Indirect Meta-Analysis of Brexpiprazole Versus Aripiprazole in the Acute Treatment of Schizophrenia

Background: Brexpiprazole and aripiprazole are atypical antipsychotics that act as partial agonists at the dopamine D2 receptor. No head-to-head trial comparing brexpiprazole and aripiprazole in the treatment of schizophrenia is available. Here, we carry out a systematic review and comparison of the ecacy and safety of brexpiprazole and aripiprazole in schizophrenia treatment. Methods: We employed an indirect meta-analysis to determine effect sizes from randomised placebo-controlled trials with brexpiprazole and aripiprazole in the acute treatment of schizophrenia. We compared responder rates, incidences of adverse events and serious adverse events, the number needed to treat (NNT) for response, number needed to harm (NNH) for adverse events or treatment discontinuation, and likelihood to be helped or harmed (LHH) as ecacy and safety indices of the two drugs. Results: Five studies for each drug were included in the analysis. Similar risk differences vs. placebo were observed for responder rates under brexpiprazole (10.2%, p = 0.0015) and aripiprazole (10.3%, p = 0.0003). Higher incidences of adverse events and serious adverse events were seen under aripiprazole compared with brexpiprazole, however, the risk differences were not statistically signicant. The NNT for response was 11 for both substances. For brexpiprazole compared with placebo, we did not nd an increase of adverse events (NNH = 27, not signicant), however, we found an increased number of adverse events for aripiprazole versus placebo (NNH = 17, p < 0.05). For both drugs, benets were encountered more often than harms, with an LHH for any adverse event of 2.41 for brexpiprazole and 1.56 for aripiprazole, respectively. Conclusions: The likelihood to be helped rather than harmed was greater with brexpiprazole compared to aripiprazole for the total rate of adverse events (ratio of brexpiprazole LHH/aripiprazole

events as documented in the respective study (AEs, TEAEs; OAEs); incidences of SAEs; discontinuations due to AE; and incidences of selected adverse events.
The effect sizes of active vs. placebo are quanti ed using number needed to treat (NNT), number needed to harm (NNH) and likelihood to be helped or harmed (LHH). NNT is an intuitive measure of the clinical impact that a treatment can have, making it a useful tool for treatment comparisons [11]. The interpretation of NNT is the number of patients to be treated with intervention A instead of intervention B to yield one additional positive outcome. In order to be useful, it is important to carefully de ne the outcome of interest before applying the concept of NNT [11].
In the case of unwanted effects, the equivalent measure is NNH. It gives the number of patients to be treated with intervention A instead of intervention B to yield one additional unwanted outcome.
Both numbers can be related using LHH to illustrate the trade-off between bene ts and harms. LHH is calculated by dividing the NNH by the NNT and can thus be interpreted as a bene t-risk ratio. In order for an intervention to cause more bene t than harm, the LHH should be greater than 1.
The dopamine D2 partial agonists aripiprazole, brexpiprazole, and cariprazine have already been compared with respect to NNT, NNH and LHH by Citrome [12] using the then available short-term data. Here, however, we present an up-to-date comparison using all available evidence from randomised controlled trials selected by systematic database search.

Methods
The review protocol was not registered.

Eligibility criteria
Randomized controlled trials of brexpiprazole and/or aripiprazole p.o. in patients with current diagnosis of acute exacerbation of schizophrenia aged ≥ 18 years were eligible for this review. The study duration was restricted to 4 or 6 weeks because a maintenance treatment for relapse prevention over 26 weeks or longer was reported in only one trial for each drug.

Information sources
The following databases with DIMDI search interface (no longer available today) were used to identify eligible brexpiprazole studies: CCTR93 (Cochrane Central Register of Controlled Trials), ME60 (Medline), and EM74 and EA08 (Embase and Embase Alert). The DIMDI search was completed by searches in public clinical trials registries and databases: WHO ICTRP (International Clinical Trials Registry Platform), ClinicalTrials.gov, Pharmnet.Bund, EU Clinical Trials Register, LIVIVO, and Pubmed. One additional study [13] was provided by Lundbeck outside the searches shortly after publication in July 2018. The following databases were used to identify eligible aripiprazole studies: LIVIVO and Pubmed. This was completed by a search in the registry ClinicalTrials.gov.

Search
The basic strategy to identify brexpiprazole studies in databases with DIMDI search interface is shown in Additional le 1.

