Long‐term effectiveness of oral second‐generation antipsychotics in patients with schizophrenia and related disorders: a systematic review and meta‐analysis of direct head‐to‐head comparisons

Abstract

Second‐generation antipsychotics (SGAs) are recommended for maintenance treatment in schizophrenia. However, comparative long‐term effectiveness among SGAs is unclear. Here we provide a systematic review and meta‐analysis of randomized trials lasting ≥⃒6 months comparing SGAs head‐to‐head in schizophrenia and related disorders. The primary outcome was all‐cause discontinuation. Secondary outcomes included efficacy and tolerability, i.e., psychopathology, inefficacy‐related and intolerability‐related discontinuation, relapse, hospitalization, remission, functioning, quality of life, and adverse events. Pooled risk ratio and standardized mean difference were calculated using random‐effects models. Across 59 studies (N=45,787), lasting 47.4±32.1 weeks (range 24‐186), no consistent superiority of any SGA emerged across efficacy and tolerability outcomes. Regarding all‐cause discontinuation, clozapine, olanzapine and risperidone were significantly (p<0.05) superior to several other SGAs, while quetiapine was inferior to several other SGAs. As to psychopathology, clozapine and olanzapine were superior to several other SGAs, while quetiapine and ziprasidone were inferior to several other SGAs. Data for other efficacy outcomes were sparse. Regarding intolerability‐related discontinuation, risperidone was superior and clozapine was inferior to several other SGAs. Concerning weight gain, olanzapine was worse than all other compared non‐clozapine SGAs, and risperidone was significantly worse than several other SGAs. As to prolactin increase, risperidone and amisulpride were significantly worse than several other SGAs. Regarding parkinsonism, olanzapine was superior to risperidone, without significant differences pertaining to akathisia. Concerning sedation and somnolence, clozapine and quetiapine were significantly worse than some other SGAs. In summary, different long‐term SGA efficacy and tolerability patterns emerged. The long‐term risk‐benefit profiles of specific SGAs need to be tailored to individual patients to optimize maintenance treatment outcomes.

Schizophrenia is a mental disorder whose course is generally characterized by repeated relapses as well as a worsening of psychopathology and social functioning, thus requiring maintenance treatment1, 2, 3. Antipsychotics are efficacious for relapse prevention in chronic and first‐episode patients4, 5, reducing relapse risk by 2‐6‐fold versus no antipsychotic treatment2, 4, 5, 6.

A previous meta‐analysis by our group, comparing second‐generation antipsychotics (SGAs) with first‐generation antipsychotics (FGAs), found that the former as a class were superior to the latter regarding relapse prevention, all‐cause discontinuation and other relapse‐related outcomes3.

Despite the importance of long‐term treatment in schizophrenia, in which the magnitude of benefits and risks of medications may be different from acute phase treatment, no comprehensive meta‐analysis of the comparative long‐term effectiveness, efficacy and safety among oral SGAs currently exists7.

Although one meta‐analysis targeted maintenance trials that compared antipsychotics with placebo2, indirect comparisons using placebo as the common comparator are not conclusive8. Further, a multiple treatment meta‐analysis, which includes indirect comparisons, is not necessarily ideal, especially when the number of trials comparing antipsychotics directly is limited and when homogeneity of these trials cannot be assured9.

Knowledge about the comparative effectiveness, efficacy and tolerability of SGAs in the long‐term treatment of schizophrenia is important7. Specifically, differences in side effect risk9, 10, 11, some of which may increase with time, need to be weighed against potential differences in long‐term effectiveness and efficacy.

Here we report the results of the first comprehensive meta‐analysis of head‐to‐head randomized controlled trials comparing two or more SGAs in the long‐term treatment of schizophrenia, aiming to assess the comparative effectiveness, efficacy and safety of these medications.

METHODS

The meta‐analysis was performed following PRISMA guidelines12.

Search and inclusion criteria

We conducted an electronic search without language restrictions using MEDLINE/PubMed, the Cochrane library, ISI Web of Science, PsycINFO, CINAHL and the US National Institutes of Health clinical trials registry (http://www.clinicaltrials.gov). The following search terms were used: antipsychotic(s); neuroleptic(s); individual names of SGAs; schizophrenia; random, randomly, randomized; and maintenance, relapse, discontinuation or long‐term. The last search was done on October 29, 2018. The electronic search was supplemented by a hand search of reference lists of relevant studies and reviews. Authors and companies were contacted to provide missing information and unpublished data.

We included randomized, head‐to‐head comparisons of oral SGAs in adults with schizophrenia or schizoaffective disorder which reported on treatment discontinuation, whether randomization occurred during the acute or maintenance phase. As we aimed to focus on the comparative long‐term effectiveness of SGAs, we only included head‐to‐head studies lasting ≥⃒6 months.

We excluded studies with >20% of non‐schizophrenia/schizoaffective disorder patients. As long‐acting injectable formulation enhances the adherence and therefore has a significant impact on long‐term outcome13, 14, we excluded studies on long‐acting antipsychotics.

The search, selection of the literature, and data extraction were conducted independently by ≥⃒2 reviewers (KH, MN, TK, CC). Disagreements were resolved by consensus.

Outcomes

The primary outcome was all‐cause discontinuation at study endpoint.

Secondary outcomes included: a) psychopathology score change, measured by the Positive and Negative Syndrome Scale (PANSS), the Brief Psychiatric Rating Scale (BPRS) or the Clinical Global Impression ‐ Severity (CGI‐S) score (mixed models or last‐observation‐carried‐forward was prioritized over observed cases analysis); b) inefficacy‐related discontinuation (as reported by the original study authors); c) intolerability‐related discontinuation (as reported by the original study authors); d) relapse (as reported by the original study authors); e) hospitalization; f) remission (as reported by the original study authors); g) functioning score; h) quality of life (QOL); and i) adverse events.

