcomes were hospitalization and all-cause discontinuation. We also compared baseline severity of patients receiving clozapine and NC-SGAs in studies reporting on the coprimary outcomes to examine if patients prescribed clozapine are more severely and/or chronically ill, which would inform the interpretation of nondifferential results. Secondary outcomes included all effectiveness and safety outcomes reported in at least 3 analyzable studies.

Data Extraction

At least 2 authors (T.M., F.M., and M.T.) independently extracted data and assessed study quality with the Newcastle-Ottawa Scale (eTable 1 in the Supplement lists the complete data).¹⁵ Non–English language articles were translated by bilingual speakers and double-checked using Google Translate.¹⁶ Numerical data reported in figures only were extracted with WebPlotDigitizer.¹⁷

For missing data, we e-mailed corresponding authors of studies published after 2005. In the event of nonresponse, they were emailed 3 times.

Data Analysis

For all outcomes, data at the longest study period (and for the most included patients in case of overlapping reports) were analyzed, preferring intent-to-treat data. For dichotomous data, we prioritized the Kaplan-Meier estimates (which better account for dropouts) over crude data. Whenever at least 3 studies provided data for one of the coprimary outcomes, we also (1) investigated the effectiveness at specific time points (6, 12, 24, and 36 months), (2) compared clozapine with specific NC-SGAs, (3) assessed the association of adjusting for follow-up duration (person-years [ie, calculating incidence rate ratios {RaRs}]) and performed post hoc analyses by focusing on studies reporting data as (4) adjusted odds ratios (ORs) and (5) hazard ratios (HRs). After comparing clozapine with specific NC-SGAs across all studies, we conducted sensitivity analyses removing studies with unequal observation periods during which an event could have occurred.

For dichotomous data, we calculated risk ratio (RR). When there was a significant difference in RR, we calculated number needed to treat/number needed to harm (NNT/NNH) from an assumed control group risk,¹⁸ which we estimated as the mean occurrence of the respective outcome in the NC-SGA group, as done before.⁷ For continuous data, we calculated the standardized mean difference (SMD) and the mean difference (MD) for cardiometabolic-related outcomes to also display the data as original units that are clinically more intuitive.

When data for multiple NC-SGAs were reported, we first pooled these NC-SGA groups. We then performed individual NC-SGA comparison analyses.

To analyze baseline illness severity (Hedges g), we used 4 data types, as done before in cohort studies of oral vs LAI antipsychotics in schizophrenia¹³ (choosing in descending order in case of multiple data). These included (1) psychopathology score (baseline Positive and Negative Syndrome Scale,¹⁹ Brief Psychiatric Rating Scale,²⁰ or Clinical Global Impressions scale²¹); (2) prior number of hospitalizations; (3) prior hospitalization days; and (4) the proportion of hospitalized patients at baseline.

We also conducted subgroup and meta-regression analyses of the coprimary outcomes to explore potential moderators and sources of biases. The subgroup analyses were conducted based on (1) baseline setting (inpatients or outpatients); (2) study design (prospective or retrospective); (3) region; (4) data source (single site, multisite, or database); (5) sample specified as treatment resistant or poorly responding (yes vs no/not reported); (6) pharmaceutical sponsorship; and (7) hospitalization cause definition (all cause or mental health). Meta-regression analyses were conducted exploring the following variables: sex (percentage), sample size, publication year, mean age, illness duration, study duration, follow-up duration, study quality (Newcastle-Ottawa Scale score), and baseline illness severity.

Regarding heterogeneity, I² and χ² P values are reported. Publication bias was assessed by Egger regression test²² and the trim-and-fill method by Duval and Tweedie.²³

All analyses were conducted using a software program (Comprehensive Meta-Analysis, version 3.3.070; Biostat) based on a random-effects model because considerable between-study heterogeneity was expected.²⁴ In this large, exploratory set of analyses, no statistical adjustments were made for multiple comparisons.

Dealing With Missing Change SD. Data

When calculating SMDs and MDs for psychopathology and safety-related outcomes, some studies reported the number, mean, and SD at baseline and end point but not change SD or convertible data to obtain the SD of the change (ie, P, t, and F value within a group). When analyzing body weight, body mass index (BMI), and triglycerides change, we imputed the pooled change SD from other eligible studies because more than half of studies reported change SD^18,25 For other outcomes, we imputed change SD using the correlation coefficient of SDs at baseline and end point¹⁸ with 3 types of the correlation coefficients (high is 0.9, medium is 0.5, and low is 0.1) and comparing the SMDs and MDs because the exact correlation coefficients were unknown (eTable 2 in the Supplement lists the complete data). Herein, we show medium correlation coefficient–based SMDs and MDs. Whenever that significance level differed from high or low correlation coefficient–based calculations, we detailed that. All tests were 2 sided with α = .05.