Study selection
Five brexpiprazole and ve aripiprazole studies were suited for the 4 or 6 weeks treatment of patients with an acute exacerbation of schizophrenia. The studies were accepted with one placebo arm and one to three active treatment arms.

Data collection process
Data were extracted from brexpiprazole and aripiprazole full publications and study results in ClinicalTrials.gov, in one case supplemented by data from a poster presentation [1]. Data were entered directly into SAS les by a member of the statistics department of the CRO, printouts of which were checked against the original sources by an independent person from data management. Any corrections were quality controlled by means of printouts of the corrected les. If CGI is not available: A response is de ned as reduction from baseline in the PANSS Total score of ≥ 30%.
If CGI is available: A response is de ned as reduction from baseline in the PANSS Total score of ≥ 30% and/or CGI-I = 1 (very much improved) or CGI-I = 2 (much improved).
Further binary outcomes are: the number of completers total rates of adverse events, de ned by 'adverse events', 'treatment-emergent adverse events', or 'other adverse events than serious adverse events' rates of serious adverse events rates of study discontinuation due to adverse events incidences of selected adverse events.
S: items of study design The following items were used to characterise the study design: trial centres, concealment / blinding, allocation / randomisation, phases / duration of phases, experimental design, type of comparison, sample size determination.

Risk of bias in individual studies
The Cochrane risk of bias tool was used to describe the risk of bias in individual studies. This tool consists of ve items: sequence generation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting, and a catch-all item called 'other sources of bias'.

Summary measure
The report used Hedges's g, odds ratios, risk ratios, and risk differences to quantify the effect sizes of summary measures.
However, we focussed on binary outcomes which allow a comparison of bene ts and harm. Therefore, only two metaanalyses (responder rates, total incidences of adverse events) are shown as examples, further analyses are based on NNT and NNH (see below).Synthesis of results Fixed effect and random effects models are used in the meta-analyses, however, the main results are based on NNT and NNH (see below).

Additional statistical analyses
Typical outcomes of placebo-controlled clinical trials are the absolute increase of responder rates (AIR) and the absolute risk reduction (ARR) of adverse events. The number needed to be treated with active drug instead of placebo for one additional patient to bene t is de ned by NNT = 1/AIR, the number needed to be treated with active drug instead of placebo for one additional patient to be harmed by NNH = 1/ARR. Signi cance tests of NNT or NNH can be performed by con dence limits. If, e.g., ARR is calculated as p 1 -p 0 (p 1 = risk of active drug, p 0 = risk of placebo) with 95% con dence limits Cl low and Cl upp , then the lower limit LL = 1/Cl upp and the upper limit UL = 1/Cl low can be used as con dence limits of NNH. The con dence and UL = 1/Cl low to +∞ (minimum value: +1) In this case we do not discuss the con dence limits but only use the relevant information that the difference between p 1 and p 0 is not signi cant if the con dence limits of NNH have different directions. If the con dence interval of NNH consists of only negative values (i.e. a signi cantly lower harm was observed with active drug compared with placebo), one should interpret the lower harm as bene t. The likelihood to be helped or harmed, de ned by LHH = NNH/NNT, can then be used to quantify the value -NNH as predictor of the responder rate. In general, the LHH describes how often bene ts are encountered more frequently than harm.
Two methods are used to calculate con dence limits LL and UL: Wald and Wilson [11][12][13][14][15][16][17][18]. Wilson should be preferred in case of ARR values near 0 or 1 and small sample sizes, because the asymptotic method of Wald provides unreliable values in these situations.

Study selection
The systematic literature review identi ed 177 records of potentially eligible trials on brexpiprazole after duplicate removal within search and 649 records of potentially eligible trials on aripiprazole without any procedure to remove duplicates within search. By means of title and/or abstract screening, 135 (brexpiprazole) and 541 (aripiprazole) records were classi ed as not relevant. The remaining records were checked by means of full text screening (brexpiprazole: n = 42; aripiprazole: n = 28) or identi ed as reviews without additional information (aripiprazole: n = 80).
The full text screening led to the exclusion of 29 brexpiprazole and 20 aripiprazole records. The remaining records reported on 6 brexpiprazole and 6 aripiprazole trials ( Figure 1, Figure 2). In both drug groups one trial dealt with the maintenance therapy in the state of stable disease after previous acute illness. Due to the small number of these trials, the maintenance therapy was not considered in the further analyses. The ve relevant studies in each drug group are shown in Table 1.

