Is treatment-resistant schizophrenia categorically distinct from treatment-responsive schizophrenia? a systematic review

Evidence suggests that treatment-resistance is a stable trait, as an early lack of response to treatment has been consistently shown to predict poor treatment outcome and diagnosis of treatment-resistance [3‐5]. First described in the 1988 Kane et al. criteria [6], a consistent minimum requirement for a diagnosis of treatment-resistance is two periods of treatment with different antipsychotics at adequate dose (variously defined), each for at least 4 weeks, without at least a 20% reduction in symptoms. This is reflected in guidelines for clozapine prescription [7, 8]: a 2014 review of clozapine prescription trends concludes that clozapine has consistently remained the gold standard for treatment-resistant schizophrenia, with all evidence-based guidelines recommending prescription “after failure of two adequate trials of two different antipsychotic agents” [9]. Recent variations have progressed from exclusively considering persistent positive symptoms to also incorporating persistent negative and cognitive symptoms [10, 11]; however positive symptoms remain a central focus as the main target of antipsychotics and the primary outcome in the early clozapine trials which defined treatment-resistance (see 2016 review highlighting role of positive symptoms [12]).

The dopamine hypothesis [13] is arguably the most well-known and well-supported neurochemical model of schizophrenia, but has been unable to explain the occurrence of treatment-resistant schizophrenia. While clinical response to antipsychotics is strongly correlated with dopamine receptor D2 occupancy for most patients, treatment-resistant patients show no clinical response even when their D2 receptor occupancy is above the therapeutic threshold [14]. Furthermore, clozapine is found to be highly effective in treatment-resistant patients [6], despite relatively low levels of D2 receptor occupancy [15]. It has been suggested that systematic differences may underlie this differential response of patients to antipsychotics [16]. In particular, that the dopamine hypothesis may not apply to treatment-resistant schizophrenia [17], where symptoms are instead driven by non-dopaminergic abnormalities, perhaps involving the glutamate system [18‐20].

Approximately a third of patients are treatment-resistant [1, 8], though some estimates are as high as 40–60% [21, 22]. Among mental illness, treatment-resistant schizophrenia is associated with some of the highest levels of impaired functioning [23], rates of hospitalisation [24], and costs to society, with one review estimating that treatment-resistance leads to an additional $34 billion in direct healthcare costs in the United States alone [25].

As Farooq et al. comment [2], classifying schizophrenia by treatment-response helps ensure that treatment-resistance remains a priority; they propose establishing clinical criteria that distinguishes these groups and then evaluating the endophenotypes and biomarkers present in these samples. Previous reviews have looked at predictors of treatment-response in schizophrenia [16], but we are unaware of any reviews that have focused exclusively on studies comparing patients with treatment-resistance (defined as above by two failed antipsychotic trials) against known treatment-responders. Therefore, we conducted a systematic review of all studies which compare treatment-resistant versus treatment-responsive patients with schizophrenia to investigate whether current evidence supports a conceptualisation of these as categorically distinct subtypes.

Our inclusion criteria were: original data comparing patients with treatment-resistant and treatment-responsive schizophrenia; all patients have a primary diagnosis of schizophrenia or schizoaffective disorder; the definition of treatment-resistance incorporates (as a minimum) two failed antipsychotic trials; the treatment-responsive group comprises of patients with a known response to non-clozapine antipsychotics; publication in a peer-reviewed journal.

Studies were excluded if: treatment-resistance was classified based on only one antipsychotic trial; treatment-responsiveness was defined solely as not meeting treatment-resistant criteria; the group labelled “treatment-responsive” included patients on clozapine; if studies of clinical variables were cross-sectional (as treatment-resistance is defined by current clinical state).

The databases searched were: Ovid (on 25^th January 2015, including Medline, Embase, PsycInfo and PsycArticle), PubMed (on 2^nd February 2015) and CINAHL (on 2^nd February 2015). The grey literature database OpenGrey was also searched on 2^nd February 2015. Search terms included synonyms for schizophrenia and treatment-resistance (Additional file 1). Relevant review articles and conference proceedings published between 2005 and 2015 were identified and reference and abstract listings were screened. All eligible studies were forward and backward screened. At the stage of full-text screening, the two journals with the highest number of potentially eligible publications were Biological Psychiatry and Schizophrenia Research; therefore, the tables of contents for all online volumes of these journals were hand-searched. Authors of all eligible studies were contacted requesting any other eligible findings, published or unpublished.

