Efficacy and acceptability of long-acting antipsychotics in acutely ill individuals with schizophrenia-spectrum disorders: A systematic review and network meta-analysis

To assess the effect of Long-acting injectable (LAI) antipsychotics in acutely ill patients, we systematically searched major databases for randomized controlled trials (RCTs) comparing LAIs with other LAIs, oral anti-psychotics, or placebo in acutely symptomatic adults with schizophrenia-spectrum disorders. Data were analyzed with a random-effects network meta-analysis. Co-primary outcomes were efficacy (mean change in psychopathology rating scales) and acceptability (all-cause discontinuations) at study endpoint. Of 25 RCTs, 19 studies tested second-generation LAIs (SGA-LAIs) and six first-generation LAIs (FGA-LAIs). Due to a disconnected network, FGA-LAIs were analyzed separately, with poor data quality. The SGA-LAIs network included 8,418 individuals (males = 63%, mean age = 39.3 years). All SGA-LAIs outperformed placebo in reducing acute symptoms at study endpoint (median follow-up = 13 weeks). They were more acceptable than placebo with the only exception of olanzapine, for which no differences with placebo emerged. Additionally, we distinguished between different LAI formulations of the same antipsychotic to explore potential pharmacokinetic differences. Most formulations outperformed placebo in the very short-term (2 weeks or less), regardless of the need for initial oral supplementation. SGA-LAIs are evidence-based treatments in acutely ill individuals with schizophrenia-spectrum disorders. Findings support the use of SGA-LAIs to manage psychopathology and improve adherence right from the acute phases of illness.


Introduction
Schizophrenia spectrum disorders (SSDs) are serious mental illnesses affecting approximately 0.9% of the population (James et al., 2018).Large network meta-analyses (NMAs) showed oral antipsychotics to be efficacious for acute psychotic symptoms (Huhn et al., 2019), and oral and long-acting injection (LAI) antipsychotics to be equally effective for relapse prevention (Ostuzzi, Bertolini, et al., 2022) in clinically stable individuals.As LAIs might have advantages over oral formulations in terms of medication adherence, timely use of LAIs has been recommended by evidence-based guidelines (Kane and Garcia-Ribera, 2009;Llorca et al., 2013).However, evidence concerning LAIs' efficacy and acceptability in the acute phase of illness is yet to be substantiated by robust evidence (Liu et al., 2015;Reymann et al., 2022).A NMA by Leucht and colleagues (Leucht et al., 2023) assessed the medium-to-long-term efficacy of oral and LAI antipsychotics in acutely ill individuals with SSD.However, only trials with at least six months of follow-up were included, and the paucity of data on LAIs prevented a comparison between the LAI and oral formulations, which were grouped in the analysis.Of relevance, a pairwise meta-analysis by Wang and colleagues (Wang et al., 2023) assessed the differential efficacy of LAIs and oral antipsychotics against placebo in acutely ill individuals with SSD, finding these formulations to be similarly efficacious.However, only trials comparing the two formulations of the same medication were included, and the pairwise approach did not allow for the comparative comparison of different antipsychotics.
Against this background, we conducted a systematic review and NMA aiming to assess the differential efficacy and acceptability of LAIs and oral antipsychotics in acutely ill individuals with SSDs.

Methods
This study was conducted and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for NMAs (Page et al., 2021) (Suppl.A).The study protocol was registered in advance on Open Science Framework (available at: htt ps://osf.io/v7ehp/),and post-hoc amendments to the protocol are described in the Suppl.B.

Search strategy
We searched the electronic databases Medline, EMBASE, PsycINFO, CENTRAL, and CINAHL from database inception to August 31, 2023, without language or study date restrictions (for search strategy, see Suppl.C).Additionally, we searched databases of regulatory agencies (i.e., FDA and EMA), and online trial registers (e.g., clinicaltrials.gov).

