Background

Schizophrenia is a severe mental illness with a prevalence of approximately 1% [1].

It is expensive to treat [2] and at least 30% of patients with this illness experience a poor long-term prognosis, characterised by residual psychotic symptoms [3], poor social functioning and a poor quality of life [4]. People with schizophrenia die on average 20 years earlier than individuals without this illness and this gap is widening [5]. One of the possible explanations for the differential mortality rate is that patients with schizophrenia have an increased risk of metabolic syndrome such as diabetes, obesity, hypertension and hypercholesterolemia [6].

Antipsychotic medication is the first line treatment for schizophrenia [7, 8]. Antipsychotic drugs are effective for the treatment of the core symptoms of schizophrenia, such as auditory hallucinations and delusions. These drugs can be divided in two main classes: first generation antipsychotics (FGA or typical antipsychotics) such as haloperidol and second-generation antipsychotics (SGA or atypical) such as risperidone, olanzapine and quetiapine. FGAs are dopamine antagonists acting on the three main pathways for this neurotransmitter, while SGAs have in general, an affinity for both dopamine and 5HT 2 receptors and are thought to be more selective towards the mesolimbic system [9]. Aripiprazole differs from other established atypical antipsychotics in being a partial agonist of dopamine receptors, and is considered by some authors to be sufficiently distinct to merit classification as a ‘third generation’ antipsychotic [10]. Schizophrenia is a chronic illness and most patients require lifelong treatment. Side-effects of antipsychotics accumulate over time. Long-term treatment with antipsychotic medication can increase the risk of diabetes, hypertension and hyperlipidaemia [11]. This state of metabolic change leading to an increased risk of cardiovascular and metabolic illness is known as metabolic syndrome. There are several definitions of metabolic syndrome [12]. The American Heart Association identifies six main components: abdominal obesity, dyslipidaemia, increased blood pressure, glucose intolerance and a pro-inflammatory and pro-thrombotic state [12]. In this overview, we use the term ‘metabolic syndrome’ to refer to the occurrence of hyperlipidaemia, diabetes or hypertension which are disorders commonly requiring treatment with medication.

Antipsychotic polypharmacy (APP) is defined as the simultaneous prescription of more than one antipsychotic medication. Patients may be prescribed more than one antipsychotic when they are deemed resistant to the effect of a single antipsychotic, they have more than one psychiatric diagnosis, the clinician is overlapping one medication while another is titrated, or an effective dose of one antipsychotic cannot be achieved because of lack of tolerance or side effects [8]. APP is not actively recommended in current clinical practice guidelines, yet it is extremely common in clinical practice, occurring in up to two-thirds of patients with psychosis [13,14,15,16].

APP is controversial because of a lack of clear evidence for treatment efficacy and the possible increased risk of side-effects over and above side-effects associated with anti-psychotic monotherapy [8, 17]. Research conducted mostly in secondary-care has produced conflicting evidence on the association between APP and metabolic syndrome, with some studies suggesting an increase and some a reduction in risk [13, 18, 19].

A scoping review conducted by this group suggested a need to collate the evidence from systematic reviews on the link between APP and metabolic syndrome to facilitate clinical decisions and stimulate new research in this area.

Aims of the study

To conduct a review of published systematic reviews to assess the current state of the evidence on the association between antipsychotic polypharmacy (APP) used for the management of schizophrenia and metabolic syndrome (defined as diabetes, hypertension, or hyperlipidaemia).

Methods

This review followed guidance published by the Centre for Reviews and Dissemination [20]. We wrote a protocol for the review with pre-specified objectives, eligibility criteria and review methods and registered it with PROSPERO (CRD42017054672) [21].

Inclusion criteria for reviews

We included systematic reviews that reported an investigation of the association between APP and metabolic syndrome (diabetes, hypertension or hyperlipidaemia) in adults with schizophrenia treated in any setting. To be inclusive, we considered any reviews and meta-analysis reports to be systematic reviews as long as they reported a systematic search when evaluating the association between APP and metabolic syndrome.

We excluded reviews that focussed on animal or laboratory studies only.

Identification and selection of reviews

Five databases (Medline, Embase, Cochrane, PsychInfo and Web of Science) were searched from inception until February 2017 to identify relevant reviews, using a systematic review filter. The search strategy for Medline is reported in the web appendix (Additional file 1). Searches were not limited by language, date, setting or publication status. An internet search using Google Scholar and screening reference lists of included publications were used to identify any additional relevant unpublished reviews. A systematic review filter was applied along with removal of duplicates to find relevant reviews. We did not search any regional databases.