All identified studies underwent title and abstract screening, then all potentially eligible studies underwent full-text screening. When full-text articles were not available or eligibility was unclear, authors were contacted where possible. If confirmation of eligibility was not possible, studies were excluded. AG did the initial screening, then RS independently re-screened a randomly selected a) 10% of titles and abstracts identified by the database search and b) 20% of full-text articles identified by title and abstract screening. Cohen’s kappa was calculated to provide a conservative measure of inter-rater reliability; this produced kappas of 0.8 and 1 respectively, indicating substantial agreement. If the second reviewer identified novel potentially eligible studies, these were re-screened collaboratively.

A standard data extraction form was used (Additional file 1). If the treatment-resistant group were sub-divided by clozapine response, the data from both the clozapine-responsive and clozapine-nonresponsive (known as ultra-treatment resistant) group were reported.

To assess study quality and risk of bias, the Newcastle-Ottawa Scale for case-control studies was adapted (Additional file 1), as has been done previously [26]. The scale was adapted in advance to suit the scope of the research question, to include items on statistical analysis and sample size, and to allow for more differentiation between studies. Studies were scored out of 5 for selection of participants, out of 2 for comparability and out of 4 for predictor ascertainment and analysis. A total score of 9 or above was deemed high quality, a score of 6–8 was deemed moderate quality, and below 5 was deemed low quality. Low quality studies were then excluded from the review.

This systematic review has been conducted and reported according to PRISMA principles and guidelines, to aid evaluation and utilisation (see Additional file 2: PRISMA checklist).

There were five neuroimaging studies covering magnetic resonance imaging (MRI), electroencephalography (EEG), positron-emission tomography (PET), and magnetic resonance spectroscopy (MRS). All report controlling for basic demographics like age and sex but only three [17, 18, 27] report controlling for other confounders such as ethnicity, weight and smoking.

Two investigated brain volume, measured by MRI. In a study of 37 New Zealand patients (19 resistant, 18 responsive), Anderson et al. reported significantly lower grey matter volumes, especially in cortical areas, for treatment-resistant patients, but no differences in whole brain or white matter volume [28]. Ultra-treatment resistant patients also showed significantly lower grey matter volume than treatment-responsive patients, with numerically larger deficits than treatment-resistant patients (though this difference was not significant). Compared to healthy volunteers, treatment-responsive patients did not significantly differ in grey matter volume, while treatment-resistant patients showed significantly lower volumes. All the treatment-resistant patients were prescribed clozapine, while the majority of treatment-responsive patients were prescribed olanzapine or risperidone. In Molina et al., 49 Spanish patients (30 resistant, 19 responsive) were studied [29]; no patients had taken clozapine prior to baseline assessments, and all patients were trialled on haloperidol to confirm resistance or responsiveness. They report that discriminant analysis demonstrated hypertrophy in occipital white matter to be the best predictor of treatment-resistance, and that significant decreases in frontal and occipital grey matter of treatment-resistant patients compared to healthy volunteers were not evident in treatment-responsive patients.

In the same study, Molina et al. also investigated P300 parameters - the late component of the event-related potential - with EEG but found no differences between responsive and resistant patients [29].

One study compared striatal dopamine synthesis capacity with PET measurement of [18 F]-DOPA uptake, in a matched-design study of 24 British patients (12 resistant, 12 responsive) [17]. No participants were prescribed clozapine at the time of scanning (though two treatment-resistant patients had been previously prescribed clozapine but experienced adverse reactions); the majority of patients in both groups were prescribed olanzapine or risperidone depot. They reported (with multiple testing correction) significantly lower dopamine synthesis capacity in treatment-resistant patients compared to treatment-responsive patients, in the whole striatum; and in the associative and limbic, but not sensorimotor, striatal subdivisions. Compared to healthy volunteers, treatment-resistant patients were not significantly different, unlike treatment-responsive patients who showed significantly higher dopamine synthesis capacity than healthy volunteers.

