The clinical and cost effectiveness of testing for cytochrome P450 polymorphisms in patients treated with antipsychotics: a systematic review and economic evaluation

Article history

The research reported in this issue of the journal was commissioned and funded by the HTA programme on behalf of NICE as project number 06/28/01. The protocol was agreed in October 2007. The assessment report began editorial review in December 2008 and was accepted for publication in April 2009. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Professor Tom Walley is Editor-in-Chief of Health Technology Assessment, although he was not involved in the editorial processes for this report.

Copyright statement

© 2010 Queen’s Printer and Controller of HMSO. This journal may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2010 Queen’s Printer and Controller of HMSO

Possibilities for individualised patient care

There is wide variability in the response of individuals to standard doses of drug therapy, which may occur as a result of interindividual differences that may be inherited (pharmacogenetics). Thus, there is growing anticipation among scientists, health-care providers and the general public that tests to identify genetic differences will be available and be used to more specifically direct the prescribing of therapeutic agents (pharmacogenetic testing), improving our ability to personalise therapies and subsequently improving clinical outcomes. 1

Genetics

There are approximately 50,000 genes in the human genome. Inherited variation in genes coding for metabolising enzymes and drug transporters (polymorphisms) may alter drug response and toxicity. Each gene is made up of a deoxyribonucleic acid (DNA) sequence. A DNA sequence consists of a double strand of DNA molecules, with these molecules made up of even smaller molecules known as nucleotides. Most of the DNA sequence is identical from one individual to the next in that the same type of nucleotide [adenosine (A), guanine (G), cytosine (C) or thymine (T)] occurs at the same locus between individuals. However, there are a small proportion of loci where the type of nucleotide varies from one individual to the next; these parts of the DNA sequence are known as single-nucleotide polymorphisms (SNPs) and they are the most common type of genetic variation in humans. As DNA exists in double strands, these nucleotides exist in pairs (one nucleotide on each strand). Alternative forms of a nucleotide that can occur at a particular locus of the genome are known as alleles.

Loci usually have two possible alternative alleles commonly known as wild type (wt) or mutant (mut), with the wt allele being the most common allele found in the general population. Thus, for example, at a given locus where it is possible to have either an adenosine or a thymine nucleotide, if the adenosine nucleotide is the most common then this would be identified as the wt allele. Genotypes are derived from the alleles [e.g. wt/wt (also known as homozygous wild type) or wt/mut (also known as heterozygous wild type)] and thus these SNPs give rise to the variation in genotype and phenotype across individuals.

Pharmacogenetic testing

Technologies used for genetic testing (commonly called genotyping) have undergone a revolution in recent years. Since the discovery of DNA, scientists have been trying to unravel the genetic knowledge and find ways of applying it for the benefit of mankind. The problem of obtaining sufficient quantities of DNA for genetic manipulation, which was the single biggest obstacle faced by molecular biologists, was solved by the very significant development of the polymerase chain reaction (PCR) by Kary Mullis in 1983. This discovery, coupled with the advent of DNA sequencing (first developed by Frederick Sanger), significantly accelerated genetic research and discovery.

Attainment of rapid speeds of DNA sequencing by modern technologies led to the complete sequencing of the human genome (Human Genome Project) and shed more light on variations that exist in individual genomes such as SNPs and copy number variations. Recent years have seen major strides taken in genotyping technologies, thus making them more robust and easier to use as well as costing less per SNP genotyped. In addition, point-of-care tests are also being developed, which in some cases may facilitate translation into a clinical environment. A summary of the most common genotyping techniques available is provided in Table 1.

To assist policy-makers in the process of making decisions regarding the use of genetic testing in the delivery of patient care, the ACCE model has been developed. Based on previously published methodologies and terminology, this collaboration between the Foundation for Blood Research and the Centers for Disease Control and Prevention (CDC) in the USA includes four key components that are required for evaluating any genetic test (and which thus give the model its name): analytical validity; clinical validity; clinical utility; and ethical, legal and social implications (Figure 1).

FIGURE 1.

ACCE evaluation process for genetic testing taken from Palomaki et al. 2 PPV, positive predictive value; NPV, negative predictive value.

Although many genetic tests are concerned with testing for diseases, increasingly pharmacogenetic tests are also being developed to predict the probability of an individual’s response to drug treatment in terms of efficacy and ADRs and as such the key components should also be generally applicable to pharmacogenetic tests.

Analytical validity in the process includes evaluation of all aspects related to the accuracy and reliability of genotype testing and includes sensitivity, specificity, quality control and robustness of the assessment process. Assessment of clinical validity begins with linking the four components of analytical validity and then assessing the other five elements identified in Figure 1. In relation to pharmacogenetics, the important outcomes to consider are the relationships between genotypes and phenotypes, with outcomes arising from the treatments currently being used in the clinical condition being considered, specifically efficacy and ADRs. Clinical utility refers to the ability to use the information from analytical and clinical validity in clinical practice. Establishing clinical utility is therefore important and should consider evidence for the use of pharmacogenetic testing to prospectively predict clinical outcomes and to modify clinical management (e.g. changing doses or switching drugs based on genotype tests). Harms associated with tests also need to be considered. These may include increased cost without impact on clinical decision-making or improvement in patient outcomes, less effective treatment with drugs, or inappropriate use of genotype information in the management of other drugs metabolised by particular enzymes.

To date, studies and reviews of pharmacogenetic tests have yielded insufficient evidence for any unequivocal benefit in terms of clinical validity or utility in a wide range of clinical areas including psychiatry (selective serotonin reuptake inhibitors for patients with non-psychotic depression3). Part of the difficulty in establishing an evidence base may be attributed to the fact that response may be multifactorial and multigenic, being dependent not only on CYP enzymes but also on phase II enzymes and differences in the drug targets. Thus, recent evidence-based recommendations issued in this field have urged further studies to be completed before testing can be recommended. 4

Nevertheless, in December 2004 the AmpliChip,^® which is a microarray-based test, became the first test to be granted market approval in the USA by the Food and Drug Administration (FDA),5 as well as in the EU. 5,6 This test is intended to identify a patient’s CYP2D6 and CYP2C19 genotype from genomic DNA extracted from a whole blood sample and thus provide a predicted metabolic phenotype. There are other technologies also available for genotyping of CYP enzymes, as well as the other genes that can influence drug response, and there is no doubt that other genotyping technologies will follow, including point-of-care tests. Currently the tests available for pharmacogenetics include HER2 (herceptin), HLA-B*5701, thiopurine methyltransferase, G6PD, factor V Leiden, the caffeine contracture test (for malignant hyperthermia) and pseudocholinesterase deficiency. Not all of these are genotypic tests; some are phenotypic.

This report has been commissioned by the National Institute for Health Research (NIHR) Health Technology Assessment (HTA) programme to address the issues of pharmacogenetic testing related to the use of antipsychotics for schizophrenia. Specifically the report addresses the issues related to the clinical and cost-effectiveness of testing for CYP polymorphisms in patients treated with antipsychotics. As such, the remainder of this report deals with the issues of pharmacogenetics related to this specific area. The report uses as its base the ACCE process and then goes on to discuss the economic implications of this new and evolving technology.

CYP enzyme system

A link between drug metabolism and drug response has been widely discussed in the literature and a significant proportion of this literature is focused on the CYP enzyme system, which has been identified as a major metabolic pathway for many drugs and a source of interindividual variability in patient response. 7,8 The CYP enzyme system contains major phase I enzymes involved in the metabolism of a number of substrates. There are 57 CYP genes in humans with each gene being named with CYP, indicating that it is part of the CYP gene family, a number associated with a specific group within the gene family, a letter representing the gene’s subfamily and a number assigned to the specific gene within the subfamily. 9 Thus, for example, CYP2D6 is gene 6 in group 2, subfamily D.

A number of SNPs in various CYP genes have been identified in recent years and several studies have shown how these SNPs affect the metabolism, safety and efficacy of various drugs, with CYP1A2, CYP2C9, CYP2C19, CYP2D6, CYP2E1 and CYP3A4 accounting for over 90% of drugs metabolised by the CYP enzyme system. Different CYP genes are involved in the metabolism of different types of drugs. For example, in oncology, the three major genes accounting for over 85% of hepatic activity are CYP1A2, CYP2D6 and CYP3A4;10 in psychiatry, several studies11–13 have shown a link between genetic polymorphisms and response to antidepressants with regards to CYP2C9, CYP2C19 and CYP2D6; whereas, for antipsychotics, CYP1A2, CYP2D6 and CYP3A4 again seem to be the most important. 14

Indeed, the CYP2D6 gene plays a primary role in the metabolism of drugs used to treat severe depression, schizophrenia, bipolar disorder and cardiovascular diseases. 15 CYP2D6 is responsible for the metabolism of 25% of all drugs on the market and polymorphisms in its gene significantly affect the metabolism of about 50% of these drugs. 16 Thus, it is of little surprise that it is probably the most extensively studied gene with regard to its impact on the metabolism of antipsychotics. 17

Drug/enzyme interactions generally result from one of two processes, enzyme inhibition or enzyme induction. The majority of drugs act as inhibitors, that is, they decrease the metabolism of substrates, which generally leads to an increase in the effect of the drug. Inducers, on the other hand, increase the metabolism of substrates, generally resulting in a decreased drug effect. 18 The CYP2D6 enzyme is the only one among the drug-metabolising CYP enzymes that cannot undergo induction and therefore genetic variation contributes largely to the interindividual variation in enzyme activity. 16

The prevalence of CYP gene polymorphisms varies across populations. Table 2 presents a summary of the frequencies of CYP2D6 alleles in various populations and also describes each allele’s predicted enzymatic function. As can be seen from Table 3, it is with reference to these classifications that the anticipated phenotype is commonly determined (when the CYP2D6 enzyme is the primary metabolic route) although it should be noted that there are a number of other classification systems being used. 19,20 Nevertheless, according to this classification system, drugs should have the intended effect in individuals with two copies of the normal functional allele. At the same dose, suboptimal responses would be expected in individuals with deficient or differing copies of functional alleles. Thus, individuals who carry copies of decreased activity or loss of function alleles are defined as poor metabolisers (PMs).

