Antipsychotic medication for elderly people with schizophrenia

Abstract

Background

A large and growing number of older people across the world suffer from schizophrenia. Recommendations for their treatment are largely based on data extrapolated from studies of the use of antipsychotic medications in younger populations. In addition most manufacturers of such medications recommend prescription of reduced doses to the elderly. The evidence base for these assumptions is unclear and raises obvious questions regarding the appropriateness of such prescribing practice.

Objectives

To find and assimilate good evidence of the effects of antipsychotic medication for treatment of schizophrenia in people over 65 years of age.

Search methods

We searched the Cochrane Schizophrenia Group's Register (May 2003). We inspected references of all included studies for further trials and contacted relevant pharmaceutical companies.

Selection criteria

All clinical randomised trials evaluating antipsychotic drugs for schizophrenia and schizophrenia‐like psychoses in older people.

Data collection and analysis

We extracted data independently. For homogenous dichotomous data, the random effects, relative risk (RR), and 95% confidence interval (CI) and, where appropriate, the numbers needed to treat (NNT) were calculated on an intention‐to‐treat basis. For continuous data, we calculated weighted mean differences (WMD).

Main results

Two hundred and fifty two elderly people with schizophrenia participated in three relevant randomised controlled studies. We were unable to extract usable data on quality of life, satisfaction, service use, or economic outcomes. One small study (n=18) compared thioridazine with remoxipride (RR leaving the study early 1.0 CI 0.07 to 13.6). A second study (n=175) compared risperidone with olanzapine. Global state 'not improved/worse' was not significantly different between treatments (n= 171, RR 1.26 CI 0.8 to 1.9); mental state PANSS total endpoint scores were also equivocal (n=171, RR 0.98 CI 0.76 to 1.26) as were all cognitive function tests. The third study (subset n=59) compared olanzapine with haloperidol and mental state change scores (BPRS WMD ‐3.60 CI ‐10.8 to 3.6; PANSS WMD ‐6.00 CI ‐18.3 to 6.3) were equivocal.

Authors' conclusions

Antipsychotics may be widely used in the treatment of elderly people with schizophrenia, however, based on this systematic review, there are little robust data available to guide the clinician with respect to the most appropriate drug to prescribe. Clearly reported large short, medium and long‐term randomised controlled trials with participants, interventions and primary outcomes that are familiar to those wishing to help elderly people with schizophrenia are long overdue.

Plain language summary

Antipsychotic medication for elderly people with schizophrenia

Since the early 1950's the mainstay of treatment for schizophrenia has been drugs such as haloperidol and chlorpromazine. Although effective in controlling voices and fixed, false beliefs, these drugs are reported as having potentially disabling adverse effects such as tremor, stiffness and slowing of movement. The newer generation of drugs are reputed to be freer from these problems. This is particularly important for older people, who are more likely to experience adverse effects. Most manufacturers recommend prescription of reduced doses in the elderly.

This review examines trials of these drugs for the treatment of elderly people with schizophrenia. We found three small, short trials. These did not have much information that was usable and we really could not draw any firm conclusions, except that such studies are possible and more are needed as a matter of urgency.

Background

The number of older people in communities throughout the world is rising. In the most recent UK census (2001) the proportion of people over 60 years of age surpassed the figures for the under 16's for the first time (21% versus 20%). There are now 9.4 million people over the age of 65 in the UK, a 51% increase since 1961, and the figure is projected to rise to 13 million by 2025. These trends are even more marked in those over 85 years who now number 1.1 million, three times more than in 1961 (UK ONS 2003). 
 
 Worldwide the situation is similar. The most recent UN statistics reveal that in the year 2000 there were 606 million people over 60 years, and by 2050 the figure is expected to be nearly 2 billion. The figures for the over 80s are expected to rise over five times from 69 million to 379 million in 2050. In most developed countries, as in Britain, the proportion of people over 60 years is about 20% but is expected to rise to 33% by 2050 with two people over 60 to every child. In less developed countries the rise will be even more marked (from 8% to 20% by 2050) (UN Popin 2003).

Schizophrenia is a common mental disorder with a lifetime risk of between 7.0 and 13.0 per 1000 (Jablensky 1986, Regier 1988, Kessler 1994). The prognosis for patients presenting with a first episode of schizophrenia is unclear, however various studies suggest that while up to a third make a good recovery, the remainder suffer from persistent symptoms or a relapsing and remitting course requiring on‐going treatment (Bleuler 1974, Ciompi 1980, Huber 1975). It is estimated that 0.1‐1.0% of the over 65 population has schizophrenia (Sajatovic 2000), including up to 12% of all nursing home residents (Tariot 1993).

Although the life course of schizophrenia into later adulthood is not well studied, it is generally recognised that, whereas psychotic symptoms tend to remit with time, negative symptoms, such as social withdrawal and emotional apathy, increase in frequency and intensity with advancing age (Sajatovic 2000). Ten percent of all first admissions over 60 years of age to psychiatric hospitals are diagnosed with late onset paranoid disorders (Naguib 1995). Late onset schizophrenia (illness onset after 40 years of age) and very late onset schizophrenia (illness onset after 60 years of age) tend to follow a relatively benign course, with predominant positive symptoms and fewer negative symptoms, although relapse is frequent, often as a consequence of medication non‐compliance (Howard 2000). Antipsychotic treatment of people with late and very late onset schizophrenia is the focus of a companion review (Arunpongpaisal 2003) and will not be covered by this one. This review includes the larger group of people who have grown older with a diagnosis of schizophrenia having been made before the age of 65.

With regard to treatment of schizophrenia, most manufacturers of antipsychotic medication recommend prescription of reduced doses to the elderly. There are logical reasons why an older person might metabolise these medications less rapidly and might be more susceptible to adverse effects. However the use of antipsychotic medications in older people with schizophrenia is largely based on data extrapolated from studies of these agents in younger populations (Katz 1999, Frenchman 1997, Devanand 1998). It is unclear how these studies relate to older patients, particularly those who have been on antipsychotic medication for many years. A number of open studies have been conducted with elderly participants with schizophrenia and suggest that the atypical antipsychotics may be of benefit (Czobor 1995, Davidson 2000).

Objectives

To estimate the effects of antipsychotic medication for treatment of schizophrenia in people over 65 years of age (this being the standard recognised division between working age and older adults in the UK).

It is also proposed to investigate whether low doses of antipsychotic medications have different effects when compared with standard or high doses for the treatment of positive and negative symptoms of schizophrenia in older patients.

Methods

Criteria for considering studies for this review

Types of studies

All relevant randomised controlled trials. Where a trial was described as 'double‐blind' but it was implied that the study was randomised, these trials were included in a sensitivity analysis. If there was no substantive difference within primary outcomes (see types of outcome measures) when these 'implied randomisation' studies were added, then they were included in the final analysis. If there was a substantive difference only clearly randomised trials were utilised and the results of the sensitivity analysis described in the text. Quasi‐randomised studies, such as those allocating by using alternate days of the week, were excluded.

Types of participants

Older people, or cohorts of adults of whom at least 80% were over 65 years of age, resident in any setting, with a diagnosis of schizophrenia as defined in individual studies of schizophrenia or other schizophrenia‐like psychosis (including delusional disorder, schizophreniform psychosis, paraphrenia and schizoaffective disorder). Studies including other diagnoses that may also present with psychotic phenomena (e.g. bi‐polar affective disorder and organic causes of altered mental state e.g. delirium and dementia) were excluded unless 80% or more of the participants had a diagnosis of schizophrenia or other schizophrenia‐like psychosis. If a study reported that some, but not over 80% of patients were over 65 years, then we attempted to contact the authors in order to obtain separate data relating to those patients who were over 65 years.

Types of interventions

1. Atypical antipsychotic drugs 
 All these drugs share the common characteristics of reduced risk of acute and chronic movement disorders compared with typical antipsychotic agents (Kerwin 1994). We included in this category both the 'new' serotonin‐dopamine antagonists (SDA's); clozapine, iloperidone, olanzapine, quetiapine, risperidone, ziprasidone, zotepine and aripiprazole, a dopamine partial agonist, together with older agents with no significant action on the serotonin receptor (e.g. amisulpiride, sulpiride) at any dose and mode or pattern of administration.

2. Typical antipsychotic drugs 
 For the purposes of this review these are any antipsychotic drugs exerting their effect through blockade of D2 receptors, without the specific characteristics described above (Seeman 1987). This includes benperidone, chlorpromazine, flupenthixol, fluphenazine, haloperidol, levomepromazine, loxapine, mesoridazine, molindone, oxypertine, pericyazine, perphenazine, pimozide, prochlorperazine, promazine, thioridazine, thiothixene, trifluoperazine and zuclopenthixol, at any dose and mode or pattern of administration.

If there were cases where a study reported the effects of a drug confirmed as being an antipsychotic through reference to the standard pharmacopoeia (Martindale 1999), but difficult to categorise it as being either typical or atypical, a psychopharmacologist, blinded to the study details, was contacted to aid allocation of category.

3. Placebo

Types of outcome measures

As schizophrenia is often a life‐long illness, and anti‐psychotic drugs are used as an ongoing treatment, we grouped outcomes according to time periods: acute treatment (less than 6 weeks), short term (less than 3 months), medium term (3‐6 months), and long term (more than 6 months).

