Clinical and health services research is continually producing new findings that may contribute to effective and efficient patient care. However, despite the considerable resources devoted to this area, a consistent observation is that the transfer of research findings into practice is unpredictable and can be a slow and haphazard process. This phenomenon is apparent across different healthcare settings, countries, and specialties, including the United Kingdom (UK) 123, other parts of Europe 4, and the United States of America (USA) 567, with obvious implications for patient care. Studies have been unable to explain this variation in terms of either patient or resource factors. Accepting that variation alone does not necessarily represent inappropriate care, a small number of studies have gone on to assess appropriateness 7 and conclude that inappropriate care delivery was occurring.

Symptoms of upper respiratory tract infections (URTIs), are common presenting features in primary care. They are frequently treated with antibiotics, and rates of antibiotic prescribing have been increasing in the UK 8. However, "the absolute benefits of using antibiotics in the treatment of sore throat are modest. Protecting sore throat sufferers against suppurative and non-suppurative complications in modern Western society can be achieved only by treating with antibiotics many who will derive no benefit." 9. Reducing antibiotic prescribing in the community by the prudent use of antibiotics is seen as one way to slow the rise in antibiotic resistance 1011 and appears safe, at least in children 12. However, our understanding of how best to achieve this is limited 13.

Implementation research is the scientific study of methods to promote the uptake of research findings, and hence to reduce inappropriate care. It includes the study of influences on healthcare professionals' behaviour, and methods to enable them to use research findings more effectively. Over the last decade, a considerable body of implementation research has been reviewed 141516. This research demonstrates that a range of interventions (e.g., reminder systems, interactive educational sessions) can be effective in changing health care professionals' behaviour. These studies have substantial heterogeneity within interventions used, targeted behaviours, and study settings that make generalising their findings to routine healthcare settings problematic. This is largely due to the absence of any underlying generalisable framework for both research studies and service settings by which to characterise individuals, settings, and interventions.

The interventions used are usually complex. The framework for phases of investigation of complex interventions suggested by the Medical Research Council (MRC) 17 illustrates the current position with respect to implementation research. Table 1 compares the stages in the evaluation of complex interventions to stages of drug evaluation. To date, most implementation research studies have involved exploratory trials (Phase II) or, more usually, definitive randomized controlled trials (RCTs) (Phase III). While the effectiveness of interventions varies across different clinical problems, contexts, and organizations 18, studies provide scant theoretical or conceptual rationale for their choice of intervention 19. The current position in the evaluation of implementation strategies is akin to exploring the anti-anginal use of an antihypertensive drug without any understanding of the pharmacodynamics of the drug or the pathophysiology of angina or hypertension, and without Phase I trials of the pharmacodynamics of the drug. Thus, this is an expensive version of trial-and-error, with no a priori reason to expect success, nor confidence in replicating success, if achieved.

To argue against the need for a better theoretical basis for choosing implementation interventions, one would have to suggest that every combination of setting, individual(s), and intervention is unique and must be examined individually – this would require thousands of evaluations and would incur prohibitive costs. The assumption that clinical practice is a form of human behaviour and can be described in terms of general theories relating to human behaviour offers the basis for a generalisable framework. Therefore, factors influencing the effectiveness of interventions could include the beliefs of the healthcare professional, or their perceived ability to control – generalisable concepts that can be used across different interventions, settings, and individuals.

In order to optimise the number of definitive RCTs studies that will be both costly and time-consuming that need to be conducted, and ensure their generalisability, it is necessary to understand and optimise the 'active ingredients' in professional behaviour change strategies and the characteristics of the settings, targeted professionals, and behaviours that might modify the effectiveness of interventions. Two approaches are necessary to achieve this. One is to develop an understanding of the factors underlying professional behaviour in order to identify what sorts of empirical antecedents should be targeted in implementation interventions (equivalent to the theoretical phase of the MRC Framework, and the subject of our previous work 20). The other is to develop an understanding of how the elements of the interventions work and can therefore be optimised (the modeling and exploratory trial phases of the MRC Framework).

Almost all of the implementation interventions conducted to date have selected interventions using intuitive/non-theory analytical or empirically successful methods. Three other methods (behavioural change technologies, targeting theoretical antecedents, and targeting empirical antecedents) have been much less developed in implementation research. However, if psychological theory is going to contribute to effective implementation, then targeting empirical antecedents and using behavioural technologies should be the optimum methods of selecting interventions. There are three additional important issues to consider: plausibility and feasibility (both in a development experiment and in service settings), and the method of delivery to maximise efficiency.

