Pharmaceutical care for elderly patients shared between community pharmacists and general practitioners: a randomised evaluation. RESPECT (Randomised Evaluation of Shared Prescribing for Elderly people in the Community over Time) [ISRCTN16932128]

Background This trial aims to investigate the effectiveness and cost implications of 'pharmaceutical care' provided by community pharmacists to elderly patients in the community. As the UK government has proposed that by 2004 pharmaceutical care services should extend nationwide, this provides an opportunity to evaluate the effect of pharmaceutical care for the elderly. Design The trial design is a randomised multiple interrupted time series. We aim to recruit 700 patients from about 20 general practices, each associated with about three community pharmacies, from each of the five Primary Care Trusts in North and East Yorkshire. We shall randomise the five resulting groups of practices, pharmacies and patients to begin pharmaceutical care in five successive phases. All five will act as controls until they receive the intervention in a random sequence. Until they receive training community pharmacists will provide their usual dispensing services and so act as controls. The community pharmacists and general practitioners will receive training in pharmaceutical care for the elderly. Once trained, community pharmacists will meet recruited patients, either in their pharmacies (in a consultation room or dispensary to preserve confidentiality) or at home. They will identify drug-related issues/problems, and design a pharmaceutical care plan in conjunction with both the GP and the patient. They will implement, monitor, and update this plan monthly. The primary outcome measure is the 'Medication Appropriateness Index'. Secondary measures include adverse events, quality of life, and patient knowledge and compliance. We shall also investigate the cost of pharmaceutical care to the NHS, to patients and to society as a whole.


Background
What is the problem to be addressed? Up to 80% of elderly people may receive inappropriate therapy [1,2]. This includes over-treatment, causing adverse drug reactions [3] and under-treatment, e.g. for atrial fibrillation [4]. In the UK over-treatment is often due to the process of repeat prescribing. About 75% of all prescribed items are 'repeats' [5], that is items issued without the patient seeing the doctor. Doctors review only about 75% of these [6]. As a result elderly people are more likely to experience drug-related adverse events [7][8][9]. Thus they are more in need of improved prescribing than younger age groups.
In England the net drug cost for elderly people (those aged 60 and over) was £2,000 million in 1998 [10]. On average they had 3.6 times more prescriptions than younger adults (those aged 15-59 years). Additionally the percentage aged 75 or over of the UK population will double by 2020 [11]. This unprecedented age distribution will have profound effects on society and its institutions, including health care [12].
In the US Hepler and Strand [13] proposed a system of 'pharmaceutical care' (PC). In this system, while doctors continue to take ultimate care of patients, pharmacists are responsible for moderating their drug care. In doing so pharmacists co-operate with doctors, patients and carers in designing, implementing and monitoring a 'pharmaceutical care plan' (PCP) [e.g. Table 1]. By involving patients in decision-making, PC aims to improve communication, promote compliance and concordance with treatment, and achieve specified therapeutic outcomes. This has the potential to reduce the drug bill and drugrelated problems, and improve patients' quality of life.
In the UK Lord Hunt (then Parliamentary Under-Secretary of State for Health) recognised that the clinical skills of pharmacists have long been underused [14]. In contrast the workload of general practitioners (GPs) has been rising in recent years. Hence using community pharmacists to manage the medication of elderly patients will rationalise the use of human resources in community health care. Furthermore US studies have shown that better prescribing reduces drug-related problems in hospital [15], drug costs [16] and thus the health care budget. However there has been insufficient information to conduct a full economic analysis. The proposed trial will provide better economic information for policy makers and the pharmacy profession in the UK.