Primary outcomes

1. Global state

1.1 Relapse

2. Mental state 
 2.1 No clinically important change in general mental state

3. Adverse effects 
 3.1 No clinically important general adverse effects

Secondary outcomes

1. Death ‐ suicide and natural causes

2. Global state 
 2.1 Time to relapse 
 2.2 No clinically important change in global state 
 2.3 Not any change in global state 
 2.4 Average endpoint global state score 
 2.5 Average change in global state scores

3. Service outcomes 
 3.1 Hospitalisation 
 3.2 Time to hospitalisation

4. Mental state 
 4.1 Not any change in general mental state 
 4.2 Average endpoint general mental state score 
 4.3 Average change in general mental state scores 
 4.4 No clinically important change in specific symptoms 
 4.5 Not any change in specific symptoms 
 4.6 Average endpoint specific symptom score 
 4.7 Average change in specific symptom scores

5. Leaving the study early 
 5.1 For specific reasons 
 5.2 For general reasons

6. General functioning 
 6.1 No clinically important change in general functioning 
 6.2 Not any change in general functioning 
 6.3 Average endpoint general functioning score 
 6.4 Average change in general functioning scores 
 6.5 No clinically important change in specific aspects of functioning, such as social or life skills 
 6.6 Not any change in specific aspects of functioning, such as social or life skills 
 6.7 Average endpoint specific aspects of functioning, such as social or life skills 
 6.8 Average change in specific aspects of functioning, such as social or life skills

7. Behaviour 
 7.1 No clinically important change in general behaviour 
 7.2 Not any change in general behaviour 
 7.3 Average endpoint general behaviour score 
 7.4 Average change in general behaviour scores 
 7.5 No clinically important change in specific aspects of behaviour 
 7.6 Not any change in specific aspects of behaviour 
 7.7 Average endpoint specific aspects of behaviour 
 7.8 Average change in specific aspects of behaviour

8. Adverse effects 
 8.1 Not any general adverse effects 
 8.2 Average endpoint general adverse effect score 
 8.3 Average change in general adverse effect scores 
 8.4 No clinically important change in specific adverse effects 
 8.5 Not any change in specific adverse effects 
 8.6 Average endpoint specific adverse effects 
 8.7 Average change in specific adverse effects

9. Engagement with services 
 9.1 No clinically important engagement 
 9.2 Not any engagement 
 9.3 Average endpoint engagement score 
 9.4 Average change in engagement scores

10. Satisfaction with treatment 
 10.1 Recipient of care not satisfied with treatment 
 10.2 Recipient of care average satisfaction score 
 10.3 Recipient of care average change in satisfaction scores 
 10.4 Carer not satisfied with treatment 
 10.5 Carer average satisfaction score 
 10.6 Carer average change in satisfaction scores

11. Quality of life 
 11.1 No clinically important change in quality of life 
 11.2 Not any change in quality of life 
 11.3 Average endpoint quality of life score 
 11.4 Average change in quality of life scores 
 11.5 No clinically important change in specific aspects of quality of life 
 11.6 Not any change in specific aspects of quality of life 
 11.7 Average endpoint specific aspects of quality of life 
 11.8 Average change in specific aspects of quality of life

12. Economic outcomes 
 12.1 Direct costs 
 12.2 Indirect costs

Search methods for identification of studies

Electronic searches

1. Electronic searching 
 We searched the Cochrane Schizophrenia Groups trials register (April 2003) using the phrase: [(((elderly* or *elderly* in abstract terms or title) or (old* or * old *) in title in abstract, index terms or title) or (aged* in index terms) of REFERENCE) or (elderly in participants of STUDY))].

This register is compiled by up‐to‐date methodical searches of BIOSIS, The Cochrane Library CINAHL, Dissertation abstracts, EMBASE, LILACS, MEDLINE, PSYNDEX, PsycINFO, RUSSMED, Sociofile, and is supplemented with hand searching of relevant journals and numerous conference proceedings (see Group Module).

Searching other resources

1. References 
 We inspected the references of all the identified studies for further trials.

2. Personal contact 
 We contacted authors of included trials for additional studies.

3. Pharmaceutical industry 
 We contacted relevant pharmaceutical companies for information about additional studies.

Data collection and analysis

1. Selection of studies 
 We independently selected study citations identified from the trial search in order to assess their relevance for inclusion according to the selection criteria. Each study citation was inspected by RM and/or WN and/or SW. If studies were identified as relevant or if there was any disagreement about inclusion, we obtained the full article. In cases of dispute we inspected the full article and attempted to reach an agreement through discussion. If this failed, we were assisted by the third reviewer to try to resolve any doubts. If agreement could still not be reached, we contacted the authors for more information and the article was allocated to the list of 'those awaiting assessment' until additional information could be acquired to resolve the situation.

2. Assessment of quality of methodology 
 We assessed the methodological quality of each study independently (RM and/or WN and/or SW). The criteria we used were based upon those described in the Cochrane Handbook (Higgins 2005). Trials were included if they were allocated to category A or B with regards to randomisation (Cochrane Handbook criteria: A=randomisation and concealment, B=just randomisation, C=improper randomisation; A & B suggest a low‐moderate risk of bias). Where disagreement arose, we attempted to resolve it through discussion; if doubts still remained, we discussed these with the third reviewer; if we were still unable to reach consensus the authors were contacted for more information and the article was allocated to the list of those awaiting assessment.

3. Data management

3.1 Data extraction 
 We (RM and/or WN and/or SW) independently extracted data from the selected trials. In cases of dispute, we discussed any doubts in order to reach agreement. If this failed the third reviewer was involved to assist but if there were still outstanding issues a decision was made to contact the authors for more information and the article was allocated to the list of those awaiting assessment. In this case, data were not entered unless additional information became available.

3.2 Intention to treat analysis 
 We excluded data from outcomes of studies where more than 50% of participants in any group were lost to follow up. In outcomes with less than 50% dropout rate, we assumed people leaving early to have had the negative outcome, except for the event of death. We analysed the impact of including outcomes with high attrition rates (25‐50%) in a sensitivity analysis. If inclusion of data from this latter group did result in a substantive change in the estimate of effect, their data were not added to trials with less attrition, but presented separately.

4. Data synthesis 
 In the assessment of outcome, two types of data were used ‐ dichotomous and continuous.

4.1 Dichotomous data 
 Dichotomous data were used wherever possible. If a continuous measure had been used and the trialist had identified an appropriate cut‐off point, we used this. In other cases, where continuous data could be made binary, we agreed a suitable cut‐off point, with one of us blind to the data. We calculated a standard estimation of the random effects risk ratio (RR) and its 95% confidence interval (CI). The random effects model enables differences between studies to be taken into account. Where possible we estimated the number needed to treat (NNT) or the number needed to harm (NNH) taking into account the risk of an event in the control group (Cates 2005). We carried out an intention‐to‐treat analysis. In the calculation, all people lost to follow‐up were assumed to have had a poor outcome.

4.2 Continuous data

4.2.1 Skewed data 
 The statistics used in meta‐analysis were designed for use with parametric data. However, the outcome data from trials of anti‐psychotic drugs are frequently skewed. To allow for this potential problem, standards were applied to all data before inclusion: (a) standard deviations (SDs) and means were reported in the paper (or were obtainable from the authors), (b) for data with finite limits, such as endpoint scale data, the standard deviation (SD) when multiplied by two was less than the mean (otherwise the mean was unlikely to be an appropriate measure of the centre of distribution (Altman 1996)). (c) if a scale started from a positive value (such as PANSS which can have values from 30 to 210) the calculation described above was modified to take the scale starting point into account. In these cases skew is present if 2SD>(S‐Smin), where S is the mean score and Smin is the minimum score. When continuous data are presented on a scale which includes a possibility of negative values (such as change on a scale), it is difficult to tell whether data are non‐normally distributed (skewed) or not. Skewed data from studies of less than 200 participants would have been entered in additional tables rather than into an analysis. Skewed data pose less of a problem when looking at means if the sample size is large and would have been entered into a synthesis. If these standards were not met, we reported these data separately.

4.2.2 Summary statistic 
 For continuous outcomes, we estimated a weighted mean difference (WMD) between groups. Again, we used a random effects model.

4.2.3 Endpoint versus change data 
 Where possible we presented endpoint data and if both endpoint and change data were available for the same outcomes then we only reported the former. In cases of change data (end‐point minus baseline) the situation is more difficult. Without individual patient data it is impossible to know if the data are skewed. Since RevMan meta‐analyses of continuous data are based on the assumption that data are normally distributed, this is problematic. Although some data were likely to be skewed, following consultation with a statistics mailing list, we entered data into RevMan to allow the available information to be summarised. This was based on the assumption that RevVan analysis is able to cope with some degree of skew within the data.

4.2.4 Valid scales 
 There are many rating scales available to measure mental‐health outcomes within trials. The quality of such scales varies, with some lacking validity and/or reliability. We subsequently set minimum standards for accepting a measurement instrument. These were that psychometric properties of the instrument must have been described in a peer‐reviewed journal (Marshall 2000) and that the instrument should either be a self‐report or completed by an independent rater or relative (not the therapist).

4.2.5 Cluster trials 
 Studies increasingly employ 'cluster randomisation' (such as randomisation by clinician or practice) but analysis and pooling of clustered data poses problems. Firstly, authors often fail to account for intra class correlation in clustered studies, leading to a 'unit of analysis' error (Divine 1992) ‐ whereby p values are spuriously low, confidence intervals unduly narrow and statistical significance overestimated ‐ causing type I errors (Bland 1997, Gulliford 1999). Where data were presented corrected by a design effect (see below) data were pooled with non‐cluster studies.

