Introduction
In the quest to increase our understanding of schizophrenia spectrum disorders, a vast and ever-expanding body of scientific literature documents the pursuit of this cause. However, robust evidence on schizophrenia remains elusive; many findings are not replicated, and conflicting results are common. In 2008, in an attempt to condense and clarify the state of the available evidence, a series of papers titled ‘Just the Facts’ assessed a selected sample of evidence from systematic reviews, meta-analyses and primary studies according to three criteria: reproducibility, durability of the findings and whether they were primary to schizophrenia (Keshavan et al. Reference Keshavan, Tandon, Boutros and Nasrallah2008; Tandon et al. Reference Tandon, Keshavan and Nasrallah2008a , Reference Tandon, Keshavan and Nasrallah b , Reference Tandon, Nasrallah and Keshavan2009, Reference Tandon, Nasrallah and Keshavan2010). Similarly, in 2009 an informal consortium (the Minnesota Consensus Group) published a summary of expert opinion outlining the key facets contributing to our understanding of schizophrenia (MacDonald & Schulz, Reference MacDonald and Schulz2009). These papers provided a valuable and comprehensive collation of the literature, but focused on selected areas of research considered most relevant. This approach, while useful for informing priority research directions, does not take a systematic approach to inclusion and objective quality assessment of the available literature, and thus risks overlooking some areas of potential importance that may not have seemed sufficiently salient.
Given the broad state of current knowledge of schizophrenia, encompassing domains of clinical phenomenology, aetiology or pathogenesis, epidemiology and intervention efficacy, it appears imperative to undertake a systematic approach to synthesising the literature to avoid any selection biases. While evidence published in the peer-reviewed literature in these research domains is more reliable than information potentially gained by searching the Internet or through personal accounts, it is limited by the quality of the studies and the quality of the data. In this context, the Schizophrenia Library was developed to provide a free, objective and comprehensive ongoing appraisal of the schizophrenia literature, highlighting key areas where evidence is particularly strong, and areas where further attention is needed (www.schizophreniaresearch.org.au/library). To ensure that the most robust evidence is presented, the Library includes only well-conducted systematic reviews, many of which incorporate meta-analyses, and employs a relatively unbiased approach through rigorous systematic methodology, highly sensitive inclusion criteria, and objective quality assessment strategies.
This method does not look at results of individual studies in isolation, but at the body of evidence gained by numbers of studies assessing the same outcomes in the same population. Critics of systematic reviews and meta-analyses suggest that they are prone to publication bias as well as other biases that may be inherent in the primary studies. They have been criticized for ignoring single important or ‘landmark’ studies that may be highly powered and based on large unbiased samples (however uncommon these may be), with meta-analyses additionally criticized for undue leniency in statistically combining effect sizes of different samples and outcomes (that is, combining ‘oranges with apples’). However, a well-conducted systematic review has the ability to make robust, generalizable conclusions exceeding those usually possible from a single primary study, with good meta-analyses potentially providing the closest estimate of the true effect size (Button et al. Reference Button, Ioannidis, Mokrysz, Nosek, Flint, Robinson and Munafò2013). A well-conducted systematic review will develop research questions and inclusion criteria a priori (potentially even publish a review protocol), determine the presence of publication bias statistically, and investigate reasons for heterogeneity via subgroup analyses of differing study quality, samples, measures and outcomes, while also considering potential sources of bias originating from the primary studies. Subgroup analyses can shine light on important reasons for differing effects of treatments, for example, across differing samples and measures. It can also help guide the design of future primary research (Borenstein et al. Reference Borenstein, Hedges, Higgins and Rothstein2009). Our approach uses qualitative assessment of the transparency with which systematic reviews report on the primary literature and, importantly, we determine the reliability of the evidence in question, using a common platform to explore the comparative strength of evidence spanning the broad research domains.
Knowledge gained solely from a compilation and critical appraisal of systematic reviews and meta-analyses is not the only pathway to the truth about schizophrenia. There is also expert consensus, selective reviews and the methods adopted by Keshavan, Tandon and colleagues (Keshavan et al. Reference Keshavan, Tandon, Boutros and Nasrallah2008; Tandon et al. Reference Tandon, Keshavan and Nasrallah2008a , Reference Tandon, Keshavan and Nasrallah b , Reference Tandon, Nasrallah and Keshavan2009, Reference Tandon, Nasrallah and Keshavan2010), which entailed a combination of all three. We were interested in what sort of picture we would get of schizophrenia if we relied solely on a rigorous evaluation of published systematic reviews. While this would avoid many sources of bias, by capturing only what has been subjected to systematic review would ignore potentially true findings in the literature that had not been subjected to such reviews. For this reason, the Schizophrenia Library highlights topics that have not yet been reviewed systematically to encourage further research and reviews in those areas.