Study selection and data extraction
We searched for randomized-controlled trials (RCTs) including adults (≥18 years old) diagnosed with SSDs (including schizophrenia, schizoaffective disorder, schizophreniform disorder, delusional disorder, and psychotic disorders not otherwise specified) according to validated diagnostic systems (i.e., DSM or ICD), either at their first or recurring episode, who were acutely ill at the time of randomization, as defined by each study.If acute symptoms were not clearly described, we assumed their presence based on clinical data (e.g., recent hospitalization) or a mean baseline score ≥75 on the Positive and Negative Syndrome Scale (PANSS), a mean baseline score ≥44 on the Brief Psychiatric Rating Scale (BPRS) and a mean baseline score ≥4 on the Clinical Global Impressions-Severity scale (CGI-S) (Busner and Targum, 2007;Leucht et al., 2005).
Only RCTs comparing LAIs with placebo, with oral antipsychotics, or between each other were included.Interventions included all available LAIs according to the WHO ATC/DDD classification (World Health Organization, 2023).RCTs randomizing to a single LAI against a mix of different oral antipsychotics were excluded (e.g.risperidone LAI vs generic oral SGA therapy).
Two authors independently assessed titles, abstracts, and full texts of potentially relevant articles, and extracted data following recommendations of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins et al., 2023) (DPol, AC).Two authors (GV, DPol) assessed the risk of bias of included studies using the Cochrane Risk of Bias tool, version 2 (RoB2) (Sterne et al., 2019).Disagreements were resolved by discussion or arbitration by a third senior author (GO, CB, CUC).

Outcomes
The two co-primary outcomes were efficacy, defined as the mean change score at validated rating scales measuring psychopathology at the end of the study, and acceptability, defined as all-cause discontinuations.
Secondary outcomes included: (a) study-defined response or imputed from symptomatology rating scales (Furukawa et al., 2005) (b) response defined as a decrease ≥30% of PANSS mean score; (c) short-term response, between 4 and 12 weeks of follow-up, giving preference to the timepoint closest to 4 weeks; (d) medium-term response, between 12 and 24 weeks of follow-up, giving preference to the timepoint closest to 12 weeks; (e) long-term response, after 24 weeks of follow-up, giving preference to the latest timepoint available; (f) inefficacy-related discontinuation; (g) intolerability-related discontinuation; (h) mean change score on validated rating scales measuring functioning at the end of the trial; (i) common antipsychotic-related adverse events, including sedation, weight gain, hyperprolactinemia, extrapyramidal symptoms, akathisia, QTc prolongation; (j) proportion of participants experiencing at least one serious adverse event at study end; (k) proportion of deceased participants by the end of the trial.As the primary analysis pooled together different LAI formulations of the same medication, which may differ in terms of frequency of administration, onset of action, and oral supplementation, we performed a post-hoc efficacy analysis separating such formulations, and pooling very short-term data (i.e., timepoint closest to two weeks), considering that differences related to pharmacokinetic properties are more apparent in the early stages, and tend to diminish over time as the therapeutic plasma concentrations of the medication are reached.

Statistical analysis
We conducted a random-effects network meta-analysis (NMA) using the R netmeta package (Shim et al., 2019) and the Stata mvmeta package (White, 2007).For dichotomous outcomes, we pooled relative risks (RRs) with 95% confidence intervals (CIs) using a strict intention-to-treat approach (i.e.all randomized individuals as the denominator), while for continuous outcomes, we pooled mean differences (MDs) or standardized mean differences (SMDs) as appropriate.In cases where studies included different doses of the same antipsychotic, we pooled them into a single arm (Higgins et al., 2023), provided they fell within a therapeutic dose range (Gardner et al., 2010).Missing data were either obtained from trial authors or imputed using validated statistical methods (Aydin and Yassikaya, 2022;Furukawa et al., 2006;Higgins et al., 2023).Heterogeneity was assessed visually and with τ2 and I2 statistics (Higgins et al., 2023).The transitivity assumption was evaluated by comparing potential effect modifiers across treatments, followed by meta-regression analyses (Cipriani et al., 2013).Inconsistency was assessed both globally and locally (Bucher et al., 1997;Shih and Tu, 2021).Treatment rankings were generated using P-scores (Rücker and Schwarzer, 2015).Confidence in evidence was evaluated using the Confidence in Network Meta-Analysis (CINeMA) methodology (Salanti et al., 2014).(see Suppl.J: E-methods for details).

Results
Altogether, 1546 records were identified after database and hand searches.After removing duplicates and screening titles and abstracts, 95 records underwent full-text assessment.Of these, 25 primary studies, including 9027 individuals, provided meta-analyzable data for at least one outcome.The flowchart of included studies is available in suppl.I (E-Fig.1), while a bibliography of included studies is available in suppl.K. Nineteen studies provided data on second-generation (SGA) LAIs, and 6 on first-generation (FGA) LAIs (Table 1).The two pharmacological classes were analyzed separately since the two networks were not connected for any of the outcomes of interest.