Where data were reported as if from a non‐cluster randomised study, and the analyses were based on the numbers of individuals, with no account taken of the clustering effect, we sought statistical advice from the MRC Biostatistics Unit, Cambridge, UK. They advised that the binary data as presented in the report should be divided by a 'design effect' and that this was calculated using the mean number of families in the groups (m) and the intra‐class correlation co‐efficient (ICC) [Design effect = 1+(m‐1)*ICC]. If authors of the study did not or were unable to supply ICC we assumed this to be 0.1 (Ukoumunne 1999).

Where clustering was not accounted for in primary studies, we presented continuous data in a table, with an (*) symbol ‐ to indicate the presence of a probable unit of analysis error. In subsequent versions of this review we will seek to contact first authors of studies to seek intra‐class correlation co‐efficients of their clustered data and to adjust for this using accepted methods (Gulliford 1999). Where cluster studies were appropriately analysed taking into account intra‐class correlation coefficients and relevant data documented in the report, synthesis of binary data with other studies was possible using the generic inverse variance technique.

5. Sensitivity analysis 
 5.1 Subgroup analysis according to population age with respect to the above outcome measures. 
 This was to have been carried out by a meta‐regression analysis, factoring in the associated dose and only for pre‐designated primary outcomes.

5.2 Subgroup analysis according to medication type (i.e. typical versus atypical antipsychotic) only for pre‐designated primary outcomes. 
 This was to have been carried out by a meta‐regression analysis, factoring in the associated drugs and only for pre‐designated primary outcomes.

6. Investigation of heterogeneity 
 Firstly, we undertook consideration of all the included studies within any comparison to estimate clinical heterogeneity. Then we used visual inspection of graphs to investigate the possibility of statistical heterogeneity. This was supplemented employing, primarily, the I‐squared statistic. This provides an estimate of the percentage of inconsistency thought to be due to chance. Where the I‐squared estimate included 75%, this was interpreted as evidence of high levels of heterogeneity (Higgins 2003). We did not summate data with 75% or greater I‐squared statistic, but presented these separately and investigated the reasons for heterogeneity.

7. Small study bias 
 The likelihood of small study bias, and perhaps publication bias, was investigated by entering all data from studies initially identified, as well as those finally selected, into a funnel graph plotting trial effect against trial size (Egger 1997).

Results

Description of studies

For substantive descriptions of the studies please see Included and Excluded Studies tables.

1. Excluded studies 
 We excluded a total of 111 studies from the review; the majority (n=86) because they failed to meet our inclusion criteria with respect to age (i.e. 80% of patients shown to be aged 65 years or over), with a further 14 studies excluded as less than 80% of participants had a stated and confirmed diagnosis of schizophrenia. Five studies were not randomised (Ayers 1960, Haggstrom 1980, Harvey 2004b, Hellewell 2000, Mohler 1970) and one reported different outcome measures than those with which this review was concerned (Andia 1998 measured plasma homovanillic acid levels in women with schizophrenia). Interventions other than the use of antipsychotic medications for the treatment of schizophrenia were used in three studies ‐ namely the use of amitriptyline‐fluphenazine combination versus placebo (Baldini 1970), and treatments for tardive dyskinesia (Bateman 1979, Glazer 1990). Two trials were withdrawn as the study drug (ziprasidone) was not granted a UK licence in time for inclusion in the study (Singh 2002, Woodruff 2002) and one was excluded as the data provided in the paper was unusable and no further details were obtainable from the authors (Barak 2000). Finally, we excluded Friedman 2003 as it was a review article.

2. Awaiting assessment 
 Four studies currently await assessment, all of which require translation, three from Chinese (Chen 2003, Lin 2002, Wang 2002), and one from Czech (Vencovsky 1975).

3. Ongoing studies 
 We have identified three ongoing studies (Lacro 2001, Matkovits‐Gupta 1999, Newcomer 2001) and contacted the authors have been contacted for more information.

4. Included studies 
 We identified three studies that met the inclusion criteria (Harvey/Jeste 2003, Phanjoo 1990, Tollefson 1997). All were described as randomised and double‐blind. Consent was described in all cases.

4.1 Length of trials 
 All trials were restricted to short‐term follow‐up only (less than 12 weeks). Harvey/Jeste 2003 occurred over an 8 week period, preceded by one week where previously prescribed antipsychotic medications were tapered off. The two other studies, Phanjoo 1990 and Tollefson 1997, were restricted to 6 weeks with the former preceded by a 2‐5 day wash‐out period (for oral antipsychotics),or discontinuation of depot antipsychotics for at least one dose interval before the start of active medication, and the latter preceded by a 2 to 9 day screening period after which they included eligible patients.

4.2 Participants 
 All three studies included participants with clearly operationalised diagnoses. In no cases were the patients restricted to those with a diagnosis of schizophrenia alone. One study included people with schizophrenia or schizoaffective disorder (Harvey/Jeste 2003), one included those with either schizophrenia, schizoaffective disorder or schizophreniform psychosis (Tollefson 1997) and one included people with either schizophrenia, paranoia, acute paranoid disorder or atypical paranoid disorder (Phanjoo 1990). Where sex was specified, the majority of participants were women whose mean age was in excess of 65 years. All studies report clear exclusion criteria. Some excluded people on the basis of disease severity (Tollefson 1997 required a minimum BPRS score of 18) and all contained within their exclusion criteria the presence of concomitant mental and/or physical disorders (a diagnosis of substance abuse or dependence during the previous 3 months, major depressive episode in the previous 6 months, major disorders of the nervous system, including dementia and those illnesses associated with abnormal involuntary movements, serious somatic illness including renal or hepatic insufficiency, QTc interval of >500msec). Harvey/Jeste 2003 excluded people judged unresponsive to oral antipsychotics (namely risperidone or olanzapine) and Phanjoo 1990 excluded those with a history of serious allergic or toxic reaction to any drugs.

4.3 Setting 
 Harvey/Jeste 2003 and Tollefson 1997 were described as multicentre, recruiting patients from both hospital and the community in all cases. Also Harvey/Jeste 2003 and Phanjoo 1990 recruited from nursing homes and 'board and care' homes.

4.4 Study size 
 Harvey/Jeste 2003 was the largest study, recruiting 175 people, whilst Phanjoo 1990 randomised only 18. Although Tollefson 1997 reports a total of 1996 patients recruited after screening, only 59 were aged 65 years or older (mean age 69 years) so only the data on these participants were included in this review.

4.5 Interventions 
 The studies used the following oral antipsychotics: risperidone (dose 1‐3 mg/day, mean 1.9 mg/day) versus olanzapine (dose 5‐20 mg/day, mean 11.1 mg/day) (Harvey/Jeste 2003); olanzapine (5‐20 mg/day, mean 12.4 mg/day) versus haloperidol (5‐20 mg/day, mean 8.7 mg/day) (Tollefson 1997) and remoxipride (200 mg/day, mean and maximum dose) versus thioridazine (mean 133 mg/day, maximum dose 200 mg/day) (Phanjoo 1990).

4.6 Outcomes 
 Improvement ‐ Harvey/Jeste 2003 defined improvement as a 20% reduction in PANSS (total) scores. From additional data obtained from the author, we were also able to present CGI data dichotomised as improved versus unchanged or worse. Tollefson 1997 presented dichotomised data for BPRS results (with a reduction of 40% or more compared with baseline scores), however, these data were on a subset of patients and it the particular proportion included was unclear. Harvey/Jeste 2003 reported data for death, somnolence, insomnia, dizziness, agitation, constipation, headache, diarrhoea and extra‐pyramidal adverse effects. Tollefson 1997 reported usable data for dry mouth and constipation. Unfortunately the improvement data from Phanjoo 1990 were not presented in a usable fashion. Much of the other improvement data presented by Tollefson 1997, and Harvey/Jeste 2003 were in the form of continuous end‐point or change, making comparisons difficult. Making data binary makes it more clinically understandable. Also it may be wasteful of time and effort, if a simple yes/no question on clinical efficacy would suffice rather than a large number of BPRS and PANSS questions. However, the validity of dichotomising is unclear.

4.6.1 Outcome scales: details of the only scales that provided usable data are shown below. Reasons for exclusions of data are given under 'outcomes' in the 'included studies' section.

4.6.1.1 Global state 
 4.6.1.1.1 Clinical Global Impression Scale ‐ CGI Scale (Guy 1976). 
 This is used to assess both severity of illness and clinical improvement, by comparing the conditions of the person standardised against other people with the same diagnosis. A seven‐point scoring system is usually used with low scores showing decreased severity and/or overall improvement. Harvey/Jeste 2003 reported change data for this scale.

4.6.1.2. Mental state 
 4.6.1.2.1. Positive and Negative Symptom Scale ‐ PANSS (Kay 1987) 
 This schizophrenia scale has 30 items, each of which can be defined on a seven‐point scoring system varying from 1 ‐ absent to 7 ‐ extreme. This scale can be divided into 3 sub‐scales for measuring the severity of general psychopathology, positive symptoms (PANSS‐P), and negative symptoms (PANSS‐N). A low score indicates lesser severity. Harvey/Jeste 2003 and Tollefson 1997 reported data for this scale.

