Efficacy and Safety of Individual Second-Generation vs First-Generation Antipsychotics in First Episode Psychosis: A Systematic Review and Meta-analysis

Abstract

Because early treatment choice is critical in first episode schizophrenia-spectrum disorders (FES), this meta-analysis compared efficacy and tolerability of individual second-generation antipsychotics (SGAs) with first-generation antipsychotics (FGAs) in FES. We conducted systematic literature search (until 12/31/2010) and meta-analysis of acute, randomized trials with ≥1 FGA vs. SGA comparison; patients in their first episode of psychosis and diagnosed with schizophrenia-spectrum disorders; available data for psychopathology change, treatment response, treatment discontinuation, adverse effects, or cognition. Across 13 trials (n=2,509), olanzapine (7 trials) and amisulpride (1 trial) outperformed FGAs (haloperidol: 9/13 trials) in 9/13 and 8/13 efficacy outcomes, respectively, risperidone (8 trials) in 4/13, quetiapine (1 trial) in 3/13, and clozapine (2 trials) and ziprasidone (1 trial) in 1/13, each. Compared to FGAs, EPS-related outcomes were less frequent with olanzapine, risperidone and clozapine, but weight gain was greater with clozapine, olanzapine and risperidone. Pooled SGAs were similar to FGAs regarding total psychopathology change, depression, treatment response, and metabolic changes. SGAs significantly outperformed FGAs regarding lower treatment discontinuation, irrespective of cause, negative symptoms, global cognition, and less EPS and akathisia, while SGAs increased weight more (p’s<0.05-0.01). Results were not affected by FGA dose or publication bias, but industry-sponsored studies favored SGAs more than federally funded studies. To summarize, in FES, olanzapine, amisulpride and, less so, risperidone and quetiapine showed superior efficacy, greater treatment persistence and less EPS than FGAs. However, weight increase with olanzapine, risperidone and clozapine and metabolic changes with olanzapine were greater. Additional FES studies including broader-based SGAs and FGAs are needed.

Introduction

Schizophrenia is a chronic, debilitating mental disorder(van Os and Kapur, 2009) with a life-time prevalence of 0.30–0.66%, increasing to 2.3% when including other psychotic disorders(Perala et al., 2007). Schizophrenia is associated with significant medical co-morbidity and mortality, with an average life-span shortening by 10–30 years(Goff et al., 2005). Illness onset typically occurs in late adolescence/young adulthood and life-time treatment is required to maintain/improve social functioning and prevent symptom relapse, which causes significant public health and economic burden. Successful first episode schizophrenia (FES) treatment is crucial to minimize social and vocational deterioration(Robinson et al., 2004).

Antipsychotics are the mainstay of schizophrenia treatment(Kane, 1999; van Os and Kapur, 2009). Typical or first-generation antipsychotics (FGAs) and atypical or second-generation antipsychotics (SGAs) are effective for positive symptoms. However, as a class, FGAs cause more extrapyramidal motor side effects (EPS) and tardive dyskinesia (TD) than SGAs, whereas SGAs generally cause more weight gain and cardiometabolic adverse effects(Leucht et al., 2009). Recent large randomized controlled trials (RCTs)(Jones et al., 2006; Lieberman et al., 2005) and a meta-analysis(Leucht et al., 2009) of FGA vs. SGA RCTs in chronic schizophrenia suggested no significant efficacy/effectiveness differences, or few non-clozapine SGA advantages with effect size differences of only 0.1–0.3. However, RCTs in chronic schizophrenia have limitations, including confounding effects of prior medication use, possible over-representation of only partially responsive or treatment non-adherent patients, and a tendency of recruiting patients with low pre-study functioning levels, potentially reducing overall treatment responsiveness. Conversely, some trials might enroll more responsive patients consenting to RCTs.

Compared to chronic patients, FES patients generally have higher response rates(Robinson et al., 1999; Tohen et al., 2000), require lower antipsychotic doses, and are more sensitive to adverse effects(Robinson et al., 2005). Hence, FES studies offer the unique opportunity to examine antipsychotic therapeutic and adverse effects in more representative patients in whom important initial treatment effects take place. Since in FES patients, therapeutic and adverse response patterns are largely unknown, treatment recommendations must be based upon research findings, rather than past treatment history. RCTs in FES have flourished in the past decade. However, there is still a debate regarding the comparative efficacy and effectiveness of FGAs vs. SGAs, because studies have yielded discrepant results. A recent meta-analysis of FES trials(Crossley et al., 2010) showed no difference in efficacy between FGAs and SGAs. However, this meta-analysis included one non-randomized trial(Saddichha et al., 2007), failed to compare individual SGAs, investigated only two adverse effects and ignored important outcomes, including positive and negative symptoms, depression, cognitive functioning, specific-cause discontinuation, long-term remission and relapse.

Therefore, we conducted a meta-analysis comparing the efficacy, effectiveness, and adverse effects of primarily individual and secondarily pooled SGAs vs. FGAs in FES patients.

Methods

Search

We conducted an electronic PubMed and Web of Science search (until 12/31/2010) for RCTs comparing FGAs vs. SGAs in first-episode patients with schizophrenia-spectrum disorders. The following key words were used: first episode, first episode psychosis, first episode schizophrenia, early psychosis, early schizophrenia, recent onset, recent onset psychosis, recent onset schizophrenia, antipsychotic(s), typical antipsychotic(s), conventional antipsychotic(s), neuroleptic(s), atypical antipsychotic(s), first-generation and second-generation antipsychotic(s). We also screened reference lists from identified papers and reviews to identify additional studies. To find unpublished studies, we searched published meeting abstracts that were likely to contain relevant studies and contacted manufacturers of SGAs. Inclusion criteria were: 1) RCT with ≥1 FGA vs. SGA comparison; 2) patients in their first episode of non-affective psychosis; 3) diagnosis of schizophrenia, schizoaffective disorder, schizophreniform disorder, brief psychotic disorder, or psychosis NOS; 4) Data available for any of the following outcomes: efficacy, treatment discontinuation, weight gain, metabolic parameters, or cognition; 5) Acute treatment study. Maintenance studies were excluded, although long-term data from acute RCTs were included.