Given that the four most common loss of function alleles (*3, *4, *5 and *6) are associated with up to 98% of the PM phenotypes, it is no surprise to find that there are ethnic differences in metaboliser status. For example, a number of studies have found that around 7% of Caucasians are PMs compared with 1% of Asians, with data for other ethnic groups less cogent. 21 However, fewer Asians metabolise CYP2D6 normally, largely because of high frequencies of the *10 allele,22 resulting in a higher prevalence of intermediate metabolisers (IM). This decreased activity allele, which is rare in Caucasian populations, has been estimated to be as high as 55% in Chinese populations. 21,23

The classification used in Table 3 is the one that is used by the first approved pharmacogenetic test, the AmpliChip. This test also tests for CYP2C19 and patients are given either an extensive metaboliser (EM) or a PM phenotype, reflecting the fact that the *1 allele is a normal activity allele whereas *2 and *3 are associated with loss of function. Other CYP2C19 alleles exist that are not tested for by the AmpliChip including *4, *5, *6, *7 and *8, and *17, which are loss of function and increased activity alleles respectively. 26

However, as noted above, CYP1A2 and CYP3A4 appear to be relatively more important genes than CYP2C19 with regard to antipsychotics. With regard to CYP1A2, the *1 allele is associated with normal activity, *1C and *1K with decreased activity and *1F with higher inducibility. 26 For CYP3A4, the *1A allele is associated with normal activity whereas data on the function of other alleles are currently lacking.

Current costs of CYP tests to the NHS

The Doctors Laboratory (TDL)15,27 currently provides the Roche AmpliChip CYP2D6/2C19 testing facility to the NHS at a cost of £300, including any administration fees and platform costs (TDL, April 2008, personal communication). The turnaround time is stated as 1–2 weeks.

It was not possible to obtain costs for other tests including any in-house laboratory tests although these are thought to be less than that of the AmpliChip.

Current usage of CYP tests in the NHS

The use of CYP tests in the NHS has not been documented but it is thought that currently they are likely to be available only on a research basis.

Schizophrenia

Mental health is recognised as a major challenge in UK clinical practice and as such it is one of the nine National Service Frameworks (NSFs). 28 Schizophrenia is described in the NSF for mental health as a severe psychotic mental illness. Although there are no symptoms that in themselves are pathognomonic of schizophrenia, it can be viewed as a clinical syndrome within which is a broad spectrum of symptoms. Schizophrenia is viewed variably as a single disease or a group of heterogeneous disorders due to the variability of presentation and patterns within its diagnostic criteria, both currently and historically. The 10th version of the International Classification of Diseases and Related Health Problems (ICD-10)29 describes schizophrenic disorders as being ‘characterised in general by fundamental and characteristic distortions of thinking and perception, and by inappropriate or blunted affect’. These have been further described as ‘positive’ symptoms such as delusions and hallucinations (reality distortion) and ‘negative’ symptoms such as lack of emotional responsiveness and lack of volition.

Schizophrenia is associated with increased mortality compared with that of the general population, with individuals with schizophrenia having an ‘all-cause’ standardised mortality ratio (SMR) of between 2 and 3. 30,31 Suicide has been shown to have a large impact on the all-cause SMR, with an SMR for suicide or unexplained violence being greater than 10, with the prevalence of suicide amongst those with schizophrenia being currently estimated at around 5%. 30,32

The lifetime prevalence of schizophrenia is currently estimated to be between 0.34% and 1%,33–35 with annual prevalence and incidence rates of around 500 per 100,000 population (0.5%)34,35 and 10–20 per 100,000 population34,36 respectively. Overall, the rates are similar in men and women but the peak incidence of onset is between 15 and 25 years in men (where the incidence rate is twice that for women) and 25 and 35 years in women (where the incidence rate is higher among women). 33,35

Although the aetiology of schizophrenia is not clear, it almost certainly involves dopamine, specifically the D2 receptor. Thus, pharmacological agents that act as dopamine antagonists (with the exception of aripiprazole, a dopamine partial agonist) and which have actions on a number of other neurotransmitters and their receptors [e.g. 5-hydroxytryptamine (5-HT)] are used alongside a number of other strategies and interventions to treat schizophrenia, comprising a total care package.

Drugs for schizophrenia can be classified into typical (first generation) and atypical (second generation) antipsychotic agents. The historical difference between the two classes is the propensity of the older typical agents to cause catalepsy [a severe extrapyramidal symptom (EPS) characterised by muscular rigidity and fixity of posture with decreased pain sensation] in rats, whereas the newer atypical agents do not. In clinical practice risperidone, olanzapine, amisulpride and quetiapine are most commonly used as first-line treatment as recommended by the National Institute for Health and Clinical Excellence (NICE). 37 Typical antipsychotics are now more commonly used as second-line treatment, either as oral medication or in depot form, or to assist in the management of severe behavioural disturbance.

Atypical antipsychotics initially appeared to have the benefit of lower levels of some ADRs, most notably movement disorders, elevation of prolactin and sedation,38,39 although this has been increasingly challenged. 40,41 The atypicals have been associated with a metabolic syndrome including weight gain, diabetes and abnormal blood lipid profiles. 42,43

Clozapine is clinically in a separate class from typical and atypical antipsychotics. Although theoretically an atypical in that it does not cause catalepsy, its ADR profile includes a significant risk of agranulocytosis to the degree that mandatory monitoring of blood counts for neutropenia are part of its licensing requirements. 44 It remains available in the UK for use in treatment-resistant schizophrenia only. 37 It produces few acute EPS and has a lower incidence of tardive dyskinesia (TD) than other antipsychotics, although other ADRs include sedation, hypersalivation and hypertension. 44–46

Regarding efficacy, the HTA review of atypical antipsychotics in schizophrenia47 noted that evidence for the effectiveness of the atypicals compared with typicals was ‘in general of poor quality, based on short term trials and difficult to generalise to the whole population of people with schizophrenia’. The more recent Clinical Antipsychotic Trials in Intervention Effectiveness (CATIE)43 and Cost Utility of the Latest Antipsychotic Drugs in Schizophrenia Study (CUtLASS) trials48 have led the chief investigators of these trials to conclude that typical antipsychotics are as good as atypical antipsychotics for many patients. 49

With advances in treatment, the prognosis of an individual with a first episode of schizophrenia is less bleak than was once thought, with approximately 20–25% of patients having no further episodes. 50–52 However, within the first year, recurrence is observed in up to 25% of patients,53 rising to almost 50% within 2 years. 50,54,55 Within 12 months it has also been found that 14% of patients are treatment resistant,56 and over 2 years’ duration 20–45% are only partially responsive to antipsychotic medication,57,58 with 5–10% of patients deriving no benefit at all. 59 However, the prevalence of treatment resistance is hard to determine given the lack of agreement on defining the term, and, as these figures also reflect treatment outcomes with typical antipsychotics, with atypical antipsychotics now also being widely available, it has been argued there is a need to reconsider what constitutes ‘non-response’. 60 Treatment resistance is broadly described in NICE Clinical Guideline 1 (Schizophrenia)37 as a ‘lack of satisfactory clinical improvement despite the sequential use of the recommended doses for 6 to 8 weeks of at least two antipsychotics’. Individuals who receive antipsychotic prophylaxis (maintenance therapy) have been found to have a better outcome than those who have antipsychotics only when symptoms are present. 61,62

Non-compliance is also often related to efficacy limitations as well as ADRs of antipsychotics,63 increasing the risk and severity of relapse (with each further episode a decline in baseline functioning can be expected64), increasing the length of hospital stay and quadrupling the risk of suicide attempts. 65 It has been found that 10 days after discharge from hospital up to 25% of individuals with schizophrenia are partially or non-compliant, rising to 50% after 1 year and 75% after 2 years. 65 However, instruments for measuring adherence and non-compliance rates have varied across studies,66 with one recent systematic review of 28 studies finding the weighted non-adherence rate to be 40.6% (weighted mean 25%, 95% CI 17.42 to 32.66)67 and another finding the mean rate of non-adherence to be between 41.2% and 49.5% in 10 and 5 studies, respectively, depending on the inclusion criteria used. 68

Factors influencing compliance have also been explored in numerous studies and reviews and, unsurprisingly, many different factors influence compliance including those that affect patient’s beliefs that medication will be efficacious and ameliorate symptoms and fears about ADRs. 69 The types of ADRs that are distressing to patients and linked to non-compliance include EPS, neuroleptic dysphoria, akathisia, sexual dysfunction and weight gain. 69

Extrapyramidal symptoms are relatively common ADRs to antipsychotic medication. They can be severe and disabling and a significant factor in an individual deciding to stop or modify treatment as prescribed. 63 Although easy to recognise, the likelihood of EPS cannot be predicted accurately because they depend on the dose, the type of drug and individual susceptibility. 44 EPS include parkinsonian symptoms (including tremor), dystonia (abnormal face and body movements), akathisia (restlessness) and TD (rhythmic involuntary movements of tongue, face and jaw). TD also reflects the underlying pathology in schizophrenia as it has been established that the presence of TD predates the advent of antipsychotic treatment of schizophrenia70 and has been observed in older individuals with schizophrenia who have never been treated with antipsychotics. 71 A recent systematic review of TD72 gave prevalence rates of 13.1% for those treated with atypical antipsychotics, 15.6% for antipsychotic-free patients and 32.4% for those treated with typical antipsychotics.

Pharmacogenetics and schizophrenia

As noted above, a number of antipsychotics (both typical and atypical) are metabolised by the CYP1A2, CYP2D6 and CYP3A4 enzymes (Table 4). It should be noted, however, that enzymes other than CYP enzymes are also involved in the metabolism of these drugs. Another key issue for clinical practice is the risk of drug interactions, which although not common have the potential to cause significant harm. Many patients receiving antipsychotics are also likely to be prescribed other medications, including other psychotropic medications, many of which will also be inhibitors of CYP enzymes. Thus, enzyme inhibition may involve competition between drugs for the enzyme binding site, increasing the likelihood or severity of drug–drug interactions. Therefore, knowledge about CYP gene polymorphisms could potentially aid the selection of a specific drug and/or guide decisions about appropriate dosing to optimise efficacy and tolerability for individual patients.