We therefore set about summarising the evidence that is currently available in the Schizophrenia Library, in an attempt to clarify what we currently know about schizophrenia and how confident we can be about this knowledge. What we can say that we know with confidence about schizophrenia is drawn from high-quality evidence; that is, evidence from systematic reviews containing large samples and robust results. What we think we know about schizophrenia is drawn from moderate-quality evidence; that is, systematic reviews with limitations in the available data including imprecision, inconsistency, smaller samples or observational study designs that may be prone to bias.
Method
Inclusion criteria and search strategy
Included in the Library are systematic reviews of studies of patients with schizophrenia and related (e.g. schizoaffective) disorders that are published in full text in English from the year 2000. Systematic reviews, by definition, contain an explicit method section with details of inclusion/exclusion criteria and search strategy. We excluded treatment guidelines, overviews, non-systematic reviews, and those systematic reviews with a high likelihood of reporting bias in that they addressed fewer than 33% of items according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Moher et al. Reference Moher, Liberati, Tetzlaff and Altman2009). Reviews were identified by searching the databases Medline, EMBASE, CINAHL, Current Contents and PsycINFO using the search strategy: exp Schizophrenia/ OR schizophreni$.tw OR exp Psychotic Disorders/ OR schizo$.tw AND review.pt and medline.tw OR meta analysis.pt OR systematic$.tw and (review$ or overview$).tw OR meta?analy$.tw OR meta analy$.tw AND Limit to yr = 2000-current. Hand searching reference lists and advance alerts was also undertaken. The decision to include or exclude reviews was conducted in duplicate, with any disagreements settled by discussion. The information contained in this paper is drawn solely from the Schizophrenia Library, but due to the large volume of reviews contained in it, we have cited in the tables only the most current reviews reporting the highest-quality evidence available for each topic. The latest Schizophrenia Library search update was conducted in February 2013, with current contents approaching 1500 reviews categorized into 450 topics. The Schizophrenia Library website (www.schizophreniaresearch.org.au/library) contains further details of all reviews meeting inclusion criteria. Reviews of genetic studies have not been included in the Library primarily because an existing website (www.szgene.org) has systematically collated the evidence from meta-analyses for genetic associations with schizophrenia (Allen et al. Reference Allen, Bagade, McQueen, Ioannidis, Kavvoura, Khoury, Tanzi and Bertram2008).
Quality assessment and data extraction
Quality assessment and data extraction were conducted in duplicate by two of the authors (S.L.M. and A.M.S.) with training in conducting systematic reviews and meta-analyses. Any disagreements were settled by discussion. The reporting transparency of all systematic reviews was assessed using the PRISMA (Moher et al. Reference Moher, Liberati, Tetzlaff and Altman2009) statement checklist, and the assessments of included reviews are available on the Library website. Systematic reviews with a high risk of reporting bias were excluded. The evidence contained in the reviews was assessed using a strategy adapted from the Grading of Recommendations Assessment, Development and Evaluation (GRADE) Working Group approach (GRADE Working Group, 2004; Guyatt et al. Reference Guyatt, Oxman, Akl, Kunz, Vist, Brozek, Norris, Falck-Ytter, Glasziou, deBeer, Jaeschke, Rind, Meerpohl, Dahm and Schünemann2011a ). GRADE was developed primarily to appraise treatment or management strategies, using evidence from both randomized controlled trials and observational studies; however, in the absence of robust quality assessment guidelines for other pooled evidence, we have applied GRADE criteria to derive a broad indication of the strength and quality of all the results. In this approach, evidence from randomized controlled trials is considered to be high quality, but may be downgraded to moderate or low quality in the presence of: (1) inconsistency of results (statistically significant heterogeneity of results across studies, assessed using I 2 which represents the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error, and/or using the Q test for within-group heterogeneity) (Guyatt et al. Reference