Data extraction and outcome variables

Data were independently extracted by two authors (JPZ, JAG); any disagreement was resolved. For missing information, first and/or last study authors were contacted requesting additional/unpublished data. Short-term outcomes at 3 months (≤13 weeks) or closest follow-up time point, and long-term outcomes at about 6–24 months were analyzed separately. Primary variables included the following three short-term outcomes: 1) all-cause discontinuation; 2) symptom reduction: changes in the Positive and Negative Syndrome Scale (PANSS)(Kay et al., 1987) or Brief Psychiatric Rating Scale (BPRS)(Overall and Gorham, 1962) total score from baseline; and 3) response rate: percentage of patients achieving “clinical responder” status at follow-up. We chose ≥50% reduction in PANSS or BPRS total scores from baseline to follow-up as the preferred definition of clinical response, because this represents a clinically significant response(Leucht et al., 2007). When this outcome was not available, study-defined response was used.

Secondary outcomes included last-observation-carried-forward (LOCF) change from baseline in: 1) positive symptoms, measured by PANSS, Scale for the Assessment of Positive Symptoms (SAPS)(Andreason, 1984), or BPRS; 2) negative symptoms, measured by PANSS or Scale for the Assessment of Negative Symptoms (SANS)(Andreason, 1983); 3) depressive symptoms, measured by a validated depression scale; 4) Clinical Global Impressions (CGI)-Severity or Improvement(Guy, 1976) score; 5) cognitive functioning, measured by a composite score extracted from multiple neuropsychological tests. Categorical outcomes included: 6) discontinuation due to inefficacy; 7) discontinuation due to intolerability; 8) discontinuation due to non-adherence/patient choice; 9) study-defined long-term (≥1 year) remission; and 10) long-term study-defined relapse and/or rehospitalization after achieving response. We also analyzed common side effects: 11) EPS; 12) akathisia; 13) use of anticholinergic drugs, benzodiazepines and beta-blockers, commonly used to manage antipsychotic side effects; 14) weight gain, as categorical and continuous variables; and 15) metabolic changes (total cholesterol, triglycerides, glucose).

The following study and patient characteristics were examined as moderators in meta-regression analyses: publication year, number of sites, blinding status, study duration, location (US, Europe, Asia, etc.), primary follow-up time point, study sponsorship (government-sponsored vs. industry-sponsored), sample size, mean age, percentage of males, inpatient vs. outpatient status at randomization, age at illness onset, duration of psychosis, diagnosis, percentage of antipsychotic-naïve patients (i.e., <2 weeks of lifetime exposure), baseline PANSS/BPRS total score, baseline CGI-S score, and mean or modal antipsychotic dose (converted to haloperidol equivalents)(Andreasen et al., 2010; Lehman et al., 2004; Woods, 2003).

Statistical analysis

Outcomes were analyzed separately using Comprehensive Meta-Analysis software version 2 (Biostat, Eaglewood, NJ). For continuous outcomes, Hedge’s g was used as the effect size measure, representing standardized groups differences. For dichotomous outcomes, risk ratio (RR) was used as the effect size measure with FGAs as the reference group. The primary focus was individual drug comparisons. Pooled SGA-FGA comparisons were conducted to allow for meaningful moderator analyses across two antipsychotic classes that were contrasted in the literature before. Pooled effect sizes were computed with a random effects model(Borenstein et al., 2009). In each meta-analysis, a study was included only once to avoid redundancy. If a study had multiple eligible comparisons, the average effect size was used in the FGA-SGA comparison. For individual antipsychotic comparisons, the FGA comparator was entered separately for each SGA.

Study heterogeneity was assessed using Q and I² statistics, with I² <25% representing low, ~50% moderate, and >75% representing high heterogeneity(Borenstein et al., 2009). Whenever heterogeneity was present, moderator and meta-regression analyses were conducted to explore the effects of the moderators described above. Sensitivity analyses were conducted to assess potential influences of any one single study on the pooled effect size. Within each meta-analysis, included studies were removed one at a time to check for significant alterations to pooled effect sizes and associated p-values. Moreover, we assessed for an influence of sponsorship bias (government vs industry) and of age group (in- or excluding the only trial conducted in adolescents, i.e., TEOSS(Sikich et al., 2008)).

Publication bias was assessed with the funnel plot, Egger’s regression test(Egger et al., 1997), and the “Trim and Fill” method(Duval and Tweedie, 2000), an iterative procedure to assess whether small, extreme included studies and/or potentially unincluded studies biased the true effect size estimate.

Results

Search and Study Characteristics

Of 776 hits, 22 papers reporting on 13 independent cohorts(Crespo-Facorro et al., 2006; de Haan et al., 2003; Emsley, 1999; Fagerlund et al., 2004; Kahn et al., 2008; Lee et al., 2007; Lieberman et al., 2003a; Lieberman et al., 2003b; Moller et al., 2008; Sanger et al., 1999; Schooler et al., 2005; Sikich et al., 2008; Wu et al., 2006) and 20 comparisons across 12 individual SGA-FGA pairs (n=2,509) were identified (Fig S1) and meta-analyzed. Three trials included in a prior meta-analysis on this topic(Crossley et al., 2010) were excluded due to non-randomized design(Saddichha et al., 2008), skewed data(Brewer et al., 2007) and absence of meta-analyzable data(Bustillo et al., 2008). The mean study duration was 32.1±36.6 (range: 6–104) weeks, mean age was 27.1 years, 69.5% were male, and the mean baseline PANSS total score was 85.8±8.8 (Table S1). Five studies each compared olanzapine vs. haloperidol(Crespo-Facorro et al., 2006; de Haan et al., 2003; Kahn et al., 2008; Lieberman et al., 2003b; Sanger et al., 1999), risperidone vs. haloperidol(Crespo-Facorro et al., 2006; Emsley, 1999; Lee et al., 2007; Moller et al., 2008; Schooler et al., 2005). The other 10 SGA-FGA comparisons were examined in one study each.