Adverse events included: weight gain (as change from baseline or proportion of patients with clinically significant increase); prolactin increase (as change from baseline or proportion of patients with hyperprolactinemia); neuromotor adverse effects, including parkinsonism assessed with the Simpson‐Angus Rating Scale or use of anticholinergics, akathisia and dyskinesia; and sedation and/or somnolence.

Data analysis

SGAs were compared individually for each outcome. We applied a “once‐randomized‐analyzed” intent‐to‐treat (ITT) endpoint analysis. In studies that followed patients even after they were switched off the originally allocated medication during the study period, we analyzed the primary outcome based only on the first medication but, for secondary outcomes, we extracted and analyzed the data as reported in the ITT sample.

Pooled risk ratio (RR) and standardized mean difference (SMD) with 95% confidence intervals (CIs) were calculated using random‐effects models15. RR values <1 indicate superiority of the first SGA for negative outcomes (such as all‐cause discontinuation, relapse, inefficacy‐related and intolerability‐related discontinuation), while RR values >1 indicate superiority for the only positive outcome, remission. For simplicity we adjusted effect sizes, so that SMDs <0 indicate superiority of the first SGA, independent of whether a lower value (e.g., psychopathology) or higher value (e.g., functioning, QOL) is a positive outcome.

Number‐needed‐to‐treat (NNT) was calculated when categorical outcome differences were significant. Heterogeneity was only inspected when ≥⃒2 studies were analyzed, using the chi‐square test (p<0.1 indicating significant heterogeneity)16 and the I² statistic (I²≥⃒50% indicating significant heterogeneity)17. For study quality assessment, we used the Jadad scale18, that provides a sum score for sensitivity analyses.

In addition, a priori‐defined subgroup analyses of the primary outcome were conducted (where ≥⃒2 studies existed), seeking to identify potential moderators, methodological biases, and whether findings extended to clinically relevant sub‐populations or treatment groups. Subgroup analyses included: a) randomization time point (acute vs. maintenance phase); b) sponsorship (medication‐specific sponsor vs. academia); c) study quality (high vs. low Jadad score)18; d) concealment (open or single‐blinded vs. double‐blinded); e) location (international/USA/Europe/Asia); f) dosing (fixed vs. flexible), and g) first episode vs. chronically ill.

Comprehensive Meta‐Analysis, version 3 (Biostat, NJ, USA) was used for all two‐tailed analyses, with alpha=0.05, without adjustments for multiple comparisons. Publication bias was assessed with the funnel plot, Egger's regression test19 and the “trim and fill” method20 for the primary outcome, whenever ≥⃒3 studies were analyzed.

RESULTS

Search and study characteristics

A total of 8,611 references were identified (Figure 1). After removing 152 duplicates, we excluded 7,823 of the remaining 8,459 references based on title/abstract inspection. Of 113 references subjected to full‐text inspection, 54 articles were dropped because of: inappropriate participants (N=17), review/editorial (N=11), no usable data (N=10), inappropriate medication (N=6), short‐term study (N=4), no/inadequate randomization (N=3), and meeting abstracts of already included studies (N=3).

Figure 1.

PRISMA flow chart

Altogether, we included 63 reports21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83 (59 randomized studies) with 45,787 participants (median: 255 participants/study, range: from 12 to 18,154) (Table 1). The mean age of the population was 37.6±7.0 years; 62.1±13.3% were male and 61.1±28.8% were white. The mean study duration was 47.4±32.1 weeks (range: 24‐186).

Table 1.