4.6.1.2.2. Brief Psychiatric Rating Scale ‐ BPRS (Overall 1962) 
 This is used to assess the severity of abnormal mental state. The original scale has 16 items, but a revised 18‐item scale is commonly used. Each item is defined on a seven‐point scale varying from 'not present' to 'extremely severe', scoring from 0‐6 or 1‐7. Scores can range from 0‐126, with high scores indicating more severe symptoms.Tollefson 1997 reported data for this scale.

4.6.1.2.3. Montgomery‐Asberg Depression Rating Scale ‐MADRS (Montgomery 1979) 
 A 65‐item comprehensive psychopathology scale was used to identify the 17 most commonly occurring symptoms in primary depressive illness. Ratings are based on 10 items, with higher scores indicating more symptoms.Tollefson 1997 reported data for this scale.

4.6.1.3. Cognitive function 
 4.6.1.3.1. Continuous Performance Test ‐ CPT (Cornblatt 1989) 
 In this test of vigilance, patients are asked to press a computer key whenever the same four‐digit target stimulus occurs twice in a row. The dependent variable is the signal detection index, which is the response sensitivity for discrimination of target and non‐target stimuli. Harvey/Jeste 2003 reported data for this scale.

4.6.1.3.2. Serial Verbal Learning Test ‐ SVLT (Morris 1989) 
 Patients are presented with a ten‐item list of words in three separate learning trials. Dependent variables are the total number of words recalled in the three learning trials and at delayed recall. Harvey/Jeste 2003 reported data for this scale.

4.6.1.3.3. Trail Making Test ‐ TMT (Spreen 1998) 
 A test of visuomotor speed (part A) and the ability to alternate between sets (part B). The time taken to complete the test is the dependent variable. Harvey/Jeste 2003 reported data for this scale.

4.6.1.3.4. Wisconsin Card Sorting Test ‐ WCST (Heaton 1993) 
 A test of executive functioning, cognitive flexibility, maintenance of a cognitive set and working memory. The dependent variables used are the number of categories completed and the total number of errors. Harvey/Jeste 2003 reported data for this scale.

4.6.1.3.5. Cognitive function ‐ Verbal Fluency Examination ‐ VFE (Lezak 1997) 
 A test of category and phonologic fluency. The dependent variables are the total scores for category and letter fluency. Harvey/Jeste 2003 reported data for this scale.

4.6.1.4. Adverse effects 
 4.6.1.4.1. Abnormal Involuntary Movement Scale ‐ AIMS (Guy 1976) 
 This has been used to assess tardive dyskinesia, a long‐term, drug‐induced movement disorder and short‐term movement disorders such as tremor. Tollefson 1997 reported data for this scale.

4.6.1.4.2. Barnes Akathisia Scale ‐ BAS (Barnes 1989) 
 The scale comprises items rating the observable, restless movements that characterise akathisia, a subjective awareness of restlessness, and any distress associated with the condition. These items are rated from 0 ‐ normal to 3 ‐ severe. In addition, there is an item for rating global severity (from 0 ‐ absent to 5 ‐ severe). A low score indicates low levels of akathisia. Tollefson 1997 reported data for this scale.

4.6.1.4.3. Extrapyramidal Syndrome Rating Scale ‐ ESRS (Chouinard 1980) 
 This consists of a questionnaire relating to parkinsonian symptoms (nine items), a physician's examination for parkinsonism and dyskinetic movements (eight items), and a clinical global impression of tardive dyskinesia. High scores indicate severe levels of movement disorder. Harvey/Jeste 2003 reported data for this scale.

4.6.1.4.4. Simpson‐Angus Scale ‐ SAS (Simpson 1970) 
 This ten‐item scale, with a scoring system of 0‐4 for each item, measures drug‐induced parkinsonism, a short‐term drug‐induced movement disorder. A low score indicates low levels of parkinsonism. Tollefson 1997 reported data for this scale.

Risk of bias in included studies

1. Randomisation 
 All included studies were said to be randomised but none described how this was undertaken. No study stated that those in charge of allocation were blind to the method. We categorised all trials 'B' (see methods section).

2. Blindness 
 All studies were described as being double‐blind, but we could find no testing of this.

3. Loss to follow up 
 Tollefson 1997 and Phanjoo 1990 analysed their data on an intention‐to‐treat basis using LOCF (last observation carried forwards) analysis. Once participants leave a study, unless the trialists continue to follow and collect data, assumptions have to be made about outcome. Although this is a common method of accounting for missing observations it could introduce bias, and may well overestimate any treatment effect.

Drop‐out rates were variable (from 11% in Phanjoo 1990, to 23% in Harvey/Jeste 2003), however these trials described the patient disposition well, clearly informing the reader of the reason the individual withdrew from the study. Tollefson 1997 did not include specific data regarding participants who left the study early.

4. Data reporting 
 Overall much of the data we found could not be used due to poor reporting. Findings which are presented as graphs, in percentiles or just reported as inexact p‐values, are often of little use to a reviewer. Many studies failed to provide standard deviations when reporting mean changes on a particular outcome measure. We are seeking further data from the first authors of relevant trials.

Effects of interventions

1. Search 
 Our initial search identified over 300 citations. Many studies were reported in different media and it was important to clarify which papers and conference proceedings related to the same study and the same patients.

It was clear from the references that most were not relevant and we went on to retrieve all material available relating to 114 studies. As described above, only three trials met our inclusion criteria. A further 49, however, included some people over 65 years of age. Tollefson 1997 included 1996 patients with schizophrenia over the age of 18 of whom 339 were aged 50 years or over, 117 were aged 60 years or over and 59 were aged 65 years or over. Patients 65 years and over subsequently made up 3% of the total included in the study (in which all participants were aged 18 years and over), 17% of participants were aged 50 years and over and 50% of patients aged 60 years and over. The data on this sub‐group were not presented separately but were kindly supplied for our use in this review by the company (Eli Lilly). Using these proportions as a guide we estimated the number of patients aged 65 years or over in these 49 studies. Approximately 450 patients aged 65 years or over may have been included in these studies. This compares with the 253 patients we were able to include in our analysis. If studies likely to have less than ten people aged 65 years and over are excluded, that may still leave eight studies with a total of over 300 patients aged 65 years or over. There were a further 12 studies who reported no upper age range. If it is assumed that, like Tollefson 1997, they included patients aged 65 years or over at a proportion of 3%, then there may be a further 127 patients in this age group that have been excluded from the review. Six of these studies could, by the same assumption, have included ten or more participants aged 65 years or over, with a total of 103 patients. Data on possibly over 400 patients in the target population of the review could subsequently have been missed because separate data for them was unobtainable.

2. COMPARISON 1: THIORIDAZINE versus REMOXIPRIDE 
 Only Phanjoo 1990 (n=18) compared the efficacy of thioridazine with remoxipride.

2.1 Leaving the study early 
 The only usable data available were on leaving the study early. The patient who left the thioridazine arm did so because of adverse effects (somnolence), whilst the person who left the remoxipride arm refused to remain in hospital (n=18, RR 1.0 CI 0.1 to 13.6).

3. COMPARISON 2: RISPERIDONE versus OLANZAPINE 
 Only Harvey/Jeste 2003 (n=175) compared the efficacy of risperidone with olanzapine

3.1 Death: suicide and natural causes 
 One patient in the olanzapine arm of the study died of a myocardial infarction during the 8 weeks of the study period. Investigators did not relate this to the medication and there were no statistically significant differences between the two groups (RR 0.34 CI 0.01 to 8.2).

3.2 Global state: not improved or worse. 
 Harvey/Jeste 2003 did not report these data in the original paper, however, we were able to obtain the data from the company (Janssen). Numbers of patients in the seven categories from very much improved, through unchanged to very much worse were presented. We were subsequently able to dichotomise this as improved versus not improved or worse. We found no statistically significant differences between the two interventions (n= 171, RR 1.26 CI 0.8 to 1.9).

3.3 Mental state: Not improved (PANSS decrease of less than 20%) 
 The trialists defined clinical improvement as a PANSS total score reduction of greater than or equal to 20%. Again these data were not published in the original paper, but information was available through direct contact with Janssen. These data revealed that this degree of improvement was achieved in 58% of the risperidone group compared with 59% of those allocated olanzapine (n=171, RR 0.98 CI 0.8 to 1.3).

3.3.1 Mental state: Average endpoint score (PANSS, high=poor) 
 Harvey/Jeste 2003 presented end point data for PANSS total (n=171, WMD 0.00 CI ‐5.45 to 5.45), together with PANSS positive subscale (n=171, WMD 0.20 CI ‐1.8 to 2.2) and the negative subscale (n=171, WMD ‐0.30 CI ‐2.3 to 1.7). This research group found no indication that either risperidone or olanzapine was more effective for this outcome.

3.4 Leaving the study early 
 Harvey/Jeste 2003 reported 'leaving the study early'. Twenty four patients discontinued treatment in the risperidone group compared with 17 in the olanzapine group (n=175, RR 1.43 CI 0.8 to 2.5).

3.5 Cognitive function 
 Trialists sub‐divided assessments of cognitive function into attention, memory and executive domains and Harvey/Jeste 2003 used two methods of assessing the attention component of cognitive function (CPT'd and Trail‐making part A). In neither case were there any statistically significant differences identified between the two interventions (CPT'd WMD ‐0.14 CI ‐0.4 to 0.10); Trail‐making part A (WMD ‐0.59 CI ‐13.7 to 12.5).