Guyatt, Oxman, Kunz, Woodcock, Brozek, Helfand, Alonso-Coello, Glasziou, Jaeschke, Akl, Norris, Vist, Dahm, Shukla, Higgins, Falck-Ytter and Schünemann2011b ); (2) imprecise (wide) confidence intervals (>0.5 in either direction for continuous measures, and >0.25 for risk/odds ratios) (GRADEpro, version 32 for Windows, (http://tech.cochrane.org/gradepro); Guyatt et al. Reference Guyatt, Oxman, Kunz, Brozek, Alonso-Coello, Rind, Devereaux, Montori, Freyschuss, Vist, Jaeschke, Williams, Murad, Sinclair, Falck-Ytter, Meerpohl, Whittington, Thorlund, Andrews and Schünemann2011c ); (3) estimated effect sizes from indirect comparisons, populations or interventions (Guyatt et al. Reference Guyatt, Oxman, Kunz, Woodcock, Brozek, Helfand, Alonso-Coello, Falck-Ytter, Jaeschke, Vist, Akl, Post, Norris, Meerpohl, Shukla, Nasser and Schünemann2011d ); and/or (4) a small number of studies reviewed that include only small samples (Cochrane Collaboration, 2008; Guyatt et al. Reference Guyatt, Oxman, Akl, Kunz, Vist, Brozek, Norris, Falck-Ytter, Glasziou, deBeer, Jaeschke, Rind, Meerpohl, Dahm and Schünemann2011a ). Conversely, evidence from observational studies is intrinsically considered low quality due to the possible effects of confounding factors, but this evidence may be upgraded to moderate or high quality if the data are consistent or precise, if effect sizes or study samples are large, or if there is a dose-dependent response (Guyatt et al. Reference Guyatt, Oxman, Akl, Kunz, Vist, Brozek, Norris, Falck-Ytter, Glasziou, deBeer, Jaeschke, Rind, Meerpohl, Dahm and Schünemann2011a ). GRADE recommendations of the assessment of potential publication bias, and the extent to which uncontrolled confounders may influence the observed effects were not conducted because these issues are not reported in most reviews. A low-quality rating does not necessarily imply poor-quality research, but is intended to highlight the need for more targeted research in these areas, conducted in accordance with transparent reporting guidelines.
Effect sizes are quantified such that statistically significant (p < 0.05) correlation coefficients and standardized mean differences of about 0.20 represent a small effect, about 0.50 a medium effect, and about 0.80 or higher a large effect (Cohen, Reference Cohen1988). Statistically significant risk and odds ratios around 1.00 represent a small effect, effects ⩾2.00 or ⩽0.50 represent a medium effect, and effects ⩾5.00 or ⩽0.20 represent a large effect (Rosenthal, Reference Rosenthal1996). Significant effect sizes that are unable to be quantified using these standardized guidelines are allocated to an ‘unclear effect sizes’ category (e.g. incidence rates, prevalence rates, non-standardized mean differences). For this paper, we have included only evidence reporting significant differences between groups for the primary outcome of interest in each review for each comparison in question (details of all results are available on the Library website), and we have grouped this evidence into three domains: treatments, physical or clinical features, and epidemiology. Tables 1–3 describe the significant findings for each domain. Note that all treatments identified here, apart from comparisons of antipsychotics with placebo, are adjunctive to ongoing antipsychotic medication. The evidence with small or unclear effect sizes is too numerous, and some may be insignificant, to warrant presentation here other than by inspection of Tables 1–3.
Results
High-quality evidence
Large effect sizes
With respect to treatments (Table 1), large effects were found for social skills training in improving social interactions, and combined pharmaceutical and psychosocial treatment programmes for reducing symptoms and relapse rates. Among the physical and clinical features there were large effects for increased striatal presynaptic dopamine function and poor cognitive functioning on several measures (Table 2). There were no large effect sizes in the area of epidemiology (Table 3).
Table 1. Treatments
rTMS, Repetitive transcranial magnetic stimulation; NMDA, N-methyl-d-aspartate; SMD, standardized mean difference; RR, risk ratio; OR, odds ratio.
a Interpretation of effect sizes. Large effect sizes = SMD ⩾0.80, RR/OR ⩾5.00 or ⩽0.20. Medium effect sizes = SMD 0.20 to 0.70, RR/OR 2.00 to 4.00 or 0.30 to 0.50. Small effect sizes = SMD <0.20, RR/OR <2.00 (Cohen, Reference Cohen1988; Rosenthal, Reference Rosenthal1996). Except where otherwise indicated, patients receiving interventions are compared with patients receiving standard care.