What are the principal research questions to be addressed?
A Is shared pharmaceutical care for elderly people in the community effective in: i improving the quality of prescribing?
ii improving patient's knowledge about their disease and medication?
iii improving compliance? iv reducing adverse events? v and thus improving quality of life? B Is shared pharmaceutical care for elderly people in the community cost-effective? Although the pharmacy profession has started to adopt this extended role, there is insufficient rigorous evidence to substantiate this practice. Hence the proposed trial will be important in underpinning these political and professional initiatives.
Most recently the Department of Health [21] has proposed that Primary Care Trusts (PCTs) across England should invest in pharmaceutical care services giving patients access to more help from pharmacists in using their medicines. Consequently pharmacy practice in the community may be completely transformed by 2004. Hence there is a 'window of opportunity' for a randomised trial to evaluate the effectiveness and cost-effectiveness of pharmaceutical care provided by community pharmacists.
More specifically RESPECT will study the co-operation needed between community pharmacists and GPs. The Royal Pharmaceutical Society of Great Britain fully supports this trial and believes it will provide vital scientific evidence for the development of the pharmacy profession within primary health care.  Table 1 illustrates its potential. Although recruitment was slow, we identified two reasons for this -the lack of research costs for GPs and the shortage of treatment costs for pharmacists. In designing and costing RESPECT we have addressed and, we believe, overcome both problems. We have also developed a training programme for community pharmacists, focusing on the clinical content of pharmaceutical care, and on collaboration with GPs.
Secondly another of us (HE) received a grant from NHS Northern & Yorkshire to deliver a training programme to build up the research capacity of East Riding and Hull community pharmacists. This funding has helped us to prepare community pharmacists to take on the proposed trial, and to overcome the problems we have identified.

Relevant systematic reviews and the need for a trial
Numerous articles have discussed pharmaceutical care and 'medicines management' (which some British researchers regard as a synonym for pharmaceutical care) [22]. Two systematic reviews [23,24] of pharmaceutical care and medicines management report that only eight papers, all from the US, report randomised studies [15,[25][26][27][28][29][30][31]. Only one was in primary health care [30]. Another review shows that none of the studies met accepted pharmaco-economic criteria [32]. All three reviews recommended more randomised trials to evaluate pharmaceutical care.
To extend these reviews in response to comments on our application to the Medical Research Council, in particular to identify newly published or current research, we adopted five distinct strategies. First we conducted our own systematic search of the Cochrane Library, Embase, EPIC and International Pharmaceutical Abstracts (IPA). We based our search strategy for Embase, EPIC and IPA on the algorithm: (pharmaceutical care OR pharmaceutical service* OR ambulatory care clinical OR ambulatory care pharmac* OR clinical pharmac* OR clinical pharmacy program* OR clinical pharmacy specialis* OR community pharmac* OR comprehensive pharmaceutical service* OR medic* management OR patient oriented service* OR pharmac* intervention* OR pharmac*-managed clinic* OR polypharmacy OR polytherapy OR repeat dispens* OR repeat medic* OR repeat prescription) AND (elderly OR aged OR old OR geriatric). We found six more randomised trials on pharmaceutical services or medicines management for the elderly [33][34][35][36][37][38]. However none evaluated pharmaceutical care for the elderly as we have defined it. Furthermore only Bond's study of repeat dispensing was conducted in British primary health care.
Secondly we searched the current European Union BIOMED database. We identified a recently completed multi-national study of pharmaceutical care for elderly patients in the community. The UK arm in Northern Ireland includes about 100 patients in the intervention group. Though only interim results were available (McElney, personal communication), preliminary findings were encouraging [39]. As health care varies across the EU, however, it is unlikely that the rest of the project is relevant to British primary health care.
Thirdly we searched the current UK National Research Register using the search term 'pharmacy'. We identified seven continuing randomised studies. In one study of patients with coronary heart disease community pharmacists are drawing up care plans [40]. Another four are trials of medication review programmes undertaken by hospital pharmacists, practice pharmacists (pharmacists working in GP surgeries) or research pharmacists rather than traditional community pharmacists (Nazareth, Reid, Thomas, Zermansky). The remaining two interventions cannot be classified as pharmaceutical care or medicines management (Barber, Peat).
Fourthly we contacted the 16 current Schools of Pharmacy in the UK to seek other pharmaceutical care projects. We identified two randomised studies at Robert Gordon University using practice pharmacists to run 'medication review clinics' in GP surgeries rather than community pharmacists in their own pharmacies. Other Schools of Pharmacy have confirmed that no study has been or is being conducted except that in Northern Ireland [39]. . This identified four more trials [41][42][43][44].
Most of the 19 completed trials have reported positive findings including improved knowledge of and adherence to treatment, improved biochemical outcomes, and even reduced mortality in heart failure. However only two addressed the economics of pharmaceutical care and medicines management [31,38].
In summary the literature suggests that pharmaceutical care can benefit patients. As the majority of randomised studies were in the US, their relevance to the UK is limited. Furthermore very few were in community pharmacies, and only two related to elderly patients in the UK [39,42]. So the evidence about the effectiveness of 'pharmaceutical care' within British primary care remains weak.
Nevertheless PC is being promoted by all the pharmaceutical organisations -National Pharmaceutical Association, Pharmaceutical Services Negotiating Committee [45], RPSGB [18,46] and UKCPA [19]. To implement this advice without more evidence would therefore create an 'evidence gap'. Usual practice would diverge from the objective evidence [47]. Nevertheless the NHS plan and its pharmacy sequel [17,21] proposed alternative contracts for "personal medical services-like schemes" such as medicines management, pharmaceutical care and repeat prescribing provided by community pharmacists. However our literature review has shown that practice or research pharmacists based in clinics, rather than community pharmacists based in their pharmacies, have conducted most of the relevant research. Hence there is an urgent need for large studies to evaluate the effectiveness of pharmaceutical care and the use of PCPs in British primary care.