Although CYP3A4 is present in much higher abundance in the liver and is involved in the metabolism of a greater number of drugs than the other CYP enzymes (50% of all marketed drugs73,74), its enzyme activity is affected more by environmental factors such as diet and concurrent medications than by inherited variations. For example, human in vivo studies have indicated considerable interindividual variability (fivefold) that can be significantly increased by deliberate modulation, i.e. inhibition and induction. 75

Similarly it remains questionable how much of the interindividual variability in CYP1A2 activity is explained by genetic polymorphisms;13 smoking in particular is thought to affect the level of CYP1A2. 76 As already noted, CYP2D6 is the only one among the drug-metabolising CYP enzymes that is not inducible but it can be inhibited by a number of drugs and hence this, together with genetic variation, contributes to the interindividual variation in enzyme activity. The association between CYP2D6 genotype and the risk of having TD has recently been reviewed in a meta-analysis77 that investigated loss of function alleles (*3, *4, *5, *6 and *7), decreased activity alleles (*10) and the *2 allele. This found that patients who were homozygotes for loss of function alleles (PMs) had a 1.64-fold greater chance of suffering TD compared with other patients with schizophrenia, but the effect was not significant (95% CI 0.79 to 3.43).

Current service costs for treating schizophrenia

The cost of care for individuals with schizophrenia is high. Davies78 estimated that 1.6% of the total national health-care budget was attributable to schizophrenia treatment. On the basis of this figure and estimated government spending on health,79 NHS expenditure on schizophrenia in 2008–9 is calculated to be in the region of £1.2 million.

Clinical effectiveness

Cost-effectiveness

Study characteristics

For all CYP polymorphisms, real-time PCR (such as LightCycler^® or TaqMan) was the most frequent genotype method studied in 13 instances (14 if triplex real-time is included). However, for CYP2D6, the most common methods were microarrays (particularly the Roche AmpliChip) in six studies followed by multiplex methods in five instances and Pyrosequencing in four. Usually single methods were used to test for a number of different alleles for each CYP although, in some instances, multiple methods were utilised (e.g. tetra-primer PCR for testing for the CYP2D6*3, *4 and *6 polymorphisms and multiplex long PCR for *5 in Hersberger et al. 92).

The most frequent methods used as a reference method for any CYP were PCR and restriction fragment length polymorphism (PCR-RFLP) analysis in 19 studies followed by sequencing in 14 studies. However, in the vast majority of these studies, only a very small number of the original samples were compared with sequencing, often as a second reference method to verify discordant cases. Allele-specific PCR (AS-PCR) was also commonly used as a reference method for CYP2D6.

The majority of studies were conducted in Europe, most often in Germany. This was particularly the case for CYP2D6, although five of the six US studies also investigated polymorphisms of this gene. This represents a higher number of American studies than for all of the other CYP polymorphisms combined.

Studies varied in size from 40 subjects being genotyped in the smallest study112 to 428 in the largest,118 although the number of samples being compared by the reference method varied from just six samples tested by AS-PCR in Oyama et al. 120 to 1400 samples in Lee et al. 94

Participant characteristics

Given that there are racial differences in function-altering polymorphisms, the most important participant characteristic to consider is ethnicity. Unfortunately, very few studies reported the ethnic origin of their subjects. However, given the high proportion of studies carried out in Europe (and in the USA for CYP2D6), it may be reasonable to assume that the majority of subjects studied were likely to have been of Caucasian origin (although given the high number of African Americans in the USA this assumption may not be correct).

Study findings

Although not all of the studies provided detailed genotype data, all those presenting any findings reported high concordance between methods (of 95% or more). This was the case no matter which CYP was genotyped or which methods were compared. No studies used exactly the same method as both method under test and reference method. In addition, given the overwhelming positive nature of all of the results regarding the analytical validity of each of the tests, it was not considered necessary to attempt to meta-analyse these findings.

A note of caution is, however, required when interpreting these findings. As noted in Chapter 2, analytical validity should include the reporting of sensitivity, specificity, quality control and robustness of the assessment process. Very few studies reported on all four aspects of analytical validity, quality control and assay robustness most usually being neglected. Similarly, very few studies actually presented results for sensitivity and specificity. It was, however, possible to calculate the sensitivity and specificity from 20 studies that presented relevant genotype data. In the vast majority of these instances, both sensitivity and specificity were 100% and, with the exception of Eriksson et al. ,89 in which specificity between Pyrosequencing™ and PCR-RFLP for CYP2D6 was only 30.8%, it was always at least 99%.

The most comprehensive detailed data were provided in studies examining the AmpliChip for CYP2D6 and CYP2C19. Compared with PCR-RFLP, for CYP2D6 there was 95.6% concordance in Heller et al. 91 (sensitivity 95%, specificity 100%), which rose to 100% when the discordant cases were compared with more sensitive methods (SNaPshot and sequencing). Similar results had also been reported for both CYP2D6 and CYP2C19 by Roche. 5,100 Melis et al. 95 used the AmpliChip as a reference method for Tag-It™ (a bead-based array). Again, concordance was high between methods (100% sensitivity and specificity for both CYP2D6 and CYP2C9) although it was stated by the authors that the Tag-It CYP2D6 assays were less robust than the CYP2C19 assays.

The findings for each CYP are summarised in Tables 7–11 and, as already noted, more detailed findings can be found in Appendix 4.

Analytical validity summary

Based on the findings presented in this review, tests for determining genotypes are highly accurate, with concordance being 100%. However, not all aspects of analytical validity have been reported in the studies (quality control and assay robustness being commonly neglected). In studies in which data were presented to calculate sensitivity and specificity, this was typically between 99% and 100% for both.

Quality assessment of included studies

All studies reported sample size, ranging from nine to 309 with a mean size of 101 (median n = 92). Compared with the typical sample sizes required to provide sufficient power to detect a range of typical genetic effect sizes for various minor allele frequencies,83 these sample sizes are all small. Further, none of the studies explained how the sample size had been chosen or stated the a priori power for detecting effect sizes of varying degrees. Therefore, it is unclear what range of effect sizes the studies were powered to detect.

In terms of selecting the variants to genotype, although only three141,143,145 did not give reasons why the gene being investigated was chosen, a minority (n = 19)126,132,137,138,146,147,149–151,154,161,162,164,165,167–170,172 explained the process of choosing which specific variants to genotype within the gene. Given the large number of possible variants to choose from within each gene, this raised the question of whether any within-study selective reporting occurred whereby several variants may have been investigated but only those found to be most significant were reported. 173 However, when studies reporting significant outcomes were subsequently analysed, around half had adequately given reasons for the specific alleles tested.

Generally, studies presented adequate information about the genotyping procedures employed; however, only three studies131,152,162 reported that genotype quality control procedures had been applied and thus it is unclear how reliable the allocated genotypes are in the remaining studies. Around half of the studies presented allele frequencies from previous studies, from which any significant problems with the genotyping procedures could have been identified.

Given that genotypes cannot always be called with sufficient confidence, some missing genotype data would not be unexpected in any study sample. In terms of the included studies, it was not always apparent from the manner in which data were reported whether any genotype data were missing. Some studies (n = 9)127,139,145,146,148,152,153,162,164 clearly specified the number of missing genotypes and two-thirds of these127,139,146,152,153,164 provided reasons for the missing data. All but two studies142,171 gave the number of patients contributing to each analysis. However, none of the studies in which missing data were apparent reported on tests of whether the genotypes were missing at random or mentioned any attempts at imputing the missing genotypes.

No study mentioned conducting specific tests for population stratification even though six128,133,139,144,164,169 were known to include patients with different ethnic backgrounds. These studies, in particular, are at risk from confounding because of population stratification. A minority (n = 21)130,132,134,138–140,146,149,152–154,161,162,164–170,172 of studies reported on a test for Hardy–Weinberg equilibrium (HWE) that can highlight problems with the genotype data. 83 When a test had been conducted, all of the variants were said to be within HWE, although it was not always clear what significance level had been referred to.

Finally, only seven studies132,135,141,143,151,162,171 failed to adequately define or justify their choice of outcomes. Four135,141,143,171 of these were presented as abstracts in which space was limited. Another162 was subsequently excluded from the analysis because on inspection of other data a number of inconsistencies were apparent (e.g. patients with the *6/*6 genotype were attributed with experiencing an EPS despite it earlier being stated that no patient in the study had this genotype). It should also be noted that, as several outcomes can be rationally chosen to assess the hypotheses of interest, it is not possible to ascertain if any studies conferred a risk of outcome reporting bias, in which several outcomes are investigated and only the most significant reported.

Study characteristics

Participant characteristics

Data analysis

The detailed findings from all of the studies are summarised below. When appropriate the results from meta-analyses are also presented.

CYP2D6

Metabolism

The findings are summarised in Table 19.

TABLE 19 - Summary of metabolism findings: clinical validity studies in patients tested for CYP2D6

CL/F, oral clearance; C_max, maximum plasma concentration; C_min, minimum plasma concentration; EM, extensive metaboliser; IM, intermediate metaboliser; NA, not applicable (cannot be calculated from data presented); PM, poor metaboliser; UM, ultrarapid metaboliser.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant type/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Brockmoller 2002131	Haloperidol clearance (litres/hour): UM (n = 5): 57.3 ± 31.7; EM (n = 106): 48.7 ± 20.9; IM (n = 56): 44.1 ± 25.4; PM (n = 5): 34.7 ± 13.7 Jonckheere–Terpstra test: p = 0.034 (n = 172)	NA
de Leon 2004133	Half-life < 3 days: PM (n = 6): 0	Half-life < 3 days: mut/mut (n = 6): 0
	Half-life ≥ 3 days: PM (n = 6): 2	Half-life ≥ 3 days: mut/mut (n = 6): 2
Jeon 2007143	CL/F (l/h): Homozygous for functional alleles: 3.17 Heterozygous for one functional and one non-functional/reduced function alleles: 2.55 Homozygous for reduced functional alleles: 1.85 Heterozygous for one reduced functional and one non-functional allele: 1.54	NA
Jerling 1996144	CL/F (l/h), mean ± SD (range):	CL/F (l/h), mean ± SD (range):
	Perphenazine: EM homozygote (n = 9): 454 ± 385 (213–1286); EM heterozygote (n = 5): 454 ± 279 (174–883); PM (n = 2): 250 ± 30 (229–271)	Perphenazine: wt/wt (n = 9): 454 ± 385 (213–1286); wt/mut (n = 5): 454 ± 279 (174–883); mut/mut (n = 2): 250 ± 30 (229–271)
	Zuclopenthixol: EM homozygote (n = 8): 95 ± 43 (38–165); EM heterozygote (n = 9): 65 ± 21 (38–95); PM (n = 3): 42±12 (31–54)	Zuclopenthixol: wt/wt (n = 8): 95 ± 43 (38–165); wt/mut (n = 9): 65 ± 21 (38–95); mut/mut (n = 3): 42 ± 12 (31–54)
Panagiotidis 2007155	Peak (Cmax), median (range) concentration (nmol/l): 0 functional alleles (n = 1): not available; 1 functional alleles (n = 8): 14.0 (3.3–67.0); 2 functional alleles (n = 16): 6.4 (1.6–19.0); 3 functional alleles (n = 1): 6 Trough (Cmin), median (range) concentration (nmol/l): 0 functional alleles (n = 1): 6; 1 functional alleles (n = 8): 10.5 (1–49); 2 functional alleles (n = 16): 4.0 (1.4–8.7); 3 functional alleles (n = 1): 3 p = 0.047	NA

It was apparent that, with the exception of the studies by de Leon et al. ,133 which reported on half-life (i.e. the amount of time required for the concentration of a drug to be halved), and Panagiotidis et al. ,155 which reported on maximum (peak) and minimum (trough) concentrations, there were no other studies that examined any of these pharmacokinetic outcomes (t_½, C_max and C_min respectively) or other parameters such as time to maximum concentration (t_max) or area under the curve (AUC). Although a number of studies used proxy measures for clearance (and were thus excluded), only three studies131,143,144 mathematically derived this outcome.