Table 2. Physical or clinical features
IQ, Intelligence quotient; BDNF, brain-derived neurotrophic factor; IFN, interferon; TGF, transforming growth factor; TNF, tumour necrosis factor; IL, interleukin; IL-IRA, interleukin receptor; sIL-2R, soluble receptor for interleukin 2; DSM, Diagnostic and Statistical Manual of Mental Disorders; ICD, International Classification of Diseases; SMD, standardized mean difference; RR, risk ratio; OR, odds ratio.
a Interpretation of effect sizes. Large effect sizes = SMD ⩾0.80, RR/OR ⩾5.00 or ⩽0.20. Medium effect sizes = SMD 0.20 to 0.70, RR/OR 2.00 to 4.00 or 0.30 to 0.50. Small effect sizes = SMD < 0.20, RR/OR <2.00 (Cohen, Reference Cohen1988; Rosenthal, Reference Rosenthal1996). Unclear effect sizes = unable to be quantified using standardized guidelines. Except where otherwise indicated, schizophrenia patients or their relatives are being compared with healthy controls.
Table 3. Epidemiology
IQ, Intelligence quotient; TNF, tumour necrosis factor; IL, interleukin; SMD, standardized mean difference; RR, risk ratio; OR, odds ratio.
a Interpretation of effect sizes. Large effect sizes = SMD ⩾0.80, RR/OR ⩾5.00 or ⩽0.20. Medium effect sizes = SMD 0.20 to 0.70, RR/OR 2.00 to 4.00 or 0.30 to 0.50. Small effect sizes = SMD <0.20, RR/OR < 2.00 (Cohen, Reference Cohen1988; Rosenthal, Reference Rosenthal1996). Unclear effect sizes = unable to be quantified using standardized guidelines. Except where otherwise indicated, schizophrenia patients are being compared with healthy controls.
Medium effect sizes
Many treatments for schizophrenia fall into this grouping. Antipsychotic drugs compared with placebo have a medium effect size in reducing symptoms and relapse rates as reported in several different meta-analyses, and second-generation antipsychotics, particularly risperidone and olanzapine, cause fewer extrapyramidal side effects than first-generation drugs, particularly haloperidol, in first-episode psychosis. Low-frequency repetitive transcranial magnetic stimulation (rTMS) reduces auditory hallucinations, and high-frequency rTMS reduces negative symptoms, as do adjunctive antidepressants. Oestrogen therapy in women reduces positive and negative symptom severity, and Ginkgo biloba reduces positive symptoms. Among psychosocial treatments, cognitive behavioural therapy reduces symptoms, especially positive symptoms, but does not appear to differ from other psychosocial treatments in this regard, while psychoeducation, intensive case management, cognitive remediation, vocational rehabilitation and integrated psychological therapy all have medium-size effects on several outcome variables as detailed in Table 1.
Physical features of medium effect size include: increased volume of basal ganglia, lateral and third ventricles, and frequency of cavum septum pellucidum; decreased whole brain volume (grey and white matter); decreased grey matter volume in several brain regions as described in Table 2; decreased levels of N-acetyl aspartate, and a variety of changes in the N400 wave on recordings of event-related potentials. In relation to clinical features, medium effect sizes are found for a range of cognitive deficits in first-degree relatives (Table 2); theory of mind deficits in clinical high-risk samples; and the relationship of neurocognition and social cognition to functional outcome. Neutral stimuli evoke greater aversive emotion and arousal in patients, co-morbid substance use disorders are associated with fewer negative symptoms, and former cannabis-using patients have better cognitive performance than those without former cannabis use.
In the area of epidemiology, there are medium effect sizes for higher rates of cannabis use during adolescence in patients, and an earlier age of onset in cannabis-using patients. High rates of current and lifetime cannabis use are recorded in patients. From a developmental perspective there are medium effect sizes for delays in walking and poorer motor skills in childhood, and lower intelligence quotient (IQ) scores in childhood and adolescence.
Moderate-quality evidence
Large effect sizes
There are no treatments that fall into this category of evidence. Among physical features there are large effects for increased serum S100B levels in patients, increased markers of infection by several organisms including human endogenous retroviruses, Chlamydophilia pneumoniae, Chlamydophilia psittaci, Toxocara and Toxoplasma gondii in patients, and reduced sensory gating (increased P50 ratio) in both patients and their first-degree relatives. Large effect sizes are also found for reduced mismatch negativity (effect size increasing with duration of illness) and minor physical anomalies. In relation to other physical and clinical features, a wide range of cognitive deficits have been confirmed in several studies as shown in Table 2, and antipsychotic-free patients have abnormal sleep patterns on several parameters (Table 2). Patients and first-degree relatives have increased neurological soft signs, and patients with cocaine use disorder have increased extrapyramidal symptoms.