Titles and abstracts of all citations from the search were independently screened by two reviewers and discrepant decisions resolved by discussion. Full text screening was then undertaken by two reviewers and disagreements resolved by a third reviewer with experience in psychopharmacology.

Data extraction and quality assessment of included reviews

Data extraction and quality assessment were conducted in duplicate and disagreements resolved by discussion. We used a standardized data extraction template and extracted the following items from included reviews: country of study; funding source; number of studies included in review; dates of search; setting (primary/secondary care); designs of reviewed studies; whether a meta-analysis was conducted; types of participants, intervention, comparator, outcome and definition of outcome; whether a formal quality/risk of bias assessment was conducted and its findings; and results or findings of the review.

We used the validated AMSTAR (A MeaSurement Tool to Assess systematic Reviews) checklist [22] for assessing reporting quality of the systematic reviews included.

Data synthesis

We carried out a narrative synthesis of the included systematic reviews with findings summarised in the text by outcome [20].

Analysis of subgroups or subsets

We planned to investigate the effects of the combination of different classes of antipsychotics, provided that sufficient papers reporting these effects were detected.

Assessment of certainty of evidence

We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework to assess the certainty of the evidence from the reviews and the strength of the recommendations [23] This approach identifies four elements which influence the certainty of the evidence: study design, study quality (risk of bias), consistency (between estimates of effect across reviews) and directness (i.e. applicability of participants, interventions, comparisons and outcomes of included reviews to the clinical question under review).

Results

The multiple database search located 12,321 citations. Complementary searching (see web appendix for details) resulted in 29 further unique citations. Removing 6341 duplicates and applying the systematic review filter resulted in 499 references. Thirty-seven of these were assessed in full text and 12 were included. One ongoing review was also identified. (See the PRISMA flow chart in Fig. 1). This review and the excluded full text reviews with reasons are reported in web appendix (Additional file 2). The most common reason for exclusion was lack of any metabolic syndrome outcomes (n = 13) followed by the review not addressing APP (n = 8), and opinion articles (n = 2). One review did not address schizophrenia.

Fig. 1
figure 1

PRISMA flow chart of review selection process

We also extracted data on BMI from the included reviews. These additional data are available in a web appendix (Additional file 3).

Description of included reviews

Twelve systematic reviews were included [24,25,26,27,28,29,30,31,32,33,34,35]. The numbers of primary studies included in the reviews ranged from 5 to 72 (median 46), although this number was not always reported. When this was the case we derived the figure from the tables and forest plots in the paper. One ongoing Cochrane review was also identified [36]. All except two reviews reported search dates. Most were from the date of inception of the databases or, in one case from 1985. The most recent searches were up until 2015 in two reviews.

The inclusion criteria in the reviews varied and methods used for inclusion were often not explicitly described. (Table 1). All reviews included diagnosed schizophrenia patient populations on antipsychotic therapy. APP was compared to antipsychotic monotherapy in six reviews. The other six did not specify the comparison. Outcomes of interest that were reported in the included reviews were lipid metabolism markers (8 reviews), diabetes or glucose metabolism markers (3 reviews), and hypertension (1 review).

Table 1 Characteristics of included systematic reviews on effects of APP

Only two included reviews provided definitions for the metabolic outcomes: Mizuno et al. 2014 [29] defined metabolic outcomes (fasting glucose, HbA1C, Total LDL (Low density lipoprotein) and HDL (High density lipoprotein) cholesterol) as ‘at endpoint as defined in individual studies’. Young et al. 2015 [25] defined metabolic syndrome as either the International Diabetes Federation (IDF) criteria for metabolic syndrome in adults and children, or National Cholesterol Education Programme (NCEP) criteria for metabolic syndrome. Similarly, dyslipidaemia was defined as at least one of the followings: total cholesterol > 200 mg/dL; HDL cholesterol 120 mg/dL; triglycerides ≥150 mg/dL; treatment for a known lipid disorder.

No reviews reported useable data for subgroup analyses.

Quality of included reviews

All the included reviews were considered at high risk of bias based on AMSTAR assessments (21). For one review (33) we could not locate a full text or a protocol and so assessments were based on abstract information alone.

In total, four reviews reported an a priori design, only two performed study selection and data extraction in duplicate, six performed a comprehensive literature search, and only five included both published and unpublished studies irrespective of language of publication.