Two studies131,155 examined clearance in patients taking haloperidol, one after oral use and one after depot injection. In the earlier haloperidol study131 of 172 patients it was found that there was a trend towards lower therapeutic efficacy with increasing number of functional alleles, with mean clearance steadily increasing with each additional functional allele. However, the numbers of patients with the PM (mut/mut) and UM phenotypes were small (n = 5 for both groups). In the later study of 26 patients,155 clearance was measured at peak and trough and patients with the wt/mut genotype appeared to have a greater median concentration than those with the wt/wt genotype (of whom one patient was classified as UM, having an even lower median concentration than the EM). The authors state, however, that the association was not statistically significant. They acknowledge that the small number of included subjects may have limited the power of this study to detect differences.

For perphenazine, Jerling et al. 144 found the mean oral clearance (CL/F) to be similar amongst wt/wt and wt/mut patients and both to be higher than that in mut/mut patients whereas for zuclopenthixol the value decreased steadily by genotype although only two patients had the mut/mut genotype; the difference in clearance between wt/wt and mut/mut was thus threefold for perphenazine and twofold for zuclopenthixol. Regression analysis showed the effect to be statistically significant for both drugs. Similarly, for aripiprazole, Jeon et al. 143 found that for each functional allele the mean value of CL/F steadily decreased, being twice as high for wt/wt as for wt/mut patients.

Efficacy

Nine studies focused on the relationship between efficacy and genotype/phenotype but as all reported outcomes differed in how they were derived it was not possible to include the data from these studies in a meta-analysis. The findings are summarised in Table 20.

TABLE 20 - Summary of efficacy findings: clinical validity studies in patients tested for CYP2D6

BPRS, Brief Psychiatric Rating Score; EM, extensive metaboliser; IM, intermediate metaboliser; NA, not applicable (cannot be calculated from data presented); PANSS, positive and negative symptoms scale; PM, poor metaboliser; UM, ultrarapid metaboliser.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Aitchison 1999126	Total number of patients refractory to treatment: UM (n = 5): 2 (40.0%); EM (n = 287): 220 (76.7%); PM (n = 16): 13 (81.3%)	Number of patients refractory to treatment: wt/wt + wt/mut (n = 292): 222 (76.0%); mut/mut (n = 16): 13 (81.3%)
Arranz 1995129	Total number of non-responders to treatment: EM (n = 115): 42 (36.8%); PM (n = 8): 4 (50.0%)	Total number of non-responders to treatment: wt/wt + wt/mut (n = 115): 42 (36.8%); mut/mut (n = 8): 4 (50.0%)
Brockmoller 2002131	PANSS (day 28–day 3), median (range): General items: UM (n = 5): +8 (–31 to +1); EM (n = 106): –9.5 (–33 to +17); IM (n = 56): –9 (–41 to +25); PM (n = 5): –17 (–33 to -4) Positive items: UM (n = 5): –10 (–15 to –8); EM (n = 106): –9 (–25 to +18); IM (n = 56): –7 (–29 to +13); PM (n = 5): –13 (–15 to –3) Negative items: UM (n = 5): –2 (–8 to +5); EM (n = 106): –3 (–27 to +27); IM (n = 56): –5 (–18 to +27); PM (n = 5):9 (–11 to –4)	NA
Hamelin 1999139	End of study BPRS scores, mean ± SD:	End of study mean ± SD BPRS scores:
	BPRS (total): *1/*1 (n = 23): 31 ± 7; *1/*4 (n = 15): 34 ± 7; *4/*4 (n = 1): 31	BPRS (total): wt/wt (n = 23): 31 ± 7; wt/mut (n = 15): 34 ± 7; mut/mut (n = 1): 31
	BPRS (+): *1/*1 (n = 23): 8 ± 4; *1/*4 (n = 15): 10 ± 4; *4/*4 (n = 1): 11	BPRS (+): wt/wt (n = 23): 8 ± 4; wt/mut (n = 15): 10 ± 4; mut/mut (n = 1): 11
	BPRS (–): *1/*1 (n = 23): 7 ± 3; *1/*4 (n = 15): 9 ± 3; *4/*4 (n = 1): 3	BPRS (–): wt/wt (n = 23): 7 ± 3; wt/mut (n = 15): 9 ± 3; mut/mut (n = 1): 3
Kakihara 2005145	Percentage improvement in scores of PANSS: *1/*1 (n = 16): 37.7 ± 15.8; *1/*10 (n = 14): 31.9 ± 24.4; *10/*10 (n = 9): 43.5 ± 20.5	Percentage improvement in scores of PANSS: wt/wt (n = 16): 37.7 ± 15.8; wt/mut (n = 14): 31.9 ± 24.4; mut/mut (n = 9): 43.5 ± 20.5
Panagiotidis 2007155	PANSS total score, median (range): Peak: 0 functional alleles (n = 1): NA; 1 functional alleles (n = 8): 53 (35–88); 2 functional alleles (n = 16): 54 (38–91); 3 functional alleles (n = 1): 38 Trough: 0 functional alleles (n = 1): 35; 1 functional alleles (n = 8): 60.5 (38–86); 2 functional alleles (n = 16): 59 (33–95); 3 functional alleles (n = 1): 38 PANSS-G score, median (range): Peak: 0 functional alleles (n = 1): NA; 1 functional alleles (n = 8): 24 (18–38); 2 functional alleles (n = 16): 28 (18–42); 3 functional alleles (n = 1): 18 Trough: 0 functional alleles (n = 1): 16; 1 functional alleles (n = 8): 25.5 (19–41); 2 functional alleles (n = 16): 28 (17–43); 3 functional alleles (n = 1): 19	NA
	PANSS-P score, median (range): Peak: 0 functional alleles (n = 1): NA; 1 functional alleles (n = 8): 7.5 (7–24); 2 functional alleles (n = 16): 8 (7–16); 3 functional alleles (n = 1): 7 Trough: 0 functional alleles (n = 1): 7; 1 functional alleles (n = 8): 8 (7–20); 2 functional alleles (n = 16): 8 (7–16); 3 functional alleles (n = 1): 7 PANSS-N score, median (range): Peak: 0 functional alleles (n = 1): NA; 1 functional alleles (n = 8): 23.5 (7–32); 2 functional alleles (n = 16): 27 (12–36); 3 functional alleles (n = 1): 13 Trough: 0 functional alleles (n = 1): 12; 1 functional alleles (n = 8): 24 (11–36); 2 functional alleles (n = 16): 24.5 (7–36); 3 functional alleles (n = 1): 13
Plesnicar 2006156	PANSS – general subscale total score: EM/IM/UM (n = 125): 23.02 ± 5.31; PM (n = 6): 23.50 ± 3.83; p > 0.05 PANSS – positive subscale total score: EM/IM/UM (n = 125): 9.35 ± 3.22; PM (n = 6): 8.83 ± 3.06; p > 0.05 PANSS – negative subscale total score: EM/IM/UM (n = 125): 13.77 ± 4.09; PM (n = 6): 17.83 ± 2.48; p = 0.017	NA
Riedal 2005157	Total number of non-responders to treatment: wild type (n = 45): 26/45 (57.8%); heterozygous (n = 6): 4 (66.7%)	Total number of non-responders to treatment: wt/wt (n = 45): 26 (57.8%); wt/mut (n = 6): 4 (66.7%); mut/mut (n = 0): 0
Wang 2007163	BPRS (% improvement): *1/*1 (n = 22): 37.49 ± 15.47; *1/*10 (n = 39): 45.32 ± 16.29; *10/*10 (n = 41): 41.31 ± 17.10	BPRS (% improvement): wt/wt (n = 22): 37.49 ± 15.47; wt/mut (n = 39): 45.32 ± 16.29; mut/mut (n = 41): 41.31 ± 17.10

Three studies126,129,157 concentrated on the number of responders to treatment. In patients taking clozapine, the response in the study by Arranz et al. ,129 as assessed by the Global Assessment Scale, was worse for those with the mut/mut genotype than for those with either the wt/wt or the wt/mut genotype. Riedal et al. 157 did not identify any patients with the mut/mut genotype but found that proportionately more patients with the wt/mut genotype than with the wt/wt genotype failed to respond, where response was defined by a difference of 30% or less in PANSS total scores between baseline and last observation. The findings from Aitchison et al. 126 must be treated with extreme caution in the context of this review because the aim of this study was to compare UMs with other phenotypes, whereas, as already described in the methods section, for the purposes of this review, UMs are considered as wt/wt and therefore no different to EMs. Furthermore, patients were also preselected into refractory and non-refractory groups and assessed retrospectively and the drug regimens in the two groups were not the same (clozapine in the refractory group versus any antipsychotic in the non-refractory group). Nevertheless, this study also found a greater proportion of patients with the mut/mut genotype to be refractory to treatment than those with the wt/wt + wt/mut genotype, although fewer patients with the UM phenotype were refractory than either EMs or PMs. However, the number of UMs and PMs combined in this study was significantly less than the number of EMs and it should be noted that response was not defined by validated criteria but by prescribing consultants.