Nine studies (69%) were double-blinded, four were open label(Crespo-Facorro et al., 2006; Fagerlund et al., 2004; Kahn et al., 2008; Wu et al., 2006). Eight studies (62%) were short-term (≤13 weeks), five lasted ≥1 year. The most commonly studied FGA was haloperidol (9 studies, 69%); and one study each included chlorpromazine(Lieberman et al., 2003a), molindone(Sikich et al., 2008), sulpiride(Wu et al., 2006), or zuclopenthixol(Fagerlund et al., 2004). The most commonly studied SGAs were risperidone (N=8) and olanzapine (N=7), followed by clozapine (N=2), and quetiapine, ziprasidone, and amisulpride(Kahn et al., 2008), (N=1 each). The number of patients per trial ranged from 24–555. At baseline, most studies reported a diagnosis of schizophrenia in >50% of patients, except for 2 trials(Emsley, 1999; Schooler et al., 2005). The mean duration of psychosis ranged from 13–18.4 months, except for one study(Wu et al., 2006) where it was 2.6 months. The mean age of illness onset ranged from 23–27 years, except for the TEOSS study(Sikich et al., 2008), which, due to its design, had a mean age of onset of 11.1 years. Pharmaceutical companies sponsored 54% of the studies, with government agency sponsorship of the remaining trials. The proportion of antipsychotic-naïve subjects at baseline ranged from 8.3–100% (mean: 62.7%). Among 9 studies with data on response, 3 used a ≥50% cut-off as the responder definition, 3 used ≥40%, and 3 used either ≥30% or ≥20%.

Primary Outcomes

Short-term all-cause discontinuation

Olanzapine (RR=0.53, p=0.001, N=5, n=689) and risperidone (RR=0.79, p=0.03, N=4, n=1,146) caused less all-cause discontinuation than haloperidol (Table 1; Fig S2). Among single trial comparisons, only amisulpride was superior to haloperidol (RR=0.63, p<0.01). Pooled SGAs had significantly lower discontinuation rates than FGAs (RR=0.74, p<0.001, N=10, n=1,952). Pooled risk difference was −0.08 (CI=−0.14~−0.02, NNT=12, p<0.01). No moderating variables were found.

Table 1.

Pooled effect sizes (95% confidence interval in parenthesis) of short-term primary outcome variables in comparison of SGAs and FGAs.

SGA	FGA	# studies	n	All-cause Discontinuation Rate (RR)	Symptom Reduction (Hedges’ g)	Response Rate (RR)
Olanzapine	Haloperidol	5	689	0.53 (0.37~0.77)**	0.26 (0.05~0.47)*	1.29 (1.05~1.58)*b
Olanzapine	Molindone	1	75	1.37 (0.82~2.29)	−0.02 (−0.47~0.43)	0.68 (0.39~1.19)
Risperidone	Haloperidol	5	1146	0.79 (0.63~0.97)*a	−0.04 (−0.19~0.11)	1.04 (0.90~1.20)c
Risperidone	Molindone	1	81	0.85 (0.46~1.54)	−0.15 (−0.58~0.28)	0.92 (0.58~1.45)
Risperidone	Zuclopenthixol	1	25	NR	0.50 (−0.29~1.28)	1.34 (0.30~5.96)
Clozapine	Chlorpromazine	1	160	0.91 (0.37~2.25)	0.15 (−0.16~0.46)	1.03 (0.88~1.20)
Amisulpride	Haloperidol	1	207	0.63 (0.47~0.85)**	0.40 (0.13~0.68)**	1.56 (1.13~2.15)**
Quetiapine	Haloperidol	1	207	0.81 (0.63~1.05)	0.26 (−0.02~0.53)	1.30 (0.92~1.84)
Ziprasidone	Haloperidol	1	185	0.89 (0.68~1.15)	0.22 (−0.07~0.51)	1.11 (0.76~1.64)
Pooled SGAs	Pooled FGAs	12	1952	0.74 (0.62~0.87)**	0.11 (−0.02~0.24)	1.13 (0.99~1.27)

Note: # studies and n are the largest possible for each comparison pair, but # of studies and n for each outcome may vary.

^*p < .05;

^**p < .01.

^aN=4, n=1146;

^bN=4, n=652;

^cN=4, n=1089.

Short-term total symptom reduction

Only olanzapine (ES=0.26, p=0.01, N=5, n=676) and amisulpride (ES=0.40, p<0.01, N=1) were superior to haloperidol (Table 1, Fig S3). Pooled SGAs showed only trend-level superiority compared to FGAs (ES=0.11, p=0.09, N=12, n=1,951). Study sponsorship significantly moderated the effect (Q=6.68, p=0.01). Government-sponsored studies trended in favor of FGAs (ES=−0.10, p>0.10, N=5, n=525), while industry-sponsored studies significantly favored SGAs (ES=0.19, p=0.007, N=7, n=1,426; Fig S4). The EUFEST trial(Kahn et al., 2008) was investigator-designed and then sought industry funding, but excluding EUFEST did not significantly alter the results. No other significant moderator emerged.

Short-term response

Only olanzapine (RR=1.29, p=0.02, N=4, n=652) and amisulpride were superior to haloperidol (RR=1.56, p<0.01, N=1, Table 1, Fig S5). Pooled SGAs were marginally better than FGAs (RR=1.13, p=0.06, N=9, n=1,724). Excluding the TEOSS study(Sikich et al., 2008) conducted in adolescents resulted in a significant RR favoring SGAs (RR=1.14, CI=1.02~1.29, p=0.026). Again, industry-sponsored studies showed higher response for SGAs (RR=1.23, CI=1.06~1.42, p=0.005), while non-industry-sponsored studies did not (RR=0.97, CI=0.81~1.16, p=0.75), with significant sponsorship effect (Q=3.97, p=0.046) (Fig S6). There was no other significant moderator.

Long-term primary outcomes

Individual SGAs vs. FGAs were similar regarding long-term all-cause discontinuation, but pooled results still trended favoring SGAs (RR=0.78, CI=0.60~1.01, p=0.06, N=5, n=1,133). Neither long-term symptom reduction (ES=0.01, CI=−0.11~0.14, p>0.10, N=4, n=953) nor long-term response (RR=1.05, CI=0.95~1.16, p>0.10, N=5, n=1,133) differed in individual or pooled SGA-FGA comparisons.