Similarly Harvey/Jeste 2003 presented data with respect to two methods of assessing the memory domain of cognitive function (Serial verbal test, total learning and Delayed recall). In neither case did data reveal significant differences between the two interventions (Serial verbal test, total learning WMD 0.34 CI ‐1.1 to 1.8; Delayed recall WMD 0.11 CI ‐0.5 to 0.7).

Harvey/Jeste 2003 used a number of tools to assess the executive domain of cognitive function (Trail making part B, WCST categories, WCST total errors, verbal fluency total). None of the tests yielded statistically significant differences between the two interventions (Trail making part B WMD ‐8.41 CI ‐26.95 to 10.1; WCST categories WMD 0.10 CI ‐0.3 to 0.5; WCST total errors WMD ‐3.93 CI ‐10.1 to 2.3; verbal fluency total WMD 0.78 CI ‐2.4 to 3.97).

4. COMPARISON 3: OLANZAPINE versus HALOPERIDOL (all short term data) 
 Only Tollefson 1997 compared the efficacy of olanzapine with haloperidol. In the original paper, data on the over 65 age group were not published separately. However, we were able to obtain data on this group from the drug company (Eli Lilly). These data are from the sub‐group of 59 elderly people (total trial size 1996).

4.1 Mental state 
 Change data revealed no statistically significant differences between olanzapine and haloperidol for BPRS (WMD ‐3.60 CI ‐10.8 to 3.6). This also applied to PANSS scores (total PANSS WMD ‐6.00 CI ‐18.3 to 6.3; PANSS Positive subscale WMD 0.00 CI ‐3.3 to 3.3; Negative subscale WMD ‐1.90 CI ‐5.4 to 1.6). MADRS scores were also equivocal (WMD ‐4.70 CI ‐10.3 to 0.9).

4.2 Adverse events 
 4.2.1 Average change scores on continuous data scales 
 Data were available for akathisia (AIMS), and extrapyramidal adverse effects (SAS and BAS). There were no statistically significant differences between olanzapine and haloperidol with respect to the frequency or severity of adverse events (BAS WMD ‐1.00 CI ‐1.8 to ‐0.2; AIMS WMD ‐0.30 CI ‐2.8 to 2.2; SAS WMD ‐2.90 CI ‐6.3 to 0.5).

4.2.2 Discrete events 
 Data were available for the incidence of dry mouth and constipation. Although the trial found a slightly increased incidence of dry mouth with olanzapine, this was not statistically significant (n=59, RR 0.34 CI 0.02 to 5.1). Nor were there statistically significant differences found between the olanzapine and haloperidol groups with respect to constipation (n=59, RR 0.51 CI 0.1 to 2.8).

Discussion

1. Limited data 
 We found few studies fulfilling our selection criteria.

All available data were short term, with trials restricting data collection to 6 or 8 weeks. Considering that schizophrenia is a relapsing and remitting disorder and drug treatments are likely to be used by people with schizophrenia for long periods, one would expect more long‐term data to be available. The short term nature of the included trials limits the applicability of the findings of this review for routine care.

Data were often not used because of high numbers leaving the study early. The degree of loss to follow‐up is common in similar studies, but rare in everyday practice. This also casts doubt on the applicability of findings to routine care.

Much global effect, mental state and cognitive function data are either missing or unusable despite the efforts we made to contact authors to acquire the necessary information. In addition, it is disappointing and remarkable that despite considerable investment in clinical trials, data remain so limited on tolerability, social functioning, death, satisfaction, quality of life, direct and indirect costs of treatment.

2. Applicability of findings 
 Two of the three included trials were international multicentre trials (Harvey/Jeste 2003 and Tollefson 1997) however they both tended to reflect a western bias in patient recruitment and this should be taken into consideration when clinicians consider the use of antipsychotic medications in different care settings. Participants were drawn from both hospital inpatient and community settings and had operationally diagnosed disorders mostly uncomplicated by co‐morbidity. The large Tollefson 1997 study may be considered to be more representative of day‐to‐day practice, as those with co‐morbid disorders were not excluded.

3. Reporting of adverse data 
 Only Tollefson 1997 reported usable adverse event data. Although trialists in the other studies stated an intention to collect such data, they either failed to report it, or reported it in an unusable form. Therefore, the potential for rare adverse effects having been ignored cannot be overlooked.

4. The search 
 Several of the studies were reported many times. Without the use of unique study identifiers this gives the impression that there are more data than actually exist and facilitates erroneous double‐counting. Although we have found no evidence to suggest that authors have attempted to conceal the fact that the large numbers of presentations relate to a limited number of studies, the inclusion of the unique company trial codes in each presentation would have facilitated piecing together the linked trials.

5. COMPARISON: THIORIDAZINE versus REMOXIPRIDE 
 One small study could be included in this comparison. Data were only available for the outcome of leaving the study early and no differences between treatments were found. Remoxipride is off the market and so this comparison is now of historical interest only. Thioridazine is now not as widely prescribed in the West as it has been in previous years and it has been taken off the market in many countries. However it is still available in many countries including India, Thailand and Chile.

6. COMPARISON: RISPERIDONE versus OLANZAPINE 
 It would have been surprising to find any differences between the two compounds in such a small study and, indeed, none were found. The one death in this study may have been a chance event and should not necessarily be interpreted as an untoward effect of either drug.

These are important drugs with many clinicians considering using them in this age group despite the recent concerns about increased risk of mortality (FDA 2005). More data are needed. This trial should be replicated so that possible differences between the compounds (such as those evident in the review of these two drugs in the younger adult population (Jayaram 2005)) are brought to light.

7. COMPARISON: OLANZAPINE versus HALOPERIDOL 
 Only 56 people of the correct age group have been involved in a trial of these two compounds. The study was too short and small to really find any important differences, although we do recognise that it was never designed to focus specifically on the elderly. With the recent difficulties of risperidone (FDA 2005), olanzapine may well be being used more widely and more studies focusing on olanzapine are awaited. Until then the advantage of this atypical antipsychotic for the elderly over the older drug haloperidol remains in doubt.

Authors' conclusions

Implications for practice.

1. For elderly people with schizophrenia. 
 For elderly people with schizophrenia, available evidence would suggest that there is little to choose between risperidone, olanzapine, haloperidol, thioridazine and remoxipride for the treatment of the mental state effects or cognitive effects of schizophrenia. Since publication of the study by Phanjoo 1990, both thioridazine and remoxipride have been withdrawn from use and are no longer an option for treatment of schizophrenia in any age group.

2. For clinicians 
 In these small trials, most recipients of antipsychotic medication were only studied for a maximum of eight weeks. Based on the data which is currently available, this systematic review can provide few reliable conclusions about the use of antipsychotic medications in the treatment of schizophrenia in the elderly. In the absence of any robust evidence, the clinician must balance the possible benefits against the potential adverse effects of treatment and continue to extrapolate from evidence which was not focused on elderly people. Important ethical questions are raised by current prescribing practices for elderly people with schizophrenia. It is difficult to know whether current practice is justified outside of a well designed, conducted and reported randomised study.

3. For managers/policy makers. 
 Atypical antipsychotics are expensive when compared with typical antipsychotics. For those making decisions on cost‐effectiveness and cost‐utility, there are currently no useful trial‐derived data to help decision making. Consequently, decisions continue to made based on opinion and habit.

Implications for research.

1. General 
 Clear and strict adherence to the CONSORT statement (Moher 2001) for all outcomes would have resulted in this review being more informative. Denominator data were not always clearly presented and some results that were described as 'significant' failed to report raw data.

2. Specific 
 We had to inspect a great number of presentations of the few trials included in this review and it is likely that many more may still be untraced. Using specific trial identifiers within the multiple publications would greatly decrease confusion over identification of the source trial.

Funders of studies may wish to make this important group of people with schizophrenia a priority for future research. Certainly, more well‐planned, conducted and reported randomised controlled trials are needed to run, over longer periods, and address important, relevant outcomes. Funding that is as free from conflicts of interest as possible is needed. High quality trial‐derived data relating to hospital admission, satisfaction with care, carer burden and important side effects are lacking. Dichotomous, clinically useful data relating to mental state and behaviour need replication and expansion. Useful cost‐effectiveness data must be a research priority.

What's new

Date	Event	Description
11 November 2009	Amended	Contact details updated.

History

Protocol first published: Issue 1, 2004
 Review first published: Issue 1, 2006

Date	Event	Description
2 October 2008	Amended	Converted to new review format.
4 September 2005	New citation required and conclusions have changed	Substantive amendment

Data and analyses

Comparison 1. THIORIDAZINE versus REMOXIPRIDE (all short term data).

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Leaving the study early	1	18	Risk Ratio (M‐H, Random, 95% CI)	1.0 [0.07, 13.64]

1.1. Analysis.

Comparison 1 THIORIDAZINE versus REMOXIPRIDE (all short term data), Outcome 1 Leaving the study early.

Comparison 2. RISPERIDONE versus OLANZAPINE (all short term data).

