Implementing antibiotic stewardship in high-prescribing English general practices: a mixed-methods study

Background Trials have identified antimicrobial stewardship (AMS) strategies that effectively reduce antibiotic use in primary care. However, many are not commonly used in England. The authors co-developed an implementation intervention to improve use of three AMS strategies: enhanced communication strategies, delayed prescriptions, and point-of-care C-reactive protein tests (POC-CRPTs). Aim To investigate the use of the intervention in high-prescribing practices and its effect on antibiotic prescribing. Design and setting Nine high-prescribing practices had access to the intervention for 12 months from November 2019. This was primarily delivered remotely via a website with practices required to identify an ‘antibiotic champion’. Method Routinely collected prescribing data were compared between the intervention and the control practices. Intervention use was assessed through monitoring. Surveys and interviews were conducted with professionals to capture experiences of using the intervention. Results There was no evidence that the intervention affected prescribing. Engagement with intervention materials differed substantially between practices and depended on individual champions’ preconceptions of strategies and the opportunity to conduct implementation tasks. Champions in five practices initiated changes to encourage use of at least one AMS strategy, mostly POC-CRPTs; one practice chose all three. POC-CRPTs was used more when allocated to one person. Conclusion Clinicians need detailed information on exactly how to adopt AMS strategies. Remote, one-sided provision of AMS strategies is unlikely to change prescribing; initial clinician engagement and understanding needs to be monitored to avoid misunderstanding and suboptimal use.

be effective than interventions with a single focus. 4,9,10 Qualitative work has identified that GPs want interventions that decrease diagnostic uncertainty, provide patientcentred care, and are easy to implement. 11 Despite this, effective interventions have not been routinely implemented, with often only temporary improvements in prescribing rates even in trial sites. 12,13 Although existing interventions target individual factors that directly influence behaviour they may fail to fully account for organisational factors that can influence intervention implementation. 14 Many AMS strategies have been available in English general practice (for example, TARGET toolkit, Antibiotic Guardian) and there have been improvements in antibiotic prescribing; however, some practices are still prescribing relatively high quantities of antibiotics, which may not be fully explained by their patient population. 1,2 These practices may benefit from more support in implementing AMS strategies. Rather than specify one strategy, giving practices a choice of approaches may help teams find what works for them. Furthermore, providing interventions that are complementary in their mechanisms of action may also provide benefit, with two or more strategies being better than either alone. 7 There is limited research on uptake and effect of AMS strategies in English general practice outside of trials. Often this research has taken the form of quality improvement initiatives in a single practice. The authors of the current study previously described the co-development of an implementation intervention, with primary care staff and citizens, for general practice to help improve use of three AMS strategies that previous trials have shown to be effective and safe. 15 This study aimed to investigate the use of the implementation intervention in high-prescribing practices and its effect on antibiotic prescribing.

METHOD Implementation intervention
The authors of the current study co-developed an implementation intervention 15 to support the use of three AMS strategies: • enhanced communication skills with or without a patient leaflet; • delayed prescriptions; and • POC-CRPTs.
The intervention was designed to be brief, provided choice in uptake of strategies, and be delivered remotely with minimal input from the research team. The implementation intervention included ( Figure 1): • identifying a champion; • holding a practice meeting to agree a practice-wide approach to implementation; • an 'antibiotic optimisation' website including: implementation support for the champion and sections on three AMS strategies for clinicians; and • physical resources: patient leaflets, POC-CRPT equipment, clinician handouts.
Each practice was offered an Afinion TM -2 analyser, 60 Afinion CRP cartridges, and 30 SureScreen CRP lateral flow tests. Practices had access to in-person training on use of the Afinion-2 analyser. Printed copies of patient leaflets and clinician handouts were provided. 15 Practice teams were advised to use the intervention materials, as they wished, and encouraged to have follow-up practice meetings. Teams were able to select which AMS strategies they wished to implement.