None of the reviews provided a list of both included and excluded studies and only six provided characteristics of included studies as required by the AMSTAR criteria. Four assessed and documented the scientific quality of included studies and three of these used the scientific quality of the included studies appropriately in formulating conclusions.

Five reviews used appropriate methods to combine studies in a meta-analysis, but none provided a conflict of interest statement or funding sources of included studies, although most did report this for the authors. Most of the review authors were supported by one or more pharmaceutical companies.

Although none of the reviews were judged to be at low risk of bias, in our opinion reviews by Mizuno et al. and Anonymous et al. were more reliable because these used an a priori protocol, duplicate selection and extraction, and comprehensive searching without limits and also considered the study quality in their findings and conclusions. (See Table 2).

Table 2 Quality of included systematic reviews (AMSTAR)

Metabolic syndrome

Metabolic syndrome was an outcome in two reviews (Table 3). Young et al. [25] reported an association between APP and metabolic syndrome but did not provide either an estimate or a reference for the source of the data. Gallego and colleagues [33] found three studies showing increased risk of metabolic syndrome in APP (without specifying individual agents) but this association did not persist after controlling for sociodemographic and lifestyle factors.

Table 3 Metabolic effects of APP for schizophrenia reported in included reviews

Hyperlipidaemias

Seven reviews reported measures of lipid metabolism [25, 27,28,29, 32,33,34,35]. All reported that lipid profiles were better with APP particularly when aripiprazole was used as the augmentation drug (n = 6).

Diabetes/ glucose metabolism disorder

Three reviews addressed these outcomes. Gallego et al. [33] did not report any data or conclusions on glucose levels or diabetes but reported that APP was associated with diabetes. The other two reviews reported measures of glucose metabolism [29, 34] where one [29] found a small non-significant improvement in HbA1C levels in APP involving aripiprazole compared to monotherapy and the other [34] a non-significant decrease in glucose levels with APP involving aripiprazole.

Hypertension

Only one review provided information on hypertension [32] and reported that the effect of APP on hypertension was the same as monotherapy.

Certainty of evidence

Applying GRADE criteria to our included reviews we found that the evidence for all the outcomes was very low quality meaning that the evidence is very uncertain. (See Table 4).

Table 4 GRADE table for APP compared to Antipsychotic monotherapy for metabolic effects in schizophrenia

Discussion

Twelve studies fulfilled the criteria for this review of reviews. This is a large body of evidence indicating the degree of continued interest in the topic of antipsychotic polypharmacy in people with schizophrenia. In the context where there are strong opposing opinions about whether APP is harmful or beneficial, this extensive body of work shows researchers’ commitment to confirm through science what may be seen as an intuitive therapeutic approach.

Overall, our findings are in line with several of the included reviews – namely that there is insufficient evidence to clearly answer the questions on the efficacy and potential harms of APP.

In general, the quality of evidence was found to be low. This was in part likely to be due to limitations in primary studies included in the reviews. Most reviews did not include a synthesis of findings (either as a meta-analysis or a narrative synthesis) and only provided descriptions of included studies. However, where the study findings were pooled in reviews, the ranges and confidence intervals around the effect were wide, indicating uncertainty. Most reviews also did not include studies from the grey literature and did not assess publication bias. While five reviews compared APP to antipsychotic monotherapy explicitly, some of the reviews did not report their comparisons. Evidence on hypertension was limited (one review) and for all other outcomes the findings were heterogeneous across reviews.

With the exception of combinations involving aripiprazole, it was not possible to ascertain whether some combinations of antipsychotics were less harmful than others, or if associations of first generation with second generation antipsychotics had a differential effect on the selected outcomes. Six reviews [27,28,29, 32,33,34] suggested a protective effect of antipsychotic combinations which included aripiprazole for dyslipidaemia and glucose metabolism, compared to other combinations and/or monotherapy. Given the quality of evidence it would be premature to conclude that in the presence of another antipsychotic, aripiprazole protects against the risk of metabolic syndrome. Yet, the findings raise an intriguing hypothesis that warrants further investigation into why risks associated with combinations containing aripiprazole differ from those containing other commonly used atypical antipsychotics. The relationship between an antipsychotic drug’s mechanism or receptor binding profile and metabolic syndrome is thought to be very complex, and likely to be multifactorial, perhaps involving interplay of dopamine, histamine, orexigenic neuropeptides, adrenergic and muscarinic receptors, and failed glucose homeostasis, with other risk factors [37].