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Death: Suicide and natural causes	1	175	Risk Ratio (M‐H, Random, 95% CI)	0.34 [0.01, 8.16]
2 Global state: Not improved or worse (CGI)	1	171	Risk Ratio (M‐H, Random, 95% CI)	1.26 [0.83, 1.94]
2.1 Sub‐category	1	171	Risk Ratio (M‐H, Random, 95% CI)	1.26 [0.83, 1.94]
3 Mental state: 1. Not improved (PANSS decrease of less than 20%)	1	171	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.76, 1.26]
4 Mental state: 2. Average endpoint score (PANSS, high = poor)	1		Mean Difference (IV, Random, 95% CI)	Subtotals only
4.1 total	1	171	Mean Difference (IV, Random, 95% CI)	0.0 [‐5.45, 5.45]
4.2 positive	1	171	Mean Difference (IV, Random, 95% CI)	0.20 [‐1.76, 2.16]
4.3 negative	1	171	Mean Difference (IV, Random, 95% CI)	‐0.30 [‐2.25, 1.65]
5 Leaving the study early	1	175	Risk Ratio (M‐H, Random, 95% CI)	1.43 [0.83, 2.47]
6 Cognitive function	1		Mean Difference (IV, Random, 95% CI)	Subtotals only
6.1 attention domain ‐ Continuous Performance Test	1	104	Mean Difference (IV, Random, 95% CI)	‐0.14 [‐0.38, 0.10]
6.2 attention domain ‐ Trail Making Part A	1	147	Mean Difference (IV, Random, 95% CI)	‐0.59 [‐13.70, 12.52]
6.3 executive domain ‐ Trail Making Part B	1	141	Mean Difference (IV, Random, 95% CI)	‐8.41 [‐26.95, 10.13]
6.4 executive domain ‐ WCST categories	1	126	Mean Difference (IV, Random, 95% CI)	0.10 [‐0.30, 0.50]
6.5 executive domain ‐ WCST total errors	1	126	Mean Difference (IV, Random, 95% CI)	‐3.93 [‐10.11, 2.25]
6.6 executive domain ‐ verbal fluency ‐ total production	1	146	Mean Difference (IV, Random, 95% CI)	0.78 [‐2.41, 3.97]
6.7 memory domain ‐ delayed recall	1	147	Mean Difference (IV, Random, 95% CI)	0.11 [‐0.45, 0.67]
6.8 memory domain ‐ Serial Verbal Test ‐ total learning	1	147	Mean Difference (IV, Random, 95% CI)	0.34 [‐1.07, 1.75]

2.1. Analysis.

Comparison 2 RISPERIDONE versus OLANZAPINE (all short term data), Outcome 1 Death: Suicide and natural causes.

2.2. Analysis.

Comparison 2 RISPERIDONE versus OLANZAPINE (all short term data), Outcome 2 Global state: Not improved or worse (CGI).

2.3. Analysis.

Comparison 2 RISPERIDONE versus OLANZAPINE (all short term data), Outcome 3 Mental state: 1. Not improved (PANSS decrease of less than 20%).

2.4. Analysis.

Comparison 2 RISPERIDONE versus OLANZAPINE (all short term data), Outcome 4 Mental state: 2. Average endpoint score (PANSS, high = poor).

2.5. Analysis.

Comparison 2 RISPERIDONE versus OLANZAPINE (all short term data), Outcome 5 Leaving the study early.

2.6. Analysis.

Comparison 2 RISPERIDONE versus OLANZAPINE (all short term data), Outcome 6 Cognitive function.

Comparison 3. OLANZAPINE versus HALOPERIDOL (all short term data).

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Mental state: 1. Average change score (BPRS, decline = good)	1	59	Mean Difference (IV, Random, 95% CI)	‐3.60 [‐10.84, 3.64]
2 Mental state: 2. Average change score (PANSS, high = poor)	1		Mean Difference (IV, Random, 95% CI)	Subtotals only
2.1 total	1	59	Mean Difference (IV, Random, 95% CI)	‐6.0 [‐18.31, 6.31]
2.2 positive	1	59	Mean Difference (IV, Random, 95% CI)	0.0 [‐3.34, 3.34]
2.3 negative	1	59	Mean Difference (IV, Random, 95% CI)	‐1.90 [‐5.42, 1.62]
3 Mental state: 3. Not improved (MADRS decrease of less than 50%)	1	59	Mean Difference (IV, Random, 95% CI)	‐4.70 [‐10.30, 0.90]
4 Adverse events: 1. Average change scores (decline = good)	1		Mean Difference (IV, Random, 95% CI)	Subtotals only
4.1 akathisia (Barnes Akathisia Scale)	1	59	Mean Difference (IV, Random, 95% CI)	‐1.0 [‐1.79, ‐0.21]
4.2 extrapyramidal (Abnormal Involuntary Movement Scale)	1	59	Mean Difference (IV, Random, 95% CI)	‐0.30 [‐2.82, 2.22]
4.3 extrapyramidal (Simpson‐Angus Scale)	1	59	Mean Difference (IV, Random, 95% CI)	‐2.9 [‐6.34, 0.54]
5 Adverse events: 2. Discrete events	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
5.1 dry mouth	1	59	Risk Ratio (M‐H, Random, 95% CI)	0.34 [0.02, 5.12]
5.2 constipation	1	59	Risk Ratio (M‐H, Random, 95% CI)	0.51 [0.09, 2.77]

3.1. Analysis.

Comparison 3 OLANZAPINE versus HALOPERIDOL (all short term data), Outcome 1 Mental state: 1. Average change score (BPRS, decline = good).

3.2. Analysis.

Comparison 3 OLANZAPINE versus HALOPERIDOL (all short term data), Outcome 2 Mental state: 2. Average change score (PANSS, high = poor).

3.3. Analysis.

Comparison 3 OLANZAPINE versus HALOPERIDOL (all short term data), Outcome 3 Mental state: 3. Not improved (MADRS decrease of less than 50%).

3.4. Analysis.

Comparison 3 OLANZAPINE versus HALOPERIDOL (all short term data), Outcome 4 Adverse events: 1. Average change scores (decline = good).

3.5. Analysis.

Comparison 3 OLANZAPINE versus HALOPERIDOL (all short term data), Outcome 5 Adverse events: 2. Discrete events.

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Harvey/Jeste 2003.

Methods	Allocation: randomised. Blindness: double. Duration: 8 weeks, preceded by a 1 week wash‐out period. Design: multicentre. Setting: hospital and community including nursing home residents and board and care patients, North America.
Participants	Diagnosis: schizophrenia and schizoaffective disorder (DSM‐IV). N=175. Consent: described. Age: over 60, mean 71 years. Sex: M 62, F 113. History: not reported. Exclusions: DSM‐IV diagnosis of substance abuse or dependence during the previous 3 months; major depressive episode as defined by DSM‐IV criteria within 6 months before screening; MMSE score of less than 18; major disorder of the central nervous system (incl. dementia); history of neuroleptic malignant syndrome; QTc interval of >500msec; any acute unstable, significant, or untreated medical illness; patients judged to be unresponsive to risperidone or olanzapine; renal, hepatic, or gastrointestinal disease that could potentially interfere with the absorption, excretion or metabolism of trial medications.
Interventions	1. Risperidone: dose range 1‐3 mg/day (mean 1.9 mg/day). N=87. 2. Olanzapine: dose range 5‐20 mg/day (mean 11.1 mg/day). N=88.
Outcomes	Leaving the study early. Cognitive function: CPT, SVLT, TMT(A+B), WCST, VFE. Mental state: PANSS (positive, negative). Global state: CGI. Adverse events: ESRS. Death. Unable to use ‐ Cognitive function: MMSE (data incomplete). Mental state: Ham‐D (data incomplete), ACS (scale not published in peer review journal).
Notes	Concomitant medications disallowed during the study period included antidepressants, anticonvulsants, mood stabilizers other antipsychotics, sedatives, anxiolytics, and other psychoactive drugs (except lorazepam during washout period, for agitation). Results reported in two main groups ‐ mental state and side‐effect data (Jeste et al): cognitive function data (Harvey et al).
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Phanjoo 1990.

Methods	Allocation: randomised. Blindness: double. Duration: 6 weeks, preceded by a 2‐5 day wash‐out period. Depot neuroleptics were stopped at least one dose interval before the start of active medication. Design: parallel, single centre. Setting: hospital in‐patients including nursing home residents and board and care patients, North America.
Participants	Diagnosis: schizophrenia, paranoia, acute paranoid disorder, atypical paranoid disorder (DSM‐IV). Consent: described. N=18. Age: over 65, range 66‐95 years. Sex: M 1, F 17. History: diagnosis as above and considered to be in need of antipsychotic treatment. Exclusions: alcohol/drug dependence; major depressive episode; severe dementia; serious somatic illness, including renal or hepatic insufficiency; illnesses associated with abnormal involuntary movements; history of allergic or toxic reaction to drugs.
Interventions	1. Remoxipride: dose 200 mg/day (mean and max dose). N=9. 2. Thioridazine: dose mean 133 mg/day, max 200 mg/day. N=9.
Outcomes	Leaving the study early. Unable to use ‐ Mental state: BPRS (no SD), NOSIE (no SD), CVAS (no SD). Cognitive function: GBS (no data). Global state: CGI (no SD). Adverse events: AIMS (no SD), SAS (no data).
Notes	Concomitant medications disallowed during the study period included antidepressants and other antipsychotics. Chloral hydrate or a short acting benzodiazepine could be given as a hypnotic at night. In the event of intolerable extrapyramidal symptoms an anticholinergic could be given and procyclidine was recommended. Non‐psychotropic medication could be given at the investigators discretion.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Tollefson 1997.