Secondary Outcomes

Positive symptoms

No significant difference emerged between SGAs and FGAs, except that amisulpride outperformed haloperidol (Table 2). Olanzapine had trend-level superiority (ES=0.26, p=0.08, N=4, n=653). Pooled SGA-FGA comparison also revealed no significant difference.

Table 2.

Pooled effect sizes (95% confidence interval in parenthesis) of short-term secondary outcome variables in comparison of SGAs and FGAs.

SGA	FGA	# studies	n	Positive Symptoms (Hedges’ g)	Negative Symptoms (Hedges’ g)	Depression (Hedges’ g)	CGI-S (Hedges’ g)	Cognitive Function (Hedges’ g)
Olanzapine	Haloperidol	4	653	0.26 (−0.03~0.54)	0.30 (0.15~0.46)**	0.28 (0.11~0.44)**	0.15 (−0.14~0.44)	0.27 (0.06~0.49)*d
Olanzapine	Molindone	1	75	0.02 (−0.43~0.47)	−0.07 (−0.52~0.38)	NR	NR	NR
Risperidone	Haloperidol	5	1136	−0.05 (−0.17~0.07)	0.03 (−0.09~0.14)	0.00 (−0.13~0.14)a	−0.14 (−0.34~0.05)b	0.23 (0.04~0.43)*c
Risperidone	Molindone	1	81	−0.06 (−0.49~0.37)	−0.10 (−0.53~0.34)	NR	NR	NR
Risperidone	Zuclopenthixol	1	25	0.35 (−0.43~1.13)	0.29 (−0.49~1.06)	NR	NR	0.63 (−0.17~1.42)
Clozapine	Chlorpromazine	1	160	0.15 (−0.16~0.46)	0.41 (0.10~0.72)**	NR	0.24 (−0.07~0.55)	NR
Amisulpride	Haloperidol	1	207	0.54 (0.27~0.82)**	0.24 (−0.03~0.52)	0.32 (0.05~0.59)*	0.38 (0.10~0.65)**	−0.01 (−0.39~0.36)
Quetiapine	Haloperidol	1	207	0.22 (−0.06~0.49)	0.32 (0.05~0.59)*	0.23 (−0.04~0.50)	0.09 (−0.18~0.36)	0.20 (−0.17~0.57)
Ziprasidone	Haloperidol	1	185	0.27 (−0.03~0.56)	0.25 (−0.04~0.54)	0.06 (−0.23~0.34)	0.18 (−0.11~0.47)	0.11 (−0.28~0.51)
Pooled SGAs	Pooled FGAs	11	1932	0.09 (−0.03~0.21)	0.16 (0.04~0.28)**	0.12 (−0.00~0.24)	0.10 (−0.09~0.29)	0.25 (0.10~0.40)**

Note: # studies and n are the largest possible for each comparison pair, but # of studies and n for each outcome may vary.

^*p < .05;

^**p < .01.

^aN=2, n=817.

^bN=2, n=406.

^cN=2, n=396.

^dN=3, n=340.

Negative Symptoms

Several SGAs outperformed their FGA comparators (Table 2), including olanzapine vs. haloperidol (ES=0.30, p<0.001, N=4, n=653), quetiapine vs. haloperidol (ES=0.32, p<0.05), and clozapine vs. chlorpromazine (ES=0.41, p<0.01). Pooled SGAs outperformed FGAs (ES=0.16, p=0.009, N=11, n=1,931). However, sponsorship bias may exist, as industry-sponsored studies showed results favoring SGAs (N=7, n=1,430, p=0.001), while independently funded studies did not (N=4, n=501, p=0.72), without significant group difference (p>0.10, Fig S7).

Depression

Only olanzapine (ES=0.28, p=0.001, N=4, n=565) and amisulpride (ES=0.32, p<0.05) outperformed haloperidol (Table 2). Pooled SGAs showed trend-level superiority over FGAs (ES=0.12, p=0.06, N=6, n=1,376). However, again, industry-sponsored studies more likely favored SGAs.

Clinical Global Impressions-Severity

Only amisulpride was superior to haloperidol (ES=0.38, p<0.01). Pooled SGAs and FGAs were also similar (N=6, n=1038, Table 2). Again, industry-sponsored studies were more likely to favor SGAs.

Global cognition

Olanzapine (ES=0.27), risperidone (ES=0.23) and pooled SGAs (ES=0.25, p=0.001, N=5, n=693) were superior to FGAs in improving cognitive composite scores at 3–6 months (Table 2, Fig S8).

Specific-cause discontinuation

Olanzapine (RR=0.38, p<0.001), amisulpride (RR=0.24, p<0.01) and pooled SGAs (RR=0.60, p=0.001, N=9, n=1,792) led to less discontinuation due to inefficacy than FGAs (Table 3). Olanzapine (RR=0.29, p=0.001, Figure S40), risperidone (RR=0.50, p=0.02), quetiapine (RR=0.13, p<0.01) and pooled SGAs (RR=0.46, p<.001, N=8, n=1,768) led to less discontinuation due to intolerability than the FGA comparator. Only quetiapine led to less discontinuation due to patient decision/non-adherence than haloperidol (RR=0.15, p<0.05), without differences in pooled SGA-FGA comparisons.

Table 3.

Pooled effect sizes (95% confidence interval in parenthesis) of short-term discontinuation rate due to specific reasons and long-term remission and relapse rates in comparison of SGAs and FGAs.