Two studies measured regional brain metabolite concentrations using MRS. Demjaha et al. [18] carried out a study of 14 patients (6 resistant, 8 responsive) and found that treatment-responders had significantly lower levels of N-acetyl aspartate (NAA) in the anterior cingulate cortex (ACC) than both treatment-resistant patients and healthy volunteers. Also, while ACC glutamate levels in treatment-resistant patients compared to treatment-responsive patients were not significantly different, treatment-resistant patients had significantly elevated ACC glutamate levels compared to healthy volunteers, while treatment-responsive patients did not. Medication status or history was not reported. Goldstein et al. [27] performed a MRS study in 31 New Zealand patients (16 resistant, 15 responsive), acquiring data in the putamen, dorsolateral prefrontal cortex (DLPFC) and ACC. In contrast to Demjaha et al., they found no group differences in glutamate, NAA or choline. However, they did find patients with treatment-resistant schizophrenia had significantly higher total glutamate + glutamine (Glx) in the putamen (with family-wise error correction) than both treatment-responsive and ultra-treatment resistant patients. There was no significant difference between healthy volunteers and treatment-resistant patients, but healthy volunteers appeared to have very similar levels to the two other patient groups. Also, ultra-treatment resistant patients had significantly lower Glx in the DLPFC than treatment-responsive patients, with treatment-resistant patients and healthy volunteers showing levels in between. All their treatment-resistant patients were prescribed and responsive to clozapine, while all treatment-responsive patients were prescribed non-clozapine atypical antipsychotics (with the majority prescribed olanzapine and risperidone).

There were nine gene-association studies, with no eligible genome-wide association study (GWAS) papers. All of them used white participants.

Two investigated polymorphisms in the brain-derived neurotrophic factor (BDNF) gene. In a study of 94 Finnish patients (51 resistant, 43 responsive), no group differences were found for the G169A polymorphism (Val66Met) or the C270T polymorphism [30]. In a study of 88 French patients (20 resistant, 68 responsive), treatment-resistant patients displayed significantly fewer long alleles of the BDNF dinucleotide repeat polymorphism than treatment-responsive patients [31].

Two studied the serotonin 2A receptor (5-HT2A) gene, specifically the T102C polymorphism. In a study of 102 Canadian patients (63 resistant, 39 responsive), a significantly higher frequency of the C/C genotype in treatment-resistant patients was reported [32] and a study of 94 Finnish patients (51 resistant, 43 responsive) replicated this association, but only in females [33].

The latter also reported a significantly higher frequency of the C/A genotype for the tryptophan hydroxylase enzyme (TPH1) gene in treatment-resistant patients, but no group differences for the guanine-nucleotide-binding protein (GNB3) gene [33].

In a study of 193 French patients (45 resistant, 148 responsive) considering the 5’UTR (ccG repeat) polymorphism of the reelin gene, a significantly higher frequency of ccG10 alleles in treatment-resistant patients was found [34].

Another study investigated the regulator of g-protein signalling (RGS4) gene in a study of 93 Finnish patients (50 resistant, 43 responsive), but found no group differences [35].

The dopamine receptor 3 (DRD3) gene, specifically the Bal I polymorphism, was examined in a study of 89 French patients (19 resistant, 70 responsive) which reported significantly less homozygosity for treatment-resistant patients [36].

A study of 38 Finnish patients (19 resistant, 19 responsive) investigated the human leukocyte antigen (HLA) genotypes and found significantly higher HLA-A1 allele frequency for treatment-resistant patients [37]. In contrast, a study of 88 Israeli patients (50 resistant, 38 responsive) conducted HLA typing and found no group difference in frequencies of different classes of antigens [38]. All treatment-resistant patients were prescribed clozapine, while all treatment-responsive patients were prescribed haloperidol.

Unfortunately, none of these findings are reported to survive multiple-testing correction: four do not report correction [31, 33, 34, 36] and two report their findings becoming non-significant after correction [32, 37].

