Clozapine, Agranulocytosis, and Benign Ethnic Neutropenia

EDITORIAL 545 Postgrad Med J: first published as 10.1136/pgmj.2004.031641 on 2 September 2005. Downloaded from

Pharmacology and toxicology ethnic groups in the Middle East, ...... including Yemenite Jews and Jordanians, have BEN.12 13 BEN has only been reported in ethnic groups that have Clozapine, agranulocytosis, and tanned or dark skin.13 Subjects with BEN do not show increased incidence of benign ethnic neutropenia infections, and their response to infections is similar to those without BEN.13 S Rajagopal ...... CLINICAL IMPLICATIONS In the United Kingdom and Ireland, the Current knowledge and clinical implications Clozaril patient monitoring service (CPMS) supervises the prescribing of clozapine and the haematological test- lozapine is an atypical antipsycho- agranulocytosis, is more common in ing (Clozaril is the brand name of tic that is effective in treatment black people.6 A white cell count spike clozapine). The CPMS uses a lower cut resistant schizophrenia.1 The of 15% or more above the immediately C off point for patients with BEN than for National Institute for Health and preceding measurement may predict the general population (table 1). A Clinical Excellence (NICE) guidelines agranulocytosis within the next ‘‘green’’ alert indicates satisfactory for schizophrenia specify that ‘‘in indi- 75 days.7 However, as these differences count, an ‘‘amber’’ alert requires a viduals with evidence of treatment between the risk factors for agranulocy- repeat FBC test while clozapine can be resistant schizophrenia, clozapine tosis and neutropenia have been extra- should be introduced at the earliest polated primarily from epidemiological continued, and a ‘‘red’’ alert warrants opportunity’’.2 studies, they may be subject to change immediate cessation of clozapine. A severe adverse effect of clozapine as further evidence, from even larger It is important for eligible subjects to that limits its more widespread use is studies, come to light. be registered with the CPMS under the agranulocytosis. Patients who are taking The exact mechanism of clozapine BEN category, so that patients belong- clozapine need to have their full blood induced agranulocytosis is unclear. It ing to certain ethnic groups do not have counts (FBC) monitored regularly, and has been postulated that clozapine is to stop clozapine unnecessarily. This has if the total white cell and/or neutrophil metabolised to a nitrenium ion.8 The great clinical ramifications, as there is counts indicate agranulocytosis, cloza- binding of this ion to neutrophils may no other antipsychotic that has compar- pine prescription must be terminated. result in agranulocytosis. Antineutrophil able efficacy to clozapine in the treat- Among certain ethnic groups, a signifi- antibodies may be involved in mediating ment of resistant schizophrenia. In cant proportion of people have a low agranulocytosis.9 Some human leuco- addition, there is evidence that some baseline neutrophil count. This is called cyte antigen (HLA) alleles, for example ethnic groups, particularly black people, benign ethnic neutropenia (BEN). This the HLA B38 phenotype in Ashkenazi may be less likely, even in the first place, 14 editorial looks at the important issues Jews,10 have been shown to be asso- to be prescribed clozapine. These fac- associated with agranulocytosis and ciated with clozapine induced agranulo- tors may combine to further worsen the BEN in patients receiving clozapine. cytosis. prognosis of an already severely debili- tating illness in this group of patients.