How will the results of this trial be used?
This study will provide policy makers and the pharmacy profession with information about the practicality and effectiveness of pharmaceutical care in British primary care. It will estimate the benefits and costs of pharmaceutical care in this context and provide information for the Department of Health to develop an appropriate remuneration formula. In short the study will provide evidence to guide the future implementation of pharmaceutical care in the UK and thus encourage 'evidence-based pharmacy practice'.

Risks to the safety of patients in the trial?
This is a pragmatic trial with very little risk to patients. Those receiving pharmaceutical care will be at no greater risk than in normal clinical practice.

Brief summary
In PC doctors continue to take ultimate care of patients and pharmacists are responsible for moderating their drug care, in particular by designing, implementing and monitoring a PCP. Our feasibility study of PC in fourteen pharmacies in Eastern Hull and East Yorkshire Primary Care Trusts (PCTs) suggested that the approach was feasible and merited rigorous evaluation. We are therefore expanding the size and scope of our research within a randomised multiple interrupted time series design. We are collecting data on prescribing and resource use from about 20 general practices and about 60 pharmacies and on health outcomes from 400 patients.

What is the proposed trial design?
The Medical Research Council (MRC) has defined 'complex interventions in health care' as "interventions comprising separate elements which seem essential to the proper functioning of the intervention although the effective ingredient is difficult to specify" [48]. PC is an archetypal complex intervention. The MRC has identified five essential steps in the evaluation of such an interventiontheory, modelling, exploratory trial, definitive randomised trial and long-term implementation [48]. A definitive trial must be generalisable in the sense that it recruits a range of patients or conditions across a number of centres. It must also be 'pragmatic' in the sense that it evaluates the complex intervention in routine clinical practice and in a form close to that in which implementation is likely in the long term [49]. In contrast exploratory trials evaluate complex interventions in controlled experimental conditions and in a prototype or idealised form [49]. Of the 19 trials summarised previously most are exploratory, few definitive. Thus it is a pragmatic multicentre trial that PC now needs.
It is difficult to design a pragmatic trial for PC that randomises individual patients [50]. Both GPs and pharmacists learn the relevant techniques cumulatively, with the result that those who use these techniques for experimental patients cannot readily withhold them from control patients. It follows that randomisation must be by 'clusters' of patients, for example patients associated with a practice or pharmacy. Unfortunately practices and pharmacies are rarely coterminous. Hence if we take practices as the clusters to be randomised, then neighbouring pharmacies will serve some patients randomised to PC and others randomised to usual care. Thus there will be contamination within pharmacies. Similarly if we take pharmacies as clusters, there will be contamination within practices. We conclude that PCTs are the most appropriate clusters for randomisation. Our feasibility study lent support to this analysis. If we select half of the PCTs at random for training in pharmaceutical care, there is a real danger that the other half will suffer from 'resentful demoralisation' [52]. At the very least, therefore, the second half should receive training at the end of the trial. Fortunately we can do even better by phasing the training in PCTs in a random order, as shown in Figure 1. Because each PCT has a period before training to compare with the period after training, each can act as its own control. This design, known as a randomised multiple interrupted time series [52], has three major advantages: a organisational -because training needs fewer resources over a longer period; b statistical -because it is more powerful; and c epidemiological -because staggering the introduction of PC across PCTs helps to protect the evaluation from potentially contaminating changes in health and other policies by ensuring that such changes affect different PCTs at different stages of the cycle of developing PC.
Our feasibility study also suggested that within each PCT we could recruit at least four general practices and about three community pharmacies associated with each practice -about 20 practices and 40 pharmacies in all. We have randomised the resulting five clusters of practices, pharmacies and patients to begin pharmaceutical care in five successive waves. All five stay in the control group until they receive the intervention in a random sequence.