Characteristics of included studies

Study	Country	Blinding status	N. patients	Randomization time point	Duration (weeks)	First episode/chronically ill	Mean age	% male	Comparison	Dose (mean, mg/day)	Jadad score
Addington et al21	International	DB	139	Maintenance	44	Chronically ill	34.6	65.5	RIS vs. ZIP	8; 114	3
Alvarez et al22	Spain	DB	50	Acute	24	Chronically ill	38.4	70.0	OLZ vs. ZIP	15; 107.4	3
Alvarez et al23;Ciudad et al24	Spain	OL	235	Maintenance	48	Chronically ill	36.5	72.3	OLZ vs. RIS	12.2; 4.9	2
Breier et al25	International	DB	548	Acute	28	Chronically ill	39.2	64.2	OLZ vs. ZIP	15.27; 115.96	3
Chan et al26	Taiwan	RB	60	Acute	24	Chronically ill	45.4	35.0	OLZ vs. RIS	4.1; 12.6	3
Chrzanowski et al27	International	OL	214	Acute	52	Chronically ill	41.5	54.0	APZ vs. OLZ	22; 14.2	2
Citrome et al28	International	DB	629	Maintenance	52	Chronically ill	41.7	69.0	LUR vs. RIS	84.7; 4.3	4
Crespo‐Facorro et al29	Spain	OL	202	Acute	52	First episode	32.0	53.5	APZ vs. QTP vs. ZIP	11.6; 311.4; 61.0	3
Crespo‐Facorro et al30	Spain	OL	174	Acute	156	First episode	27.3	62.1	OLZ vs. RIS	12.9; 3.4	1
de Arce Cordón et al31;Gaebel et al32	International	OL	711	Maintenance	104	Chronically ill	41.6	57.8	APZ vs. QTP	15.1; 413.4	2
Deberdt et al33	USA	DB	133	Maintenance (enriched design)	26	Chronically ill	44.0	NR	OLZ vs. QTP	16.9; 439.7	3
Durgam et al34	International	DB	120	Acute	26	Chronically ill	39.6	59.2	ASN vs. OLZ	Fixed dose: 5 or 10; 15	4
Fleischhacker et al35	International	DB	488	Acute	46	Chronically ill	36.6	56.8	APZ vs. OLZ	23.0; 15.4	4
Kahn et al36	International	OL	498	Acute	52	First episode	26.0	60.0	AMI vs. OLZ vs. QTP vs. ZIP	450.8; 12.6; 498.6; 107.2	3
Kane et al37	International	DB	566	Acute	28	Chronically ill	37.8	67.8	APZ vs. OLZ	19.3; 16.7	3
Keefe et al38	International	DB	414	Acute	52	Chronically ill	39.1	71.3	OLZ vs. RIS	12.3; 5.2	3
Kern et al39	USA	OL	255	Acute	2 6	Chronically ill	40.0	64.5	APZ vs. OLZ	NR	2
Kinon et al40	USA	DB	346	Acute	24	Chronically ill	41.1	65.9	OLZ vs. QTP	15.6; 455.8	4
Kinon et al41	USA	DB	394	Acute	24	Chronically ill	41.6	62.9	OLZ vs. ZIP	Fixed dose:10 or 15 or 20;80 or 120 or 160	3
Kishi et al42	Japan	RB	44	Acute	24	Chronically ill	39.5	40.9	APZ vs. BLO	11.5; 10.3	4
Kumar et al43	India	DB	71	Maintenance	48	Chronically ill	40.7	50.7	OLZ vs. RIS	14.4; 5.8	3
Lecrubier et al44	France	DB	244	Maintenance	26	Chronically ill	37.4	68.6	AMI vs. OLZ	Fixed dose; 150; 5 or 20	3
Lieberman et al45	USA	DB	1,460	Acute	78	Chronically ill	40.6	72.3	OLZ vs. QTP vs. RIS vs. ZIP	20.1; 543.4; 3.9;112.8	3
Liu et al46	China	OL	80	Acute	52	First episode	29.5	0.00	QTP vs. RIS	420; 3.4	3
Loebel et al47;NCT0078969848	International	DB	327	Maintenance	52	Chronically ill	37.6	66.8	LUR vs. QTP	NR	4
McEvoy et al49	USA	OL	99	Acute	26	Chronically ill	39.7	81.0	CLO vs. OLZ vs. QTPvs. RIS	332.1; 23.4; 642.9;4.8	2
McEvoy et al50	USA	DB	400	Acute	52	First episode	24.5	73.0	OLZ vs. QTP vs. RIS	11.7; 506; 2.4	3
McQuade et al51	International	DB	317	Acute	26	Chronically ill	38.4	72.0	APZ vs. OLZ	25.1; 16.5	3
Meltzer et al52	International	RB	980	Acute	104	Chronically ill	37.1	61.4	CLO vs. OLZ	274.2; 16.6	2
Meltzer et al53	USA	DB	40	Acute	26	Chronically ill	36.8	67.5	CLO vs. OLZ	564; 33.6	4
Mortimer et al54	International	DB	377	Acute	24	Chronically ill	37.8	65.0	AMI vs. OLZ	504; 13	5
Naber et al55	Germany	DB	114	Acute	26	Chronically ill	34.0	61.0	CLO vs. OLZ	209; 16.2	3
Naber et al56;NCT0060075657	International	OL	798	Acute	52	Chronically ill	39.7	58.2	QTP vs. RIS	NR	3
Németh et al58	International	DB	461	Maintenance	26	Chronically ill	40.5	57.4	CAR vs. RIS	Fixed dose: 3 or 4 or 5 or 6; 3 or 4 or 6	5
Noordsy et al59	USA	DB	107	Maintenance	24	Chronically ill	42.0	82.2	OLZ vs. RIS	Range: 2.5‐30; 1‐10	1
Parabiaghi et al60	Italy	OL	300	NR	52	Chronically ill	42.7	58.0	APZ vs. OLZ	19.7; 13.7	3
Purdon et al61	Canada	DB	65	Maintenance	54	Chronically ill	28.9	70.6	OLZ vs. RIS	11.00; 6.00	4
Ritchie et al62	Australia	OL	66	Acute	186	Chronically ill	69.5	28.8	OLZ vs. RIS	NR	2
Sanz‐Fuentenebroet al63	Spain	OL	30	Acute	52	First episode	24.5	70.0	CLO vs. RIS	220.45; 5.43	2
Schnell et al64	Germany	DB	30	NR	52	Chronically ill	29.0	86.7	CLO vs. ZIP	225; 200	3
Schoemaker et al65	International	DB	440	Maintenance	96	Chronically ill	36.9	55.5	ASN vs. OLZ	13.4; 13.4	3
Schooler et al66	USA	DB	107	Acute	29	Chronically ill	41.9	79.4	CLO vs. RIS	456.7; 6.8	4
Sechter et al67	International	DB	310	Acute	26	Chronically ill	38.4	55.0	AMI vs. RIS	683; 6.92	3
Schreiner et al68	International	OL	459	Acute	26	Chronically ill	38.2	58.0	OLZ vs. PAL	11.6; 6.9	3
Simpson et al69	USA	DB	126	Maintenance	26	Chronically ill	NR	NR	OLZ vs. ZIP	12.6; 135.2	2
Strom et al70	International	OL	18,154	Acute	52	Chronically ill	41.1	55.0	OLZ vs. ZIP	NR	2
Stroup et al71	USA	DB	444	Acute	26	Chronically ill	40.8	69.0	OLZ vs. QTP vs. RIS vs. ZIP	20.5; 565.2; 4.1; 115.9	3
Stroup et al72	USA	DB	115	Acute	78	Chronically ill	40.8	77.0	OLZ vs. QTP vs. RIS	20.7; 586.1; 3.7	3
Thomas et al73	International	OL	9,809	Acute	Mean: 564.0; 489.6 days	Chronically ill	38.3	55.3	RIS vs. SER	Range: 2‐8; 12‐20	3
Tran et al74	International	DB	339	Acute	28	Chronically ill	36.2	64.9	OLZ vs. RIS	17.2; 7.2	3
Tunis et al75	USA	OL	450	Acute	52	Chronically ill	43.0	63.0	OLZ vs. RIS	13.49; 4.95	2
Wani et al76	India	OL	62	Maintenance (enriched design)	24	Chronically ill	29.8	62.9	APZ vs. OLZ	NR	1
Zhang et al77	China	OL	254	Acute	52	First episode	26.4	61.0	APZ vs. PAL vs. ZIP	NR	2
NCT0014549678	International	DB	468	Maintenance	26	Chronically ill	42.9	73.9	ASN vs. OLZ	NR	3
NCT0020610279	USA	OL	1,098	Maintenance	104	NR	NR	58.8	QTP vs. RIS	Range: 200‐800; 2‐8	3
NCT0021283680	International	DB	481	Maintenance	26	Chronically ill	40.5	68.2	ASN vs. OLZ	NR	2
NCT0023637981	International	DB	59	Maintenance	24	Chronically ill	39.7	NR	OLZ vs. RIS	Range: 5‐20; 2‐6	3
NCT0057328782	USA	RB	14	Acute	24	First episode	22.4	57.1	CLO vs. RIS	Range: 12.5‐100; 0.5‐5.0	1
NCT0080210083	USA	RB	12	Acute	28	Chronically ill	29.0	61.9	APZ vs. OLZ	NR	2