Methods	Allocation: randomised. Blindness: double. Duration: 6 weeks, preceded by 2‐9 day screening period after which eligable patients were included. Design: multicentre. Setting: hospital and community, Europe and North America.
Participants	Diagnosis: schizophrenia, schizoaffective disorder and schizophreniform psychosis (DSM‐III‐R). Consent: described. N=59*. Age: >18, of whom 59 were 65 years or older, mean 69 years. Sex: male and female, numbers not specified. History: minimum BPRS score of 18 and/or intolerant of current antipsychotic, excluding haloperidol. Exclusions: not reported.
Interventions	1. Olanzapine: dose range 5‐20 mg/day (mean 12.4 mg/day). N=44. 2. Haloperidol: dose range 5‐20 mg/day (mean 8.7 mg/day). N=15.
Outcomes	Mental state: BPRS (total), MADRS, PANSS (total), PANSS (positive), PANSS (negative) Adverse events: AIMS, BAS, SAS, dry mouth, constipation. Unable to use ‐ Adverse events: akathisia (no SD), back pain (no data), rhinitis (no data), tremor (no SD), weight gain (endpoint data only).
Notes	*subset of N=1998.
Risk of bias
Bias	Authors' judgement	Support for judgement
Allocation concealment?	Unclear risk	B ‐ Unclear

Diagnostic tools 
 DSM III‐R and DSM‐IV ‐ Diagnostic Statistical Manual version 3 Revised and version 4

Rating Scales:

Cognitive function ‐ 
 CPT ‐ Continuous performance test 
 GBS ‐ Gotheries, Brane & Steen (rating scale for dementia) 
 MMSE ‐ Mini mental state examination 
 SVLT ‐ Serial verbal learning test 
 TMT ‐ Trail making test 
 VFE ‐ Verbal fluency examination 
 WCST ‐ Wisconsin card sorting test

Global rating scales ‐ 
 CGI ‐ Clinical Global Impressions

Mental state ‐ 
 BPRS ‐ Brief Psychiatric Rating Scale 
 CVAS ‐ Crichton visual analogue scale 
 Ham‐D ‐ Hamilton depression scale 
 MADRS ‐ Montgomery‐Asberg Depression Rating Scale 
 NOSIE ‐ Nurses observation scale for in‐patient evaluation 
 PANSS ‐ Positive and Negative Syndrome Scale

Side effects ‐ 
 ACS ‐ Anticholinergic symptom survey 
 AIMS ‐ Abnormal Involuntary Movement Scale 
 BAS ‐ Barnes Akathisia Scale 
 BMI ‐ Body mass index 
 ESRS ‐ Extrapyramidal Syndrome Rating Scale 
 SAS ‐ Simpson‐Angus Index ‐ for neurological side effects