What are the planned trial interventions? Intervention group
The experimental intervention is PC, in which pharmacists co-operate with doctors, patients and carers in designing, implementing and monitoring a PCP for example Table 1. The London and Bradford schools of pharmacy are providing five blocks of training, each consisting of one workshop for community pharmacists on PC for the elderly and one joint workshop for pharmacists and GPs on how to work together. GPs and community pharmacists in the same PCT attend the same training block (Figure 1)). The workshops encourage them to adopt a problem-based approach to the application of pharmaceutical care. In particular they enable pharmacists to assess patients' pharmaceutical care, to withdraw unwanted medicines, and to draw up PCPs. The training also enables GPs to work with PCPs. Hence these workshops are accredited for both professions. They also take account of experiences with our feasibility study. i Failure to receive drug: the patient has a medical problem that resulted from his or her not receiving a drug.

Train + Base
Training for GPs and pharmacists + Measure baseline outcomes prior to the start of intervention (2 month chase-up) Intervention: Intervention period.

MAI
Extract retrospective data from GP practices (e.g. MAI).

M3
Measure outcomes at 3 months from the start of intervention, allowing up to 2 months to chase questionnaire returns.

M12:
Measure outcomes at 12 months from the start of intervention: indicated in table by M12, allowing 2 months chase up.

Final Data Collection:
Measure outcomes at 30-35months (roughly 2 ½ years) after the start of recruitment. Home visits + MAI data. Analysis and Reporting: Data analysis and reporting.
iii Improper drug selection: the patient has a drug indication but is taking the wrong drug.
iv Sub-therapeutic dosage: the patient has a problem that is being treated with too little of the correct drug.
v Overdosage: the patient has a medical problem that is being treated with too much of the correct drug.
vi Adverse drug reaction: the patient has a medical problem that stems from an adverse reaction to a drug.
vii Drug interaction: the patient has a medical problem that is the result of a drug-drug or drug-food interaction.
viii Drug use without indication: the patient is taking a drug for which there is no medically valid indication.
D Establish therapeutic goals for each drug-related problem with patient and GP.
E Identify feasible alternative treatments.
F Select the best pharmaceutical solution and decide with patient the best drug, formulation and dose.
G Agree pharmaceutical care plan with both patient and GP (e.g. Table 1).
H Implement and monitor plan.
I Follow up and measure outcome.
In addition pharmacists educate patients and, if appropriate, carers about the indication for each medication and its use, and withdraw unwanted medicines with patients' consent. If patients need compliance aids such as dosette boxes or reminder charts, the pharmacists provide these services. They continue to update and implement the PCP, and monitor outcome at least monthly in association with patients and their GPs.
The patient has a Study Membership Card to remind hospital staff to contact the patient's community pharmacist when the patient is admitted to and discharged from hospital. The community pharmacists can then change the patient's PCP accordingly and avoid drug-related problems arising from miscommunication after hospital discharge. If patients are unable to visit the pharmacy, for example after hospital discharge, pharmacists should arrange a domiciliary visit to provide the pharmaceutical care.
We are monitoring the process of pharmaceutical care by collecting PCPs and analysing pharmacists' time sheets, continuing education records and professional development portfolios.
Although patients have the right to take their prescriptions to any pharmacy for dispensing, there is evidence that more than 80% consistently use the same pharmacy. To facilitate pharmaceutical care, we encourage, but not require, them to do so. As PC is time-consuming, however, it is unlikely that a second pharmacy would duplicate the plan. As the potential use of more than one pharmacy is an inevitable part of this complex intervention, this pragmatic arrangement will enhance external validity.
If a trained pharmacist leaves a participating pharmacy, we shall give individual training to his or her successoranother example of the pragmatic nature of this trial.

Control group
Until they receive training and join the intervention group, community pharmacists provide their usual dispensing service and so act as controls. At this stage pharmacists do not know which patients have consented to join the trial.

