Treatment resistant schizophrenia: a comprehensive survey of randomised controlled trials

Schizophrenia is a common serious mental health condition affecting approximately 1% of the population [1]. The course of illness is variable with a minority fully recovering from an initial episode whilst most with have a relapsing remitting course [2]. Antipsychotic medication has been part of the standard care of schizophrenia since the introduction of chlorpromazine in 1952. Up to 60% of people with schizophrenia will respond to antipsychotic medication but about 1 in 3 people have an illness that is "treatment resistant" [3] and this group suffer, cause problems to their carers and society, and are enormously resource-intensive - often for decades [4].

Treatment resistant schizophrenia (TRS) has not been consistency defined within the literature [5]. In a landmark randomised trial John Kane and colleagues investigated the effects of clozapine compared with chlorpromazine for people with TRS [6]. In this study TRS was defined as "at least three periods of treatment with antipsychotics from at least two different classes at adequate doses for an adequate period time with no relief and no period of good functioning over the last five years". Kane's trial led to clozapine's reintroduction to common use and the criteria used to define TRS in this study remain some of the most cited in randomised trials concerning TRS [7].

Clozapine's use is restricted partly due to serious adverse effects including blood disorders and cardiac toxicity. Many newer agents or approaches have been the focus of TRS studies with the hope of having similar efficacy to clozapine but better tolerability and safety - although, on average, other medications seem less effective than clozapine [8,9]. Whilst clozapine is the only licensed drug for TRS, clinicians do often try other approaches such as poly-pharmacy and high dose prescribing [10] before its prescription with an average 47.7 months delay in use of clozapine [11]. Up to 30% of people whose illness has been designated as resistant to treatment will have an inadequate response to clozapine and more interventions may be indicated [12]. Interventions in TRS include not only pharmacological interventions but also psychological and 'physical' interventions such as electroconvulsive therapy (ECT). This important subgroup of people is an active area of evaluative research and we know of no overview and taxonomy.

Small trials can lead to imprecise results where there are wide 95% confidence intervals (CIs). Wide CIs can lead to difficulties in making clinical recommendations especially where these cross the line of no effect suggesting a treatment is potentially ineffective. As study size increases so does precision of results leading to a more accurate estimate of the effect size. It has also been shown that trials containing small numbers of randomised participants can lead to treatment effects being overestimated and that these can decrease as more data becomes available [13]. This may be due to a variety of factors such as low methodological quality but may also be due to a failure of the randomisation to balance differences between control and intervention groups in terms of prognosis due to random error. The GRADE recommendations take account of this and suggest rating down recommendations based on imprecise results or studies containing small numbers of participants [14].

The Cochrane Schizophrenia Group (CSG) register was searched in October 2012. This register is compiled by methodical searches of 70 different biomedical databases including BIOSIS, CINAHL, Dissertation abstracts, EMBASE, LILACS, MEDLINE, PSYNDEX, PsycINFO, RUSSMED, Sociofile and is supplemented with hand searching of relevant journals and numerous conference proceedings. This strategy attempts to reduce the risk of publication bias. A detailed account of the group's search strategy is available [15].

The search terms used were: *Treatment?resist*, *non?respon*, *non?remission*, *non?remitter*, *Therapy?Resist*, *treatment?refract*, *medication?resistan*, *drug?resistan*.

All abstracts returned by this search were then inspected for relevance. If there was any doubt from the abstract alone, the full paper was obtained for clarification. Studies not concerning therapeutic interventions, not mentioning treatment resistant schizophrenia, or that were not randomised were excluded. Given the lack of consensus criteria for diagnosing TRS a broad definition was adopted whereby a paper was included if the study population was deemed to be treatment resistant by the authors. We also included papers where the authors did not explicitly mention treatment resistance but there was evidence that all participants had tried pharmacological interventions but these had not been effective or the patients were intolerant of them prior to the study intervention. There was no requirement for the papers to report any particular outcome measure or length of follow-up to be included. Multiple reports of single trials were grouped to avoid double counting. A single trial may appear in numerous publications and, if not corrected for, could introduce spurious precision by being counted over and over again. Trials were then grouped by type of intervention and data extracted on number of participants. Data were extracted onto an electronic database. The data extraction was undertaken independently by DS. The primary outcome of interest was the number of participants in a study so to produce an indication how much RCT evidence there was for any particular intervention. Relevant Cochrane reviews were identified and inspected.