Results

The initial database search produced 12 895 hits. After removing 4449 duplicates, 8446 articles were screened, and 8015 were excluded by title and abstract screening. Of the remaining 431 articles and adding 2 articles through manual search, 68 articles from 63 individual cohort studies (n = 109 341) were meta-analyzed (the Preferred Reporting Items for Systematic Reviews and Meta-analyses [PRISMA] flowchart and study references and characteristics are available in eTable 3 and eFigure 1 in the Supplement). Thirty-three studies had a prospective design (n = 34 434), 30 studies had a retrospective design (n = 74 907), and the number of patients per study ranged from 22 to 30 387 (median, 167). The mean (SD) age was 38.8 (6.5) years, 60.3% (13.7%) were male, illness duration was 11.0 (5.1) years, and study duration was 19.1 (23.3) months (range, 1-132 months; median, 12 months). Fourteen (22.2%) of the 63 studies (n = 7229) included patients characterized as treatment-resistant or suboptimal responders.

Main Analyses of Coprimary Outcomes

Clozapine was associated with a significantly lower hospitalization risk than NC-SGAs (19 studies^{26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43} [n = 49 453]; mean [SD] age, 38.8 [5.2] years; mean [SD] study duration, 20.0 (7.9) months; RR, 0.817; 95% CI, 0.725-0.920; P = .001; NNT, 18; 95% CI, 12-40), with significant heterogeneity (I² = 67.1%; P < .001) (Figure 1). Clozapine was also associated with a significantly lower risk of all-cause discontinuation than NC-SGAs (16 studies^{29,34,35,36,39,43,44,45,46,47,48,49,50,51,52,53} [n = 56 368]; mean [SD] age, 39.8 [5.0] years; mean [SD] study duration, 23.3 (21.7) months; RR, 0.732; 95% CI, 0.639-0.838; P < .001; NNT, 8; 95% CI, 6-12), with significant heterogeneity (I² = 92.5%; P < .001) (Figure 2).

Figure 1. Forest Plot of Risk Ratio for Hospitalization^{26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43}.

The mean risk for the nonclozapine second-generation antipsychotics (NC-SGAs) (31.9%) was used as the assumed control group risk. White lines within the pooled risk estimate boxes represent that the corresponding 95% CI is narrower than the pooled effect that depends on the size of the sample. NNT indicates number needed to treat; SOHO, Schizophrenia Outpatient Health Outcomes.

^aData from the Kaplan-Meier estimates.

^bMean, 9.8 (17.0 for clozapine and 7.3 for NC-SGAs).

^cMean, 21.7 (23.2 for clozapine and 21.6 for NC-SGAs).

Figure 2. Forest Plot of Risk Ratio for All-Cause Discontinuation^{29,34,35,36,39,43,44,45,46,47,48,49,50,51,52,53}.

The mean risk for the nonclozapine second-generation antipsychotics (NC-SGAs) (51.7%) was used as the assumed control group risk. White lines within the pooled risk estimate boxes represent that the corresponding 95% CI is narrower than the pooled effect that depends on the size of the sample. IC-SOHO indicates Intercontinental Schizophrenia Outpatient Health Outcomes; NNT, number needed to treat; and SOHO, Schizophrenia Outpatient Health Outcomes.

^aMean, 21.7 (23.2 for clozapine and 21.6 for NC-SGAs).

^bData from the Kaplan-Meier estimates.

^cMean, 9.8 (17.0 for clozapine and 7.3 for NC-SGAs).

Publication Bias

There was no indication of publication bias for both coprimary outcomes. Using the trim-and-fill method by Duval and Tweedie to adjust for potentially missing publications, 1 data point was imputed for hospitalization, but the imputation did not change the results (eFigure 2 in the Supplement).

Baseline Illness Severity

Of 28 studies analyzed for coprimary outcomes, including 7 overlapping studies, 17 provided baseline illness severity and/or chronicity data. Patients receiving clozapine had significantly greater baseline illness severity (17 studies^{26,27,28,29,30,31,32,33,37,38,39,46,49,50,51,52,53} [n = 38 766]; Hedges g, 0.222; 95% CI, 0.013-0.430; P = .04) (Figure 3).

Figure 3. Forest Plot of Hedges g for Baseline Illness Severity of Studies With Coprimary Outcomes^{26,27,28,29,30,31,32,33,37,38,39,46,49,50,51,52,53}.