AMI – amisulpride, APZ – aripiprazole, ASN – asenapine, BLO – blonanserin, CAR – cariprazine, CLO – clozapine, LUR – lurasidone, OLZ – olanzapine, PAL – paliperidone, QTP – quetiapine, RIS – risperidone, SER – sertindole, ZIP – ziprasidone, DB – double‐blind, OL – open label, RB – rater‐blinded, NR – not reported

Forty‐six studies included multiple‐episode patients, eight included exclusively first‐episode patients, four included exclusively treatment‐resistant patients (all clozapine studies), and one did not report the number of episodes of included patients79. Thirty‐four studies were double‐blind, 20 were open‐label, and five had masked raters. Forty studies were sponsored by pharmaceutical companies, 18 were publicly funded, and funding was uncertain in one study77.

The number of studies with each individual SGA were: 43 for olanzapine, 27 for risperidone, 15 for quetiapine, 12 for ziprasi‐done, 12 for aripiprazole, eight for clozapine, four for amisulpride, four for asenapine, two for lurasidone, two for paliperidone, one for blonanserin, one for cariprazine, and one for sertindole.

Thirty‐nine studies (66.1%) randomized patients in the acute phase, eighteen (30.5%) in the maintenance phase, while the randomization time point was uncertain for two studies (3.4%)60, 64. Two studies33, 76 utilized an enriched design, in that patients stabilized on drug A were randomized to continued treatment or switch to drug B. Two studies70, 75 had a “naturalistic” follow‐up design, in that switches off the originally assigned drugs were allowed.

Eleven studies reported on relapse, and six on remission. The definition of relapse varied, with only two studies using the same criteria28, 47. Three8, 31, 37 out of six studies reporting on remission used Andreasen et al's criteria84.

Primary outcome measure: all‐cause discontinuation

Across 59 studies, the pooled effect sizes of individual SGA pairs concerning all‐cause discontinuation are shown in Figure 2.

Figure 2.

Results of comparisons of all‐cause discontinuation in meta‐analysis of second‐generation antipsychotics (SGAs). The first drug is the one written on the left side of the graph, and the comparator is written in the row of comparison. AMI – amisulpride, APZ – aripiprazole, ASN – asenapine, BLO – blonanserin, CAR – cariprazine, CLO – clozapine, LUR – lurasidone, OLZ – olanzapine, PAL – paliperidone, QTP – quetiapine, RIS – risperidone, SER – sertindole, ZIP – ziprasidone, M‐H RR – Mantel‐Haenszel risk ratio.

Clozapine had a significantly lower all‐cause discontinuation as compared with quetiapine (one study, N=64, RR=0.59, 95% CI: 0.42‐0.83, p=0.002) and risperidone (four studies, N=216, RR=0.74, 95% CI: 0.57‐0.95, p=0.020, I²=5.1%). Olanzapine had a significantly lower all‐cause discontinuation as compared with paliperidone (one study, N=459, RR=0.64, 95% CI: 0.46‐0.90, p=0.010), quetiapine (eight studies, N=1,942, RR=0.79, 95% CI: 0.71‐0.89, p<0.001, I²=55.8%), risperidone (16 studies, N=3,131, RR=0.88, 95% CI: 0.83‐0.93, p<0.001, I²=0.0%), and ziprasidone (eight studies, N=20,225, RR=0.82, 95% CI: 0.77‐0.87, p<0.001, I²=37.0%). Risperidone had a significantly lower all‐cause discontinuation as compared with sertindole (one study, N=9,809, RR=0.83, 95% CI: 0.80‐0.86, p<0.001) and ziprasidone (three studies, N=906, RR=0.90, 95% CI: 0.83‐0.98, p=0.012, I²=0.0%).

Other significant differences included the following: significantly lower all‐cause discontinuation for amisulpride vs. quetiapine (one study, N=208, RR=0.63, 95% CI: 0.44‐0.89, p=0.009); significantly higher all‐cause discontinuation for aripiprazole vs. olanzapine (eight studies, N=2,117, RR=1.17, 95% CI: 1.05‐1.30, p=0.006, I²=28.8%); significantly higher all‐cause discontinuation for lurasidone vs. risperidone (one study, N=629, RR=1.17, 95% CI: 1.02‐1.35, p=0.027); and significantly higher all‐cause discontinuation for quetiapine vs. ziprasidone (four studies, N=1,064, RR=1.12, 95% CI: 1.01‐1.25, p=0.031, I²=47.0%).