What are the proposed arrangements for allocating patients to trial groups?
The PCT is the natural unit within which to train and support community pharmacists and GPs in PC care and to allocate this intervention in sequence. One of us (PDC) wrote to all practices in the five PCTs, outlining the project, making clear the nature of the working agreement between practices, associated community pharmacies, and the project team, and inviting expressions of interest. From those responding in each PCT who are close to potential participating pharmacies, we recruited a sample of at least four practices after stratifying for number of partners and about 60 associated pharmacies. When appropriate we recorded reasons given by practices and pharmacies for not taking part. Recruited practices then helped us recruit samples of about 35 eligible patients, depend on the size of the practice, the sampling method varies: a) For smaller practices (consisting of one or two partners) with less than 100 patients there was no sampling. For practices with over 100 eligible patients sampling (stratified by age and the number of drugs) was completed. A list of eligible patients was obtained from the practice computer record and from them a random sample of up to 100 patients was obtained. This list was given to the GPs to obtain their consent for researchers to approach those patients concerned. Following this each patient was written to and where appropriate a meeting arranged with a member of the research team. The decision to participate in the trial was therefore the patients choice.
b) For larger practices (consisting of three or more partners) -sampling (stratified by age and total number of drugs) was undertaken if there were more than 200 eligible patients. A list of eligible patients was obtained from the practice computer and the GPs were then asked to provide consent for researchers to approach a maximum of 200 patients. Again the decision to participate in the trial involved choice on the part of those patients concerned.
There is conflicting evidence whether age is a risk factor for drug-related adverse events [55][56][57]. Fortunately stratification is a technique that generally yields statistical gains, but not costs.
In principle we did not restrict the pharmacies that could volunteer for RESPECT. This pragmatic approach enhances external validity, especially if pharmacists can volunteer to provide PC in future. In practice we sampled general practices after stratifying by size. This limited the final sample of participating pharmacies to those associated with sampled practices. Whether a pharmacy is independent or part of a chain is a potential covariate in our eventual multi-level analysis. Both our palliative pharmaceutical care study [58] and our feasibility study have shown that single-handed pharmacists can provide PC. However both studies have also shown that resource constraints limit the numbers of patients in PC at any time. The NHS treatment costs available within RESPECT have helped to address this hurdle.

What are the proposed methods for protecting against other sources of bias?
Blinding is not possible for pharmacists, GPs or patients because the nature of the intervention requires their full knowledge. However pharmacists do not know which patients have been recruited into the trial until they receive training and join the intervention group. Furthermore patient questionnaires are self-administered or administered by the research pharmacist and trial co-ordinator rather than pharmacists or GPs. Finally the assessors of one of the main outcome measures, the Medication Appropriateness Index will remain blind throughout.
The trial co-ordinator and research pharmacist used practice computer records to identify potential trial participants and asked GPs to give consent. Although GPs withheld consent for some patients, the research staff asked for reasons. (To identify, and if necessary adjust for, selection bias we shall also compare the personal characteristics of all consented and non-consented patients.) The research staff then invited potential participants into the trial using practice headed letters and visits to practice or home.

What are the planned inclusion and exclusion criteria?
Inclusion criteria Patients aged more than 75 years with repeat prescriptions for five or more drugs (excluding drugs taken only when required), who are living at home, well oriented in time and place, and able to give their consent to take part. Abbreviated Mental Test is used to confirm patients' mental state, patients score 7 or above (out of 10) will be included. Their GPs' consent is also necessary.

Exclusion criteria
Patients in residential or nursing homes. Patients who took part in our feasibility study of vulnerable elderly people or who normally use a pharmacy which has refused to participate in the trial. Patients with memory impairment who score 6 or below on the Abbreviated Mental Test.

What is the proposed duration of treatment period?
12 months (Figure 1)

What is the proposed frequency and duration of follow up?
We plan to observe patients for 36 months between recruitment and final visit (Figure 1). We have asked them to provide data at recruitment, at the beginning of pharmaceutical care and three and 12 months thereafter, and at the final visit 36 months after recruitment. We have asked their practices and pharmacies to provide access to their medical records at the end of the study.