IC-SOHO indicates Intercontinental Schizophrenia Outpatient Health Outcomes; NC-SGAs, nonclozapine second-generation antipsychotics; and SOHO, Schizophrenia Outpatient Health Outcomes.

Detailed Analyses of Coprimary Outcomes

Adjusting for follow-up duration by person-year calculation, clozapine was associated with a significantly lower risk than NC-SGAs for hospitalization (11 studies [4526 person-years]; RaR, 0.615; 95% CI, 0.482-0.786; P < .001; NNT, 8; 95% CI, 6-14) and for all-cause discontinuation (5 studies [14 392 person-years]; RaR, 0.542; 95% CI, 0.367-0.802; P = .002; NNT, 5; 95% CI, 3-10). These results are summarized in Table 1 and eFigure 3 in the Supplement.

Table 1. Results of Detailed Analysis of Coprimary Outcomes for Clozapine vs NC-SGAs.

Outcome	No. of Studies	No. of Patients	Effect Size		Heterogeneity		Effect Size, NNT (95% CI)a
			Risk Ratio (95% CI)	P Value	I2, %	P Value
Hospitalization (vs Specific NC-SGAs)
Clozapine vs risperidone	16	17 697	0.871 (0.745-1.018)	.08	61.6	.001b	NA
Clozapine vs olanzapine	11	20 338	0.936 (0.765-1.146)	.52	76.4	<.001b	NA
Clozapine vs quetiapine fumarate	7	8180	0.727 (0.570-0.927)	.01b	80.2	<.001b	14 (9-53)
Clozapine vs aripiprazole	3	4529	0.698 (0.559-0.872)	.002b	46.1	.16	20 (14-47)
Clozapine vs amisulpride	3	2443	0.777 (0.532-1.137)	.19	75.3	.02b	NA
Hospitalization (at Specific Period)
At 6 mo	9	1203	0.791 (0.462-1.354)	.39	59.9	.01b	NA
At 12 mo	16	36 532	0.803 (0.699-0.922)	.002b	63.6	<.001b	21 (14-52)
At 24 mo	10	13 107	0.804 (0.653-0.989)	.04b	51.0	.03b	13 (7-221)
All-Cause Discontinuation (vs Specific NC-SGAs)
Clozapine vs risperidone	14	20 836	0.702 (0.615-0.800)	<.001b	87.0	<.001b	7 (5-10)
Clozapine vs olanzapine	13	27 212	0.792 (0.669-0.939)	.007b	92.9	<.001b	10 (7-33)
Clozapine vs quetiapine fumarate	10	8642	0.654 (0.490-0.874)	.004b	97.0	<.001b	6 (4-15)
Clozapine vs aripiprazole	3	4070	0.712 (0.338-1.501)	.37	99.0	<.001b	NA
Clozapine vs amisulpride	3	2375	0.694 (0.495-0.971)	.03b	85.7	.001b	6 (4-57)
All-Cause Discontinuation (at Specific Period)
At 6 mo	6	32 230	0.779 (0.564-1.076)	.13	95.6	<.001b	NA
At 12 mo	9	39 913	0.649 (0.530-0.795)	<.001b	95.1	<.001b	6 (5-11)
At 24 mo	5	33 862	0.707 (0.540-0.926)	.01b	94.4	<.001b	7 (5-28)
At 36 mo	4	12 120	0.836 (0.688-1.016)	.07	65.2	.04b	NA
Adjusted for Person-Years
Hospitalization, person-years	11	4526	Rate ratio: 0.615 (0.482-0.786)	<.001b	54.1	.02b	8 (6-14)
All-cause discontinuation, person-years	5	14 392	Rate ratio: 0.542 (0.367-0.802)	.002b	83.7	<.001b	5 (3-10)
Hazard Ratio
All-cause discontinuation	5	NA	Hazard ratio: 0.670 (0.464-0.967)c	.03b	96.9	<.001b	NA
Adjusted Odds Ratio
Hospitalization	3	NA	Adjusted odds ratio: 0.841 (0.620-1.141)d	.27	41.5	.18	NA

Abbreviations: NA, not applicable; NC-SGAs, nonclozapine second-generation antipsychotics; NNT, number needed to treat.

^aNumbers needed to treat were only calculated when the risk ratio analysis was significant (P < .05). The mean risk for the NC-SGAs was used as the assumed control group risk.

^bP < .05.

^cHazard ratio for hospitalization risk was not calculated due to only 2 studies.

^dAdjusted odds ratio for all-cause discontinuation risk was not calculated due to only 2 studies.