SGA	FGA	# studies	n	Due to inefficacy (RR)	Due to intolerability (RR)	Due to patient choice/non-adherence (RR)	Long-term Remission Rate (RR)	Long-term Relapse Rate (RR)
Olanzapine	Haloperidol	5	689	0.38 (0.25~0.59)**	0.29 (0.14~ 0.59)**a	0.64 (0.16~2.57)b	1.57 (1.06~2.32)*d	1.00 (0.77~1.31)f
Olanzapine	Molindone	1	75	0.69 (0.18~2.67)	0.86 (0.33~2.23)	4.00 (0.89~18.01)	NR	NR
Risperidone	Haloperidol	4	1146	0.88 (0.51~1.53)	0.50 (0.28~0.89)*	0.78 (0.53~1.15)c	1.40 (0.79~2.46)e	0.77 (0.64~0.95)*g
Risperidone	Molindone	1	81	0.78 (0.23~2.70)	0.61 (0.22~1.71)	1.95 (0.38~10.06)	NR	NR
Clozapine	Chlorpromazine	1	160	NR	NR	NR	1.03 (0.88~1.20)	1.20 (0.38~3.77)
Amisulpride	Haloperidol	1	207	0.24 (0.12~0.49)**	0.47 (0.20~1.08)	0.69 (0.31~1.52)	2.35 (1.45~3.83)**	0.73 (0.37~1.41)
Quetiapine	Haloperidol	1	207	1.06 (0.73~1.55)	0.13 (0.03~0.55)**	0.15 (0.04~0.65)*	1.41 (0.82~2.44)	1.05 (0.54~2.01)
Ziprasidone	Haloperidol	1	185	0.67 (0.41~1.09)	0.47 (0.19 ~1.16)	0.77 (0.34~1.75)	1.65 (0.95~2.86)	0.32 (0.11~0.89)*
Pooled SGAs	Pooled FGAs	9	1792	0.60 (0.43~0.82)**	0.46 (0.31~0.68)**	0.84 (0.57~1.24)	1.26 (0.99~1.60)	0.84 (0.72~0.99)*

Note: # studies and n are the largest possible for each comparison pair, but # of studies and n for each outcome may vary.

^*p < .05;

^**p < .01.

RR = Relative Risk.

^aN=4, n=665.

^bN=4, n=606.

^cN=3, n=963.

^dN=3, n=582.

^eN=1, n=119.

^fN=3, n=394.

^gN=3, n=663.

Long-term remission and relapse rates

Regarding long-term remission, olanzapine (RR=1.57, p=0.03) and amisulpride (RR=2.35, p=0.001) were superior to haloperidol. Pooled SGAs (N=4, n=740) showed only trend-level advantage over FGAs (RR=1.26, p=0.06) (Table 3). However, risperidone (RR=0.77, p=0.01), ziprasidone (RR=0.32, p=0.03) and pooled SGAs (RR=0.84, p=0.04, N=6, n=1,092, Table 3), had lower relapse rates than FGAs.

Long-term discontinuation due to inefficacy was lower with olanzapine than haloperidol (RR=0.51, CI=0.27~0.95, p=0.04, N=3, n=582), but pooled SGA-FGA comparisons were non-significant (RR=0.66, CI=0.37~1.17, p=0.16, N=5, n=1295). Regarding long-term discontinuation due to intolerability, olanzapine outperformed haloperidol (RR=0.31, p<0.001) and SGAs outperformed FGAs (RR=0.49, CI=0.32~0.75, p=0.001, N=5, n=1295).

Adverse Events

EPS and akathisia-related outcomes

EPS was less frequent and severe in patients on several SGAs than their FGA comparators, including olanzapine (ES=−0.69, p<0.001, N=4, n=609) and risperidone (ES=−0.33, p<0.001, N=3, n=588) compared to haloperidol, and clozapine compared to chlorpromazine (ES=−0.72, p<0.01), although all patients on chlorpromazine received prophylactic benztropine (Table 4). Pooled SGAs produced significantly less EPS than FGAs (ES=−0.43, p<0.001, N=9, n=1,338). Two significant moderators emerged: More recent studies had smaller SGA-FGA differences in EPS (b=0.04, p=0.02). Conversely, higher patient age was associated with larger effect sizes (b=−0.04, p=0.006). At long-term follow-up, SGAs still produced less EPS (RR=0.42, CI=0.24~0.73, p=0.002, N=2, n=319).

Table 4.

Pooled effect sizes (95% confidence interval in parenthesis) of short-term common side effects and co-medications treating side effects in comparison of SGAs and FGAs. A negative effect size and an RR below 1 indicate worse side effects from FGAs.

SGA	FGA	EPS (Hedges’ g)	Akathisia (Hedges’ g)	Co-meds: Anticholinergerics (RR)	Co-meds: Benzodiazepines (RR)	Co-meds: Beta-blockers (RR)
Olanzapine	Haloperidol	−0.69 (−1.02~−0.35)** (N=4, n=609)	−0.61 (−0.79~−0.42)** (N=4, n=611)	0.21 (0.09~0.51)** (N=3, n=445)	0.83 (0.71~0.96)* (N=3, n=445)	0.11 (0.03~0.40)** (N=1, n=251)
Olanzapine	Molindone	0.14 (−0.31~0.59) (N=1, n=75)	−0.35 (−0.81~0.10) (N=1, n=75)	0.31 (0.13~0.76)**a (N=1, n=75)	0.51 (0.24~1.11) (N=1, n=75)	0.85 (0.25~2.92) (N=1, n=75)
Risperidone	Haloperidol	−0.33 (−0.51~−0.16)** (N=3, n=588)	−0.29 (−0.52~−0.06)* (N=2, n=406)	0.71 (0.45~1.11) (N=3, n=591)	0.88 (0.73~1.06) (N=2, n=408)	0.62 (0.22~1.69) (N=1, n=289)
Risperidone	Molindone	0.21 (−0.22~0.64) (N=1, n=81)	−0.28 (−0.71~0.16) (N=1, n=81)	0.76 (0.44~1.31)a (N=1, n=81)	1.05 (0.62~1.79) (N=1, n=81)	0.54 (0.14~2.13) (N=1, n=81)
Risperidone	Zuclopenthixol	−0.72 (−1.52~0.08) (N=1, n=25)	−0.71 (−1.51~0.09) (N=1, n=25)	NR	NR	NR
Clozapine	Chlorpromazine	−0.72 (−1.04~−0.41)** (N=1, n=160)	NR	NRb	NR	NR
Amisulpride	Haloperidol	0.24 (−0.34~0.82) (N=1, n=163)	−0.46 (−1.00~0.08) (N=1, n=163)	NR	NR	NR
Quetiapine	Haloperidol	0.21 (−0.37~0.80) (N=1, n=167)	−0.33 (−0.83~0.17) (N=1, n=167)	NR	NR	NR
Ziprasidone	Haloperidol	0.28 (−0.33~0.89) (N=1, n=142)	−0.28 (−0.82~0.26) (N=1, n=142)	NR	NR	NR
Pooled SGAs	Pooled FGAs	−0.43 (−0.64~−0.22)** (N=9, n=1338)	−0.48 (−0.62~−0.34)** (N=7, n=998)	0.47 (0.29~0.77)** (N=6, n=999)	0.84 (0.75~0.95)** (N=5, n=816)	0.37 (0.12~1.12) (N=3, n=618)

Note:

^*p < .05;

^**p < .01.