First-generation antipsychotics
The six studies on FGA-LAIs included 609 individuals (age=35.5 years, male=59%).The mean follow-up was 37.3 weeks.Five out of six Regarding efficacy, three studies and 106 individuals contributed to the analysis.The RoB2 showed "high" risk of bias for 67% of studies, and "some concerns" for 33% (Suppl.E.1).Both the efficacy and acceptability network were poorly populated and connected, with no closed loops, preventing the assessment of heterogeneity and inconsistency.Both trifluoperazine OS, and FLUS-LAI were more efficacious than FLUPH-LAI, while no differences in acceptability emerged with any of the other treatments (Suppl.G).All comparisons had a "very low" certainty of evidence (Suppl.H.4-H.5).
As for secondary outcomes, a network meta-analysis was feasible only for extrapyramidal symptoms, which showed FLUS-LAI to be better tolerated than FLUPH-LAI.The analysis was however based on a sparsely populated and connected network, for which inconsistency and heterogeneity could not be measured (Suppl.F.4.1).For the remaining secondary outcomes, a meta-analytical approach was not feasible, and results of single studies are reported in the Suppl.G.3-G.4.

Second-generation antipsychotics
All 19 studies contributed to the two co-primary outcomes.These studies included 8418 individuals, 63% of whom were male, with a mean age of 39.3 years.The mean follow-up duration was 16.6 weeks.Sixteen out of 19 (84.2%) studies were double-blind, and 11 out of 19 (57.9%) included a placebo as a comparator.LAIs included paliperidone palmitate 1-monthly (PALI-LAI; N = 11, n = 2818, mean age 39.9 years, Regarding efficacy (mean change at psychopathology rating scales), 19 studies and 6876 participants contributed to the analysis.The RoB2 showed that the risk of bias was "high" for 15.8% of studies, "low" for 21.1%, and "some concerns" for 63.1% (Suppl.E.1).The efficacy network was well connected, with 5 out of 7 treatments included in closed loops (Fig. 1).The transitivity assumption was not violated for any of the potential effect modifiers analyzed (Suppl.F.6).All treatments outperformed placebo, with moderate certainty of evidence (according to CINeMA) for aripiprazole LAI (ARI-LAI), olanzapine LAI (OLA-LAI), risperidone LAI (RIS-LAI), paliperidone LAI (PALI-LAI), and low certainty for aripiprazole OS (ARI-OS) and olanzapine OS (OLA-OS) (Fig. 1).NNTs for LAIs ranged from 3.2 (ARI-LAI) to 4.7 (PALI-LAI).When compared head-to-head, no differences emerged, except for ARI-LAI outperforming PALI-LAI (very low certainty) (Table 2).This analysis was characterized by moderate heterogeneity (τ 2 =0.03;I 2 =69.6%), while consistency was preserved both globally (p = 0.82) and locally (Suppl.F.1). Sensitivity analyses yielded results largely consistent with the primary analysis in terms of effect sizes of treatments and degree of heterogeneity, except for the analysis excluding studies for which the standard deviation was imputed, where heterogeneity notably  decreased (I 2 =33.6%).After excluding placebo-controlled trials, heterogeneity decreased (I 2 =40.5%), and no significant differences emerged between treatments (Suppl.F.5.1).
Regarding the acceptability outcome, 19 studies and 8418 participants contributed to the analysis.The risk of bias was "high" for 10.5% of the studies, "low" for 47.4%, and carried "some concerns" for 42.1% (Suppl.E.2).Most treatments outperformed placebo, with moderate certainty for ARI-LAI, RIS-LAI, and PALI-LAI, and low certainty for ARI-OS, while no differences emerged for OLA-LAI (very low certainty) and OLA-OS (low certainty) (Fig. 1).NNTs of LAIs ranged from 4.2 (ARI-LAI) to 6.4 (PALI-LAI).When compared head-to-head, no differences emerged between LAIs, except for RIS-LAI outperforming PALI-LAI (Table 2).This analysis was characterized by low heterogeneity (τ 2 =0.008;I 2 =36.7%), and consistency was preserved globally (p = 0.08), although three out of 7 comparisons showed local inconsistency (Suppl.F.2). Sensitivity analyses yielded results largely consistent with the primary analysis in terms of effect sizes of treatments, degree of heterogeneity and inconsistency, which however decreased after excluding trials on individuals with recent-onset disease (p = 0.58) and placebo-controlled trials (p = 0.36).
Meta-regression analyses for the co-primary outcomes did not detect any potential effect modifier among the variables of interest (Suppl.F.6).