Performing analyses at specific time points, clozapine was associated with a significantly lower risk of hospitalization at 12 months (RR, 0.803; 95% CI, 0.699-0.922; P = .002; NNT, 21; 95% CI, 14-52) and 24 months (RR, 0.804; 95% CI, 0.653-0.989; P = .04; NNT, 13; 95% CI, 7-221) but not at 6 months (RR, 0.791; 95% CI, 0.462-1.354; P = .39) (Table 1). There were too few analyzable studies at 36 months. Clozapine was associated with a significantly lower risk of all-cause discontinuation at 12 months (RR, 0.649; 95% CI, 0.530-0.795; P < .001; NNT, 6; 95% CI, 5-11) and 24 months (RR, 0.707; 95% CI, 0.540-0.926; P = .01; NNT, 7; 95% CI, 5-28) but only nominally at 6 months (RR, 0.779; 95% CI, 0.564-1.076; P = .13) and 36 months (RR, 0.836; 95% CI, 0.688-1.016; P = .07).

Compared with specific NC-SGAs, clozapine was associated with a significantly lower hospitalization risk than quetiapine fumarate (RR, 0.727; 95% CI, 0.570-0.927; P = .01; NNT, 14; 95% CI, 9-53) and aripiprazole (RR, 0.698; 95% CI, 0.559-0.872; P = .002; NNT, 20; 95% CI, 14-47) but not risperidone (RR, 0.871; 95% CI, 0.745-1.018; P = .08), olanzapine (RR, 0.936; 95% CI, 0.765-1.146; P = .52), or amisulpride (RR, 0.777; 95% CI, 0.532-1.137; P = .19) (Table 1). For all-cause discontinuation, clozapine was associated with a significantly lower risk than risperidone (RR, 0.702; 95% CI, 0.615-0.800; P < .001; NNT, 7; 95% CI, 5-10), olanzapine (RR, 0.792; 95% CI, 0.669-0.939; P = .007; NNT, 10; 95% CI, 7-33), quetiapine (RR, 0.654; 95% CI, 0.490-0.874; P = .004; NNT, 6; 95% CI, 4-15), and amisulpride (RR, 0.694; 95% CI, 0.495-0.971; P = .03; NNT, 6; 95% CI, 4-57) but not aripiprazole (RR, 0.712; 95% CI, 0.338-1.501; P = .37). After removing studies with unequal observation periods, clozapine was associated with a significantly lower risk of hospitalization than all NC-SGAs, including risperidone, except for olanzapine (RR, 0.914; 95% CI, 0.732-1.142; P = .43), and with a lower risk of all-cause discontinuation than all other NC-SGAs, except for aripiprazole (RR, 0.712; 95% CI, 0.338-1.501; P = .37) and amisulpride (RR, 0.792; 95% CI, 0.598-1.049; P = .10) (eTable 4 in the Supplement).

Subgroup and Meta-Regression Analyses

Across 7 subgroup analyses for hospitalization risk (eTable 5-1 in eTable 5 in the Supplement) and 6 subgroup analyses for all-cause discontinuation (eTable 5-2 in eTable 5 in the Supplement), only for all-cause discontinuation 2 variables (treatment setting and study design) differed significantly across subgroups. Clozapine was associated with greater advantages than NC-SGAs in mixed inpatients and outpatients and in retrospective database studies (eTable 5-2 in eTable 5 in the Supplement). In addition, none of the meta-regression analyses for both coprimary outcomes yielded significant moderators (eTable 6 in the Supplement).

Secondary Outcomes

Secondary outcomes were recorded. These results are summarized in Table 2 and eFigure 4 in the Supplement.

Table 2. Results of Secondary Outcomes.