Secondary outcomes

Across 23 SGA comparisons concerning psychopathology, based on 32 studies, the following nine significant differences emerged: aripiprazole was superior to quetiapine and ziprasidone; clozapine was superior to quetiapine and risperidone; lurasidone was superior to quetiapine; olanzapine was superior to paliperidone and risperidone; and paliperidone was superior to aripiprazole and ziprasidone (Figure 3).

Figure 3.

Results of comparisons of psychopathology scores in meta‐analysis of second‐generation antipsychotics (SGAs). The first drug is the one written on the left side of the graph, and the comparator is written in the row of comparison. AMI – amisulpride, APZ – aripiprazole, ASN – asenapine, BLO – blonanserin, CAR – cariprazine, CLO – clozapine, LUR – lurasidone, OLZ – olanzapine, PAL – paliperidone, QTP – quetiapine, RIS – risperidone, SER – sertindole, ZIP – ziprasidone, SMD – standardized mean difference.

Across 26 comparisons concerning intolerability‐related discontinuation, based on 50 studies, the following significant differences emerged: quetiapine was superior to amisulpride; risperidone was superior to clozapine, quetiapine and sertindole; and ziprasidone was superior to clozapine (Figure 4).

Figure 4.

Results of comparisons of intolerability‐related discontinuation in meta‐analysis of second‐generation antipsychotics (SGAs). The first drug is the one written on the left side of the graph, and the comparator is written in the row of comparison. AMI – amisulpride, APZ – aripiprazole, ASN – asenapine, BLO – blonanserin, CAR – cariprazine, CLO – clozapine, LUR – lurasidone, OLZ – olanzapine, PAL – paliperidone, QTP – quetiapine, RIS – risperidone, SER – sertindole, ZIP – ziprasidone, M‐H RR – Mantel‐Haenszel risk ratio.

Across 20 comparisons concerning inefficacy‐related discontinuation, based on 47 studies, the following significant differences emerged: aripiprazole was superior to quetiapine; clozapine was superior to risperidone; lurasidone was superior to quetiapine; and olanzapine was superior to aripiprazole, quetiapine and ziprasidone (Figure 5).

Figure 5.

Results of comparisons of inefficacy‐related discontinuation in meta‐analysis of second‐generation antipsychotics (SGAs). The first drug is the one written on the left side of the graph, and the comparator is written in the row of comparison. AMI – amisulpride, APZ – aripiprazole, ASN – asenapine, BLO – blonanserin, CAR – cariprazine, CLO – clozapine, LUR – lurasidone, OLZ – olanzapine, PAL – paliperidone, QTP – quetiapine, RIS – risperidone, SER – sertindole, ZIP – ziprasidone, M‐H RR – Mantel‐Haenszel risk ratio.

Across 11 comparisons concerning relapse, only one significant difference emerged: the superiority of olanzapine over risperidone. Across 13 comparisons concerning hospitalization, clozapine was superior to olanzapine, and lurasidone and risperidone were superior to quetiapine. Across six comparisons concerning remission, lurasidone was superior to quetiapine, and quetiapine was superior to risperidone. Across 12 comparisons concerning functioning, aripiprazole was superior to quetiapine, cariprazine was superior to risperidone, and clozapine was superior to olanzapine. Across 11 comparisons concerning QOL, there were no significant SGA‐pair differences.

Twenty‐five comparisons based on 46 studies were meta‐analyzed for weight gain. Amisulpride, aripiprazole, quetiapine, risperidone, paliperidone and ziprasidone were superior to olanzapine; amisulpride, cariprazine, lurasidone and ziprasidone were superior to risperidone; paliperidone was superior to aripiprazole; and ziprasidone was superior to paliperidone and quetiapine (Table 2).

Table 2.