Secondary efficacy outcomes
Table 3 and Suppl.F.3 show secondary efficacy outcomes.All 4 SGA-LAIs outperformed placebo in reducing both positive and negative symptomatology scores at PANSS subscales.OLA-LAI, RIS-LAI, and PALI-LAI were more effective than placebo in reducing general psychopathology symptoms.ARI-LAI appeared more effective than PALI-LAI in reducing positive symptoms, while RIS-LAI was more effective than PALI-LAI in reducing general psychopathology symptoms.No other differences between treatments were observed.
All treatments outperformed placebo in terms of response, both study-defined and defined as a reduction of 30% or more in PANSS total score reductions and as inefficacy-related discontinuation.ARI-LAI, RIS-LAI, and PALI-LAI outperformed placebo regarding short-term response (1-3 months).Data on medium-term response (4-6 months) were available only for RIS-LAI and PALI-LAI, which outperformed placebo.ARI-LAI, RIS-LAI, and PALI-LAI outperformed placebo also in terms of mean change score at functioning rating scales, while no data were available for other treatments.Apart from the response according to any definition, which showed relevant inconsistency, no relevant issues emerged in terms of consistency and heterogeneity.The lack of data concerning quality of life prevented a meta-analytical approach.

Tolerability outcomes
Table 3 and Suppl.F.4 show secondary tolerability outcomes.ARI-LAI, RIS-LAI, and PALI-LAI outperformed placebo regarding intolerability-related discontinuation.RIS-LAI was associated with more extrapyramidal symptoms than placebo.All treatments, except for ARI-OS, increased body weight measured both as a continuous and as a dichotomous outcome compared to placebo.Compared to placebo,   PALI-LAI increased serum prolactin, and RIS-LAI and PALI-LAI induced hyperprolactinemia.RIS-LAI, PALI-LAI, and OLA-OS outperformed placebo in terms of severe adverse events.Compared to placebo, no difference emerged in the number of deaths and QTc prolongation.Except for prolactin increase (dichotomous), which showed relevant inconsistency, no relevant issues emerged in terms of inconsistency and heterogeneity.