Outcome	No. of Studies	No. of Patients	Effect Size		Favorsa	Heterogeneity		Effect Size		Heterogeneity
			SMD (95% CI)	P Value		I2, %	P Value	MD (95% CI)	P Value	I2, %	P Value
Effectiveness-Related Outcomes
Psychopathology-related outcomes
Overall symptoms	10	7767	−0.302 (−0.572 to −0.032)	.03b	Clozapine	74.8	<.001b	NA	NA	NA	NA
Positive symptoms	8	8111	−0.253 (−0.574 to 0.069)	.12	NS	76.3	<.001b	NA	NA	NA	NA
Negative symptoms	9	8153	−0.227 (−0.568 to 0.113)	.19	NS	79.5	<.001b	NA	NA	NA	NA
General symptoms	5	357	−0.050 (−0.351 to 0.251)	.75	NS	45.4	.12	NA	NA	NA	NA
Depressive symptoms	8	8212	−0.008 (−0.178 to 0.163)	.93	NS	30.9	.18	NA	NA	NA	NA
Cognitive symptoms	3	7818	−0.124 (−0.235 to −0.014)	.03b	Clozapine	<0.1	.60	NA	NA	NA	NA
Clinical Global Impressions scale severity	4	170	−1.182 (−2.243 to −0.122	.03b	Clozapine	87.7	<.001b	NA	NA	NA	NA
Hospitalization-related outcomes
Length of hospitalization, d	3	2587	0.464 (−0.084 to 1.011)	.10	NS	64.1	.06	NA	NA	NA	NA
Frequency of hospitalizationc	5	17 747	Rate ratio: 1.180 (1.034 to 1.346)	.01b	NC-SGAs	27.0	.24	NA	NA	NA	NA
Other outcomes
Suicide attempt or self-injurious behavior	4	19 700	Risk ratio: 0.672 (0.432 to 1.046)	.08	NS	13.2	.33	NA	NA	NA	NA
Study-defined response rate	6	9182	Risk ratio: 1.066 (0.976 to 1.164)	.16	NS	6.2	.38	NA	NA	NA	NA
Safety and/or Tolerability–Related Outcomes
Cardiometabolic-related outcomes
Body weight, kg	9	5329	0.308 (0.057 to 0.560)	.02b	Clozapine	81.1	<.001b	1.70 (0.31 to 3.08)	.02b	81.1	<.001b
BMI	8	8606	0.317 (0.079 to 0.554)	.009b	Clozapine	75.5	<.001b	0.96 (0.24 to 1.68)	.009b	77.8	<.001b
Waist circumference, cm	3	453	0.272 (−0.042 to 0.586)	.09	NS	37.7	.20	1.21 (−0.78 to 3.20)	.23	51.1	.13
Systolic blood pressure, mm Hg	3	437	0.247 (0.034 to 0.459)	.02b	Clozapine	<0.1	.37	2.22 (0.15 to 4.28)	.04b	9.3	.33
Diastolic blood pressure, mm Hg	3	437	0.336 (0.111 to 0.561)	.003b	Clozapine	9.0	.33	1.92 (0.03 to 3.81)	.047b	53.8	.12
Total cholesterol, mg/dL	4	433	0.297 (−0.186 to 0.781)	.23	NS	75.3	.007b	11.06 (−7.66 to 29.78)	.25	72.8	.01b
Triglycerides, mg/dL	7	725	0.162 (−0.041 to 0.365)	.12	NS	23.9	.25	11.66 (2.93 to 20.38)	.009b	<0.1	.59
Glucose, mg/dL	7	880	0.486 (0.230 to 0.743)	<.001b	Clozapine	55.6	.04b	8.09 (4.56 to 11.62)	<.001b	9.6	.18
Insulin, μIU/mL	3	394	0.391 (0.115 to 0.667)	.005b	Clozapine	<0.1	.41	3.50 (−1.04 to 8.04)	.13	36.2	.21
HOMA-IR	4	422	0.382 (0.119 to 0.645)	.004b	Clozapine	<0.1	.49	0.99 (0.01 to 1.96)	.047b	15.6	.31
Metabolic syndrome risk	3	378	Risk ratio: 1.630 (0.937 to 2.834)	.08	NS	23.7	.27	NA	NA	NA	NA
Weight gain risk	5	4793	Risk ratio: 1.075 (0.751 to 1.537)	.69	NS	77.3	.001b	NA	NA	NA	NA
Type 2 diabetes risk	5	5539	Risk ratio: 1.777 (1.229 to 2.570)	.002b	Clozapine	<0.1	.87	NNH: 27 (13 to 90)d	NA	NA	NA
Other safety-related outcomes
Death	5	45 272	Risk ratio: 0.920 (0.708 to 1.194)	.53	NS	26.8	.24	NA	NA	NA	NA
EPS risk or anticholinergic use	7	9734	Risk ratio: 0.640 (0.412 to 0.996)	.048b	NC-SGAs	62.1	.02b	NNH: −16 (−1359 to −10)d	NA	NA	NA
EPS score	3	306	0.068 (−0.179 to 0.314)	.59	NS	<0.1	.72	NA	NA	NA	NA

Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); EPS, extrapyramidal symptoms; HOMA-IR, homeostatic model assessment of insulin resistance; MD, mean difference; NA, not applicable; NC-SGAs, nonclozapine second-generation antipsychotics; NNH, number needed to harm; NS, not significant; SGA, second-generation antipsychotic; SMD, standardized mean difference.