Setting and participants
The aim was to recruit 8-10 practices from the 20% of highest antibiotic prescribing practices in England (based on antibiotic items per Specific Therapeutic group Agesex Related Prescribing Unit [STAR-PU] from ePACT data in 2018) 16 and in areas local to the research team. Practices were contacted by email or post. Practices that

How this fits in
An intervention to support the implementation of three evidence-based antimicrobial stewardship strategies was evaluated in nine high antibiotic prescribing general practices in England. General practice teams received intervention materials and chose to use them in substantially different ways in real-life settings, outside of trial conditions. Antimicrobial stewardship strategies are complex interventions that require sufficient understanding and engagement by clinicians for successful adoption and use, to obtain the full benefit in reducing antibiotic prescribing. This study highlights that remote, one-sided delivery of AMS strategies should be done cautiously to avoid misunderstanding and suboptimal use. expressed interest were selected to ensure variation in location (region and urban/ rural), number and type of healthcare professionals (HCPs), and local area deprivation (based on the overall English Index of Multiple Deprivation 2015 by postcode). 17 Practices were offered £1000 after study set-up and another £1000 at the end of the study when at least 70% of eligible HCPs had completed surveys at each timepoint.

Data collection and analysis
Practice-level antibiotic prescribing. The primary outcome was total antibiotic prescriptions per practice, as reported in the NHS Business Service Authority dataset of all prescribing centres in England, summarised over time (count/ month). 18 Forty-five practices from the same clinical commissioning groups as intervention practices were selected as a control. Practices were matched on pre-intervention trends in overall antibiotic prescribing rate, practice list size, and prevalence of comorbidities (asthma, cancer, chronic kidney disease, cardiovascular disease, and diabetes). A difference-in-difference analysis was used to estimate intervention effects, comparing change in the differences in observed outcomes between intervention and control groups, across pre-intervention and postintervention periods.
Use of intervention materials. Website use was monitored through Google Analytics. The website address was known only to the practice teams. Practices could request additional CRP cartridges/tests or printed materials. Orders from each practice were recorded.
Surveys. Surveys were sent at three timepoints: baseline, 2 months, and 12 months. Surveys asked about views on antibiotic prescribing, the three strategies, and satisfaction with the intervention materials (Supplementary Information S1). The Consolidated Framework for Implementation Research and the Normalisation Measure Development questionnaire were used to guide question development. 18,19 HCPs consented at the start of each survey. Associations between responses at baseline and follow-up surveys were assessed using c 2 tests.
Interviews. The plan was to interview two HCPs from each practice at 6 and 12 months, to make it feasible for practice teams to participate in the qualitative interviews. The person liaising with the study team in each practice identified participants. Interviews explored views of AMS strategies, intervention materials, and antibiotic prescribing (Supplementary Information S2). Interviewees gave verbal consent before each interview. Interviews were audio-recorded and transcribed  verbatim; field notes were also made. The first and the third author used deductive framework analysis and developed an a priori framework based on the topics of interest. 20 Transcripts were coded, using NVivo software (version 12), to assign data to pre-existing categories, informed also by field notes. Data that did not fit these categories were given their own categories and the framework developed.

RESULTS
Ninety-seven practices were invited and 15 expressed interest (15% response rate). Nine practices participated ( Table 1). The study ran from November 2019 for 12 months. The COVID-19 pandemic from March 2020 onwards had an impact on UK general practice and study activities paused after this date. Results are focused on the period up to March 2020.

Antibiotic prescription data
Data from between September 2018 and March 2020 were analysed for nine intervention and 45 control practices. It was assumed implementation occurred in December 2019, giving practices 4 weeks from the start of the study to adopt the AMS strategies they chose. The mean number of antibiotic items per month for the intervention group was 331 (SD 174) and 367 (SD 182) pre-and post-implementation, respectively, and 340 (SD 172) and 374 (SD 189), respectively, in control practices ( Figure 2).
A time series plot of the total number of antibiotic items prescribed by each intervention practice over the study period indicated that in five practices prescribing increased after the intervention period and in four prescribing stayed relatively the same (Supplementary Figure S1). There was no evidence of differences in covariate distributions between intervention and control groups, or pre-and postimplementation (Supplementary Table S1). In the difference-in-difference regression, there was no evidence of an effect of the implementation intervention on total antibiotic prescribing (Supplementary  Table S2), or prescribing of individual antibiotics (Supplementary Table S3).