There was one study of familial genetic loading and prevalence in relatives, which studied the prevalence of schizophrenia spectrum disorders, cluster A personality disorders and long-term psychiatric care in relatives, and calculated family loading scores and morbidity risks [39]. In a study of 71 Canadian patients (35 resistant, 36 responsive), first and second degree relatives of treatment-resistant patients had a significantly higher morbidity risk of schizophrenia spectrum disorders compared to relatives of treatment-responsive patients, and a significantly higher familial-loading score was found for treatment-resistant patients. While treatment-resistant patients demonstrated significantly greater risk than healthy volunteers, treatment-responsive patients did not differ significantly. No group differences were found for risk of cluster A personality disorders or long-term psychiatric care in relatives.

Two studies investigated whether clinical variables were associated with subsequent treatment response or resistance.

There was one prospective study [40] of Positive and Negative Syndrome Scale (PANSS) scores² in which patients with no previous regular antipsychotic use and an illness onset within the last five years were randomised to receive either first or second generation non-clozapine antipsychotics in an open trial. Those that did not respond after one trial were then switched to another antipsychotic, and those that failed two trials were deemed treatment-resistant (4 patients in total). In the study of 17 Brazilian patients, treatment-resistance at twelve weeks was predicted by lower baseline total PANSS score. This study also measured improvement in the first 2 weeks of antipsychotic treatment and did not find this to be predictive of treatment-resistance.

Meltzer et al. studied age of onset [41]. In a study of 322 American patients (196 resistant, 126 responsive), treatment-resistant patients had an earlier age of onset.

There were two papers looking at neurocognitive function.

Joober et al. examined 75 Canadian patients (39 resistant, 36 responsive) [42]. Treatment-resistant patients had greater deficits in verbal ability and language, verbal memory and learning, and visual memory, but there were no significant differences for visual-spatial ability, abstraction and concept formation, visual motor processing and sustained attention. Treatment-resistant patients were prescribed a combination of typical and atypical antipsychotics (including clozapine), while all but four treatment-responsive patients were prescribed typical antipsychotics.

De Bartolomeis et al. investigated verbal memory, working memory, motor speed, verbal fluency, processing speed and executive functions, in a study of 41 Italian patients (19 resistant, 22 responsive) [43]. The only significant difference reported was lower verbal memory scores in treatment-resistant patients. Medication status or history separated by group is not reported but prescription of first and second-generation antipsychotics was not significantly different between treatment-resistant and responsive patients.

There was one study measuring demographic characteristics. Meltzer et al.– described above - studied gender and ethnicity [41]. The study demonstrated a significantly higher frequency of white patients in the treatment-resistant group compared to treatment-responsive, but no gender difference.

This is the first systematic review directly addressing whether treatment-resistant schizophrenia is categorically different from treatment-responsive schizophrenia, a question with substantial implications for research and clinical practice. It is somewhat unusual for a systematic review to address a theoretical question like this, but the benefits of comprehensiveness, reduced bias, and replicability remain desirable. To ensure that findings were not biased towards research fields that unduly influenced the conclusions, a broader search strategy than is conventional was necessary. This provides a comprehensive overview allowing the emergence of a pattern across research fields. It also highlights the areas lacking in quantity or quality of research. Another strength of this review is the focused definitions of treatment-resistance and treatment-responsiveness; they relate to clinical practice for diagnosing and treating treatment-resistance and provide a comparison between two groups of patients with distinctly different responses to antipsychotics.

However, this study does have limitations. Firstly, while every effort was made to seek out all available research there is a stronger chance of publication bias for non-RCT studies, as there are less protective measures (such as prospective registration of trials) in place; however, we did identify several studies reporting null findings (e.g., 4 out of 9 gene-association studies reported null results). Secondly, despite not excluding non-English language papers, papers were largely Western and participants were largely white. Thirdly, significant heterogeneity remained in definitions of treatment-resistance and treatment-responsiveness and in the populations and methods used to recruit participants, making comparison different. Fourthly, in comparing based on treatment outcome, antipsychotic response and resistance are not readily separable from group differences in illness symptomatology or severity. Also, the binary comparison of treatment-resistant patients to confirmed treatment-responders arguably ignores some of the complexity of the disorder. Not only can treatment-response be considered along a spectrum [44], but schizophrenia is heterogenous in more than just treatment-outcome and likely encompasses a spectrum of disorders [45]. However, focusing on this comparison of patient groups with distinctly different responses to antipsychotics is a strong starting point for addressing the heterogeneity, one likely to provide larger effect sizes and one which provides information with significant clinical relevance. Finally, the strict definitions of treatment-resistant and treatment-responsive patients mean that potentially relevant studies investigating markers of treatment-resistance using different definitions have been excluded. The rest of the discussion incorporates particularly pertinent studies from this broader literature.