Secondary
Patients' knowledge, compliance and concordance, practice-reported (and therefore more serious) adverse events, and self-assessed health outcome

Economic
Cost of treatment to NHS, patients and society as a whole

How will the outcome measures be measured at follow-up? Medication Appropriateness Index (MAI)
The MAI was devised by Hanlon [59] to measure the appropriateness of medication, validated by Samsa [60] and Schmader [61], and successfully used within a randomised trial by Hanlon [15]. The resulting score depends on the number of drugs regularly taken by the patient and the assessed appropriateness of each. In this study we shall use blinded and independent (though paid) clinical pharmacists to assess the appropriateness of drugs prescribed within the trial. Clinical pharmacologists and gerontologists will validate a sample of these data. They will use the information available from patient medical records at recruitment, at the beginning of pharmaceutical care and three and 12 months thereafter, and at the final visit 36 months after recruitment. The research pharmacist and trial co-ordinator will collect these data from the records of participating practices, including their repeat prescription indices, electronically wherever possible. By designing the panel assessment forms for electronic scanning we shall quickly identify patients about whom the assessors disagree. They will then meet to resolve such differences. Although the MAI is an intermediate outcome, it is correlated with the absence of adverse reactions and interactions. We shall investigate whether it is also correlated with better patient outcomes. Few British studies have used the MAI. We shall therefore use psychometric methods [62] to revalidate it for the UK. Meanwhile we are undertaking a feasibility study to compare scores within and between expert professionals.

Knowledge, compliance, adverse events and health outcomes
Research staffs are monitoring knowledge, compliance, adverse events and self-assessed health outcomes in all trial patients over the duration of the study. Depending on the mental state and accessibility of the patient they use patient-completed questionnaires, telephone interviews or home interviews -at recruitment, at the beginning of pharmaceutical care and three and 12 months thereafter, and at the final visit 36 months after recruitment. There is evidence that method of completion affects health outcome measures [63,64]. To prevent this affecting the internal validity of the trial we shall check that the proportions of patients using the three modes of completion are consistent across both PCTs and time. If necessary we shall use analysis of covariance to adjust for any imbalance in these proportions.

Compliance and patients' knowledge of medication
We ask patients to complete a questionnaire to measure compliance, originally devised by researchers at Keele University. Before its use in RESPECT we are enhancing and revalidating it Pharmacists ask patients how they take each drug during pharmaceutical care. We shall calculate the proportion of medications for which their response agrees with the directions on the records in the pharmacy or practice [15].
For patients with at least six months' continuous drug records we shall calculate expected percentage compliance from the expected finish date for each prescription and the date on which the next prescription was filled. We are using general practice data to validate pharmacy records, particularly important during the control period.
We use tablet counts as a further measure of compliance. We ask pharmacists to request patients to bring all tablets to the pharmacy at the start and end of the intervention. The pharmacist conducts a tablet count on both occasions. If a patient is housebound, the pharmacist visits the patient's home. Pharmacists also ask patients about the indications for the medication, using questions like "what is this drug for?" or "how does this drug help?" To measure patients' knowledge of medication we shall calculate the proportion of drugs for which they give a correct response, based upon their medical problems and accepted indications for prescribing each drug [15].
To validate pharmacists' tablet counts and patients' knowledge about medication, researchers conduct further tablet counts and ask samples of patients about indications for medications on two separate occasions. At recruitment and at the start of the intervention they visit a 1 in 5 stratified random sample of patients at home and count their supply of tablets. This sample of approximately 140 patients consists of 70 housebound patients and 70 patients able to visit the practice. They visit a second stratified random sample of 140 patients and count their tablets at the start of the intervention and at either three or twelve months thereafter, selected at random.

Adverse events
Throughout the study we collect data from general practices on serious adverse events, consultations in primary care, and hospital referrals and admissions. We define a serious adverse event as one that carries a risk of death or persistent or significant disability or incapacity, or requires hospitalisation or medical or surgical intervention. Though there is evidence that patients can enhance the measurement of mild adverse events, [65][66][67] pharmaco-vigilance in the UK traditionally focuses on practice-reported adverse events, which are generally more serious, and provide a more cost-effective monitoring system.

Health outcomes
We shall use two instruments that are acceptable to elderly patients. To measure general health status we shall use the SF-36. To underpin cost-utility analysis we shall use the EQ5D [68], acceptable to elderly people if one omits the 'thermometer'.