SI conversion factors: To convert cholesterol level to millimoles per liter, multiply by 0.0259; triglycerides level to millimoles per liter, multiply by 0.0113; glucose level to millimoles per liter, multiply by 0.0555; and insulin level to picomoles per liter, multiply by 6.945.

^aFor effectiveness-related outcomes, “favors” denotes “superiority” (ie, better effectiveness); for safety and/or tolerability–related outcomes “favors” denotes “inferiority” (ie, higher burden of adverse effects).

^bP < .05.

^cNumber of hospitalizations per person-years.

^dNumber needed to treat and number needed to harm were only calculated when the risk ratio analysis was significant (P < .05). The mean risk for the NC-SGAs was used as the assumed control group risk.

Effectiveness-Related Outcomes

In 11 effectiveness-related analyses, clozapine was associated with a significant reduction in overall symptoms (SMD, −0.302; 95% CI, −0.572 to −0.032; P = .03), cognitive symptoms (SMD, −0.124; 95% CI, −0.235 to −0.014; P = .03), and Clinical Global Impressions scale severity (SMD, −1.182; 95% CI, −2.243 to −0.122; P = .03) but not positive symptoms (SMD, −0.253; 95% CI, −0.574 to 0.069; P = .12), negative symptoms (SMD, −0.227; 95% CI, −0.568 to 0.113; P = .19), general symptoms (SMD, −0.050; 95% CI, −0.351 to 0.251; P = .75), and depressive symptoms (SMD, −0.008; 95% CI, −0.178 to 0.163; P = .93) (Table 2). Only for cognitive symptoms, there was a difference in statistical significance depending on the pre-post correlation coefficients (no significance when using low correlation coefficients) (eTable 2-1 in eTable 2 in the Supplement). Clozapine was associated with a significantly increased number of hospitalizations (RaR, 1.180; 95% CI, 1.034-1.346; P = .01), but only numerical differences favoring clozapine but missing statistical significance were found regarding hospitalization days (SMD, 0.464; 95% CI, −0.084 to 1.011; P = .10) and suicide attempt or self-injurious behavior (RR, 0.672; 95% CI, 0.432-1.046; P = .08) (Table 2).

Safety-Related Outcomes

Among 16 safety-related analyses, clozapine was significantly associated with an increase in body weight (SMD, 0.308; 95% CI, 0.057-0.560; P = .02; MD, 1.70; 95% CI, 0.31-3.08 kg; P = .02), BMI (SMD, 0.317; 95% CI, 0.079-0.554; P = .009; MD, 0.96; 95% CI, 0.24-1.68; P = .009), systolic blood pressure (SMD, 0.247; 95% CI, 0.034-0.459; P = .02; MD, 2.22; 95% CI, 0.15-4.28 mm Hg; P = .04), diastolic blood pressure (SMD, 0.336; 95% CI, 0.111-0.561; P = .003; MD, 1.92; 95% CI, 0.03-3.81 mm Hg; P = .047), triglycerides (SMD, 0.162; 95% CI, −0.041 to 0.365; P = .12; MD, 11.66; 95% CI, 2.93-20.38 mg/dL; P = .009), glucose (SMD, 0.486; 95% CI, 0.230-0.743; P < .001; MD, 8.09; 95% CI, 4.56-11.62 mg/dL; P < .001), insulin (SMD, 0.391; 95% CI, 0.115-0.667; P = .005; MD, 3.50; 95% CI, −1.04 to 8.04 μIU/mL; P = .13), homeostatic model assessment of insulin resistance (SMD, 0.382; 95% CI, 0.119-0.645; P = .004; MD, 0.99; 95% CI, 0.01-1.96; P = .047), and type 2 diabetes risk (RR, 1.777; 95% CI, 1.229-2.570; P = .002; NNH, 27; 95% CI, 13-90) (Table 2) (to convert triglycerides level to millimoles per liter, multiply by 0.0113; glucose level to millimoles per liter, multiply by 0.0555; and insulin level to picomoles per liter, multiply by 6.945). Clozapine was associated with significantly lower extrapyramidal symptoms (EPS) risk or anticholinergic use (RR, 0.640; 95% CI, 0.412-0.996; P = .048; NNH, −16; 95% CI, −1359 to −10). There were no significant differences for waist circumference, total cholesterol, metabolic syndrome risk, study-defined weight gain risk, and EPS score between clozapine and NC-SGAs.