Results of meta‐analysis for adverse events

Outcome	Comparison	n	N	RR/SMD	95% CI		p	I 2 (%)
					Lower limit	Upper limit
Akathisia	ASN vs. OLZ	1	89	–0.21	–2.00	1.58	0.818	–
	CAR vs. RIS	1	460	0.15	–0.18	0.49	0.361	–
	CLO vs. OLZ	1	58	0.44	–1.26	2.14	0.614	–
	CLO vs. QTP	1	54	–0.97	–2.03	0.08	0.071	–
	CLO vs. RIS	1	54	0.30	–1.41	2.00	0.735	–
	LUR vs. RIS	1	608	0.13	–0.04	0.30	0.131	–
	OLZ vs. QTP	2	201	–0.46	–1.66	0.75	0.459	51.2
	OLZ vs. RIS	3	548	–0.08	–0.32	0.17	0.552	17.2
	OLZ vs. ZIP	2	725	–0.11	–0.28	0.05	0.184	0.0
	QTP vs. RIS	3	1277	0.16	–0.56	0.89	0.657	65.4
	QTP vs. ZIP	1	190	0.26	–0.42	0.93	0.458	–
	RIS vs. ZIP	1	193	–0.17	–0.97	0.64	0.683	–
Dyskinesia	AMI vs. OLZ	1	356	–0.11	–0.32	0.09	0.281	–
	AMI vs. RIS	1	310	0.02	–0.21	0.24	0.886	–
	ASN vs. OLZ	1	89	–1.46	–3.25	0.33	0.109	–
	CLO vs. OLZ	2	88	–0.21	–0.71	0.29	0.416	0.0
	CLO vs. QTP	1	44	0.47	–0.76	1.69	0.456	–
	CLO vs. RIS	1	45	1.01	–0.61	2.64	0.222	–
	OLZ vs. QTP	3	234	–0.35	–0.76	0.07	0.099	0.0
	OLZ vs. RIS	7	698	–0.02	–0.19	0.15	0.790	0.0
	OLZ vs. ZIP	2	701	–0.03	–0.19	0.13	0.726	0.0
	QTP vs. RIS	4	1,301	0.23	–0.28	0.74	0.375	58.8
	QTP vs. ZIP	1	165	0.52	0.05	0.99	0.030	–
	RIS vs. ZIP	1	156	0.10	–0.44	0.65	0.709	–
Parkinsonism	AMI vs. OLZ	2	562	0.26	–0.34	0.86	0.399	77.6
	AMI vs. QTP	1	179	0.30	–0.18	0.79	0.219	–
	AMI vs. RIS	1	310	0.07	–0.15	0.29	0.539	–
	AMI vs. ZIP	1	162	0.03	–0.43	0.50	0.887	–
	APZ vs. BLO	1	44	–0.41	–1.74	0.92	0.546	–
	APZ vs. OLZ	3	1,483	0.06	–0.27	0.38	0.737	76.5
	APZ vs. QTP	2	497	–0.10	–0.45	0.25	0.585	26.6
	APZ vs. ZIP	1	124	–0.07	–0.57	0.43	0.776	–
	ASN vs. OLZ	2	529	0.08	–0.90	1.06	0.867	16.0
	CAR vs. RIS	1	460	–0.23	–0.61	0.15	0.233	–
	CLO vs. OLZ	3	201	0.13	–0.18	0.45	0.402	0.0
	CLO vs. QTP	1	53	–0.75	–1.90	0.40	0.200	–
	CLO vs. RIS	1	54	0.30	–1.41	2.00	0.735	–
	LUR vs. RIS	1	621	–0.19	–0.46	0.08	0.169	–
	OLZ vs. QTP	5	1,126	–0.08	–0.51	0.36	0.725	51.7
	OLZ vs. RIS	9	1,934	–0.28	–0.44	–0.12	0.001	28.3
	OLZ vs. ZIP	5	1,808	–0.10	–0.23	0.03	0.129	0.0
	QTP vs. RIS	4	1,953	–0.26	–0.60	0.08	0.133	60.5
	QTP vs. ZIP	4	971	–0.19	–0.55	0.18	0.323	44.1
	RIS vs. ZIP	2	725	0.40	–0.23	1.03	0.214	66.6
Body weight gain	AMI vs. OLZ	3	742	–0.40	–0.54	–0.25	<0.001	0.0
	AMI vs. QTP	1	127	–0.06	–0.41	0.29	0.749	–
	AMI vs. RIS	1	195	–0.46	–0.83	–0.10	0.013	–
	AMI vs. ZIP	1	115	0.36	–0.02	0.74	0.066	–
	APZ vs. OLZ	5	1,413	–0.63	–0.81	–0.44	<0.001	31.7
	APZ vs. PAL	1	134	0.37	0.03	0.71	0.034	–
	APZ vs. QTP	2	501	–0.06	–0.47	0.35	0.774	53.5
	APZ vs. ZIP	2	264	0.63	–0.07	1.32	0.077	82.3
	APZ vs. BLO	1	44	0.09	–0.50	0.68	0.770	–
	ASN vs. OLZ	4	1,447	–0.39	–0.86	0.08	0.107	88.0
	CAR vs. RIS	1	431	–0.29	–0.48	–0.10	0.003	–
	CLO vs. OLZ	4	1,167	–0.33	–0.80	0.13	0.161	83.0
	CLO vs. QTP	1	54	0.02	–0.61	0.64	0.957	–
	CLO vs. RIS	3	96	–0.32	–0.78	0.14	0.172	0.0
	LUR vs. QTP	1	111	–0.13	–0.54	0.28	0.526	–
	LUR vs. RIS	1	621	–0.48	–0.65	–0.31	<0.001	–
	OLZ vs. PAL	1	449	0.49	0.31	0.68	<0.001	–
	OLZ vs. QTP	8	1,592	0.42	0.21	0.62	<0.001	69.1
	OLZ vs. RIS	11	1,646	0.37	0.19	0.55	<0.001	58.5
	OLZ vs. ZIP	6	1,509	0.74	0.62	0.85	<0.001	9.6
	PAL vs. ZIP	1	132	0.62	0.27	0.97	0.001	–
	QTP vs. RIS	8	2,813	0.01	–0.06	0.09	0.701	0.0
	QTP vs. ZIP	4	871	0.24	0.10	0.38	0.001	0.0
	RIS vs. SER	1	9,809	–0.61	–2.37	1.16	0.501	–
	RIS vs. ZIP	3	800	0.22	0.07	0.37	0.003	0.0
Prolactin increase	AMI vs. OLZ	1	105	0.63	0.24	1.03	0.002	–
	AMI vs. QTP	1	84	0.62	0.18	1.07	0.006	–
	AMI vs. ZIP	1	71	1.05	0.53	1.57	<0.001	–
	APZ vs. OLZ	4	1,686	–1.09	–1.63	–0.54	<0.001	84.4
	APZ vs. QTP	1	382	–0.23	–1.83	1.38	0.783	–
	ASN vs. OLZ	1	89	0.07	–0.47	0.61	0.804	–
	CLO vs. OLZ	1	55	–0.29	–0.87	0.30	0.333	–
	CLO vs. QTP	1	52	0.39	–0.24	1.02	0.229	–
	CLO vs. RIS	1	50	–1.62	–2.36	–0.88	<0.001	–
	LUR vs. RIS	1	554	–0.56	–0.74	–0.38	<0.001	–
	OLZ vs. QTP	6	996	0.13	0.01	0.26	0.040	0.0
	OLZ vs. RIS	7	1,225	–1.05	–1.23	–0.87	<0.001	40.7
	OLZ vs. ZIP	5	1,510	0.06	–0.16	0.27	0.596	73.1
	QTP vs. RIS	8	2,131	–1.24	–1.59	–0.90	<0.001	84.9
	QTP vs. ZIP	3	659	0.03	–0.41	0.47	0.890	82.9
	RIS vs. SER	1	9,809	0.00	–0.88	0.88	1.000	–
	RIS vs. ZIP	2	596	0.93	0.75	1.10	<0.001	0.0
Sedation and/or somnolence	AMI vs. OLZ	1	377	0.99	0.46	2.16	0.989	–
	AMI vs. RIS	1	310	0.69	0.29	1.65	0.407	–
	APZ vs. BLO	1	44	0.50	0.05	5.12	0.559	–
	APZ vs. OLZ	5	1,802	0.64	0.38	1.09	0.099	68.0
	APZ vs. QTP	1	119	1.39	0.60	3.24	0.442	–
	APZ vs. ZIP	1	124	1.34	0.60	3.00	0.479	–
	ASN vs. OLZ	3	1,038	0.89	0.66	1.22	0.477	0.0
	CAR vs. RIS	1	460	0.69	0.30	1.59	0.385	–
	CLO vs. OLZ	1	956	1.86	1.54	2.23	<0.001	–
	CLO vs. RIS	1	14	5.00	0.77	32.57	0.092	–
	LUR vs. RIS	1	621	0.76	0.52	1.12	0.166	–
	OLZ vs. PAL	1	459	2.85	1.29	6.31	0.010	–
	OLZ vs. QTP	4	1,220	0.95	0.83	1.10	0.531	0.0
	OLZ vs. RIS	7	1,656	1.14	0.99	1.32	0.064	0.0
	OLZ vs. ZIP	2	766	1.78	0.84	3.75	0.130	79.5
	QTP vs. RIS	6	3,095	1.46	1.09	1.96	0.010	78.1
	QTP vs. ZIP	3	861	1.49	0.89	2.48	0.129	56.7
	RIS vs. ZIP	3	906	1.35	0.94	1.95	0.104	41.4

