By 2008 of 65 practices in Sandwell Primary Care Trust, case finding had already been completed in 12 (six as part of the pilot project); in a further 25, case finding had started or a start date had already been agreed; two practices did not have available consultation rooms in the practice building, leaving 26 practices for randomisation. The total population registered with these practices was approximately 125,000 and from the pilot study it was anticipated that 5% (about 6,250) would be untreated high risk patients.
There were six practices in Solihull Care Trust in whom the targeted case finding programme could be implemented. The total population registered in these practices was 18,938 and it was anticipated that 945 (5%) would be untreated high risk patients.
Therefore overall 32 practices, representing 32 separate clusters, were included in the study with a combined estimated 7,247 untreated people at high risk of CVD.
Randomisation was stratified by primary care trust. In Sandwell PCT, after discussion with the case finding programme manager, it was agreed that 11 practices were a higher priority for the case finding intervention and the remaining 15 practices were a lower priority. It was therefore decided to stratify the practices into two groups for randomisation: early intervention and late intervention. A table of random numbers was generated to determine the order in which practices should receive the project nurse led intervention in the early intervention group and a second table for practices in the late intervention group.
The six practices in Solihull were of different characteristics, two were single handed and two were located very near each other and served similar populations. They were therefore stratified into three pairs (two single handed, two similar location and two remaining practices) for randomisation. A table of random numbers was generated to determine which of each pair should have the intervention first and which second.
All patients aged 35 to 74 who were not currently on the CVD or diabetic registers and who were not currently receiving antihypertensive or statin treatment (no prescription within 90 days) were included in the programme. From this were selected all those whose estimated ten-year CVD risk ≥20%.
Practice software systems were used to identify patients with a ten-year CVD risk ≥20%. In Sandwell PCT all practices subscribe to Clinical Manager software manufactured by MSDi
]. This software produces lists of patients meeting pre-specified criteria (e.g. diabetic patients due for a blood pressure measurement). The manufacturers incorporated an additional module into the software to calculate ten-year CVD risk for all untreated patients without CVD using the Framingham risk equations
]. The following method was used to deal with missing risk factor data. If smoking status was missing the patient was assumed to be a non smoker; if blood pressure or cholesterol values were missing a default value was used, based on the average for an untreated person of that age, sex, smoking status and diabetic status. This followed a previously described method
]. In Sandwell PCT this allowed every practice to produce a list of untreated patients at ≥20% ten-year CVD risk (high risk patients).
In Solihull PCT practice data was extracted using the search facility in the electronic patient records and exported to an Excel spreadsheet. Ten year CVD risk (high risk patients) was calculated in the spreadsheet using the same method and patients at ≥20% ten-year CVD risk (high risk patients) were identified for the practices.
Participating practices reviewed the list of high risk patients identified and excluded those who had died, moved practice and or who the GP judged were unsuitable for CVD prevention (e.g. those with a terminal illness).
The intervention in Sandwell was delivered by project nurses trained in the management of either diabetes or coronary heart disease. In Solihull prescribing advisors, trained pharmacists who were licensed to prescribe, delivered the intervention.
In the intervention practices the project nurse (Sandwell) or prescribing advisor (Solihull) arranged for a letter along with an information sheet about the case finding project to be sent from the GP inviting patients for cardiovascular assessment. The letter specifically mentioned that patients might be offered lifestyle advice and medication ( Additional file
and Additional file
). The patient was given the time and date of an appointment for their CVD assessment and asked to reschedule if this was unsuitable. If they did not attend or reschedule the appointment two further attempts were made to contact patients either by letter or telephone.
Patients who attended underwent a full cardiovascular risk factor assessment, including enquiry about smoking, measurement of blood pressure, blood tests for fasting glucose, total cholesterol and HDL cholesterol. Each patient’s ten-year CVD risk was calculated using the Framingham equation as recommended in UK guidelines based on their measured risk factors and their eligibility for treatment was determined
]. If they were eligible for antihypertensive or statin treatment the project nurse/prescribing advisor discussed the risks and benefits of treatment with the patient and the general practitioner was informed so that treatment could be initiated. If appropriate the patient could also be referred to local smoking cessation services, for advice on physical activity or for dietary advice about weight loss.