Home interviews
Rather than interview all respondents face to face on all occasions, we invite able respondents to complete postal questionnaires, and conduct interviews by telephone whenever feasible. To this end we have developed criteria to guide the completion mode appropriate to each respondent. We shall also take account of completion mode in our analysis. In particular we shall compare the validity, reliability and responsiveness of each data collection method [62,69].

Will health services research issues be addressed?
The economic evaluation will take the form of an incremental cost-utility analysis from the perspective of both the NHS and society in general. The main focus of the economic evaluation will be the EQ5D. We shall also analyse other outcomes, in particular the SF-36 and MAI, to establish whether the choice of measure affects the results of the economic evaluation.
Given the difficulty of disentangling the costs of pharmaceutical care and its sequelae from other costs, economic analysis will include all health care costs, including hospital admissions and referrals, ambulance journeys and consultations in primary care and at the pharmacy.
We shall extract data from practice records on medications prescribed, number and type of consultations in the practice or at home, outpatient referrals and length of any inpatient stay. We shall also extract data from pharmacy records on medications dispensed. All these data will relate to the period from 12 months before the intervention until 24 months after completion of recruitment. In addition pharmacists contributing to RESPECT will complete a questionnaire at baseline to identify the average time spent dealing with patients like those to be recruited into RESPECT, both in dispensing medications and providing advice and support. They will also record prospectively the contact and non-contact time spent on the PCP for each patient, as part of that PCP. Patients and carers will provide data on the costs to themselves in both time and money of seeking pharmaceutical and other health care, as well as estimates of time off from normal activities. They will also provide data on the carer support they normally receive, including help with taking their medication and using compliance aids. These data from patients will serve to validate and enhance the estimates of health care use derived from practice records. The costs associated with training general practitioners and pharmacists will also be assessed, including fixed costs, consumables, patients' time, locum cover and travel costs.
Later sensitivity analysis will allow us to assess the extent to which the cost-effectiveness of pharmaceutical care depends on the process of implementation, especially on the relative increase in pharmacist workload.
We shall derive an annual cost of management for the control and intervention period for each patient and use these data to calculate an incremental cost for each patient. As patients serve as their own control, if a patient dies during the control period they will not be included in any of the analysis. If the patient dies during the intervention period, their data will be included in the analysis.
Censored data will be adjusted accordingly using appropriate techniques.
Cost-effectiveness will be assessed via the ratio of additional cost to additional benefit, measured according to the change in the EQ5D between the control and intervention period. The confidence region surrounding the average incremental cost effectiveness ratio will be estimated using appropriate statistical techniques such as the bootstrap method and Monte-Carlo simulation. Stochastic analyses such as this will allow a cost effectiveness acceptability curve to be generated. This will represent a plot of the probability of management with a PCP being more cost-effective than usual care under alternative assumptions about the threshold cost per QALY. Based on the average incremental cost, cost-effectiveness ratio and the uncertainty surrounding the estimates, we will explore the implications of a widespread implementation of pharmaceutical care.