Significant (p<0.05) results are in bold prints. RR – risk ratio, SMD – standardized mean difference, AMI – amisulpride, APZ – aripiprazole, ASN – asenapine, BLO – blonanserin, CAR – cariprazine, CLO – clozapine, LUR – lurasidone, OLZ – olanzapine, PAL – paliperidone, QTP – quetiapine, RIS – risperidone, SER – sertindole, ZIP – ziprasidone. Effect sizes for sedation and/or somnolence are expressed in RR, others in SMD. SMD <0 and RR<1 indicate superiority of the first medication.

Prolactin increase was meta‐analyzed in 16 comparisons based on 21 studies. Clozapine, lurasidone, olanzapine, quetiapine and ziprasidone were superior to risperidone; aripiprazole and quetiapine were superior to olanzapine; olanzapine, quetiapine and ziprasidone were superior to amisulpride (Table 2).

Parkinsonism was meta‐analyzed in 20 comparisons based on 28 studies: olanzapine was superior to risperidone. Dyskinesia was meta‐analyzed in 11 comparisons based on 13 studies: ziprasidone was superior to quetiapine. Akathisia was meta‐analyzed in 11 comparisons based on 9 studies: no significant differences emerged. Sedation and/or somnolence were meta‐analyzed in 17 comparisons based on 27 studies: olanzapine and paliperidone were superior to clozapine, and risperidone was superior to quetiapine.

Subgroup analyses for primary outcome

In subgroup analyses, the significance of the primary results was altered in 49/267 (18.4%) analyses, but most subgroups were very small both in number of studies and patients. Comparative effectiveness patterns were mostly consistent in high‐quality studies and double‐blind trials.

Regarding industry sponsorship, results showing a specific drug's inferiority were neutralized when three of 43 medication‐specific manufacturer‐sponsored studies were included. In contrast, one outcome showing superiority of olanzapine was neutralized when one manufacturer‐funded study was included.

Regarding blinding, some results changed when we restricted the analyses to open label or blinded studies. Restricting the analyses to only blinded studies, 5/39 results that showed statistical significance became non‐significant. Restricting the analyses to only open label studies, 1/39 non‐significant results became statistically significant.

None of the other potential effect‐moderators addressed in subgroup analyses revealed a clear pattern of effect. There were no subgroup analyses in which the direction of the results was reversed.

Publication bias

Publication bias for all‐cause discontinuation was assessed by funnel plot. In nine of eleven comparisons with ≥⃒3 studies, the funnel plot was asymmetrical. Subsequently, we applied the trim‐and‐fill method to adjust for potential publication bias, and found that the effect sizes were similar after adjustment, and that the significance for RRs did not change, except for two comparisons. Quetiapine was not different in observed values but became inferior to risperidone in adjusted values (original RR=1.07, 95% CI: 0.98‐1.18; adjusted RR=1.11, 95% CI: 1.00‐1.24). Quetiapine was significantly inferior in observed values, but became not different from ziprasidone in adjusted values (original RR=1.12, 95% CI: 1.01‐1.25; adjusted RR=1.08, 95% CI: 0.98‐1.19).

DISCUSSION

In this first comprehensive meta‐analysis of comparative effectiveness, efficacy and tolerability of SGAs in the long‐term treatment of schizophrenia, including 59 studies and 45,787 participants, no consistent superiority of any single antipsychotic across multiple outcome domains was observed.

Regarding all‐cause discontinuation, clozapine, olanzapine and risperidone were superior to several other SGAs, whereas quetiapine was inferior to several other SGAs. Regarding psychopathology, clozapine and olanzapine were superior to several other SGAs, while again quetiapine as well as ziprasidone were inferior to several other SGAs. Regarding functioning, QOL and remission, data were sparse.