Patients at highest risk were invited first followed by those at next highest risk until all high-risk patients (≥20% ten-year CVD risk) had been invited. The project nurse (or prescribing advisor) then repeated the process in the next practice in a sequence determined by the randomisation. Because the practices varied in size this meant that the number of weeks spent at each practice would be expected to vary from one practice to another. It was anticipated that it would take about 18 months to complete the implementation of targeted case-finding in all Sandwell practices and about 12 months in Solihull practices.
Since there is good evidence for the effectiveness of CVD prevention, it is reasonable to use uptake of treatment as evidence of effectiveness. Prescriptions of antihypertensive drugs or statins are recorded in electronic primary care records, therefore routine data could be used to assess outcomes.
The primary outcome is the number of high-risk patients started on at least one preventive treatment: an antihypertensive drug or a statin. Antihypertensive drugs and statins are defined by the chapter codes in the British National Formulary
Secondary outcomes include the number of patients who have cardiovascular risk factors assessed (blood pressure, lipid levels); the number of patients referred to services for lifestyle advice (smoking cessation services, advice on physical activity, dietetic advice on weigh loss); changes cardiovascular risk factors (blood pressure, lipid levels); cardiovascular events (new diagnoses of angina, MI, TIA or stroke).
Follow up and data collection
Data extraction will be undertaken as a single data extract at least one year after targeted case-finding has been implemented in the last practice. This allows for at least a year of follow up. The number of untreated high-risk patients who are started on treatment will be assessed at the start of each step in the stepped wedge trial. An outline of the data extracted is given in Table
Data to be extracted from electronic primary care records at the end of the study
The characteristics of the populations served by the GP practice will be summarised, including, primary care trust, list size, age of practice population, gender, ethnicity and practice Index of Multiple Deprivation score.
] These characteristics will be summarised by numbers and percentages, means and standard deviations or medians and inter-quartile ranges as appropriate. We will then report the number of patients excluded in the study by each of the inclusion criteria (age, CVD status and medication use), and also the number of patients meeting the study inclusion criteria.
Fidelity of uptake will be explored by reporting, the number of patients invited, the number of patients seen by the nurse/prescribing advisor and the proportion who were subsequently confirmed to have a CVD risk ≥20%.
The characteristics of the patients included in the invited population will also be summarised. These characteristics will include gender, age, blood pressure, lipids, glucose, medication use (at baseline), cardio-vascular disease, smoking status, Framingham risk score, and referrals to smoking cessation, physical advice and dietetic advice services. These characteristics will be summarised by their values at baseline (2008) and will be compared between those not attending; and for those attending stratified by whether their calculated risk was greater than 20%.
The primary aim of the study is to evaluate whether there is a difference in proportion of eligible patients on the relevant prescription medication before and after the practice was exposed to the intervention. In statistical terms this null-hypothesis (no difference) can be tested using a mixed logistic regression model with binary outcome (prescription of the appropriate medication). Important independent variables to consider are the clustering effect (i.e. effect of GP practice), calendar time effect (since the intervention is sequentially rolled-out) and an indicator of intervention exposure for each practice at each time point; in an to adjustment for other characteristics . These models will be fitted using population averaged models using GEE methods in STATA, allowing for clustering and adjusting for individual level covariates and any cluster level covariates where available. Population averaged models as opposed to random effects models (also known as marginal models) will be used as within the framework of cluster randomised trials, random effects models both lack appropriate interpretation and might be biased
]. The covariates to be included in the adjustment will be pre-specified and will include practice size and patient level characteristics (age, sex, diabetic status, smoking status, blood pressure, total cholesterol and HDL cholesterol); GP practice will be included as a random effect; time point as a fixed effect; and exposure or non-exposure as the main intervention effect. Null hypotheses for secondary outcomes take a similar form to that for the primary outcome. Analysis of the secondary outcomes will take a similar form to that described for the primary outcome.
The primary outcome will be considered significant at the 5% level and so 95% confidence intervals (CIs) reported; whereas secondary outcomes will be deemed significant at the 1% level (and so 99% CIs reported). This difference in levels of significance, give more weight to the primary outcomes. Analysis will be stratified by primary care trust: the analysis of the project nurse led intervention in Sandwell and the prescribing advisor led intervention in Solihull will be carried out separately.