The remaining six studies all used PANSS or Brief Psychiatric Rating Scores (BPRS) to measure efficacy. Using total BPRS scores in patients using any antipsychotic, Hamelin et al. 139 found little difference between patients with the wt/wt, wt/mut or mut/mut genotypes, although only one patient possessed this last genotype. Brockmoller et al. 131 found that, for haloperidol, PMs (mut/mut) fared better than EMs, IMs or UMs (wt/wt or wt/mut) on median changes in general, positive and negative items on the PANSS. The score for UMs on the general items scale was notably different to the scores for EMs and IMs, in the other direction [+8 compared with between –9 (EMS) and –17 (IMs) for the other genotypes], but was similar to the scores for EMs and IMs for positive and negative items scale scores. Plesnicar et al. 156 found end of study PANSS scores for patients on long-term maintenance antipsychotic treatment to be similar between patients with the mut/mut genotype and those with the other two genotypes. In Panagiotidis et al. ,155 the median PANSS scores at both peak and trough for patients taking haloperidol injections were similar for UMs and PMs but higher for EMs and IMs, suggesting lower efficacy in these phenotypes, although extreme caution is required in interpreting this finding as only one patient was reported as having either the PM or UM phenotype. For risperidone, the mean percentage improvement in PANSS scores was lowest in the wt/mut group and highest in the mut/mut group in Kakihara et al. ;145 however, using BPRS, Wang et al. 163 found little difference between the groups.

Adverse drug reactions

In total, 30 studies were found examining the relationship between ADRs and genotype. Results for each type of ADR are presented in the following sections.

Tardive dyskinesia

A total of 14 studies quantified the number of patients with TD by genotype, with data from up to 13 included in the meta-analysis (but only between nine and 11 for any given comparison because of the manner in which these studies grouped their genotypes) (Table 21). Most of the studies measured the occurrence of TD using the validated Abnormal Involuntary Movement Scale (AIMS) and stipulated that patients were taking any typical or any antipsychotic.

TABLE 21 - Summary of findings for tardive dyskinesia: clinical validity studies in patients tested for CYP2D6

AIMS, Abnormal Involuntary Movement Scale; EM, extensive metaboliser; IPD, individual patient data; NA, not applicable; PM, poor metaboliser; TDRS, Tardive Dyskinesia Rating Scale.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Patient excluded from analysis.

Study	Outcomes as reported	‘Standardised outcome’a
Andreassen 1997127	Total number of patients with TD: EM homozygote (n = 61): 30 (49.2%); EM heterozygote (n = 29): 16 (55.2%); PM (n = 10): 5 (50.0%)	Total number of patients with TD: wt/wt (n = 61): 30 (49.2%); wt/mut (n = 29): 16 (55.2%); mut/mut (n = 10): 5 (50.0%)
	AIMS score – all patients, mean ± SD: EM homozygote (n = 61): 5.1 ± 4.0; EM heterozygote (n = 29): 5.8 ± 5.3; PM (n = 10): 6.8 ± 6.3	AIMS score – all patients, mean ± SD: wt/wt (n = 61): 5.1 ± 4.0; wt/mut (n = 29): 5.8 ± 5.3; mut/mut (n = 10): 6.8 ± 6.3
Arthur 1995130	Individual patient data presented	AIMS score – patients with TD, mean ± SD: wt/wt (n = 8): 5.7 ± 1.9; wt/mut (n = 7): 8 ± 5.2; wt/wt + wt/mut (n = 15): 7 ± 4.1; mut/mut (n = 1): 13
Ellingrod 2002137	Total number of patients with TD: *1/*1 (n = 11): 3 (27.3%); *1/*3 + *1/*4 (n = 26): 12 (46.2%); *3/*3 + *4/*4 (n = 0): 0	Total number of patients with TD: wt/wt (n = 11): 3 (27.3%); wt/mut (n = 26): 12 (46.2%); mut/mut (n = 0): 0
	Smokers with TD: *1/*1 (n = 5): 1 (20.0%); *1/*3 + *1/*4 (n = 9): 7 (77.8%); *3/*3 + *4/*4 (n = 0): 0	Smokers with TD: wt/wt (n = 5): 1 (20.0%); wt/mut (n = 9): 7 (77.8%); mut/mut (n = 0): 0
	Non-smokers with TD: *1/*1 (n = 6): 2 (33.3%); *1/*3 + *1/*4 (n = 17): 5 (29.4%); *3/*3 + *4/*4 (n = 0): 0	Non-smokers with TD: wt/wt (n = 6): 2 (33.3%); wt/mut (n = 17): 5 (29.4%); mut/mut (n = 0): 0
	AIMS score in smokers, mean ± SD: *1/*1 (n = 5): 1.23 ± 1.56; *1/*3 + *1/*4 (n = 9): 5.8 ± 4.3; *3/*3 + *4/*4 (n = 0): NA	AIMS score in smokers, mean ± SD: wt/wt (n = 5): 1.23 ± 1.56; wt/mut (n = 9): 5.8 ± 4.3; mut/mut (n = 0): NA
	AIMS score in non-smokers, mean ± SD: *1/*1 (n = 6): 1.7 ± 2.25; *1/*3 + *1/*4 (n = 17): 1.2 ± 2.17; *3/*3 + *4/*4 (n = 0): NA	AIMS score in non-smokers, mean ± SD: wt/wt (n = 6): 1.7 ± 2.25; wt/mut (n = 17): 1.2 ± 2.17; mut/mut (n = 0): NA
Fu 2006138	Total number of patients with TD: TT (n = 50): 37 (74.0%); CT (n = 64): 30 (46.9%); CC (n = 35): 15 (42.9%)	Total number of patients with TD: wt/wt (n = 50): 37 (74.0%); wt/mut (n = 64): 30 (46.9%); mut/mut (n = 35): 15 (42.9%)
	AIMS score – patients with TD, mean ± SD: TT (n = 37): 6.32 ± 2.62; CT (n = 30): 6.90 ± 2.83; CC (n = 15): 7.87 ± 3.60	AIMS score – patients with TD, mean ± SD: wt/wt (n = 37): 6.32 ± 2.62; wt/mut (n = 30): 6.90 ± 2.83; mut/mut (n = 15): 7.87 ± 3.60
Inada 2003140	Total number of patients vulnerable to TD with *2: WW (n = 234): 30 (12.8%); WM (n = 68): 11 (16.2%); MM (n = 7): 0	Total number of patients vulnerable to TD with *2: NA
	Total number of patients vulnerable to TD with *10: WW (n = 78): 10 (12.8%); WM (n = 97): 13 (13.4%); MM (n = 39): 4 (10.3%)	Total number of patients vulnerable to TD with *10: wt/wt (n = 78): 10 (12.8%); wt/mut (n = 97): 13 (13.4%); mut/mut (n = 39): 4 (10.3%)
Jaanson 2002142	Total number of patients with TD: EM homozygote (n = 35): 6 (17.1%); EM heterozygote (n = 13): 4 (30.8%); PM (n = 4): 1 (25.0%)	Total number of patients with TD: wt/wt (n = 35): 6 (17.1%); wt/mut (n = 13): 4 (30.8%); mut/mut (n = 4): 1 (25.0%)
Kapitany 1998146	Total number of patients with TD: EM homozygote (n = 28): 13 (46.4%); EM heterozygote (n = 16): 13 (81.3%); PM (n = 1): NAb	Total number of patients with TD: wt/wt (n = 28): 13 (46.4%); wt/mut (n = 16): 13 (81.3%); mut/mut (n = 1): NAb
	TDRS score – all patients, mean ± SD: EM homozygote (n = 28): 7.6 ± 5.94; EM heterozygote (n = 16): 11.6 ± 6; PM (n = 1): NAb	TDRS score – all patients, mean ± SD: wt/wt (n = 28): 7.6 ± 5.94; wt/mut (n = 16): 11.6 ± 6; mut/mut (n = 1): NAb
Lam 2001147	Total number of patients with TD: wt and heterozygous (n = 40): 19 (47.5%); mut homozygote (n = 36): 19 (52.8%)	Total number of patients with TD: wt/wt + wt/mut (n = 40): 19 (47.5%); mut/mut (n = 36): 19 (52.8%)
	Male patients with TD: wt and heterozygous (n = 26): 16 (61.5%); mut homozygote (n = 18): 6 (33.3%)	Male patients with TD: wt/wt + wt/mut (n = 26): 16 (61.5%); mut/mut (n = 18): 6 (33.3%)
	Female patients with TD: wt and heterozygous (n = 14): 3 (21.4%); mut homozygote (n = 18): 13 (72.2%)	Female patients with TD: wt/wt + wt/mut (n = 14): 3 (21.4%); mut/mut (n = 18): 13 (72.2%)
Liou 2004149	Total number of patients with TD: TT (n = 87): 39 (44.8%); CT (n = 81): 47 (58.0%); CC (n = 48): 27 (56.3%)	Total number of patients with TD: wt/wt (n = 87): 39 (44.8%); wt/mut (n = 81): 47 (58.0%); mut/mut (n = 48): 27 (56.3%)
	Male patients with TD: TT (n = 54): 20 (37.0%); CT (n = 47): 29 (61.7%); CC (n = 32): 21 (65.6%)	Male patients with TD: wt/wt (n = 54): 20 (37.0%); wt/mut (n = 47): 29 (61.7%); mut/mut (n = 32): 21 (65.6%)
	AIMS score – patients with TD, mean ± SD: TT (n = 39): 12.0 ± 6.0; CT (n = 47): 8.8 ± 4.1; CC (n = 27): 11.3 ± 6.1	AIMS score – patients with TD, mean ± SD: wt/wt (n = 39): 12.0 ± 6.0; wt/mut (n = 47): 8.8 ± 4.1; mut/mut (n = 27): 11.3 ± 6.1
Lohmann 2003150	Total number of patients with TD: ≥ 2 functional alleles (n = 68): 31 (45.6%); 1 functional alleles (n = 34): 15 (44.1%); 0 functional alleles (n = 7): 4 (57.1%)	Total number of patients with TD: wt/wt (n = 68): 31 (45.6%); wt/mut (n = 34): 15 (44.1%); mut/mut (n = 7): 4 (57.1%)
Nikoloff 2002152	Total number of patients with TD: *1/*1 (n = 24): 11 (45.8%); *1/*2 (n = 15): 8 (53.3%); *1/*2 (n = 4): 1 (25.0%); all wild/wild (n = 43): 20 (46.5%) *1/*10B (n = 82): 46 (56.1%); *1/*41 (n = 3): 1 (33.3%); *2/*10B (n = 16): 10 (62.5%); *2/*41 (n = 1): 0; all wild/decreased (n = 102): 57 (55.9%) *2/*14 (n = 1): 1 (100%); *1/*5 (n = 7): 6 (85.7%); all wild/loss (n = 8): 7 (87.5%) *10B/*10B (n = 42): 22 (52.4%); *10B/*41 (n = 3): 2 (66.7%); all decreased/decreased (n = 45): 24 (53.3%) *10B/*5 (n = 4): 2 (50.0%); all decreased/loss (n = 4): 2 (50.0%)	Total number of patients with TD: wt/wt (n = 43): 20 (46.5%); wt/mut (n = 110): 64 (58.2%); mut/mut (n = 49): 26 (53.1%)
Ohmori 1998153	Total number of patients with TD: *1/*1 (n = 26): 4 (15.4%); *1/*10 (n = 43): 9 (20.9%); *10/*10 (n = 30): 11 (36.7%)	Total number of patients with TD: wt/wt (n = 26): 4 (15.4%); wt/dec (n = 43): 9 (20.9%); dec/dec (n = 30): 11 (36.7%)
	AIMS score – all patients, mean ± SD: *1/*1 (n = 26): 1.54 ± 1.78; *1/*10 (n = 43): 2.00 ± 2.01; *10/*10 (n = 30): 3.31 ± 3.69	AIMS score – all patients, mean ± SD: wt/wt (n = 26): 1.54 ± 1.78; wt/dec (n = 43): 2.00 ± 2.01; dec/dec (n = 30): 3.31 ± 3.69
Ohmori 1999154	Total number of patients with TD: wt/wt (n = 67): 18 (26.9%); wt/m (n = 26): 5 (19.2%); m/m (n = 6): 1 (16.7%)	Total number of patients with TD: wt/wt (n = 67): 18 (26.9%); wt/mut (n = 26): 5 (19.2%); mut/mut (n = 6): 1 (16.7%)
	AIMS score – all patients, mean ± SD: wt/wt (n = 67): 2.46 ± 2.88; wt/m (n = 26): 2.00 ± 2.27; m/m (n = 6): 2.17 ± 2.41	AIMS score – all patients, mean ± SD: wt/wt (n = 67): 2.46 ± 2.88; wt/mut (n = 26): 2.00 ± 2.27; mut/mut (n = 6): 2.17 ± 2.41
Plesnicar 2006156	Total number of patients with TD: non-PM (n = 125): 22 (17.6%); PM (n = 6): 1 (16.7%)	Total number of patients with TD: wt/wt + wt/mut (n = 125): 22 (17.6%); mut/mut (n = 6): 1 (16.7%)
	AIMS score – all patients, mean ± SD: non-PM (n = 125): 5.44 ± 3.9; PM (n = 6): 5.16 ± 3.4	AIMS score – all patients, mean ± SD: wt/wt + wt/mut (n = 125): 5.44 ± 3.9; mut/mut (n = 6): 5.16 ± 3.4