Discussion
To our knowledge, this is the largest systematic review and the first NMA comparing the efficacy and tolerability of LAIs in acutely ill individuals with SSDs.
Data on both second-generation (SGA) and first-generation (FGA) long-acting injectable antipsychotics (LAIs) yielded separate networks, precluding a single network meta-analysis (NMA).Further, the quality of data on FGA-LAIs was poor, as it included mostly old and small studies comparing medications seldom used in current clinical practice.
As for efficacy, all SGAs outperformed placebo in reducing symptoms, with effect sizes ranging from large (ARI-LAI: Cohen's d 0.87, NNT 3.2) to moderate (PALI-LAI: Cohen's d 0.56; NNT 4.7) supported by "moderate" certainty of evidence according to the CINeMA approach.No relevant differences emerged between LAIs (except ARI-LAI outperforming PALI-LAI), and between oral and LAI antipsychotics.Overall, these results were supported by sensitivity and meta-regression analyses, as well as secondary efficacy outcomes.The post-hoc analysis on the very short-term efficacy generally confirmed that LAI formulations not requiring oral supplementation had an efficacy profile comparable to both oral antipsychotics and LAIs with slower onset of action, requiring oral supplementation.Achieving a timely response and assuring treatment adherence is crucial to minimize the duration of symptoms and the risk of relapse and associated detrimental phenomena, e.g., structural brain damage, treatment-resistance, and functional impairment (Lin et al., 2021;Takeuchi et al., 2019).LAIs can rapidly achieve therapeutic levels, and certain formulations do not require oral supplementation, thus they could be a valid option to manage acute symptoms, especially in settings where adherence to oral treatment is suspected to be suboptimal.
In terms of acceptability, ARI-LAI, PALI-LAI, and RIS-LAI were associated with fewer all-cause discontinuation than placebo (moderate certainty).Compared to placebo, ARI-LAI, PALI-LAI, and RIS-LAI were associated with fewer drop-outs due to adverse events; all SGA-LAIs were associated with an increased risk of sedation and weight gain compared to placebo; RIS-LAI and PALI-LAI were associated with prolactin increase; RIS-LAI was associated with increased risk of extrapyramidal symptoms; RIS-LAI and ARI-LAI were associated with increased risk of akathisia.As no major differences were observed between LAIs and OAPs, we did not confirm the findings of Wang et al. (2023) that some adverse events were less pronounced with LAIs than with oral formulations.Results for the first primary outcome (mean change at psychotic symptomatology rating scales at study endpoint) are reported in bottom-left part of the table, with standardized mean differences (SMDs) and 95% confidence intervals (CIs).SMDs lower than 0 favor the column-defining treatment.Results for the secondary outcome (drop-outs due to any causes) are reported in the upper-right part of the table, with risk ratios (RR) and 95% confidence intervals (CIs).RR lower than 1 favor the column-defining treatment.These findings should be interpreted considering some limitations.Firstly, we pooled together different LAI formulations of the same antipsychotic in most analyses, despite pharmacokinetic differences regarding the onset of action and need for oral supplementation, which may have introduced heterogeneity.Although we performed a post-hoc analysis on the very short-term efficacy of different LAI formulations, which was overall consistent with primary results, this analysis large (ARI-LAI: Cohen's d 0.87, NNT 3.2), was characterized by an impoverished and scattered network, losing precision and overall certainty of evidence.Second, for some studies, we assumed the presence of acute symptoms although this was not clearly reported by authors.We based this choice on clinically recognized cut-offs of commonly used rating scales (LEUCHT et al., 2005) .Further, a sensitivity analysis removing such studies, yielded results broadly consistent with the primary analysis.Third, we included two RCTs (Fleischhacker et al., 2012;Li et al., 2011) that allowed oral supplementation beyond the standard first three weeks for people taking RIS-LAI, potentially inflating the efficacy estimates of this arm.However, sensitivity analyses excluding such RCTs did not change the interpretation of the results.Fourth, the overall risk of bias was high for many studies.Again, after removing these RCTs through sensitivity analyses, primary results did not remarkably change.
In addition, the data on FGAs were poor in both quantity and quality, making it difficult to draw firm conclusions.The lack of available evidence is an important limitation because FGA-LAIs, such as haloperidol LAI and fluphenazine LAI are still widely used worldwide and may be the only LAI available in several settings, including low-and middleincome countries (Ostuzzi, Gastaldon, et al., 2022).
Regrettably, there has been no trial that directly compared FGA-LAIs against SGA-LAIs or placebo, making it impossible to draw comparisons between these two drug classes.Hence, our choice to concentrate on SGA-LAIs is driven solely by the available evidence and should not be misconstrued as an indication of the superior efficacy of SGA-LAIs over FGA-LAIs.
Moreover, the results for oral antipsychotics should be interpreted with caution because we did not include RCTs comparing oral antipsychotics with each other.In particular, OLA-OS performed strikingly poorly in assessing very short-term efficacy, based on a single small (Huang et al., 2018) in which participants randomized to OLA-OS showed a slower improvement in symptoms compared with PALI-LAI, although no differences between the two treatments emerged at the end of the study.Finally, as we included few RCTs with an oral comparator, the certainty of evidence was generally poorer for oral antipsychotics compared to LAIs.Still, no clear differences emerged between SGA-LAIs and their oral counterparts, which is in line with evidence on remitted individuals (Ostuzzi, Bertolini, et al., 2022;Schneider-Thoma et al., 2022) Despite of these limitations, results from this study significantly updates and extends previous literature on this topic (Leucht et al., 2023;Wang et al., 2023), demonstrating that most of the SGA-LAIs currently marketed in Europe and in the U.S. can be used effectively for the management of acute psychotic symptoms, helping to optimize treatment adherence from the earliest stages of disease.

Table 1
Characteristics of included studies*.

Table 2
Netleague table of co-primary outcomes.

Table 3
Secondary efficacy and tolerability outcomes.