RR = Relative Risk.

^aIn the TEOSS study, all patients on molindone were also given benztropine 1mg daily by study design, but some patients received additional benztropine due to side effects. Calculation of ES was based on number of patients on molindone received additional benztropine.

^bIn Lieberman 2003 clozapine vs chlorpromazine study, all patients on chlorpromazine also took benztropine 2mg twice daily by study design.

Short-term akathisia was also less likely with olanzapine (ES=−0.61, p<0.05) and risperidone (ES=−0.29, p<0.05), and with pooled SGAs vs. FGAs (ES=−0.48, p<0.001, N=7, n=998) (Table 4). At 1–2-year follow-up, akathisia was still less severe/prevalent with SGAs (ES=−0.33, CI=−0.48~−0.19, p<.001, N=4, n=930).

Compared to haloperidol, patients on olanzapine took less anticholinergics (RR=0.21, p<0.001, N=3, n=445), benzodiazepines (RR=0.83, p=0.02, N=3, n=445), and beta-blockers (RR=0.11, p<0.01, N=1, n=251) (Table 4). Olanzapine was also superior to molindone regarding less anti-cholinergic co-administration (RR=0.31, p<0.01), although all patients on molindone received prophylactic benztropine (Table 4). Pooled SGAs were associated with less anticholinergic (RR=0.47, p<0.01, N=6, n=999) and benzodiazepine use (RR=0.84, p<0.01, N=5, n=999) (Table 4). Moderator analyses revealed that in open-label studies more patients on FGAs took anticholinergics than in double-blind studies. Significantly less anticholinergic use with SGAs than FGAs was associated with smaller sample size, younger age, male sex, and longer follow-up. SGAs and FGAs did not differ regarding beta-blocker use.

At 1–2-year follow-up, SGAs were still associated with less anticholinergic use than FGAs (RR=0.55, CI=0.34~0.88, p=0.01, N=4, n=1,083). Conversely, benzodiazepine use did not differ between SGAs and FGAs (RR=0.91, CI=0.80~1.03, p=0.14, N=3, n=820). Only one study reported long-term beta-blocker use (n=555), showing lower use with risperidone than haloperidol (RR=1.48, CI=0.26~0.88, p=0.02).

Weight and metabolic outcomes

Olanzapine increased weight significantly more than haloperidol, molindone, and sulpiride (ES=0.61–3.56) (Table 5). Risperidone also caused significantly more weight gain than haloperidol and molindone (ES=0.22–0.93). Clozapine was associated with more weight gain than sulpiride (ES=4.95) in one study. Pooled SGAs caused more weight gain than FGAs (ES=0.65, p<0.001, N=7, n=1,059). Moderator analysis revealed that larger differences in weight gain between SGAs and FGAs were associated with shorter follow-up time, smaller sample size, younger age, female sex, and schizophrenia diagnosis. While at long-term follow-up, SGAs were still associated with more weight gain than FGAs (N=5, n=996), the ES was about halved (pooled ES=0.33, CI=0.07~0.59, p=0.01). Likewise, weight gain ≥7% was significantly more likely with olanzapine (RR=3.31, p<0.01) and risperidone (RR=1.61, p<0.01) than haloperidol, and with SGAs than FGAs (RR=2.26, CI=1.33~3.69, p=0.001, NNH=4, N=3, n=733) (Table 5). Weight gain ≥7% during long-term follow-up was still more likely with SGAs than FGAs, but the RR was smaller (RR=1.45, CI=1.17~1.79, p=0.001, N=3, n=778, NNH=5.35).

Table 5.

Pooled effect sizes (95% confidence interval in parenthesis) of short-term weight gain and metabolic changes in comparison of SGAs and FGAs. A negative effect size and an RR below 1 indicate worse side effects from FGAs.

SGA	FGA Comparator	Weight Change (Hedges’ g)	Weight Gain (>7%) (RR)	Glucose change (Hedges’ g)	Cholesterol change (Hedges’ g)	Triglyceride change (Hedges’ g)
Olanzapine	Haloperidol	0.61 (0.35~0.87)** (N=4, n=572)	3.31 (1.83~5.98)** (N=2, n=362)	0.08 (−0.17~0.32) (N=3, n=496)	0.17 (−0.00~0.35) (N=3, n=501)	0.20 (−0.25~0.65) (N=2, n=280)
Olanzapine	Molindone	1.77 (1.24~2.30)** (N=1, n = 75)	NR	−0.02 (−0.70~0.66) (N=1, n=35)	1.02 (0.30~1.75)** (N=1, n=35)	0.51 (−0.20~1.21) (N=1, n=33)
Olanzapine	Sulpiride	3.56 (2.70~4.42)** (N=1, n=53)	NR	−1.21 (−1.79~−0.63)** (N=1, n=53)	5.12 (4.01~6.23)** (N=1, n=53)	3.32 (2.49~4.15)** (N=1, n=53)
Risperidone	Haloperidol	0.22 (0.03~0.40)* (N=2, n=443)	1.61 (1.25~2.09)** (N=2, n=485)	−0.40 (−0.83~0.03) (N=1, n=84)	−0.07 (−0.50~0.36) (N=1, n=84)	−0.03 (−0.47~0.41) (N=1, n=78)
Risperidone	Molindone	0.93 (0.48~1.3)** (N=1, n=81)	NR	0.03 (−0.55~0.60) (N=1, n=0.92)	−0.46 (−1.04~0.12) (N=1, n=45)	0.32 (−0.28~0.92) (N=1, n=42)
Risperidone	Sulpiride	−3.86 (−4.73~−3.00)** (N=1, n=58)	NR	−1.99 (−2.61~−1.36)** (N=1, n=58)	−1.36 (−1.93~−0.80)** (N=1, n=58)	−1.18 (−1.74~−0.63)** (N=1, n=58)
Clozapine	Chlorpromazine	NR	NR	0.25 (−0.06~0.56) (N=1, n=160)	NR	NR
Clozapine	Sulpiride	4.95 (3.92~5.97)** (N=1, n=59)	NR	−1.54 (−2.12~−0.97)** (N=1, n=59)	3.07 (2.32~3.82) (N=1, n=59)	5.02 (3.98~6.05)** (N=1, n=59)
Amisulpride	Haloperidol	0.12 (−0.20~0.43) (N=1, n=155)	NR	−0.03 (−0.30~0.24) (N=1, n=207)	0.19 (−0.08~0.47) (N=1, n=207)	0.34 (0.06~0.61)* (N=1, n=207)
Quetiapine	Haloperidol	0.30 (−0.01~0.62) (N=1, n=157)	NR	−0.11 (−0.38~0.16) (N=1, n=207)	−0.03 (−0.30~0.25) (N=1, n=207)	0.15 (−0.12~0.42) (N=1, n=207)
Ziprasidone	Haloperidol	−0.12 (−0.46~0.22) (N=1, n=132)	NR	−0.36 (−0.65~−0.06)* (N=1, n=185)	0.02 (−0.27~0.31) (N=1, n=185)	0.10 (−0.19~0.39) (N=1, n=185)
Pooled SGAs	Pooled FGAs	0.65 (0.33~0.98)** (N=7, n=1059)	2.26 (1.33~3.69)** (N=3, n=733)	−0.20 (−0.61~0.22) (N=6, n=749)	0.46 (−0.01~0.92) (N=5, n=593)	0.71 (−0.02~1.44) (N=4, n=371)