In the meta-analyses no significant differences were found between genotypes for either Caucasian or Asian populations (Appendix 5, Figure 6). However, there was a significant amount of heterogeneity among the Asian studies. It should be noted that the meta-analyses included data from Lohmann et al. ,150 in which seven UMs were classified with EMs (of whom three developed persistent TD), and an unknown number of UMs (and thus an unknown number with TD) from Plesnicar et al. 156

Sensitivity analysis was conducted to include only the studies that tested for *10 and no other CYP. 138,147,149 This increased heterogeneity and the effect was again non-significant for all comparisons (data not presented).

Sensitivity analysis was also carried out excluding one study150 that did not use AIMS to define/measure TD but rather the Tardive Dyskinesia Rating Scale (TDRS). This produced almost identical findings to the original analysis (data not presented).

Further sensitivity analyses were also carried out for study type. When only cross-sectional studies were included,127,138,140,150,154,158 heterogeneity was again increased while the odds ratios (ORs) varied slightly from those in the original analysis but remained non-significant (data not presented). However, the inclusion of only prospective studies137,142,146,152 decreased heterogeneity to 0% and, for two comparisons (wt/mut versus wt/wt and mut/mut + wt/mut versus wt/wt), significant findings were found [OR 2.08 (95% CI 1.21 to 3.57) and OR 1.83 (95% CI 1.09 to 3.08) respectively; Figure 2].

FIGURE 2.

Meta-analysis for patients tested for the CYP2D6 genotype – total number of patients with tardive dyskinesia: (a) wt/wt vs mut/mut; (b) wt/wt vs wt/mut; (c) wt/wt vs mut/mut + wt/mut; (d) wt/wt + wt/mut vs mut/mut. Sensitivity analysis by study type (only prospective studies included).

The only study not included in the meta-analysis was that by Ohmori et al. 154 This was because it only genotyped *1 and *2, both of which are being considered as wt alleles for the purposes of this review for reasons explained in the methods section. Here the proportion of patients with TD was highest amongst those with the *1/*1 genotype and lowest among those with the *2/*2 genotype, although there was only one patient with this genotype. Following regression analysis the authors concluded that there was no association of the CYP2D6*2 genotype with the occurrence of TD.

Seven studies also assessed TD severity, which was usually measured using AIMS with only one study using the TDRS. Data from five of these studies could be included in the meta-analysis (but only between two and four for any given comparison because of the manner in which these studies grouped their genotypes, comprising between 136 and 264 patients). Significantly, the weighted mean difference (WMD) AIMS score was in favour of the wt/wt genotype compared with the mut/mut genotype [WMD 1.80 (95% CI 0.40 to 3.19)] (Figure 3). In the sole study that measured TD severity using the TDRS,146 comprising 45 patients, the mean scores favoured patients with the wt/wt genotype compared with the wt/mut genotype.

FIGURE 3.

Meta-analysis for patients tested for the CYP2D6 genotype – mean ± SD AIMS score for all patients: (a) wt/wt vs mut/mut; (b) wt/wt vs wt/mut; (c) wt/wt vs mut/mut + wt/mut.

It was not possible to include data from Plesnicar et al. 156 in the meta-analysis because this study of 131 patients only compared patients with the PM phenotype with non-PMs (including seven UMs). No significant difference in AIMS score was found between these groups in this study. Data from Ohmori et al. 154 were also excluded because this study only genotyped *1 and *2 and no significant differences were found across groups and the authors concluded that there was no association between the CYP2D6*2 genotype and the AIMS score.

A further two studies measured AIMS only in patients who had TD and data from these were also included in the meta-analysis (an overall patient population of between 118 and 153 depending on the genotypes being compared). Although no significant differences were found, the WMD AIMS score was in the direction of favouring wt/mut compared with mut/mut (Appendix 5, Figure 7).

Finally, Ellingrod et al. 137 compared AIMS scores by genotype (wt/wt and wt/mut) in 14 smokers and 23 non-smokers – differences were only significant in smokers, in whom the mean AIMS score was much higher in the wt/mut group.

Parkinsonism

Seven studies examined the relationship between parkinsonism and genotype. Five of these reported the total number of patients with parkinsonism and four of these reported mean Simpson–Angus Scale (SAS) scores. Patients were taking at least one typical antipsychotic in all of the studies. The findings are summarised in Table 22.

TABLE 22 - Summary of findings for parkinsonism: clinical validity studies in patients tested for CYP2D6

EM, extensive metaboliser; ESRS, Extrapyramidal Symptoms Rating Scale; NA, not applicable; PM, poor metaboliser; SAS, Simpson–Angus Scale; TDRS, Tardive Dyskinesia Rating Scale.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Andreassen 1997127	Total number of patients with parkinsonism: EM homozygote (n = 61): 23 (37.1%); EM heterozygote (n = 29): 13 (44.8%); PM (n = 10): 2 (20.0%)	Total number of patients with parkinsonism: wt/wt (n = 61): 23 (37.1%); wt/mut (n = 29): 13 (44.8%); mut/mut (n = 10): 2 (20.0%)
	SAS score – all patients, mean ± SD: EM homozygote (n = 61): 0.37 ± 0.35; EM heterozygote (n = 21): 0.40 ± 0.36; PM (n = 10): 0.56 ± 0.74	SAS score – all patients, mean ± SD: wt/wt (n = 61): 0.37 ± 0.35; wt/mut (n = 21): 0.40 ± 0.36; mut/mut (n = 10): 0.56 ± 0.74
Culav-Sumic 2001132	Total number of patients with parkinsonism: EM (n = 43): *1/*1 (n = 43): 13 (30.2%) IM (n = 23): *4/*1 (n = 20): 11 (52.6%); *6/*1 (n = 3): 1 (33.3%) PM (n = 5): *4/*4 (n = 5): 4 (80.0%)	Total number of patients with parkinsonism: wt/wt (n = 43): 13 (30.2%); wt/mut (n = 20): 12 (52.2%); mut/mut (n = 5): 4 (80.0%)
Jaanson 2002142	Total number of patients with parkinsonism: EM homozygote (n = 35): 20 (57.2%); EM heterozygote (n = 13): 8 (61.5%); PM (n = 4): 4 (100.0%)	Total number of patients with parkinsonism: wt/wt (n = 35):20 (57.2%); wt/mut (n = 13): 8 (61.5%); mut/mut (n = 4): 4 (100.0%)
Kakihara 2005145	SAS score – all patients, mean ± SD: *1/*1 (n = 16): 2.6 ± 2.0; *1/*10 (n = 14): 2.0 ± 1.7; *10/*10 (n = 9): 1.3 ± 1.5	SAS score – all patients, mean ± SD: wt/wt (n = 16): 2.6 ± 2.0; wt/mut (n = 14): 2.0 ± 1.7; mut/mut (n = 9): 1.3 ± 1.5
Panagiotidis 2007155	Median (range) ESRS parkinsonism score – peak: 3 functional alleles (n = 1): 5; 2 functional alleles (n = 16): 7 (0–13); 1 functional alleles (n = 8): 3 (0–12); 0 functional alleles (n = 1): NA Median (range) ESRS parkinsonism score – trough: 3 functional alleles (n = 1): 3; 2 functional alleles (n = 16): 5 (1–19); 1 functional alleles (n = 8): 3 (0–12); 0 functional alleles (n = 1): 2	NA
Plesnicar 2006156	Total number of patients with parkinsonism: non-PM (n = 125): 18 (14.4%); PM (n = 6): 0	Total number of patients with parkinsonism: wt/wt + wt/mut (n = 125): 18 (14.4%); mut/mut (n = 6): 0
	SAS score, mean ± SD: non-PM (n = 125): 1.35 ± 3.1; PM (n = 6): 0.16 ± 0.41	SAS score, mean ± SD: wt/wt + wt/mut (n = 125): 1.35 ± 3.1; mut/mut (n = 6): 0.16 ± 0.41
Scordo 2000158	Total number of patients with parkinsonism: UM (n = 6): 0; EM homozygote (n = 65): 20 (30.8%); EM heterozygote (n = 44): 14 (31.8%); PM (n = 4): 3 (75.0%)	Total number of patients with parkinsonism: wt/wt (n = 71: 20 (28.2%); wt/mut (n = 44): 14 (31.8%); mut/mut (n = 4): 3 (75.0%)
	SAS score – all patients, mean ± SD: hom (n = 65): 4.8 ± 1.9; mut (n = 48): 5.1 ± 1.7	SAS score – all patients, mean ± SD: wt/wt (n = 65): 4.8 ± 1.9; wt/mut + mut/mut (n = 48): 5.1 ± 1.7

For the total number of patients with parkinsonism, it was possible to include data from between four and five studies of between 233 and 470 patients in the meta-analyses depending on the genotypes being compared. Nevertheless, the number of patients with the mut/mut genotype included in the meta-analyses was still small (n < 30).