As clozapine induced agranulocytosis http://pmj.bmj.com/ CLOZAPINE AND OTHER HAEMATOLOGICAL is an idiosyncratic reaction,8 it is diffi- AGRANULOCYTOSIS ABNORMALITIES cult to predict and to identify high risk Agranulocytosis occurs in about 1% of Clozapine is associated with increased patients. Also, as it is a comparatively patients taking clozapine.34 risk of eosinophilia, particularly in rare phenomenon occurring in less than Neutropenia is seen in about 3%.4 The 11 women. Eosinophilia typically occurs 1% of subjects, the number of reported risk of both agranulocytosis and neu- between weeks 3 and 5 of treatment and cases is not adequate to clearly identify tropenia is highest between 6 weeks and resolves spontaneously without need for specific risk factors; general risk factors 18 weeks after starting clozapine treat- specific treatment. Clozapine is also on September 28, 2021 by guest. Protected copyright. such as increasing age, female sex, etc, ment.4 Hence, in the United Kingdom associated with anaemia, lymphopenia, are not robust enough to change deci- and Ireland, weekly FBC monitoring is leucocytosis, and thrombocytopenia.8 mandatory for the first 18 weeks, after sion making in individual patients. which it is done fortnightly until the Therefore, clinicians should continue to end of the first year, and every four BENIGN ETHNIC NEUTROPENIA remain vigilant against this potentially weeks thereafter. In the USA, FBC is BEN has been defined as ‘‘the occur- fatal side effect of clozapine in all the monitored weekly for the first six rence of neutropenia, defined by nor- patients prescribed this drug, especially months and fortnightly thereafter. mative data in white populations, in in the first few months of treatment. Not all risk factors are the same for individuals of other ethnic groups who are otherwise healthy and who do not ACKNOWLEDGEMENTS agranulocytosis and neutropenia; this 12 implies that there may be distinct have repeated or severe infections’’. Dr Tony Wong, staff grade psychiatrist, South mechanisms for the two disorders. A About 25% to 50% of Africans and some London and Maudsley NHS Trust. low baseline white cell count has been associated with future neutropenia but Table 1 CPMS alert ranges for subjects with BEN (ranges for non-BEN subjects) not agranulocytosis.5 The risk of agra- 9 9 nulocytosis increases with age,36 while Alert colour WCC 610 /l Neutrophils 610 /l 6 that of neutropenia decreases with age. Green .3.0 (.3.5) .1.5 (.2.0) Agranulocytosis is more common in Amber 2.5–3.0 (3.0–3.5) 1.0–1.5 (1.5–2.0) women.3 It is more than twice as Red ,2.5 (,3.0) ,1.0 (,1.5) frequent in Asians as in the white CPMS, Clozaril patient monitoring service; BEN, benign ethnic neutropenia; WCC, white cell count. population.6 Neutropenia, but not

www.postgradmedj.com 546 EDITORIAL Postgrad Med J: first published as 10.1136/pgmj.2004.031641 on 2 September 2005. Downloaded from

Postgrad Med J 2005;81:545–546. and secondary care. Clinical guideline 1. London: research and implications for drug development. doi: 10.1136/pgmj.2004.031161 NICE, 2002. CNS Drugs 1997;7:139–58. 3 Alvir JM, Lieberman JA, Safferman AZ, et al. 9 Claas FH. Drug-induced agranulocytosis: review Correspondence to: Dr S Rajagopal, South Clozapine-induced agranulocytosis. Incidence of possible mechanisms, and prospects for London and Maudsley NHS Trust, Adamson and risk factors in the United States. N Engl J Med clozapine studies. Psychopharmaclogy (Berl) 1989;99(suppl):S113–17. Centre for Mental Health, St Thomas’s Hospital, 1993;329:162–7. 4 Atkin K, Kendall F, Gould D, et al. Neutropenia 10 Meged S, Stein D, Sitrota P, et al. Human London SE1 7EH, UK; Sundararajan. leukocyte antigen typing, response to [email protected] and agranulocytosis in patients receiving clozapine in the UK and Ireland. Br J Psychiatry neuroleptics, and clozapine-induced Funding: none. 1996;169:483–8. agranulocytosis in jewish Israeli schizophrenic 5 Gillman K. Paradoxical pattern of haematological patients. Int Clin Psychopharmacol Competing interests: none. risk with clozapine. Authors’ reply. Br J Psychiatry 1999;14:305–12. 2000;177:88. 11 Banov MD, Tohen M, Friedberg J. High risk of eosinophilia in women treated with clozapine. 6 Munro J, O’Sullivan D, Andrews C, et al. J Clin Psychiatry 1993;54:466–9. Active monitoring of 12,760 clozapine 12 Haddy TB, Rana SR, Castro O. Benign ethnic REFERENCES recipients in the UK and Ireland. Beyond neutropenia: what is a normal absolute neutrophil pharmacovigilance. Br J Psychiatry 1 Kane J, Honigfeld G, Singer J, et al. count? J Lab Clin Med 1999;133:15–22. 1999;175:576–80. Clozapine for the treatment-resistant 13 Shoenfeld Y, Alkan ML, Asaly A, et al. Benign schizophrenic. A double-blind comparison 7 Alvir JM, Lieberman JA, Safferman AZ. Do white- familial leukopenia and neutropenia in different with chlorpromazine. Arch Gen Psychiatry cell count spikes predict agranulocytosis in ethnic groups. Eur J Haematol 1988;41:273–7. 1988;45:789–96. clozapine recipients? Psychopharmacol Bull 14 Kuno E, Rothbard AB. Racial disparities in 2 National Institute for Clinical Excellence. 1995;31:311–14. antipsychotic prescription patterns for patients Schizophrenia: core interventions in the treatment 8 Pirmohamed M, Park K. Mechanism of clozapine- with schizophrenia. Am J Psychiatry and management of schizophrenia in primary induced agranulocytosis. Current status of 2002;159:567–72.