Study characteristics
The

Risk of bias within studies
The methods of blinding and treatment allocation as well as the justi cation of anticipated effect size were often not speci ed. Furthermore, in three aripiprazole studies the variability of outcome data was repeatedly missing.

Synthesis of results
The synthesis of results is represented for responder rates, incidences of adverse events, and incidences of serious adverse events. Additional analyses are focused on NNT and NNH (see below).
Meta-analyses and indirect comparison of brexpiprazole and aripiprazole with respect to responder rates and total incidences of adverse events as well as serious adverse events The meta-analysis of responder rates in the acute treatment of schizophrenia with a pool of approved and recommended doses of brexpiprazole and aripiprazole provided for both drugs signi cant superiority versus placebo of 10.2% In the meta-analyses, I 2 was used as estimator of heterogeneity.
Absolute increase of the responder rates and number needed to treat (AIR, NNT) The absolute increase of the responder rates (AIR) was 0.093 in both active drug study groups compared with placebo. This corresponds to NNT = 11 for both brexpiprazole and aripiprazole.
Both con dence intervals according to Wald as well as Wilson demonstrated signi cant effect sizes upon brexpiprazole and aripiprazole ( Table 2).
Absolute risk reduction and number needed to harm (ARR, NNH) In the brexpiprazole studies, no signi cant differences were observed between active drug and placebo in the rates of adverse events because the con dence intervals were compatible with higher adverse event rates upon active drug as well as lower adverse event rates upon active drug as compared with placebo (Table 3). Negative values of the absolute risk reduction (ARR) or the number needed to be treated for one additional patient to be harmed (NNH) indicate a lower adverse event rate upon active drug compared with placebo.
In the aripiprazole studies, the positive values of both LL and UL (LL: lower con dence limit; UL: upper con dence limit) indicate a signi cant increase of the adverse event risk upon aripiprazole compared with placebo in these studies ( Table 3).
Rates of discontinuation due to adverse events (ARR, NNH) In the case of brexpiprazole studies, the rate of discontinuation due to adverse events was signi cantly lower with active drug compared with placebo. In the aripiprazole studies, comparable rates of discontinuation due to adverse events were observed with active drug and placebo ( Table 4).

Incidences of selected adverse events (ARR, NNH)
Caused by different directions of the effects, adverse events could appear as 'risks' or 'bene ts' upon active drug as compared with placebo ( If the reduction of the adverse event schizophrenia with active drug versus placebo serves as predictor for the absolute increase of response, the higher accordance was observed under brexpiprazole compared with aripiprazole.