Patients with the mut/mut or wt/mut genotype were significantly more likely to develop parkinsonism than patients with wt/wt (OR 1.64, 95% CI 1.04 to 2.58) (Figure 4). It should be noted that this meta-analysis includes in the wt/wt group six patients from Scordo et al. 158 who were classified as UMs (none of whom had developed parkinsonism) and an unknown number of patients (and thus those with parkinsonism) from Plesnicar et al. 156

FIGURE 4.

Meta-analysis for patients tested for the CYP2D6 genotype – total number of patients with parkinsonism: (a) wt/wt vs mut/mut; (b) wt/wt vs wt/mut; (c) wt/wt vs mut/mut + wt/mut; (d) wt/wt + wt/mut vs mut/mut.

In two studies the criteria for measuring parkinsonism were either unknown132 or known to be different from those in the other studies142 and so sensitivity analyses were carried out removing these studies. In these sensitivity analyses none of the effect sizes was statistically significant and the new OR comparing mut/mut or wt/mut with wt/wt was now 1.21 (95% CI 0.69 to 2.14) (data not presented). A further sensitivity analysis was carried out that included only the three cross-sectional studies. 127,132,158 Again, none of the effects was statistically significant and the heterogeneity increased for all.

Regarding mean SAS score, no consistency in the results was found, with some studies reporting higher scores for wt/wt and others for mut/mut or wt/mut + mut/mut.

Acute dystonia

Data from both studies128,158 that examined dystonia in 195 patients taking any antipsychotic in relation to genotype were included in the meta-analysis and the findings are summarised in Table 23. No significant effect was found for any of the genotypes (Appendix 5, Figure 9) although the numbers of patients with the mut/mut genotype was small (n = 9). Furthermore, it should be noted that Scordo et al. 158 also included six UMs (two of whom had acute dystonia) who have been included here with the wt/wt patients and so the results should be treated with caution.

TABLE 23 - Summary of findings for acute dystonia: clinical validity studies in patients tested for CYP2D6

EM, extensive metaboliser; PM, poor metaboliser; UM, ultra rapid metaboliser.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Armstrong 1997128	Total number of patients with acute dystonia: wt homozygote (n = 43): 4 (9.3%); wt heterozygote (n = 28): 5 (17.9%); mut homozygote (n = 5): 0	Total number of patients with acute dystonia: wt/wt (n = 43): 4 (9.3%); wt/mut (n = 28): 5 (17.9%); mut/mut (n = 5): 0
Scordo 2000158	Total number of patients with acute dystonia: UM (n = 6): 2 (33.3%); EM homozygote (n = 65): 14 (21.5%); EM heterozygote (n = 44): 6 (13.6%); PM (n = 4): 1 (25.0%)	Total number of patients with acute dystonia: wt/wt (n = 71): 16 (22.5%); wt/mut (n = 44): 6 (13.6%); mut/mut (n = 4): 1 (25.0%)

Akathisia

Two studies127,156 in which a total of 231 patients were taking any typical antipsychotic quantified the number of patients with akathisia. As Plesnicar et al. 156 combined all patients who did not have the mut/mut genotype (including an unknown number of patients with the UM phenotype) then the study findings were also pooled in this manner (Appendix 5, Figure 10). Based on this meta-analysis, the number of patients with akathisia did not significantly differ between those having the mut/mut genotype and those who did although heterogeneity between the studies was large and effect sizes were in opposite directions. Plesnicar et al. 156 was the only study to measure severity and found no significant difference between the patients with the mut/mut genotype and those without. Heterogeneity may have been explained by either differences in study design or differences in the gender mix of these two studies. The findings are summarised in Table 24.

TABLE 24 - Summary of findings for akathisia: clinical validity studies in patients tested for CYP2D

EM, extensive metaboliser; PM, poor metaboliser.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Andreassen 1997127	Total number of patients with akathisia: EM homozygote (n = 61): 10 (16.4%); EM heterozygote (n = 29): 5 (17.2%); PM (n = 10): 0	Total number of patients with akathisia: wt/wt (n = 61): 10 (16.4%); wt/mut (n = 29): 5 (17.2%); mut/mut (n = 10): 0
Plesnicar 2006156	Total number of patients with akathisia: non-PM (n = 125): 6 (4.8%); PM (n = 6): 1 (16.7%) Barnes Scale score – all patients, mean ± SD: non-PM (n = 125): 0.27 ± 1.3; PM (n = 6): 0.0	Total number of patients with akathisia: wt/wt + wt/mut (n = 125): 6 (4.8%); mut/mut (n = 6): 1 (16.7%) Barnes Scale score – all patients, mean ± SD: wt/wt + wt/mut (n = 125): 0.27 ± 1.3; mut/mut (n = 6): 0.0

General chronic movement disorder

One study128 in which 76 patients were taking any antipsychotic examined the association between genotype and chronic movement disorders, which were defined as experiencing either parkinsonism or TD, or both. A much higher proportion of patients with the mut/mut genotype than with either the wt/wt or the wt/mut genotype experienced such disorders but the number of mut/mut patients was small (n = 5). The findings are summarised in Table 25.

TABLE 25 - Summary of findings for general chronic movement disorders: clinical validity studies in patients tested for CYP2D6

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Armstrong 1997128	Total number of patients with chronic movement disorders: homozygous wt (n = 43): 18 (41.9%); heterozygous wt (n = 28): 13 (46.4%); homozygous mut (n = 5): 4 (80.0%)	Total number of patients with chronic movement disorders: wt/wt (n = 43): 18 (41.9%); wt/mut (n = 28): 13 (46.4%); mut/mut (n = 5): 4 (80.0%)

Extrapyramidal symptoms in general

Six studies focused on the relationship between EPS and genotype/phenotype. It was not possible to include data from any of these in a meta-analysis because each study measured or reported the data differently. The findings are summarised in Table 26.

TABLE 26 - Summary of findings for extrapyramidal symptoms in general: clinical validity studies in patients tested for CYP2D6

EM, extensive metaboliser; EPS, extrapyramidal symptoms; ESRS, Extrapyramidal Symptoms Rating Scale; NA, not applicable; PM, poor metaboliser; SAS, Simpson–Angus Scale; UM, ultra-rapid metaboliser.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

In one table there are *3/wt and *6/*6 genotypes in patients with EPS but there are no schizophrenic patients with these genotypes in an earlier table of patient demographics (or indeed, any patients with *6/*6 at all).

Study	Outcomes as reported	‘Standardised outcome’a
Arthur 1995130	Individual patient data presented	SAS score – patients with EPS, mean ± SD: wt/wt (n = 8): 2.5 ± 2.6; wt/mut (n = 7): 3.9 ± 4.6; wt/wt + wt/mut (n = 15): 3.4 ± 3.9; mut/mut (n = 1): 20
Brockmoller 2002131	Median EPS sum score: UM (n = 5): 9; EM (n = 106): 4; IM (n = 56): 5; PM (n = 5): 8 Median EPS sum score stratified for comedication with biperiden: UM (n = 5): 7; EM (n = 106): 6; IM (n = 56): 5; PM (n = 5): 12	NA
Inada 2003140	Total number of patients vulnerable to EPS with *2: wt/wt (n = 234): 20 (8.5%); wt/mut (n = 68): 13 (19.1%); mut/mut (n = 7): 5 (71.4%)	Total number of patients vulnerable to EPS with *2: NA
	Total number of patients vulnerable to EPS with *10: wt/wt (n = 78): 10 (12.8%); wt/mut (n = 97): 16 (16.5%); mut/mut (n = 39): 6 (15.4%)	Total number of patients vulnerable to EPS with *10: wt/wt (n = 78): 10 (12.8%); wt/mut (n = 97): 16 (16.5%); mut/mut (n = 39): 6 (15.4%)
Panagiotidis 2007155	Peak ESRS parkinsonism score, median (range): 3 functional alleles (n = 1): 8; 2 functional alleles (n = 16): 7.5 (0–18); 1 functional allele (n = 8): 3.5 (0–17); 0 functional alleles (n = 1): NA Trough ESRS parkinsonism score, median (range): 3 functional alleles (n = 1): 2; 2 functional alleles (n = 16): 4.5 (1–18); 1 functional allele (n = 8): 3.5 (0–20); 0 functional alleles (n = 1): 2	NA
Scordo 2000158	Total number of patients with EPS: UM (n = 6): 3 (50.0%); EM homozygote (n = 65): 33 (50.8%); EM heterozygote (n = 44): 23 (52.3%); PM (n = 4): 4 (100.0%)	Total number of patients with EPS: wt/wt (n = 71): 36 (50.7%); wt/mut (n = 44): 23 (52.3%); mut/mut (n = 4): 4 (100.0%)
Topic 2000162	Inconsistent datab	NA

Haloperidol was taken by patients in at least three of the studies131,155,162 and possibly also in the other three,130,140,158 which stipulated that patients were taking any antipsychotic. Three studies140,158,162 quantified patients with this ADR and three130,131,155 assessed the severity. However, one162 of the studies quantifying EPS has been excluded from the analysis in this review for reasons discussed earlier in this chapter (see Quality assessment of included studies).