The aims are 1) To develop interventions to change beliefs that have already been identified as antecedents to antibiotic prescribing for sore throats, and 2) to experimentally evaluate these interventions to identify those which have the largest impact on behavioural intention and behavioural simulation.

We will use the management of uncomplicated sore throat in general practice as the clinical focus for this work. Participants will be general medical practitioners.

Two interventions will be developed to address the discriminant beliefs in the prescribing of antibiotics for sore throat. Appropriate intervention components will be selected from a number of available evidence-based behavioural technologies. The design of the interventions will incorporate these techniques and will be further informed by the empirical findings of our previous studies. The interventions will be evaluated in a 2 × 2 factorial randomised controlled trial delivered in a postal questionnaire survey.

Our previous work 21 has identified eight discriminant beliefs that distinguish between GPs who do (intenders) and do not intend (non-intenders) to prescribe antibiotics for patients with uncomplicated sore throat (Table 2). We will make the assumption that altering these beliefs will change intentions to manage URTI without prescribing antibiotics, and we will therefore design the interventions to change these beliefs. Therefore, it will be possible to test the assumption empirically by applying a mediational analysis to explain intervention effects. Two theory-based interventions that incorporate behaviour change technologies will be designed to promote the management of URTI presenting in primary care without prescribing antibiotics.

Two outcome measures will be assessed, behavioural intention and behavioural simulation. We will measure behavioural intention using the standard methods used in investigations based on the theory of planned behaviour using rating scales of likelihood, frequency, or agreement with statements or questions about intention (e.g. Out of the next 10 patients you see with acute sore throat, how many do you intend to prescribe antibiotics for? Score 0 – 10). To measure behavioural simulation, participants will be asked to respond to written scenarios describing patients presenting with sore throat in general practice. The scenarios will reflect the range of patients and clinical features that present in general practice informed by qualitative work conducted in our previous work 21. Participants will be asked to write on a simulated set of notes the relevant management they would use.

We will examine whether the interventions affect the discriminant antecedents identified in the previous theory of planned behaviour study (Table 2). We have piloted these methods successfully 2931. The results will be explored using the Baron and Kenny methodology for mediational analyses 32, which incorporates the Sobel test, to ascertain the extent to which these antecedent beliefs mediated effects on outcomes within these experiments. Where possible, the measurement will be made twice, with these process measures assessed both before and following the intervention. The time between measurements will be six to eight weeks.

The experimental materials will be delivered by post. The experiment will be embedded within a questionnaire survey which will be administered twice, once before the intervention and once immediately following the intervention. Based on our previous experience, we plan that subjects will receive a letter of invitation, a set of instructions, and individually packaged set of materials for measuring behavioural simulation and intention that they will be asked to read in this order. On the second administration of the survey, they will also receive the intervention which they will be asked to complete prior to completing the process and outcome measures. Two reminders will be mailed to non-responding clinicians. Given our experience of the response rate in our previous study 31, we plan to offer a £10 incentive to each subject to increase response rates 3334.

In a 'definitive trial', there is inherent variability in the number of patients who consult with each condition, and the characteristics of these patients vary from doctor to doctor and from year to year. By giving all subjects in the experiment the same context in which to examine behavioural intention, we have eliminated these two sources of variation. Therefore, if we use the same outcome in both the trial and in the IME, we would expect its standard deviation to be smaller in the IME than in the trial. Thus, a given shift in outcome (difference between two groups) represents a much larger effect size (difference in outcome divided by the standard deviation) in the IME than in the trial. Thus, if a trial were to produce a moderate effect size we might expect a large effect size in the IME. The IME will be powered to detect difference between each of the active intervention groups and the control group. Using standard methods for a continuous outcome, we need 50 subjects per group to have 80% power of detecting an effect size of 0.8 using a significance level of 2.5%, giving a total sample size of 200 for the experiment. We will over-sample, using an initial sample of 800, to ensure that we achieve this final sample size. This will be adjusted in the light of the impact of the incentive. Groups will be compared using methods appropriate for comparing independent samples (t-tests to compare two groups, analysis of covariance to compare groups adjusting for differences in baseline performance).

The study has ethical approval from the Northern and Yorkshire Multi-Centre Research Ethics committee. (REC Reference: 05/MRE03/11).