In epidemiological research there are large effects for increased exposure to Toxoplasma gondii antibodies and maternal diabetes in utero, for low birth weight and for experience of childhood adversities. There are similar effect sizes for rates of schizophrenia in black Caribbean and African migrant groups and their descendants, particularly those living in white communities. Patients have increased mortality due to a variety of causes (Table 3), increased risk of visual impairment and higher rates of tobacco smoking.
Medium effect sizes
In the area of treatment there are medium effect sizes for adjunctive non-steroidal anti-inflammatory drugs (especially for reducing positive symptoms), crisis intervention in reducing family disruption, and clozapine for treatment resistance compared with ‘typical’ antipsychotics. Similar-level effects are found for social skills therapy for improving general psychopathology and relapse rates, Ginkgo biloba for improving negative symptoms, music therapy for improving global state, and family interventions for reducing relapse rates and improving functioning.
Among physical features there are medium effect sizes for reduced brain-derived neurotrophic factor levels in serum, regardless of medication status; increased markers for Borna disease virus; increases in a variety of immune system molecules, some of which are normalized by antipsychotic drugs (Table 2); and reduced P300 amplitude and increased P300 latency in patients and their first-degree relatives. For clinical features, first-episode psychosis patients have poor processing speed; lower levels of insight are associated with increased severity of positive and negative symptoms (but not depressive symptoms), and people at high genetic and clinical risk for psychosis show impaired olfactory identification. Patients show a reduced pain response with increased sensory threshold.
In epidemiological studies there are medium effect sizes for exposure to urbanicity, emergency caesarean section, congenital malformations and uterine atony. There is increased reporting of childhood central nervous system (CNS) viral infections, and increased rates of schizophrenia in first- and second-generation immigrants (especially coming from developing countries). Internalized stigma is related to several adverse subjective and objective phenomena (Table 3). Lifetime risk of suicide is 1.8% (5.6% in the earlier stages of illness), and several clinical factors are significantly associated with suicide (Table 3). Increased symptom severity is associated with decreased objective and subjective quality of life. Longer duration of untreated psychosis is related to poorer clinical and social outcomes and poorer treatment response. Rates of relapse of positive symptoms are 28% at 1 year and up to 54% at 3 years, and are associated with substance use, poor treatment adherence, high levels of critical family comments and poor pre-morbid adjustment. Physicians are the most likely point of first contact, but the most common source of referral to mental health care is the emergency services.
Discussion
The summary of evidence described above is unique in that it presents a thorough overview of our current knowledge about schizophrenia while providing an objective quality assessment of this knowledge. It was expected that our methodology would yield some results that are ‘well known or somewhat trivial’ as well as others that are not so well known, unexpected and potentially important.
However, knowledge is not the same as comprehension or understanding, and this knowledge base requires careful reflection if the field is to move closer towards understanding. In doing so it is necessary to consider the sources of this knowledge, to appraise the level of certainty warranted by research findings, and carefully weigh the evidence in the face of a preponderance of medium to small effect sizes and lower-quality evidence. In biomedical research, bias is the primary source of false-positive findings (Ioannidis, Reference Ioannidis2005). The consequences of bias are high rates of refutation, conflicting or contradictory results, and failure of replication. More highly powered, large studies can help reduce the occurrence of such outcomes, and low-bias meta-analyses may also help in approximating true effects (Ioannidis, Reference Ioannidis2005; Button et al. Reference Button, Ioannidis, Mokrysz, Nosek, Flint, Robinson and Munafò2013). The latter is the approach we have taken here, a critical appraisal of well-conducted systematic reviews usually based on well-designed studies.
So, what are the facts? On the whole, the findings are not surprising, although it is very striking that there are solid positive findings, with large and medium effect sizes, for numerous adjunctive psychosocial treatments. The magnitude of these treatment effects exceeds the medium effect size obtained for antipsychotics compared with placebo. Adjunctive biomedical treatments, including rTMS, oestrogen, Ginkgo biloba, antidepressants and non-steroidal anti-inflammatories, also provide clear benefits in the medium effect size range. This is not to say that these adjunctive treatments are as effective as antipsychotics on their own, but when combined with antipsychotics, on average, they have an additional large- or medium-size effect. These are robust findings, some of which were quite unexpected (e.g. Ginkgo biloba), but all have passed the same objective, quality assessment tests as the expected findings such as those on the effectiveness of antipsychotic medications over placebo. Also striking are various markers of infection, inflammation or altered immunological parameters that parallel the findings of increased exposure to infection in utero and during childhood that are reported in epidemiological studies. The potential aetiological or pathophysiological role of these processes in schizophrenia can clearly not be ignored. We can be certain that patients have relatively poor cognitive functioning (as do first-degree relatives, but in comparatively attenuated form) and subtle, but diverse and widely replicated structural brain changes, while several electrophysiological measures are certainly altered, namely P50, P300, N400, and mismatch negativity. The presence of minor physical anomalies, neurological soft signs (also in first-degree relatives), and certain sensory changes (reduced olfactory and pain sensitivity), complete the picture of a disorder that irrefutably involves widespread neural dysfunction to a degree apparently not shared by other major psychiatric disorders, with a causal role for altered inflammatory and/or immunological processes that may be infective in origin, at least in some cases, yet is responsive to biomedical and psychosocial therapeutic influences of a wide variety of types.