Regarding intolerability‐related discontinuation, risperidone was superior and clozapine was inferior to several other SGAs. However, it should be kept in mind that discontinuation due to adverse events often includes inefficacy‐related adverse events in modern trials and, therefore, this outcome does not purely reflect tolerability.

When broken down into individual adverse events, superiority/inferiority patterns became clearer in some domains. For example, olanzapine was associated with more body weight gain than all other non‐clozapine SGAs, whereas ziprasidone was less so than other SGAs; and amisulpride and risperidone raised serum prolactin level more than other SGAs. Furthermore, sedation and/or somnolence were more common during long‐term treatment with clozapine and quetiapine.

We focused on head‐to‐head comparisons for the current meta‐analysis. The relative lack of direct head‐to‐head maintenance comparisons may raise interest in conducting a network meta‐analysis. However, while such methodology using indirect comparisons can create rankings, the very lack of so many comparisons and the heterogeneity of the studies conducted in different populations and over several decades are likely to introduce relevant biases that are not present in meta‐analyses of direct head‐to‐head trials9.

In fact, comparing our results with those from Zhao et al85, who conducted a network meta‐analysis of relapse prevention studies in stable patients with schizophrenia that also included first‐generation and long‐acting injectable antipsychotics, some differences emerge. For example, for relapse prevention, the only significant result involving an SGA was olanzapine's superiority over chlorpromazine and haloperidol, whereas we found olanzapine to be superior to risperidone (although based on one trial only). Furthermore, regarding all‐cause discontinuation, we observed a significant superiority of olanzapine over aripiprazole, paliperidone, quetiapine, risperidone and ziprasidone in direct comparisons, while Zhao et al, including indirect comparisons, found olanzapine only superior to aripiprazole. Thus, we believe that restricting the meta‐analysis exclusively to randomized head‐to‐head comparisons yields more precise results.

What are the implications of our findings for the choice of SGA in the long‐term treatment of schizophrenia? First, we must consider the magnitude of the effect sizes for all‐cause discontinuation. Since these ranged from medium to large, we believe that they are clinically meaningful, especially during the important maintenance treatment phase2, 7, 86, 87. The results regarding psychopathology roughly matched the findings for all‐cause discontinuation, in that clozapine and olanzapine were superior to several other SGAs, whereas quetiapine seemed inferior, this time together with ziprasidone. However, the findings of divergent adverse effect outcomes, with particular disadvantages for clozapine, olanzapine and risperidone, highlight the fact that it is crucial to not view efficacy and effectiveness in isolation of tolerability. For example, clozapine and olanzapine are among the medications with some of the most problematic adverse effects, including weight gain and metabolic abnormalities10, 88 as well as, in the case of clozapine, blood dyscrasias89. Given such inconsistent results in the different outcome categories, the importance of a balanced medication choice based on each patient's own situation should be emphasized.

Regarding the comparative effectiveness of clozapine and olanzapine, we found similar results in the maintenance treatment of schizophrenia. Even in studies targeting treatment‐refractory patients, the effect sizes were similar. Since a network meta‐analysis of short‐term trials in refractory patients did not find superiority of clozapine vs. olanzapine, risperidone and ziprasidone90, which may have been driven by use of suboptimal clozapine doses or inclusion of non‐refractory patients, further high‐quality, short‐ and long‐term, head‐to‐head trials of clozapine vs. other SGAs are needed.

Several limitations of this study need to be considered. Most comparisons relied on relatively few head‐to‐head trials. As many as 139 of all 250 comparisons were based on one study only, but we only meta‐analyzed outcomes for which at least two head‐to‐head trials provided data. The number of patients per trial was also often small, and dose equivalencies used across studies might not have been balanced or consistent. Furthermore, the limited number of studies reduced the power of our exploratory subgroup analyses. Additionally, only six and eleven studies reported remission and relapse as an outcome, respectively. However, since psychopathology, treatment response and functioning can worsen with repeated relapse87, 91, information on comparative remission and relapse risk with individual antipsychotics is important.

The randomization point in the included studies differed, i.e., some studies randomized patients during the acute phase, and others during the maintenance phase. Moreover, some studies included exclusively treatment‐refractory patients, whereas some others included exclusively first‐episode patients. Relapse and remission definitions varied across studies. Moreover, two of the included studies had an enriched design, and two allowed switches after randomization, which could have affected the results. Such heterogeneity of the study design as well as patient populations introduces biases. However, we assessed the impact of patient and study design characteristics as potential moderators by conducting subgroup analyses.

Finally, although the effectiveness of long‐acting injectable antipsychotics (LAIs) in the long‐term treatment of schizophrenia is clearly important92, we excluded LAI studies, as this aspect has already been comprehensively meta‐analyzed13, 14, 93. Including LAIs in this meta‐analysis, which are not available for all SGAs, would have further increased the heterogeneity of samples and methods, the complexity of the analyses and the interpretation of the results.

In conclusion, results from this meta‐analysis suggest that there are some significant differences in the effectiveness, efficacy and tolerability among SGAs in the long‐term treatment of schizophrenia. Clozapine, olanzapine and risperidone seem to be superior to several other SGAs regarding all‐cause discontinuation, while quetiapine seems to be inferior. Regarding psychopathology scores, clozapine and olanzapine seem to be superior to several other SGAs, while quetiapine and ziprasidone seem to be less effective. Regarding discontinuation due to adverse events, only risperidone was superior and clozapine was inferior to several other SGAs.

Due to the limited number of head‐to‐head trials, the comparative effectiveness of some SGAs is unclear, and results need to be interpreted cautiously whenever they were based on few trials. Thus, a sufficiently larger database involving many SGAs and including detailed effectiveness and tolerability outcomes is desirable to further guide the evidence‐based long‐term treatment of patients with schizophrenia. In particular, identifying predictors of beneficial outcomes with specific antipsychotics would further enhance the ability to personalize treatments.