Discussion

In this systematic review and meta-analysis, we examined various outcomes related to the effectiveness and safety of clozapine vs oral NC-SGAs from 63 nonrandomized cohort studies and 109 341 patients. We found that clozapine was associated with a significant reduction in hospitalization risk by 18% (NNT, 18) and all-cause discontinuation by 27% (NNT, 8), despite greater illness severity and/or chronicity in patients treated with clozapine. The results were not altered by study quality or publication bias. In pairwise comparisons and studies with equal observation time, clozapine was associated with lower hospitalization risk than all NC-SGAs with such data, except for olanzapine, while clozapine was associated with reduction in all-cause discontinuation compared with all NC-SGAs, except for aripiprazole and amisulpride. We were unable to identify significant moderators and mediators of the advantage of clozapine regarding hospitalization risk, indicating that the advantage of clozapine applies across broad patient, setting, and study design features. The same was true for clozapine’s advantage regarding all-cause discontinuation, except that clozapine was associated with a significantly larger effect size in retrospective database studies where all eligible patients are included, no consent occurs, and observation biases are absent.

In addition, clozapine was also associated with a significantly greater reduction than NC-SGAs regarding Clinical Global Impressions scale severity (effect size, −1.182) and overall symptoms (effect size, −0.302) and cognitive symptoms (effect size, −0.124) but not positive, negative, or depressive symptoms. While reducing EPS or anticholinergic use risk by 36% (NNH, −16), clozapine was associated with greater risk of cardiometabolic-related outcomes, including increased body weight, BMI, triglycerides (MD only), glucose, insulin (SMD only), insulin resistance, and type 2 diabetes risk (78%; NNH, 27).

The observed effectiveness-related advantages of clozapine in the meta-analyzed cohort studies are in contrast to meta-analytic results of masked RCTs,^7,8 which provided little or inconsistent evidence for the superiority of clozapine vs SGAs. Because participants in RCTs do not necessarily represent real-world patients or settings, the results from RCTs cannot readily be generalized to clinical practice.¹⁰ Therefore, a meta-analysis of cohort studies can provide complementary information about clozapine’s association with influence, effectiveness, and safety and tolerability.¹³ Cohort studies of longitudinal and registry databases also have the advantage of longer-term follow-up than RCTs.

Hospitalization was not evaluated in recent meta-analyses of masked RCTs.^7,8 One meta-analysis of 3 RCTs (n = 369) and 19 observational studies (n = 44 349) investigated the association of clozapine with hospitalization, showing also that clozapine was associated with a significant reduction in the proportion of individuals hospitalized compared with NC-SGAs (13 studies [n = 29 559]; RR, 0.75; 95% CI, 0.67-0.83; P < .001).⁵⁴ Although in our much larger meta-analysis clozapine was not associated with a significantly lower risk than NC-SGAs for hospitalization at 6 months, this result may reflect a type II error because there were fewer events at 6 months than at later periods, and 6 months may be too short to reliably evaluate hospitalization risk. Unexpectedly, clozapine was associated with a higher frequency of hospitalizations than NC-SGAs. However, because clozapine-treated patients had greater illness severity, they probably had been hospitalized more frequently before clozapine initiation. Only an average of 40% of clozapine-treated patients with TRS are considered responders⁵⁵; therefore, it is likely that a subgroup of the most severely ill patients treated with clozapine contributed to the higher number of hospitalizations, despite lower overall hospitalization risk in the entire group of clozapine-treated patients. Because adherence was neither reported nor assured, it is possible that some of the severely ill patients receiving clozapine were not sufficiently adherent.

Clozapine was also associated with a lower risk of all-cause discontinuation, reflecting effectiveness. Reduced all-cause discontinuation in cohort studies is inconsistent with the lack of difference in meta-analyses of masked RCTs.^7,8 Considering that clozapine has been associated with serious adverse effects and requires blood monitoring,⁵⁶ continuation reflects treatment acceptability, although a “surveillance association” may influence outcomes. Clozapine is considered a last-resort treatment for TRS.¹ Therefore, clozapine might have been continued, despite minimal effectiveness. In analyses at specific periods, there were no statistically significant differences in all-cause discontinuation between clozapine and NC-SGAs at 6 and 36 months, but these analyses were likely underpowered; early adverse effect–related discontinuations with clozapine may also have had a role.

Numerical advantages for associations favoring clozapine regarding study-defined response and suicide attempt or self-injurious behavior missed statistical significance. Only 3 to 6 studies contributed such data, likely yielding insufficient power.

Regarding tolerability, clozapine was associated with a lower risk of EPS, consistent with short-term trials in acutely ill patients,⁶ but was associated with significantly greater cardiometabolic risk than NC-SGAs in 9 of the 13 specific meta-analyzable adverse effects. This finding is consistent with clozapine being one of the most cardiometabolically problematic antipsychotics.^{57,58,59,60,61} Therefore, a careful risk-benefit evaluation is needed before initiating clozapine, and cardiometabolic parameters (body weight, fasting glucose and lipid levels, and glycated hemoglobin levels) should be monitored regularly.