Discussion
Here, we present an indirect meta-analysis comparing the e cacy and safety of brexpiprazole with aripiprazole. We made indirect comparisons based on responder rates, incidences of adverse events, NNT, NNH, and LHH calculated based on data from all randomised controlled trials available to date. Previous analyses had included short-term studies mentioned in the respective product labeling [12].
NNT is a useful tool, showing effect sizes in an intuitive way as a proxy for the clinical relevance of e cacy results. In order to be meaningful, the outcome of interest must be carefully de ned and should be important to the patient [11]. Here, it means the number of patients that need to be treated with brexpiprazole or aripiprazole instead of placebo for one additional patient to experience a 30% reduction in PANSS total score (or to achieve a CGI-I value of 1 or 2). A 30% PANSS reduction can be considered a clinically meaningful difference in real-world practice, re ected by CGI-I scores of at least 3 [19].
For unwanted outcomes, the equivalent measure is NNH. Again, it is important to rst identify clinically meaningful negative outcomes before applying the concept of NNH [11]. Here, we calculated NNH for several negative outcomes of clinical interest. Correspondingly, our results are interpreted as the number of patients to be treated with brexpiprazole or aripiprazole instead of placebo for one additional patient to a) experience any adverse event; b) discontinue treatment because of an adverse event or c) experience a speci c adverse event common in schizophrenia treatment.
Both numbers can be related using the concept of LHH, the ratio of NNT and NNH. Here, we calculated LHH only for selected NNHs.
We used two different methods for calculating con dence limits by Wald and Wilson [14,15] that both led to the same results regarding signi cance/non-signi cance, showing the robustness of the analysis.
The NNTs found here for brexpiprazole and aripiprazole were both 11, indicating comparable e cacy of both drugs. Usually, an NNT of 10 or lower is considered clinically relevant [11], however, this may vary depending on the indication of interest. The relatively high NNTs found here may also re ect high rates of placebo response in the trials that provided the data source. In general, increasing placebo responses are a problem in schizophrenia trials [20,21]. In addition, the relatively high NNT for brexpiprazole could be caused by the inclusion of the subtherapeutic dose of 1 mg in this meta-analysis. Eventually, it is noted that regulatory authorities throughout the world have approved both brexpiprazole and aripiprazole for the treatment of schizophrenia showing that both drugs are e cacious for the treatment of this disease.
In his previous analysis, Citrome found smaller but comparable NNTs of 7 and 8 for brexpiprazole and aripiprazole, respectively [12]. In an exploratory, open label study, brexpiprazole showed a higher response rate (60.9%) than aripiprazole (48.5%), using the same response criteria that were used here [3]. Taken together, the available data indicate that the e cacy of brexpiprazole is at least comparable to that of aripiprazole.
The NNHs found in this study were mostly high and not signi cantly different from placebo. The number of patients needed to treat with brexpiprazole for any additional adverse event to occur was 27 and not signi cantly different from placebo, whereas for aripiprazole, it was 17 and signi cantly different from placebo. Considering individual adverse events, the NNHs found here are mostly in the high double-digit or triple-digit range, which are acceptable numbers for a good tolerability pro le [11]. Only the NNHs for akathisia, headache and nausea upon aripiprazole treatment were signi cantly different from placebo.
These data indicate better tolerability for brexpiprazole over aripiprazole.
In some instances, even negative NNHs were found, indicating a bene t rather than a harm for the active substance. This was the case for discontinuation because of adverse events upon brexpiprazole as well as several adverse events.
For some speci c unwanted outcomes, we additionally calculated LHH. We found that brexpiprazole treatment was 2.4 times more likely to lead to a 30% PANSS reduction than any adverse event, compared with a likelihood of 1.6 for aripiprazole, also indicating better overall tolerability for brexpiprazole. The advantages for brexpiprazole were even more pronounced for akathisia and nausea, with a bene t/risk-ratio of 12.7 and 6.2, respectively. Only for weight gain, aripiprazole had a better LHH than brexpiprazole, but brexpiprazole was still 2.7 times more likely to reduce PANSS by 30% than to cause a weight gain of at least 7%. For the adverse event of schizophrenia, both drugs yielded a bene t compared with placebo. In this case, the LHH gives a measure of how well the avoidance of schizophrenia as an adverse event correlates to a treatment response. This correlation was stronger for brexpiprazole.
In his 2015 analysis, Citrome found comparable NNH and LHH values [12]. However, these were based on fewer data, namely only studies from the product labelling.
In a separate analysis, Citrome focused on sedating and activating adverse events as they were reported in the product labelling of several antipsychotics. These types of events require special attention because they may be especially strongly associated with nonadherence [10]. In Citrome's analysis, brexpiprazole was one of only two antipsychotics out of 11 classi ed as neither activating or sedating [22]. This is in line with our results, indicating an advantage for brexpiprazole especially when considering akathisia.
In an exploratory, open-label trial with brexpiprazole and aripiprazole, a lower incidence of akathisia was found in the brexpiprazole group (9.4% vs. 21.2%) [3], which is in line with our results. Both drugs elicited similar overall rates of adverse events in this trial (57.8% vs. 63.6%).
The present meta-analysis indirectly compares brexpiprazole and aripiprazole using data from randomised clinical trials.
Here, we analysed data from short-term trials of maximally 6 weeks, and our nding should ideally be con rmed using data from long-term treatment, where treatment outcome may be de ned differently (e.g. avoiding relapse) and some adverse events may resolve over time.
If possible, the results found here should be con rmed by a randomised head-to-head trial of brexpiprazole and aripiprazole.

Conclusions
We used all published randomised, placebo-controlled short-term clinical trials studying brexpiprazole and aripiprazole to establish an indirect comparison using responder rates, incidences of adverse events, NNT, NNH, and LHH. We found that brexpiprazole and aripiprazole show comparable e cacy, while brexpiprazole shows advantages in terms of tolerability.     ns BREX = brexpiprazole; ARI = aripiprazole; ARR = absolute risk reduction; LL = lower 95% con dence limit; UL = upper 95% con dence limit; Sign. = Signi cance; X = p < 0.05; ns = not signi cant