Second, the epidemiological data furnish evidence for the role of developmental factors in schizophrenia by identifying medium to large effects for pregnancy and birth complications, prenatal Toxoplasma gondii exposure, developmental motor delays, lower child and adolescent (pre-morbid) IQ, childhood CNS viral infections, and childhood adversities. In addition, epidemiological research confirms the effects of immigration (first and second generation), membership of racial and ethnic immigrant groups, and urbanicity on risk for schizophrenia. The causal influence of pre-morbid cannabis use and its role in bringing forward the onset of schizophrenia is also confirmed, yet paradoxically, cannabis use is associated with better cognitive performance in clinical studies. Epidemiological studies also identify incidence and prevalence figures, increased mortality and suicide rates, and confirm that longer duration of untreated psychosis is associated with poorer outcome. Overall, the epidemiological findings suggest a combination of intrinsic (genetic?) and extrinsic (environmental) factors operating during development somewhere between conception and adolescence.
Does this knowledge about schizophrenia increase our understanding of this condition? We would assert that schizophrenia is a disorder of widespread neural dysfunction accompanied by widespread psychological effects that responds moderately well to psychosocial and biomedical therapies. It has an onset and course shaped by a combination of intrinsic and extrinsic factors, that involve anomalies evolving in the early stages of human development and entail, in a proportion of cases, altered inflammatory and/or immunological processes, possibly infective in origin. If these conclusions are true and constitute an advance in our understanding of schizophrenia, it is a very modest advance indeed.
Limitations
We cannot control for bias at the study level, nor at the review level, which may have entailed manipulation in the analyses or in selective or distorted reporting of results. Further, summary effects from meta-analyses may be modestly inflated (Button et al. Reference Button, Ioannidis, Mokrysz, Nosek, Flint, Robinson and Munafò2013). Some reviews may be prone to sampling bias; for example, 90% of the sample included in the Ginkgo biloba review were from China (Singh et al. Reference Singh, Singh, Kar and Chan2010b ). We have attempted to address potential bias by including only systematic reviews that achieved a satisfactory level of reporting transparency. As we have not included unpublished reviews, reviews published only in abstract form, and reviews published in languages other than English, the results may be subject to publication and language bias. While all attempts were made to be objective during quality assessment, some subjectivity is unavoidable, particularly for reviews not reporting assessment criteria.
In order to facilitate comparison across a range of disciplines, detail has been minimized and represented here solely in terms of effect sizes and standardized quality ratings. This has the benefit of allowing appraisal of research from diverse disciplines in order to highlight findings of substance; however, it has the limitation of losing a vast amount of additional information. Specifics of sample characteristics, secondary comparisons and outcomes, findings that are not statistically significant but may be clinically significant, discipline-specific jargon, and even citations that laid the groundwork for the more recent reviews were necessarily omitted in order to present only the key outcomes relating to the most up-to-date research in the field. Moreover, the exclusion from the Schizophrenia Library those reviews assessing the associations of genetic risk factors for schizophrenia precludes any consideration of this important literature in this summary of evidence; however, the SZGene database (www.szgene.org) comprehensively collates the extant literature for schizophrenia risk associated with individual polymorphisms. Finally, replicated and robust studies that have not been the subject of a systematic review are not taken into account here, while ‘fashionable’ topics may be more readily reviewed, potentially biasing our results.
Conclusion
We have presented a succinct summary of the vast available evidence encompassing domains of clinical phenomenology, biology, epidemiology and therapeutics in relation to schizophrenia. We have summarized the more robust evidence and concluded that while our knowledge of schizophrenia is very substantial, our understanding of it remains limited.
Declaration of Interest
None.