One study158 reported that around half of the patients carrying the wt/wt (including UMs) or wt/mut genotype developed EPS (defined as having any one of acute dystonia, TD or parkinsonism) but that all mut/mut patients had EPS, albeit the number of patients with this last genotype was only four. The other study also found that significantly more patients with the mut/mut genotype were vulnerable to EPS than patients with the wt/wt or wt/mut genotype. 140

The mean SAS scores for patients with EPS were found to be lowest in the wt/wt group and highest in the mut/mut group in one study,130 a finding echoed by median EPS sum scores in a later study,131 although, here, when the EPS sum score was not stratified for comedication with biperiden (which is taken to alleviate ADRs associated with some antipsychotics such as stiffness, tremors, spasms and poor muscle control) the score for UMs was similar to that for PMs. The most recent study155 reported the median Extrapyramidal Symptoms Rating Scale (ESRS) scores at peak and trough; however, there was only one mut/mut patient in this study and data were only available at trough for this patient making comparisons problematic.

Adverse drug reactions in general

Among patients taking any antipsychotic, one study139 of 39 patients focused on the association between genotype and mean number of ADRs as assessed by the SAFTEE (Systematic Assessment For Treatment Emergent Effects), which is a technique for the systematic assessment of side effects. This study found that neither the wt/mut or mut/mut genotype differed statistically from the wt/wt genotype in relation to disease severity or number or severity of ADRs. The findings are summarised in Table 27.

TABLE 27 - Summary of findings for ADRs in general: clinical validity studies in patients tested for CYP2D6

ADR, adverse drug reaction.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Hamelin 1999139	Number of ADRs, mean ± SD: *1/*1 (n = 23): 2 ± 2; *1/*4 (n = 15): 4 ± 5; *4/*4 (n = 1): 1	Number of ADRs, mean ± SD: wt/wt (n = 23): 2 ± 2; wt/mut (n = 15): 4 ± 5; mut/mut (n = 1): 1
	Number of ADRs and severity scores (ADR × severity), mean ± SD: *1/*1 (n = 23): 3 ± 3; *1/*4 (n = 15): 6 ± 7; *4/*4 (n = 1): 1	Number of ADRs and severity scores (ADR × severity), mean ± SD: wt/wt (n = 23): 3 ± 3; wt/mut (n = 15): 6 ± 7; mut/mut (n = 1): 1

Agranulocytosis

Dettling et al. 134 quantified the number of subjects with clozapine-induced agranulocytosis by genotype (wt/wt, wt/mut and mut/mut) in a sample of 108 patients. This study found that the occurrence of clozapine-induced agranulocytosis was similar in each group. The findings are summarised in Table 28.

TABLE 28 - Summary of findings for agranulocytosis: clinical validity studies in patients tested for CYP2D6

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Dettling 2000134	Total number of patients withclozapine-induced agranulocytosis: 3 active genes (n = 4): 1 (25.0%); 2 active genes (n = 69): 21 (30.4%); 1 active gene (n = 30): 8 (26.7%); 0 active genes (n = 5): 1 (20.0%)	Total number of patients with clozapine-induced agranulocytosis: wt/wt (n = 73): 22 (30.1%); wt/mut (n = 30): 8 (26.7%); mut/mut (n = 5): 1 (20.0%)

QTc prolongation

Thioridazine-induced QTc prolongation was assessed in relation to genotype in one study of 91 patients. 159 This study provided no evidence that patients with any particular genotype are at increased risk of QTc prolongation. The findings are summarised in Table 29.

TABLE 29 - Summary of findings for QTc prolongation: clinical validity studies in patients tested for CYP2D6

EM, extensive metaboliser; IM, intermediate metaboliser; PM, poor metaboliser.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Thanacoody 2007159 and 2003160	QTc prolongation (ms), mean ± SD: EM (n=51): 425 ± 29; IM (n = 31): 427 ± 22; PM (n = 9): 411 ± 41	QTc prolongation (ms), mean ± SD: wt/wt (n = 51): 425 ± 29; wt/mut (n = 31): 427 ± 22; mut/mut (n = 9): 411 ± 41

Weight gain

Two prospective studies examined the association between genotype and weight gain in patients taking olanzapine136 and risperidone. 148 Differences in the outcomes measured made it impossible to include these data in a meta-analysis. In a small study136 of only 11 patients it was found that those with a wt/mut genotype taking olanzapine experienced a statistically significantly larger percentage change in body mass index than the wt/wt group. Derived from multiple linear regression analysis, the other study148 of 29 patients estimated the difference in body weight to be greater in the patients with a wt/wt genotype compared with a wt/mut genotype than in those with a wt/wt genotype compared with a mut/mut genotype. The findings are summarised in Table 30.

TABLE 30 - Summary of findings for weight gain: clinical validity studies in patients tested for CYP2D6

BMI, body mass index; EM, extensive metaboliser; IM, intermediate metaboliser; NA, not available; UM, ultra rapid metaboliser.

For ease of comparison, outcomes have been summarised as wt/wt (wild type/wild type), wt/mut (wild type/mutant) or mut/mut (mutant/mutant).

Study	Outcomes as reported	‘Standardised outcome’a
Ellingrod 2002136	Baseline BMI (kg/m2), mean ± SD: *1/*1 (n=6): 28 ± 4.2; *1/*3 or *1/*4 (n = 5): 24 ± 4.0; *3/*3 or *4/*4 (n = 0): NA	Baseline BMI (kg/m2), mean ± SD: wt/wt (n = 6): 28 ± 4.2; wt/mut (n = 5): 24 ± 4.0; mut/mut (n = 0): NA
	End-point BMI (kg/m2), mean ± SD: *1/*1 (n = 6): 31.8 ± 4.1; *1/*3 or *1/*4 (n = 5): 31.4 ± 6.9; *3/*3 or *4/*4 (n = 0):NA	End-point BMI (kg/m2), mean ± SD: wt/wt (n = 6): 31.8 ± 4.1; wt/mut (n = 5): 31.4 ± 6.9; mut/mut (n = 0): NA
Lane 2000148	Difference in body weight (kg): C/C (n = 29) vs C/T (n = 37): –1.138; C/C (n = 29) vs T/T (n = 50): –0.799	Difference in body weight (kg): wt/wt (n = 29) vs wt/mut (n = 37): –1.138; wt/wt (n = 29) vs mut/mut (n = 50): –0.799

Clinical validity summary

Half of the studies included in this review genotyped for CYP2D6, a quarter for CYP1A2 and the rest for other CYP polymorphisms. Around half of the studies were prospective and around half were cross-sectional. This can make combining data into a meta-analysis problematic. When possible, sensitivity analyses were carried out to include studies of the same study type. In the majority of studies, patients were taking any antipsychotic, most often typical antipsychotics. Therefore, not all of the drugs included in any given study may have been metabolised by the CYP being investigated.

ADR outcomes were most commonly investigated with only a handful of studies measuring efficacy or metabolism using pharmacokinetic parameters. Given the multiple CYP enzymes involved in metabolism of the antipsychotics, there was variation in the CYP alleles investigated between studies, with no study undertaking a comprehensive assessment of the variants in all CYP isoforms. It is difficult to generalise the efficacy findings because of the small number of studies and a lack of any consistent effect being evident across the studies. ADR findings were also generally contradictory. Outcomes that could be included in the meta-analysis by genotype were the number of patients with TD and the mean AIMS scores, the number of patients with parkinsonism, the number of patients with acute dystonia and the number of patients with akathisia. The only significant findings were that, for CYP2D6, patients included in prospective studies were at increased risk of TD if they had the wt/mut and mut/mut + wt/mut genotypes compared with those with the wt/wt genotype; the WMD AIMS score was significantly in favour of the wt/wt genotype compared with the mut/mut genotype; and patients with the wt/wt genotype were significantly less likely to develop parkinsonism than patients with the mut/mut + wt/mut genotypes. Most, if not all, of the patients in these two TD meta-analyses were taking typical antipsychotics.

Clinical utility summary

There is currently a lack of evidence for clinical utility. The only known study findings have yet to be published in a comprehensive or peer-reviewed manner, and, because of the small size of the study, extreme caution must be taken in interpreting these findings.

Economic review

Methods for the review are presented in Chapter 3. No evidence relating to the costs and benefits of CYP testing for prescribing antipsychotics was identified. Therefore, we expanded our search to identify published literature on the costs and benefits of CYP testing for prescribing in the field of psychiatry as a whole. The aim of broadening the search was to identify the key issues that may be relevant to our decision problem.

Modelling CYP testing for prescribing antipsychotics

Although the technology being considered is a relatively simple diagnostic test, the potential implications are large and the means of representing the various aspects of the decision to employ the pharmacogenetic test are not straightforward. Figure 5 illustrates the high-level design structure for a mathematical model required to carry out the assessment, and indicates that four distinct modules are involved, each with its own assumptions and data needs.

FIGURE 5.

Outline structure for an economic model to assess CYP pharmacogenetic testing for schizophrenia prescribing.

In principle the first two modules (‘pharmacogenetic test’ and ‘clinical effects’) may be readily constructed but require the results of clinical trials relevant to the specific treatments and patient populations involved. The findings from Chapters 4 and 5 indicate that such data are currently very limited. The third (‘translational’) module depends on more empirical studies of clinician behaviour, and to date such information does not exist (see Chapter 6). Additionally, the NICE schizophrenia guidelines recommend the use of risperidone, olanzapine, quetiapine, amisulpride and zotepine for both initiation and acute episodes, but only risperidone and olanzapine are metabolised by CYP2D6, neither of which are that important, as the former has an active metabolite and the latter is primarily metabolised by other isoenzymes. This means that results from the test could, at best, only inform the choice of two out of the five currently recommended drugs. So, from a practical standpoint, there is only a slight incentive to carry out the test. As neither the evidence nor the guidelines support clinicians’ use of CYP test results to determine the most appropriate treatment strategy for patients with schizophrenia it appears to be premature to attempt a full economic modelling and evaluation exercise for this technology at this time. There are clearly important knowledge gaps that should be remedied by primary research.

When these deficiencies in the evidence have been addressed it will be necessary to develop a suitable disease and economic ‘schizophrenia module’ that can encompass all relevant aspects of schizophrenia and its treatment in the UK. To assist in future economic evaluations of CYP testing and similar technologies for schizophrenia prescribing, in the following sections we consider the necessary features of such a schizophrenia model and then use these to assess the suitability of published models for this decision problem.