What is proposed sample size and justification for power calculations?
After losses to follow up the final sample size will be about 400, viz 20 practices × 20 patients on average or 40 pharmacies × 10 patients on average. It is difficult to estimate the power of this complex design to detect changes in the MAI, the principal outcome measure. To simplify this calculation we discount the likely advantages conferred by time series analysis and estimate the power of analysing the change in the MAI between baseline and three (or 12) months.
The 400 or more patients completing the proposed trial will not be a simple random sample, but a sample from 40 or so clusters -the patients who use each of the participating pharmacies. If these pharmacies are at all consistent in their approach to pharmaceutical care, findings will be more homogeneous within them than between them. The effect of this is to reduce the power of the trial by the factor [1 + ICCC × (n -1)], where ICCC is the 'intra-cluster correlation coefficient' and n is the average cluster size [70].
Although we know of no data on ICCCs for the MAI, we have access to the full data set of the North of England Study of Standards and Performance in General Practice (1992), [71] including data on practitioners' adherence to clinical guidelines analogous to the MAI. Since none of those intra-practice CCs exceeded 0.1 and few exceeded 0.05, we base our preliminary estimate of the power of RESPECT on an intra-pharmacy CC of 0.1. Since the average cluster size is 10 (viz 400 divided by 40), the effective sample size of RESPECT for the MAI could be as low as 210 (viz 400 divided by 1.9).
An effective sample size of 210 would yield 80% power when using a significance level of 5% to detect a difference in MAI between intervention and control groups of 0.4 of a standard deviation (SD) [72] -usually regarded as a moderate effect size. Although we cannot yet estimate the SD of the MAI within RESPECT, Hanlon's trial reported that the MAI had an SD close to 10 points [15]. Under conservative assumptions about loss to follow up and intra-pharmacy correlation, therefore, RESPECT can detect a change of four points in the MAI. In contrast Hanlon's trial achieved a change of five points [15]. The grant-holders, who include four clinicians, therefore judge that a change of four points is both achievable and clinically important.
We cannot yet estimate the SD of EQ-5D scores within RESPECT. In the general population the EQ-5D has an SD of the order of 16 points [73]. Although we know of no data on ICCCs for the EQ-5D, none of the intra-practice CCs for patient outcomes in the North of England Study of Standards and Performance in General Practice (1992) [71] exceeded 0.01. We therefore base our preliminary estimate of the power of the RESPECT trial to detect changes in EQ-5D on an IPCC of 0.01. So RESPECT should have an effective sample size of at least 370 for outcomes (viz 400 divided by 1.09). This would yield 80% power when using a significance level of 5% to detect a change in EQ-5D scores equivalent to 0.3 standard deviations, ie about five points. Although the applicants do not expect much change in generic outcomes following the introduction of pharmaceutical care, they judge that a change of five points would indeed be clinically meaningful.

What is the planned recruitment rate?
There are nearly 150 practices in the East Riding and Selby & York PCT, all computerised and linked to the NHS network through the East Riding Health Authority. From a survey of four practices we established that there are more than enough patients meeting the inclusion criteria for the trial. In one practice of 8,000 patients, for example, there were 180 men over 75, of whom 25% took five or more different drugs. The numbers and proportion of women were greater. We therefore expect no difficulty in recruiting 35 patients from each practice.

Are there likely to be any problems with compliance?
Compliance with the intervention is unlikely to be a problem, either by professionals or by patients. We did not encounter any problems with professional compliance in a previous study of pharmaceutical care within palliative care [58]. If a pharmacist leaves a participating branch, however, we shall ask for a permanent replacement and give the replacement pharmacist individual training. Fur-thermore patient compliance with drugs is an outcome rather than a threat.

What is the likely rate of loss to follow-up?
From East Riding data we estimate that losses from mortality (about 5%), leaving home for institutional care or elsewhere (about 10%), and non-response to questionnaires (about 20%) will not exceed 40% in total.

How many centres will be involved?
Twenty general practices -four within each of the five PCTs likely to take part. With the encouragement of the former East Riding & Hull Health Authority (ERHHA) the four PCTs within ERHHA have agreed to support the study and meet excess treatment costs. The Selby and York PCT has also agreed to do so.

What is the proposed type of analysis?
All analyses will be by 'intention to treat'. Whenever possible we shall include patients lost to follow up by using all available data, including those from pharmacies' patient medication records and general practices' repeat prescription indices. We shall analyse the five interrupted time series by the methods expounded by Box and Jenkins (1976) [74]. Within each time series the analysis will in effect compare the process and outcome data collected before training in pharmaceutical care with those collected after that training. Because the definitive analysis combines these sub-analyses, it is much more robust against external events like changes in health policy than a single time series would have been. Nevertheless because the five PCTs differ in patient characteristics, notably socio-demographic characteristics, the time series models for the four main outcome measures will take as covariates the known risk factors for drug-related adverse events We shall analyse the five multiple interrupted time series by the latest Statistical Analysis System Econometrics and Time Series (SAS/ETS) procedures [75]. In doing so we shall distinguish between four levels -PCT, practice, pharmacy and patient.

Are there any planned subgroup analyses?
To guide the implementation of PC we shall use analysis of covariance to compare effectiveness and cost-effectiveness between patients at low risk of adverse events and those at high risk. The covariates will be age, total number of drugs, and oral anti-coagulants, hypoglycaemic and non-steroidal anti-inflammatory drugs [76].

What is the proposed frequency of analyses (including any interim analyses)?
Although we shall monitor the incidence of adverse events throughout the trial, we shall analyse only once ( Figure 1).