Despite greater cardiometabolic adverse effects than with almost all other antipsychotics, mortality rates with clozapine have not increased but even decreased,^62,63 including cardiometabolic-related mortality,⁶⁴ likely due to effectiveness-related indirect benefits on healthy lifestyle and monitoring and management of physical health risk factors and comorbidities. However, in contrast to a recent meta-analysis reporting a significant association with a reduction in all-cause mortality with clozapine vs other antipsychotics,⁶⁵ we did not observe this advantage. Reasons for this difference could include that in the other meta-analysis⁶⁵ first-generation antipsychotics and NC-SGAs were pooled, and the significant finding only emerged in studies where patients used clozapine continuously (RR, 0.56; 95% CI, 0.36-0.85; P = .007) and not in studies where patients could have stopped clozapine (“ever use”) (RR, 0.74; 95% CI, 0.38-1.45; P = .376), while we did not make this distinction due to too few studies comparing clozapine specifically with NC-SGAs. Moreover, the follow-up duration in our meta-analysis was shorter (median, 1 year vs 5.4 years⁶⁵), and we did not adjust the results by person-years.

Characteristics of studies in our meta-analysis of cohort studies differ in 4 ways from the 3 most recent meta-analyses.^7,8,54 These include (1) lack of randomization in all studies, increasing generalizability of included patients but introducing a channeling bias of more severely ill patients being treated with clozapine (creating a conservative bias that strengthens the outcomes); (2) longer duration of our trials vs the meta-analyses that included RCTs only (median, 12 months vs 11 weeks⁷ and 12 weeks⁸); (3) inclusion of more recently published evidence (up to 2018 vs 2012,⁷ 2009,⁸ and 2016⁵⁴); and (4) inclusion of more study participants comparing clozapine specifically with NC-SGAs (eg, 7767 vs 1319 individuals⁷ and 1429 individuals⁸ for overall symptom reduction and 49 453 vs 29 559 individuals⁵⁴ for hospitalization in the sole meta-analysis that included a subsample of cohort studies and that analyzed only hospitalization-related outcomes).

Limitations

This meta-analysis has several limitations. Although risk estimates adjusting for time (HRs) and other relevant covariates would have been ideal, too few such data were available for meta-analysis. For example, only 3 studies provided adjusted ORs (eFigure 3 in the Supplement). Nevertheless, because clozapine is more likely prescribed to patients with more severe treatment resistance, the lack of adjusting for baseline illness severity creates a conservative bias against clozapine. Therefore, clozapine’s association with many better effectiveness outcomes is a sign of the robustness of the association, and effect sizes are an underestimate if anything. Furthermore, despite significant heterogeneity across studies and results, few if any significant moderators or mediators emerged in sensitivity, subgroup, and meta-regression analyses of both coprimary outcomes. Not only clozapine’s influence and effectiveness but also the frequency of clinical contacts and monitoring could alter the lower risk of hospitalization and all-cause discontinuation, but we lacked data to control for frequency of clinical contacts.

Conclusions

This comprehensive meta-analysis of cohort studies, reflecting clinical practice more than RCTs, found clozapine to be associated with better effectiveness outcomes than NC-SGAs, despite more severely ill patients being treated with clozapine, but with significantly greater risk of cardiometabolic adverse outcomes, both of which require consideration when making treatment choices. Because cohort studies have inherent biases, large pragmatic RCTs with broad inclusion and minimal exclusion criteria that better reflect real-world practice are needed.

Supplement.

eTable 1. Results of Secondary Outcomes Using Different Types of Pre/Post Coefficients

eTable 2. Study and Patient Characteristics With Analyzed Outcomes

eTable 3. Sensitivity Analyses of Co-Primary Outcomes Removing Studies With Unequal Observation Periods for Comparing Clozapine With Specific Non-Clozapine SGAs

eTable 4. Subgroup Analyses for Co-Primary Outcomes

eTable 5. Results of Meta-Regression Analyses for Co-Primary Outcomes

eTable 6. Newcastle-Ottawa Quality Assessment Scale of Studies With Co-Primary Outcomes

eFigure 1. PRISMA Diagram of the Systematic Literature Search

eFigure 2. Funnel Plots for Co-Primary Outcomes

eFigure 3. Forest Plots of Detailed Analyses of Co-Primary Outcomes

eFigure 4. Forest Plots of Analyses of Secondary Outcomes