Note:

^*p < .05;

^**p < .01.

RR = Relative Risk.

Olanzapine, risperidone, and clozapine were associated with lower glucose change than sulpiride in one study (Wu et al., 2006). Compared to haloperidol, glucose changes were similar with olanzapine, risperidone, amisulpride, and quetiapine haloperidol; only ziprasidone was significantly better than haloperidol in one study (ES=−0.36, p<0.05) (Table 5). Pooled SGAs and FGAs were similar (ES=−0.16, CI=−0.57~0.25, p=0.44, N=6, n=749). Heterogeneity across studies was high (Q=36.10, p<0.001, I²=86.15%). Even when excluding an outlier (Wu et al., 2006), pooled ES remained non-significant. No SGA-FGA difference emerged regarding long-term glucose change (N=3).

Regarding short-term total cholesterol change, olanzapine was significantly worse than molindone and sulpiride, and marginally worse than haloperidol (ES=0.17, p=0.051, N=3, n=501) (Table 5). Risperidone was not different from haloperidol, but significantly better than sulpiride. Pooled SGAs were associated with a marginally larger total cholesterol increase than FGAs (ES=0.46, p=0.053, N=5, n=593). Heterogeneity was large (Q=24.64, p<0.001, I²=83.76%). Excluding Wu et al(Wu et al., 2006) resulted in a much smaller ES (ES=0.15, p=0.08). No significant SGA-FGA difference emerged regarding long-term cholesterol change (N=2).

Only few studies reported on triglyceride changes. Olanzapine and clozapine were worse than sulpiride, and amisulpride was worse than haloperidol. However, risperidone was better than sulpiride (Table 5). Pooled SGAs showed marginally greater short-term triglyceride increase than FGAs (ES=0.71, p=0.057, N=4, n=371). Heterogeneity was high (Q=25.54, p<0.001, I²=88.25%). Excluding Wu 2006, the pooled ES became much smaller (ES=0.20, p=0.08).

Discussion

In this comprehensive meta-analysis of head-to-head trials, several SGAs outperformed FGAs in FES patients: olanzapine, amisulpride and risperidone were associated with significantly lower treatment discontinuation rates; olanzapine and amisulpride were superior regarding dropout due to inefficacy; and olanzapine, risperidone and quetiapine led to less dropouts due to intolerability. Only quetiapine had less discontinuation due to non-adherence/patient choice. Pooled together, SGAs vs FGAs had lower treatment discontinuation rates due to any cause (NNT=12), inefficacy and intolerability, reducing these events by 26%, 40% and 54%, respectively (Fig S9). Regarding total symptom reduction, only olanzapine and amisulpride outperformed FGAs, and FGA-SGA class differences were non-significant. Moreover, olanzapine outperformed haloperidol on negative symptoms, depression, global cognition and long-term remission; amisulpride outperformed haloperidol regarding positive symptoms, depression, CGI severity and long-term remission; and clozapine and quetiapine, each, outperformed chlorpromazine and haloperidol regarding negative symptoms, while risperidone and ziprasidone had lower relapse rates. Pooled SGAs outperformed FGAs on negative symptoms (ES=0.16), global cognition (ES=0.25) and relapse (Fig S9).

Interestingly, some of the important factors in treating first episode psychosis such as haloperidol equivalent dosing and percentage of drug-naïve patients were not significant in moderator analysis of any outcome. The same was true for blinding status. Notably, however, government-sponsored studies tended to favor FGAs, whereas industry-sponsored studies tended to favor SGAs in total symptom reduction and response rate. While being a post hoc finding, it is possible that in industry-sponsored studies raters had a subtle bias against FGAs, especially when EPS unmasked drug assignment, but more independently funded studies are needed.

Results of drug-induced adverse events were robust, and effect sizes were large. Olanzapine, risperidone, clozapine and pooled SGAs caused significantly less EPS, akathisia, and/or co-treatment with anticholinergics and benzodiazepines than FGAs. Conversely, clozapine, olanzapine, risperidone and pooled SGAs showed larger weight gain than FGA comparators, but without significantly greater increases in glucose and lipid parameters, except for isolated findings with clozapine, olanzapine and amisulpride, at least in these shorter-term trials.