Single subject design generalisability. I would argue that a ...... study of this kind cannot provide any reassurance that we can extrapolate the findings to other patients. One could Should the single subject design be say, to other similar patients, but what does similar mean in this context? regarded as a valid alternative to the Demographic, physiological, and diag- nostic similarity are of little relevance randomised controlled trial? here, the only similarity that matters relates to propensity to respond to the R G Newcombe treatment in question, and this can neither be observed nor ensured...... Conversely, the conventional large clinical trial relates to patients drawn from a For debate. population defined by well defined eligibility criteria, and random alloca- n an accompanying article Janine ongoing CPD, and clinical experience tion ensures groups are comparable

Janosky sets out the case for the use facilitate the recognition of patients who within limits of chance variation in http://pmj.bmj.com/ Iof single subject designs.1 I was asked are not ‘‘average’’ and for whom current respect of all possible variables, includ- by my colleague Dr John Mayberry, the evidence based guidelines, which are ing counterfactual treatment response. editor of the journal, to referee this optimised for patient populations, may This is what justifies applying the paper, but felt it would be more appro- not be optimal. How to decide on conclusions of the trial to patients at priate to respond to it, largely to management for a specific patient may large who fulfil the eligibility criteria stimulate debate on this issue. I would be problematic. When the issue relates used in the trial. suggest that the proper applicability of to maintenance treatment, the single The other key issue relates to drawing on September 28, 2021 by guest. Protected copyright. single subject designs is much narrower patient design certainly has a role. For an ‘‘obvious’’ conclusion from a limited than this article would imply. I would example, the patient may have two dataset. This is shaky on two counts, furthermore warn readers of the dan- coexisting conditions for which the relating to clinical lability and statistical gers of a view that if left to grow therapeutic requirements conflict. methodology. Dr Janosky refers to the unchecked could result in an important Another context is polypharmacy—per- patient ‘‘in need of lower fasting blood undermining of the dominance of the haps the patient is currently taking four glucose values’’—but there is such a multi-patient randomised clinical trial drugs, and the clinician suspects that thing as regression towards the mean that is now, with very strong justifica- one could be withdrawn without (strictly, a misnomer, regression tion, accepted as the cornerstone of diminution of therapeutic effect. towards the mode would be a more apt evidence based clinical practice—with In some parts of the article, including description). The inference that the serious consequences for the choice of the six listed ‘‘possible research ques- ‘‘switch’’ in figure 1 is real is strongly appropriate management for future tions’’, Dr Janosky clearly implies that dependent on a presupposition that patients. The two key issues are equivo- the research issue only applies to a patients don’t just ‘‘switch’’ sponta- cation regarding the ambit of the single specific patient. In other places, a neously in this way. Perhaps this is subject design, and the robustness of broader scope is implied by phrases reasonable in diabetes—it would not be the inference to be drawn from data such as ‘‘unique study populations’’, for remitting/relapsing conditions such such as figure 1 in Janosky’s paper. ‘‘choosing the patient to participate’’, as inflammatory bowel disease or multi- It is well accepted that clinical exper- and ‘‘typical in terms of the practice ple sclerosis, and certainly not for tise is needed to apply the findings of demographics (and) for disease presen- thyroid disease or bipolar disorder. large clinical trials to the individual tation and progression’’. Dr Janosky What Dr Janosky terms the ‘‘primary patient. The doctor’s initial training, concedes that there is an issue of limited A-B single subject design’’, as used here,