The initial electronic search returned 1307 references (Figure 1). 560 of these were relevant to this review. After excluding duplicates, non-randomised studies and studies not concerning a therapeutic intervention there were 268 included studies. Relevant interventions tested in randomised trials fell into four broad categories (Tables 1, 2, 3 and 4). The average size of trials was 64.8 (SD 61.6, range 7-526, median 56, mode 60) and the most commonly evaluated treatments are cognitive behavioural therapy, electroconvulsive therapy, or transcranial magnetic stimulation supplementing a standard care which would include antipsychotic treatment, risperidone supplementation of clozapine, or a list of antipsychotics (most notably clozapine) as stand-alone treatment for people whose illness had been designated resistant to treatment (Table 4). Cochrane reviews, not specifically focusing on clozapine, provide extensive coverage of the data from within broad groups of trials, but few of these leading reviews provide specific consideration of this important sub-group of people (Additional file 1: Table S1).

The James Lind Alliance has recently focused on gaining a UK consensus for prioritisation of research topics in schizophrenia for clinicians, researchers and patients. This broad process results in a top ten topic list and top of that list was "What is the best way to treat people with schizophrenia that is unresponsive to treatment?" [45]. Whilst we have not assessed the quality of the research this survey does give a broad taxonomy of how much evaluative research has been carried out investigating interventions for people with treatment resistant schizophrenia. We are aware of discrepancies between the numbers we present and the study count in some key systematic reviews - almost invariably with the paper published review containing considerably less studies than we have listed in the tables (for example [46]). This may be a function of the reviewers selecting studies of highest quality, and issues of currency and comprehensiveness of searches.

It is good to see how evaluative effort has some focuses - such on the evolution of adjunctive cognitive behavioural therapy, or use of clozapine - but it still seems that there are considerable efforts made by pioneering and entrepreneurial researchers and funders on trials focusing on any number of treatments. Common approaches to treatment are under-researched. For example, poly-pharmacy is prevalent and may be used in preference to initiating clozapine in people whose illness is resistant to treatment [47]. There are, however, only four studies investigating combinations of non-clozapine antipsychotics (total of 297 participants). Also, evidence for augmenting antipsychotics with anti-depressant medication seems lacking as no individual drug had more than one study and these trials included only 244 participants in total.

The great majority of trials right across this area seem grossly underpowered to find any clinically important outcome. Understandably, much evidence synthesis has been undertaken but key maintained reviews (Cochrane reviews) have largely been too broad to provide synthesis of evidence for this important sub-group of people. When that is not the case, and synthesis of all trials relevant to those with treatment resistant illness is undertaken and presented, these reviews report salutary lessons on the limited clinical conclusions to be drawn from even the totality of evidence [16-44].

The thousands of people with treatment resistant schizophrenia who have invested time, effort and trust may be heartened to know that some direction has emerged from the not quite so random activity. The management of treatment resistant schizophrenia does remain controversial [48] but guidelines have some consistency in supporting use of clozapine [49,50] and psychosocial approaches for which service users have helped generate the greatest body of evidence. Participants in trials may be far less pleased to see how often they have been persuaded to give informed consent to trials that are never likely to really provide clinically important outcomes.

Too often funders and researchers do not seem to communicate or collaborate. We do not need more individual trials of 100 people unless part of a network of other studies. Such studies are likely to overestimate the effects of experimental treatments [51] and are unlikely to give a precise enough estimate of effect size and may lead to Type I or Type II errors [13,52]. Two arm randomised trials with a total of 300 participants have power to clearly show a difference of 20% between groups for binary outcomes such as 'better' or 'not better' (α 0.05, β 20%). We suggest that there are enough studies of about 100 participants to act as pilots for larger trials. Underpowered studies leave everyone depending on interpretation of scales such as BPRS [53] that are rarely used in clinical practice. Many such scales are proxy measures for real clinical outcomes. There are difficulties in determining whether changes in, for example, BPRS scores translate into clinically meaningful changes for patients with evidence suggesting higher cut-offs are required [54]. GRADE guidelines would also suggest downgrading recommendations based on small numbers of data [14] and only 33 studies contained more than 100 participants. This meant lots of participants were being enrolled onto studies that were unlikely to change clinical practice.

Systematic reviews should make data accessible on this important sub-group of people. This area of care is of concern to everyone [45] but too often good reviews leave data inaccessible. With maintenance of these reviews this shortcoming could be addressed.