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Altman 1973	Allocation: randomised. Participants: people with schizophrenia, <80% were over 65.
Andia 1998	Allocation: randomised. Participants: people with schizophrenia, < 80% were over 65.
Angst 1973	Allocation: randomised. Participants: people with schizophrenia, < 80% were over 65.
Ayers 1960	Allocation: not randomised.
Azorin 2001	Allocation: randomised. Participants: people with schizophrenia and poor response to previous treatment, age range 18‐65 yrs.
Baldini 1970	Allocation: randomised. Participants: people with schizophrenia and depressive symptoms, age range 17‐82 yrs. Interventions: amitriptyline‐fluphenazine combination versus placebo (not antipsychotic alone).
Bamrah 2000	Allocation: randomised. Participants: people with dementia and psychosis (not schizophrenia).
Barak 2000	Allocation: randomised. Participants: people with schizophrenia. Interventions: risperidone versus typical antipsychotics. Outcomes: no usable data (unable to obtain further details from authors).
Barak 2002	Allocation: randomised. Participants: people with schizophrenia age range 63‐83; mean age quoted differently in abstract, 72.7 (sd 5.9), and methods, 69.2 (sd 6.1); assuming a normal distribution, the abstract figure would suggest that 80% of participants were over 65 years, whereas the methods figure would not support this; we were unable to contact the author to clarify this.
Bateman 1979	Allocation: randomised Participants: people with tardive dyskinesia secondary to long term neuroleptics. Interventions: treatment for tardive dyskinesia not schizophrenia.
Beasley 2003	Allocation: randomised. Participants: people with schizophrenia or schizoaffective disorder, age range 18‐65 yrs.
Berman 1995	Allocation: randomised. Participants: people with schizophrenia, age range 57‐77 yrs, (proportion over 65 not stated, no information provided by authors).
Bora 1968	Allocation: randomised. Participants: people with schizophrenia, age range 50‐80 yrs, <80% over 65.
Bossie 2003	Allocation: randomised. Participants: people with schizophrenia or schizoaffective disorder, participants <65 years of age.
Brauzer 1968	Allocation: randomised. Participants: people with schizophrenia, <80% over 65.
Breier 1999	Allocation: randomised. Participants: people with schizophrenia, < 80% over 65.
Buruma 1982	Allocation: randomised. Participants: people with tardive dyskinesia and underlying psychiatric syndromes, age range 39‐70 years (not clear if diagnosis is schizophrenia). Interventions: tiapride and placebo (studied with respect to their effect on tardive dyskinesia, not as a treatment for schizophrenia).
Cantillon 1998	Allocation: randomised. Participants: people with schizophrenia, age not stated (unclear if 80% over 65, no information provided by the authors).
Canuso 2003	Allocation: randomised. Participants: people with schizophrenia, age range 18‐65 years.
Cassano 1975	Allocation: randomised. Participants: people with schizophrenia, age range 19‐75 yrs (mean 38 yrs, < 80% over 65).
Chouinard 1979	Allocation: randomised within blocks. Participants: people with schizophrenia; no information regarding age. Interventions: alpha‐methyldopa versus rubidium versus tryptophan‐benserazide with chlorpromazine.
Chouinard 1983	Allocation: randomised. Participants: people with schizophrenia, age range 18‐57 years.
Citrome 2001	Allocation: randomised. Participants: people with schizophrenia and schizoaffective disorder, age range 18‐60 years.
Cody 1977	Allocation: randomised. Participants: people with schizophrenia, aged over 14. Interventions: various antipsychotic medications offered at nominal cost versus full price, no further details.
Conley 2001	Allocation: randomised. Participants: people with schizophrenia and schizoaffective disorder, age range 18‐64 years.
Coryell 1998	Allocation: randomised. Participants: people with schizophrenia, age over 17 years (mean 32).
Crow 1986	Allocation: randomised. Participants: people with first episode schizophrenia, age over 59 years, no further details.
Csernansky 2002	Allocation: randomised. Participants: people with schizophrenia and schizoaffective disorder, age range 18‐65 years.
Daniel 1999	Allocation: randomised. Participants: people with schizophrenia and schizoaffective disorder, age range 18‐67 yrs, <80% over 65.
de Jong 1965	Allocation: randomised. Participants: people with schizophrenia, age range 35‐75 yrs, <80% over 65.
Delwaide 1979	Allocation: unclear. Participants: people with bucco‐lingual‐facial dyskinesias. Interventions: thioperazine, tiapride and placebo; given to treat dyskinesia not as a treatment for schizophrenia.
Denijs 1973	Allocation: randomised. Participants: people with schizophrenia, age range 26‐80 years, <80% over 65.
Dick 1975	Allocation: randomised. Participants: people with schizophrenia, age range 20‐66 years, <80% over 65.
Docherty 2002	Allocation: randomised. Participants: people with schizophrenia or schizoaffective disorder, age "elderly", no further details ‐ not clear if 80% over 65.
Dolnak 2001	Allocation: randomised. Participants: people with schizophrenia, age 18‐65 years.
Ekblom 1974	Allocation: randomised. Participants: people with schizophrenia, age range 15‐60 years.
Elie 1975	Allocation: randomised. Participants: people with schizophrenia, age over 60 years, no further details.
Emsley 1999	Allocation: randomised. Participants: people with schizophrenia, age not stated.
Emsley 2000	Allocation: randomised. Participants: people with schizophrenia, age range 18‐75 yrs, <80% over 65.
Evans 2000	Allocation: randomised. Participants: people with alzheimer's disease and psychosis (not schizophrenia).
Feldman 2003	Allocation: randomised. Participants: people with schizophrenia, age 50‐65.
Fitzgerald 1969	Allocation: randomised. Participants: people with psychosis (unclear if >80% with diagnosis of schizophrenia), age range 22‐68 yrs, <80% over 65.
Friedman 2003	Allocation: not randomised, review article.
Garry 1966	Allocation: randomised. Participants: people with schizophrenia, age range 23‐72 yrs, <80% over 65.
Gerlach 1988	Allocation: randomised. Participants: people with tardive dyskinesia ‐ antipsychotics given to investigate their effects on TD, not given as a treatment for schizophrenia.
Glazer 1985	Allocation: randomised. Participants: people with schizophrenia and tardive dyskinesia (TD) whose TD worsened on withdrawal of usual medication. Interventions: molindone or haloperidol given to investigate their effects on TD ‐ not given as a treatment for schizophrenia.
Glazer 1990	Allocation: randomised. Participants: people with schizophrenia and tardive dyskinesia, age range 18‐70 years. Interventions: molindone and haloperidol given to mask tardive dyskinesia ‐ not as a treatment for schizophrenia.
Glazer 1997	Allocation: randomised. Participants: people with schizophrenia, age over 18 years, <80% over 65.
Gregor 1999	Allocation: randomised. Participants: people with schizophrenia, age unclear <80% over 65.
Gross 1974	Allocation: randomised. Participants: people with schizophrenia, age range 21‐67 yrs, <80% over 65.
Haggstrom 1980	Allocation: not randomised.
Hanlon 1964	Allocation: randomised. Participants: people with schizophrenia, age range 21‐74 yrs, <80% over 65.
Harris 1992	Allocation: randomised ("where possible"). Participants: peoples with dementia, schizophrenia, mood disorders and other conditions <80% with schizophrenia.
Harris 1997	Allocation: randomised. Participants: people with schizophrenia and related disorders, age over 45 years, <80% over 65.
Harvey 2002	Allocation: randomised. Participants: people with schizophrenia or schizoaffective disorder, age not stated (later correspondence reveals 'insignificant numbers over 65').
Harvey 2003	Allocation: randomised. Participants: people with schizophrenia or schizoaffective disorder, age range 18‐64 years.
Harvey 2004a	Allocation: randomised. Participants: people with schizophrenia or schizoaffective disorder, age range 18‐55 years.
Harvey 2004b	Allocation: not randomised.
Heck 2000	Allocation: randomised. Participants: people with schizophrenia, age range 23‐68 yrs, <80% over 65.
Hellewell 1998a	Allocation: randomised. Participants: people with schizophrenia, age not stated, unclear if 80% over 65.
Hellewell 1998b	Allocation: randomised. Participants: unknown number of participants with schizophrenia, age not stated, not clear if 80% over 65.
Hellewell 2000	Allocation: open label extension of a randomised controlled trial.
Hirsch 1975	Allocation: randomised. Participants: people with schizophrenia, age under 67, <80% over 65.
Howanitz 1999	Allocation: randomised. Participants: people with schizophrenia, age at least 55 yrs, not clear if 80% over 65.
Jakovljevic 1999	Allocation: randomised. Participants: people with schizophrenia, age range 18‐65 years.
Johnstone 1997	Allocation: randomised. Participants: people with schizophrenia, age range 15‐65 years.
Judah 1959	Allocation: randomised. Participants: people with schizophrenia or other diagnosis, median age 63 yrs, <80% over 65.
Kinon 1998	Allocation: randomised. Participants: people with schizophrenia, age range 23‐72 yrs, <80% over 65.
Kinon 2000	Allocation: randomised. Participants: people with schizophrenia, age range 23‐72 yrs, <80% over 65.
Kinon 2001	Allocation: randomised. Participants: people with schizophrenia, age mean 69 yrs, not clear if 80% over 65 and unable to contact authors for clarification.
Kinon 2003	Allocation: randomised. Participants: elderly people with schizophrenia ‐ no further details, unable to contact authors for clarification.
Kopala 2003	Allocation: randomised. Participants: people with schizophrenia, age not specified, no further details, unable to contact authors for clarification.
Lovett 1987	Allocation: randomised. Participants: people with chronic brain syndromes (unclear how many might have schizophrenia or related disorders and unable to contact authors for separate information on participants with schizophrenia or related disorders); age over 60 (unclear if 80% over 65 and unable to contact authors for separate information on older participants).
MacMillan 1986	Allocation: randomised. Participants: people with schizophrenia, age range 15‐65 years.
McInnes 1978	Allocation: randomised. Participants: people with schizophrenia, <80% over 65.
Meltzer 1997	Allocation: randomised. Participants: people with schizophrenia, <80% over 65.
Mohler 1970	Allocation: not randomised.
Moller 1997	Allocation: randomised. Participants: people with schizophrenia, age range 20‐68 yrs, <80% over 65.
Mullen 1999	Allocation: randomised. Participants: people with schizophrenia, age reported as 'including elderly over 65 years', <80% over 65.
Naukkarinen 2000	Allocation: randomised. Participants: people with schizophrenia, age range 18‐70 yrs, <80% over 65.
Niskanen 1974	Allocation: randomised. Participants: people with schizophrenia, age range 15‐60 years.
Nussbaum 1964	Allocation: randomised. Participants: people with schizophrenia, age range 18‐70 yrs, <80% over 65.
Pigott 2003	Allocation: randomised. Participants: people with schizophrenia, age range 18‐77 yrs, <80% over 65.
Povlsen 1987	Allocation: randomised. Participants: people with tardive dyskinesia, age range 47‐80 years. Interventions: treatments for TD not schizophrenia.
Quinn 1984	Allocation: randomised. Participants: people with Huntington's disease or tardive dyskinesia.
Rapaport 2001	Allocation: randomised. Participants: people with schizophrenia, age range 18‐65 years.
Reznik 1998	Allocation: randomised. Participants: people with schizophrenia plus obsessional symptoms, age range 27‐67 years. Intervention: neuroleptic + fluvoxamine versus neuroleptic alone.
Ritchie 2003	Allocation: randomised. Participants: people with schizophrenia, age over 60, mean 69 (sd 6.2), assuming a normal distribution <80% are over 65.
Ruskin 1991	Allocation: randomised. Participants: people with schizophrenia, age over 50 yrs, mean age 60, <80% over 65.
Salganik 1998	Allocation: randomised. Participants: people with schizophrenia, age range 60‐78 yrs, mean 66.6 (sd 5.08), assuming a normal distribution <80% are over 65.
Schiele 1975	Allocation: randomised. Participants: people with schizophrenia, age range 29‐74 yrs, <80% over 65.
Sharma 1991	Allocation: randomised. Participants: people with schizophrenia, age range 30‐81 yrs, <80% over 65.
Silverstone 1984	Allocation: randomised. Participants: people with schizophrenia, age range 19‐68 yrs, <80% over 65.
Singh 2002	Allocation: randomised. Participants: people with schizophrenia. Trial withdrawn because ziprasidone had not been granted a licence in the UK in time.
Speller 1997	Allocation: randomised. Participants: people with schizophrenia, age range 35‐76 yrs, <80% over 65.
Stotsky 1977	Allocation: randomised. Participants: people with schizophrenia (80%), age range 18‐66 yrs, <80% over 65.
Sutton 2001	Allocation: randomised. Participants: 39 patients with schizophrenia, age range 50‐65 years.
Svestka 1972	Allocation: randomised. Participants: people with schizophrenia, paraphrenia and schizoaffective disorder, age range 17‐81 yrs, <80% over 65.
Svestka 1987	Allocation: not stated. Participants: people with schizophrenia/schizoaffective disorder/paraphrenia, age range 18‐56 years.
Tench 1990	Allocation: randomised. Participants: people with schizophrenia, age range 18‐70 yrs, <80% over 65.
Tourlentes 1958	Allocation: randomised. Participants: people with schizophrenia, age range 45‐74 yrs, <80% over 65.
Tran 1999	Allocation: randomised. Participants: people with schizophrenia, age range 18‐65 years.
Vangtorp 1968	Allocation: unclear. Participants: people with schizophrenia, senile psychoses (unclear if 80% with schizophrenia or related psychoses), age not stated, <80% over 65.
Vasquez‐Gomez 2001	Allocation: randomised. Participants: people with schizophrenia or affective psychosis (unclear if >80% with schizophrenia or related psychoses), age range 18‐83 yrs, <80% over 65.
Vereecken 1972	Allocation: randomised. Participants: people with schizophrenia, age range 22‐75 yrs, <80% over 65.
Vianna 1975	Allocation: probably randomised. Participants: people with schizophrenia, age 16 and over, <80% over 65.
Walker 1983	Allocation: randomised. Participants: people with schizophrenia, age range 23‐67 yrs, <80% over 65.
Wetzel 1991	Allocation: randomised. Participants: people with schizophrenia, age range 18‐68 yrs, <80% over 65.
Woggon 1978	Allocation: randomised. Participants: people with schizophrenia, age range 18‐68 yrs, <80% over 65.
Woodruff 2002	Allocation: randomised. Participants: people with schizophrenia. Trial withdrawn because ziprasidone had not been granted a licence in the UK.
Yamawaki 1996	Allocation: unclear. Participants: unclear ‐ 35 people with schizophrenia in treatment arm, age unclear ‐ described as elderly, unclear if 80% over 65.

Characteristics of ongoing studies [ordered by study ID]

Lacro 2001.

Trial name or title	Antipsychotic treatment in late life schizophrenia
Methods
Participants	Older patients with schizophrenia
Interventions	Haloperidol vs Risperidone for 12 weeks
Outcomes	Clinical response Cognitive performance Health related quality of life
Starting date	Not stated
Contact information
Notes

Matkovits‐Gupta 1999.

Trial name or title	Managing psychotic disorders through balance receptor blockade: the Zomaril tm clinical programme
Methods
Participants	Patients with psychotic disorders
Interventions	Iloperidone vs placebo and/or active control
Outcomes	PANS scale;Clinical Global Impression of Improvement;Clagary Depression Scale;Psychotic Anxiety Scale;Cognitive functioning tests and quality of life scales
Starting date	Not stated
Contact information
Notes

Newcomer 2001.

Trial name or title	Neurochemical control of memory in aging schizophrenics
Methods
Participants	Older patients with schizophrenia
Interventions	Not stated
Outcomes	Not stated
Starting date	Not stated
Contact information
Notes

Contributions of authors

Richard Marriott ‐ took the lead in starting the review and collaborated with the co‐reviewers in writing the protocol and carrying out the review.

Wendy Neil ‐ collaborated in writing the protocol and carrying out the review.

Susie Waddingham ‐ collaborated in writing the protocol and carrying out the review.

Sources of support

Internal sources

• Leeds Mental Health Teaching NHS Trust, UK.

External sources

• No sources of support supplied

Declarations of interest

None.

Edited (no change to conclusions)