www.postgradmedj.com COMMENTARIES 547 Postgrad Med J: first published as 10.1136/pgmj.2004.031641 on 2 September 2005. Downloaded from is particularly vulnerable to criticism— of this magnitude would not be glucose units) would be much more while it is the simplest within subject regarded as strongly convincing. A directly interpretable for clinical impor- design, it is the least adequately con- decision rule approach is more relevant. tance than a relative measure such as D. trolled, and effective blinding is unlikely This might relate to a pre-agreed clini- Clinicians are forever ‘‘trying’’ patients to be achieved. The methodological cally importantly large difference— with different treatments. Use of a single issues arising in the familiar multi- although this would share the non- subject design, with additional rigour subject crossover design are well known. robustness problem. Alternatively, one ensured by multiple periods, randomisa- Borrowing terminology commonly used could abandon conventional hypothesis tion of treatments to periods and blind- in that context, the observed treatment testing with a low a level and opt for a ing, and perhaps some statistical analysis, difference could equally be interpreted ‘‘pick the winner’’ approach with is certainly one stage more formal and as a period effect, or could be distorted implied equal a and b rates. This rigorous. But we should not imagine it is by carry-over. In the example given, the corresponds more closely to the less anything more than that: we can only treatment difference could be consider- formalised ‘‘trial and error’’ course of validly draw conclusions about that one ably confounded by seasonal differ- events that commonly occurs in clinical patient, in their present state, it would be ences. practice. very risky to extrapolate to ostensibly Furthermore, with regard to statistical Dr Janosky’s stance contrasts quite ‘‘similar’’ cases. 2 methodology, what is the implied cut off sharply with that taken by Guyatt et al. Postgrad Med J 2005;81:546–547. between a ‘‘real’’ difference and one This highly informative review of the doi: 10.1136/pgmj.2004.031641 that could be attributable to chance? For appropriate use of single subject trials the data as shown, an unpaired two was restricted to double blind, rando- Correspondence to: Professor R G Newcombe, Centre for Health Sciences Research, sample t test would give a highly mised, multiple crossover designs aim- Department of Epidemiology, Statistics and significant p value, around 0.0001 here, ing to optimise management of a Public Health, Wales College of Medicine, but we do not have sufficient evidence specific patient. Decisions about efficacy Cardiff University, Heath Park, Cardiff CF14 to decide whether an assumption of were based on a combination of a signed 4XN, UK; [email protected] Gaussian distributional form is reason- standardised difference measure of able, without considerable extrapolation effect size D and a single tailed p value. REFERENCES from data on others. The non-para- Although my reservation above concern- 1 Janosky JE. Use of the single subject design for metric Mann-Whitney test is robust, ing unquestioned use of parametric practice based primary care research. Postgrad and yields a two sided exact p value of methods still applies. Furthermore, an Med J 2005;81:549–51. 2 Guyatt GH, Keller JL, Jaeschke R, et al. The n-of-1 1 in 35 or 0.029. This is below 0.05, but effect size criterion expressed in abso- randomized controlled trial: clinical usefulness. much less extreme, normally a p value lute terms (for example, fasting blood Ann Intern Med 1990;112:293–9.

Single subject design about the probable effects of an inter- ...... vention on a population, or the effects on another, inevitably unique, person. Other potential problems with single Single subject trials in primary care subject trials are problems of bias and http://pmj.bmj.com/ the determination of statistical signifi- N A Francis cance. Effects that are likely to lead to bias in this type of trial include regres- ...... sion to the mean and ‘‘carry-over’’ Lack of generalisability limits use. effects. Values towards the extremes are likely to normalise on repetition for rom a GP’s perspective, the accom- data that can readily be applied to statistical reasons, an effect that is on September 28, 2021 by guest. Protected copyright. panying article by Janine Janosky others. The author does mention that described as regression to the mean. on the single subject design is both the generalisability of results from this This will occur without any clinical F 1 interesting and stimulating. This type type of study is limited, but goes on to change in the subject, and in a single of design, as Dr Janosky highlights, is suggest that if a subject that is ‘‘repre- subject trial is likely to lead to a false infrequently used in research and has sentative of the general type of patients impression of treatment effect when some potential advantages. Most nota- for which this intervention would be none may be present. Secondly, the bly, it is the only type of design that can used’’ then the results become more author describes using a washout period provide information about effects at an generalisable. A person can be chosen between different phases of treatment. individual level. There are obvious ben- that has a certain disease at a certain However, treatments can have lasting efits in formalising what all GPs do on a stage and with certain sociodemo- effects and it can be difficult to distin- day to day basis, namely observing the graphic characteristics. But is this per- guish whether any prior treatments or effects of individual treatments on son really representative? How do we indeed the rotation of treatments plays a individual patients. However, the article know exactly which variables are rele- part in any observed effect. suggests a scope and potential for the vant to the effect being shown? And Furthermore, once a subject has ‘‘chan- n = 1 trial that I would take issue with, how can we judge ‘‘biological represen- ged’’, how can you ever really know and the author fails to adequately tation’’. For example, are there aspects what you are comparing to? Can change describe the limits and disadvantages of a person’s metabolism that influences revert back to its original state in every of this type of design. the way they respond to a drug? way? After all, you can only ever be a While single subject designs have the Furthermore, demonstrating an effect virgin once! potential of examining effects at an in a person, even if the person is similar Ways of attempting to minimise bias individual level, they do not provide to a population, provides little evidence in n = 1 trials, as the author points out,