Sample size & number of practices
Since the analysis is to be stratified by primary care trust sample size calculations were also stratified by trust. However, since this is a pragmatic evaluation the study has a limited sample size to those practices which agreed to participate in the evaluation. In the Solihull trust only 6 practices agreed to participate and it was not expected that this would provide adequate power. We therefore based power calculations on the Sandwell part of the study only. In the pilot study approximately 5% to 6% of registered patients in each practice were found to be untreated high risk patients. In Sandwell this means we expect to invite between 6,241 and 7,489 untreated high-risk patients from a total registered population of 124,820. From this fixed study size, it is possible to estimate the difference that will be detectable (difference between patients exposed and not exposed to the intervention in proportion of eligible patients medicated). These patients are spread across 26 clusters, each with a conservative estimate of average size of 240 (assuming 5% of patients eligible); but with some variation between practices sizes with coefficient of variation of 0.74.
The difference detectable depends on the level of intra-cluster correlation (ICC), the variation in sizes of practices, and the proportion of patients medicated in the control arm. Estimates of ICC would ideally come from other similar studies, but in the absence of such evidence we are guided by a review of estimates of ICCs which found that ICC values are typically between 0.02 and 0.1
]. The stepped wedge nature of this study should mean that impact of intra-cluster correlations are lower than in conventional cluster trials
], so these values can be viewed as conservative. The pilot study estimated that the proportion of eligible patients started on medication over the course of one year is around 13%.
So, using several different estimates of likely values of ICCs and current proportions of patients medicated, we have estimated what values for outcomes post intervention could be detected (all for 80% power, 5% significance and coefficient of variation of 0.74). We have used conventional cluster RCT power calculation methods since this study is a modification of the conventional stepped wedge design and so provide conservative estimates
For example, if the current proportion of patients medicated is about 13%, and values for intra cluster correlation low (ICC of 0.01) then this study would be powered to detect an increase in proportion of high risk patients started on treatment to 19% (a relative risk of 1.46); with an ICC of 0.05 the study would detect an increase to 21% (relative risk 1.62); with an ICC of 0.1 the study would detect an increase to 26% (relative risk 2.00). If however the current proportion of patients medicated was closer to 20% and the ICC much higher (say 0.1) then this study would provide 80% power to detect an increase to 40% (that is almost a 20% absolute percentage increase).
An economic evaluation will also be carried out. Costs will be imputed from prescribing and consultation rates determined from data extracts. A Markov model will be constructed to determine long term impacts on costs and health outcomes (QALYs) using the effects of treatment in the short term to determine the long term impact on QALY life expectancy and health service costs.
Alongside the trial, qualitative research will explore the barriers and facilitators to the implementation of the programme. In combination, the qualitative and quantitative components of the full study will generate knowledge not only on the potential value, but also the feasibility and acceptability, of rolling out the case-finding intervention more widely in the NHS. Birmingham and the Black Country provide the opportunity to explore attitudes to and experience of primary prevention within an ethnically and economically diverse urban community. These findings may also be transferable to other similar programmes for prevention of serious disease in primary care. Interview data will be collected from the implementers of the programme, including the research team, and from patients that have attended.
The aims are:
To explore the implementation process, using in-depth interviews, and propose theories about what implementation configurations need to be in place for successful implementation.
To explore the patients’ experience of the intervention and how it fits into their broader health beliefs and health behaviours, using subject-produced photo elicitation
] (Harper 2002) in the context of an open-ended interview. This will enable an appraisal of the acceptability of the intervention along with insights into barriers and facilitators to adherence.
Advice on the need for ethical approval was sought from the National Research Ethics Service. Implementation of the targeted case-finding programme was being rolled out as rapidly as was practical. The order in which the implementation was being carried out in each practice had been determined in advance by randomisation. No additional investigations or measurements were being undertaken and no patient identifiable data would leave the NHS. We were therefore advised that this did not need ethical approval as it was evaluation of a service development [ Additional file
Although interviews with patients and clinical staff for the qualitative evaluation could also be considered as part of the service evaluation we sought and obtained ethical approval from the University of Birmingham ethics committee for this part of the study [ Additional file
]. Information sheets and consent forms were provided for interview participants who were clinical staff or patients [ Additional file
, Additional file
, Additional file
and Additional file