While aripiprazole differs from other established atypical antipsychotics by being a partial agonist at the dopamine D2 receptor, the possibility of reduced risk of metabolic syndrome in combination treatment may relate to its action on the serotonin system rather than the dopamine system. One area which has received much attention for atypical antipsychotics known to carry a relatively high risk of metabolic complications, such as clozapine and olanzapine, is their high affinity for serotonin 5-HT2C receptors. It has been postulated [38] that while aripiprazole acts on orexin and histamine systems that might be protective, its key pharmacological property may be its partial agonist activity at the serotonin 5-HT1A receptor which may counterbalance the problematic effects exerted through the 5-HT2C receptor. This benefit would apply not only in offsetting its own actions on 5-HT2C, but also those of co-prescribed antipsychotics with high affinity for this receptor.

No synthesis of effects considering dose equivalents of antipsychotics was found. The use of dose equivalents allows for comparison of dosage between different drugs. It is possible that APP is only harmful relative to monotherapy when the final equivalent dose is excessive and not when it is kept within established therapeutic ranges [13]. Some reviews [28, 31,32,33, 35] addressed APP which included clozapine. Clozapine can cause leukopenia [39] and should only be used after an ineffective trial with two other antipsychotics [8]. Furthermore, people treated with clozapine combinations have more chronic illness than other patients because they are considered to be treatment-resistant, and this factor may confound the occurrence of poorer physical health outcomes within this sub-group.

Limitations

We searched multiple databases and employed complementary approaches to ensure no relevant published reviews were missed. The relatively large number of reviews detected indicates that this is an active research area. Our inclusion and search criteria were broad and therefore include heterogeneous patient populations, APP interventions (combinations), comparators and study designs. While this makes our findings generalizable, these should be interpreted with caution due to the limited data and quality of the included reviews. For example, the lack of uniform comparisons with monotherapy (specific AP drug, any AP drug, atypical or typical AP) added to problems interpreting the findings of the same outcomes across reviews. Including a broad range of systematic reviews provides a more complete picture than a single systematic review and also allows examination of any conflicting findings across this evidence base [40].

There was no evidence from the reviews on combinations involving other atypical drugs that might theoretically present a lower risk for metabolic syndrome, such as the second generation antipsychotics ziprasidone, asenapine, lurasidone and the recently introduced ‘third generation antipsychotic’ brexpiprazole. Most of these medications are relatively new in the market and more time may be needed until relevant reviews reach the publication stage. Since we did not search for primary studies we are not able to say with certainty whether this is an overlooked area in systematic reviews or in empirical research. This gap in evidence needs addressing in future to enable robust comparisons of atypical drugs available today.

Primary-care is an important part of the care pathway for schizophrenia where most stable patients are managed [41], although most prescribing of both antipsychotic monotherapy and poly-therapy is initiated in secondary care. None of the included reviews reported results separately by setting (primary and secondary). This an important gap in the evidence. Primary-care data has been used recently in long term follow up studies of psychosis treatment [42,43,44,45,46,47,48,49], however, none addressed APP and its consequences.

The evidence appeared to be of low quality for all outcomes (Table 4), partly because of the high risk of methodological and publication bias and also because the effects of APP were variable across reviews.

Although we searched for grey literature we did not request unpublished reviews or missing data from authors, which limits the comprehensiveness of our review especially if negative or inconclusive reviews are not published [50, 51]. Furthermore, we did not t search regional databases and therefore some potentially eligible reviews that are not indexed in major international databases may have been missed. However, considering the limitations in included reviews we consider that unpublished or missed reviews, if any, would also suffer from the same limitations due to the low quality primary studies. Due to the same limitations we did not pool results of the reviews statistically [52]. The lack of summarised or individual study data in full, prevented us from carrying out any additional analyses. In addition, the synthesis of review level data is complex and requires careful consideration of overlap of primary studies included in several reviews and this is not always possible or practical [40]. Therefore, we refrained from re-analysis and only relied on a narrative synthesis to derive our conclusions.

Based on the current evidence, we cannot definitively conclude that APP increases the risk for metabolic syndrome in schizophrenia, nor that it is safe, relative to antipsychotic monotherapy. It is imperative that this question is investigated in a robust prospective study before any key clinical recommendations are made. Future empirical studies should include sufficiently powered samples and adequate follow up periods with clearly defined comparison groups and outcomes to identify at risk subgroups and whether safer regimens for schizophrenia exist. Until better evidence is available, we advise that clinicians should err on the side of caution when considering prescribing APP.