www.postgradmedj.com 548 COMMENTARIES Postgrad Med J: first published as 10.1136/pgmj.2004.031641 on 2 September 2005. Downloaded from include randomisation, blinding, and it not possible that some other, unmea- 2 Larson EB, Ellsworth AJ, Oas J. Randomized clinical trials in single patients during a 2-year multiple treatment phases. An effect sured variable changed during the period. JAMA 1993;270:2708–12. that is observed by a blinded observer, course of this study, and that this was in a blinded patient, each time a the cause of the change in measured treatment is randomly introduced, and fasting blood glucose? For example, the that disappears each time the treatment subject may have independently taken is withdrawn, is more likely to represent up a new form of exercise. AUTHOR’S COMMENTS a true treatment effect. However, trials In conclusion, Janosky’s paper on I am pleased to read the commentaries that entail randomisation, blinding, and single subject trials has relevance to that accompany my article (see page multiple phases are likely to be more both GP clinicians and researchers. In 549). Each of the commentaries raises difficult to implement in primary care, primary care, making treatment deci- relevant issues that serve to illuminate and more costly. Larson et al estimate sions with patients and then monitoring the strengths and limitations of single the costs of formal n = 1 trials at $450 to their response to those treatments is a subject research designs. As with all $500 per patient.2 This is much less than daily occurrence. This, on occasion, methodological designs, each affords large scale RCTs, but still a significant entails a number of phases, where either strengths and limitations to answer cost for a primary care clinician inter- different treatments or no treatment are tailored research questions. The research ested in evaluating a treatment on a tried. Formalising this process, and questions best answered, through the application of a single subject research patient. Furthermore, if the single sub- consideration of the use of blinding or design, are questions that garner their ject trial is repeated in a number of placebo treatments, provides opportu- interest in the potential to have patients to try and prove generalisabil- nities for primary care clinicians to high internal validity while acknowl- ity, the costs could easily add up to the assess the effectiveness of treatments edging that the external validity is most sort of amount that a small scale RCT on a person in a more comprehensive probably weak. The single subject would cost. A series of n = 1 trials and rigorous fashion. Conducting formal n = 1 studies, compared with the design can be easily contrasted with a combined could thus be seen as a long, informal treatment trial, can however, clinical trial or true randomised experi- drawn out, ‘‘sequential RCT’’. Some have significant time, cost, and ethical mental study that aims to examine comments from the author on the considerations. Given these constraints, effectiveness where the strength of the feasibility and methods for combining it is unlikely that this type of trial will design is its strong external validity. The the results of multiple single subject have wide applicability in everyday results of these clinical trials provide trials, and the applicable lessons that general practice. Furthermore, it must conclusions regarding effectiveness on have been learnt from cross over trials, be emphasised that the results from this the average treatment effects for only would prove very useful for those inter- type of study can generally not be the studied populations; however at ested in conducting studies of this kind. applied to anyone other than the person times, the results from these clinical The working example given by the that has been studied, and that this trials might not always be applicable author shows fasting blood glucose design is particularly vulnerable to in determining the most effective values before and after a comprehensive certain types of bias as described above. treatment for an individual patient. intervention for diabetes management. A single subject research design, Postgrad Med J 2005;81:547–548. This is a two phase, A-B, design that although limited for external validity, doi: 10.1136/pgmj.2005.032581 shows fasting blood glucose values after provides an opportunity to examine the intervention as being lower than Correspondence to: Dr N Francis, Department the applicability of the study findings of General Practice, Health Centre, Llanederyn, those before the intervention. The to a specific patient. As methodologists, http://pmj.bmj.com/ author concludes that, ‘‘it seems that Cardiff CF23 9PN, UK; [email protected] having an arsenal of research designs the intervention was effective in low- REFERENCES with strengths and limitations iden- ering the measured fasting blood glu- tified for each affords us a comprehen- cose in this subject.’’ Is this a fair 1 Janosky JE. Use of the single subject design for practice based primary care research. Postgrad sive means to answer diverse research conclusion from this type of study? Is Med J 2005;81:549–51. questions. on September 28, 2021 by guest. Protected copyright.