First episode patients are especially sensitive to drug-induced side effects including EPS and weight gain. Therefore, the lower EPS risk with most SGAs, also found in chronic patients(Fischer-Barnicol et al., 2008), is beneficial. Interestingly, in more recent studies, the ES favoring SGAs has been diminishing. This is likely attributable to lower FGA comparator doses and less uniform use of haloperidol. The fact that older subjects showed greater FGA-SGA differentiation is consistent with the fact that young and pediatric patients are at particular risk for EPS, even with SGAs(Correll et al., 2006).

Consistent with data in chronic patients, continuous and categorical “significant” weight gain outcomes disfavored most SGAs being more pronounced in younger patients, females, and schizophrenia patients. That schizophrenia emerged as a risk factor suggests that illness severity and/or negative symptoms might also play a role in weight gain. The lack of glucose differences (except for a favorable outcome with ziprasidone) is not surprising, since (pre)diabetes generally emerges only after a period of insulin resistance(De Hert et al., 2011), and insulin levels were not measured. Moreover, since mostly short-term results were available, longer-term risk for more distal outcomes, such as diabetes and cardiovascular illness, were not available in first-episode patients, except for a data base study(Nielsen et al., 2010). Interestingly, SGA-FGA effect size differences for weight and lipid outcomes declined with longer follow-up, suggesting that non-medication effects, such as unhealthy lifestyle, the underlying illness, environment and, possibly, genetics, start playing a relevant role.

Taken together, these meta-analytic results confirm that SGAs are not a homogeneous class, being associated with different efficacy and side effect profiles(Kane and Correll, 2010; Leucht et al., 2009). Notably, like a prior meta-analyses in chronic schizophrenia(Crossley et al., 2010), we also found small effect size differences favoring amisulpride and olanzapine and, less so, risperidone compared to FGAs. The superior outcome regarding a global cognitive index with olanzapine and risperidone compared to FGAs, even given at relatively low doses, is noteworthy, being inconsistent with chronic schizophrenia data(Keefe et al., 2007). The potential for greater treatment differences in earlier illness phases, although unclear to date, deserves follow-up, as cognitive dysfunction has been associated with poor functional outcomes(Brekke et al., 2007). However, the superiority of olanzapine, risperidone and amisulpride, but not of other SGAs, compared to FGAs could be due to a cohort effect, in that these three SGAs were studied earlier, at a time when higher haloperidol were commonly used, causing potentially higher dropout rates. In fact, in our analyses, EPS differences between SGAs and FGAs diminished over time. Moreover, during the past decade, effect sizes have decreased in schizophrenia and CNS trials in general(Correll et al., 2011). Nevertheless, neither FGA comparator doses nor year of study publication emerged as significant moderators in any analyses. Another interesting finding is that clozapine was not more efficacious or effective than chlorpromazine in FES patients. This is inconsistent with chronic schizophrenia data, but confirmed the notion that clozapine should be reserved for treatment refractory schizophrenia due to its significant side effect profile. However, these results were based on one single study conducted in Chinese patients; yet additional FES studies with clozapine are unlikely to be conducted due to its significantly more severe side effect burden and the overall greater responsiveness of FES patients compared to chronic patients.

Nevertheless, overall, effect sizes of the efficacy differences, either in pooled or individual analyses, were relatively small, with the exception of significantly lower all-cause and specific-cause discontinuation rates with SGAs. This suggests that FES patients who are generally more treatment responsive than chronic patients have a reasonable chance of benefiting from any antipsychotic treatment. By contrast, side effect differences were larger. Therefore, antipsychotic choice should take into consideration the safety profile of each agent and the patient’s willingness to accept specific adverse effect clusters. Moreover, the robust finding of lower treatment discontinuation with SGAs is quite relevant, given that engagement and continued antipsychotic treatment are crucial for relapse prevention, remission, recovery, and other beneficial outcomes(Kane and Correll, 2010). It is unclear, however, if the lower treatment discontinuation was affected by less frequent EPS and akathisia compared FGAs, as patient choice-related discontinuation did not differ between SGAs and FGAs, except for quetiapine that has particularly low EPS and akathisia rates. Nevertheless, the coding of reasons for treatment discontinuing treatment mostly lacked detail. Furthermore, despite the neuromotor adverse effect advantage of SGAs, significant weight gain induced by SGAs, especially olanzapine and clozapine, but also risperidone, is of major concern for patients’ long-term health(Correll et al., 2009). More studies are needed to better understand predictors of antipsychotic-induced weight gain, including genetic markers.

Several limitations of the present study require consideration. The number of relevant RCTs in first-episode patients was small and some outcomes were not reported by all studies. Only one study was available for amisulpride, quetiapine and ziprasidone, and studies were missing for aripiprazole, asenapine, iloperidone, lurasidone and paliperidone. Therefore, specific drug recommendations should be viewed with caution. Moreover, studies included in any meta-analysis are heterogeneous, but we used random effects models and performed sensitivity and moderator analyses to deal with this limitation. In addition, haloperidol was used as the SGA comparator in 9/13 studies and was frequently used in relatively high doses. This might have caused higher EPS and discontinuation rates, although haloperidol dose was not a significant moderator in our analyses. Moreover, most data were available for short-term outcomes and data on tardive dyskinesia were generally lacking. Also, unmeasured medication non-adherence may have affected these findings. Finally, head-to-head trials comparing SGAs with SGAs in FES were not included, being beyond the scope of this meta-analysis.

In summary, in FES, pooled SGAs were either similarly effective or modestly better than FGAs regarding several efficacy and tolerability outcomes, being associated with greater weight gain. Among individual SGAs, amisulpride and olanzapine and, to a lesser degree, risperidone were most consistently superior to their respective FGA comparator, but weight and metabolic problems were also greater with olanzapine. Overall, risperidone appears to be reasonably efficacious and is associated with relatively benign side effects, so it should be considered as a first line therapy for first-episode schizophrenia. However, studies with aripiprazole and newer SGAs that generally have less metabolic liability(De Hert et al., 2012) are clearly needed and mid-potency FGAs should strongly be considered as comparators. Furthermore, since SGAs were significantly superior for negative symptom, global cognition and long-term relapse outcomes compared to FGAs, future studies need to study the real world functional implications of these potential differences and include subjective well-being and cost-effectiveness outcomes to inform broader health care strategies and resource allocation.