One study identified indicates that treatment-resistant schizophrenia may be differentiated from treatment-responsive schizophrenia by higher familial loading scores and greater risk of schizophrenia in relatives; in fact, the study found that treatment-responsive patients did not show the significantly higher risk compared to healthy volunteers that treatment-resistant patients did [39]. An earlier study looking at a single antipsychotic trial similarly found that first-degree relatives of non-responders to haloperidol had higher lifetime risk for schizophrenia spectrum disorders [67]. More recent studies have found associations between a history of clozapine treatment and a higher polygenic risk score [68] and treatment-resistance and higher genome-wide burden of rare duplications [69] (but neither compared to confirmed treatment-responders and were thus not included in this review). These cumulatively suggest that treatment-resistant schizophrenia may be more heritable, with stronger genetic influence.

One study identified indicates that treatment-resistance may be differentiated from treatment-responsive schizophrenia by predominance of the HLA-A1 allele [37] but another found null results for a difference in HLA antigen types [38].

Wider literature has indicated the importance of the immune system in treatment-response, with two studies finding that treatment-resistant patients show significantly higher levels of serum IL-6 levels than healthy volunteers (with non-resistant patients presenting with intermediate levels) [70, 71] (neither were included in this review because their non-resistant patients were not confirmed as treatment-responsive). Similarly, two GWAS’s investigating treatment response have also implicated variants associated with the immune system: one compared treatment-resistant patients to healthy volunteers and found an association with polymorphisms of SLAMF1, a gene associated with lymphocyte activation [72]; while another looked at response to a single antipsychotic trial and found an association with RTKN2, believed to be involved in lymphopoiesis [73].

Two papers identified indicate that treatment-resistant patients have greater deficits in neurocognitive domains, with a replicated result in verbal memory [42, 43]; this would be consistent with previous studies that have long reported associations between worsened memory and treatment-resistance [74]. However, cross-sectional studies suffer from symptom severity potentially confounding the results.

The one identified prospective study on PANSS scores reported that a lower score at baseline predicted treatment-resistance at 12 weeks, and found no association between improvement in the first 2 weeks of antipsychotic treatment and later treatment-response [40]. These are surprising findings, especially the latter as a 2014 review of treatment response (looking only at single-antipsychotic trials, and thus not included in the review) found that early lack of response was one of the most robust predictors of later lack of response [75], and a 2015 meta-analysis (again looking at only single-antipsychotic trials) found that lack of response at two weeks had high specificity and positive predictive validity for predicting later non-response [76]. It may be that early lack of response predicts later non-response for specific antipsychotics but not non-response in general (as characterised by treatment-resistance); however, this seems unlikely and it is much more plausible that the reported lack of association in the present review is due to the small sample size of the identified study.

One identified study indicated that treatment-resistance is associated with an earlier age of onset [41], consistent with multiple studies reporting that early age of onset is predictive of poor outcomes [4, 77] and a review by Kessler et al. [78] which reports that early age of onset is associated with more severe, persistent and treatment-refractory conditions across psychiatric disorders.

The one identified study indicates a greater proportion of white patients in treatment-resistant patient groups. This finding has been replicated by another study [79] (not included because it did not compare to confirmed treatment-responsive patients) but the implications of this are unclear. It has been fairly well-established that the excess of schizophrenia in black African populations typically seen is due to (as yet undetermined) non-genetic factors [80], so these findings suggest that treatment-resistant schizophrenia is less influenced by these environmental factors; this would be consistent with the above section suggesting increased heritability in the treatment-resistant patient group. Similarly, a large Danish epidemiological study [81] recently found that established environmental risk factors for schizophrenia such as urban environment did not predict treatment-resistance, and in fact were negatively correlated with treatment-resistance.