Survey and interview participants
Practices identified 81 HCPs to complete surveys ( Table 2). The baseline survey was completed October to November 2019, 2-month follow-up between December and January 2019-2020. The 12-month survey is not reported here as it was conducted after March 2020.
Thirteen HCPs participated in interviews: nine in February-March 2020 and eight in October-November 2020 ( Table 2). Nine participants completed one interview; four completed interviews at both timepoints. Interviews lasted 18-39 min (mean 28 min).

Prescribers' views on antibiotic prescribing
Survey data indicated that views on antibiotic prescribing changed little between baseline and 2-month follow-up  ( Supplementary Table S4). However, when asked about their prescribing at 2 months, most prescribers (28/35, 80%) believed their antibiotic prescribing had improved since the start of the study.

Engagement with implementation and the three AMS strategies
All practices confirmed they had identified an antibiotic champion, completed their practice meeting, attended POC-CRPT training, and had received intervention materials. Eight practices accepted Afinion and SureScreen POC-CRP equipment. Practice J opted out of using Afinion POC-CRP equipment as they did not consider it feasible for clinicians to share one machine. Table 3 shows an overview of implementation engagement based on interviews. It was found that survey responses sometimes contrasted with interview data (discussed below), and interview data were prioritised; these gave more detail as to how a practice engaged.
Champions were asked to have familiarised themselves with all materials,  HCPs, a n prescribers, n) prescribers, n) n (job title) n (job title) including the implementation section of the website, to have held a practice meeting discussing the three AMS strategies, and to have chosen which strategies to use. Practices differed in how champions did these activities.

Champions and practice meetings
There were 11 champions across nine practices (Table 3). Seven champions answered the survey at 2 months. Six agreed that they were able to engage and encourage their colleagues to use the intervention resources; one was neutral.
Champions were satisfied with how each AMS strategy was being implemented in their practice: communication strategies with leaflets (5/7); POC-CRPTs (7/7); and delayed prescriptions (7/7 In interviews, only one champion (Practice J) reported spending time on the website. Other interviewees were either not aware of the website or had only briefly looked at it, reporting that they already felt familiar with the content. Instead, champions had focused on the physical materials that had been posted to practices.
Some interviewees thought that the website was aimed at patients rather than themselves. Despite this, participants generally thought a website was an appropriate format for them to access information easily. Time was felt to be the main barrier to use: Delayed prescriptions. At baseline most prescribers were confident that they could explain a delayed prescription to a patient and this did not change at 2 months (Supplementary Table S4). At 2 months, most prescribers reported using delayed prescriptions (29/35, 83%).
Most prescribers (25/31, 81%) agreed that increased use of delayed prescriptions had helped to reduce antibiotic use in their practice (of the six prescribers who did not use delayed prescriptions, two still answered the question about the effect of delayed prescriptions in their practice). Prescribers used various formats: gave to patient with advice to delay (19/29, 66%), post-dated prescription (12/29, 41%), asked to collect from agreed location (8/29, 28%), and contact practice again (6/29, 21%).
In contrast to the survey responses, interviewees reported that they did not think delayed prescriptions were useful and did not use them frequently or at all. Clinicians felt patients would take antibiotics immediately regardless of what they were told, and discussed delayed prescription formats as ways of preventing access: Interviewees in four practices (A, E, F, G) reported that the Afinion machine was used by one person. Some practices had one GP referring patients to a nurse or ambulatory clinician to have a POC-CRPT and one had one GP doing tests on his own patients: 'No one's been trained to do it, none of the GPs really know how to do it, so I'm the only one that's trained, so if I'm not here, they're not able to do it, and if I'm here, then that's when I'll do the testing for them.' (A1, nonprescriber) In all four practices only one GP's prescribing was being influenced by the POC-CRPT; other prescribers did not participate in using, or referring patients to, the test. Practice B had also allocated the Afinion machine to a nurse but only recorded two tests being conducted. Other practices had POC-CRPT equipment available to all staff but reported infrequent use. In practice D, the Afinion machine was not used at all (except as part of training) because they could not find a suitable place to keep it.
Interviewees reported that they carried out POC-CRPTs mostly on patients presenting with cough, but some practices also included patients with other conditions, indicating mission creep. (Interview participants mentioned using POC-CRPTs for: polymyalgia rheumatica, chronic obstructive pulmonary disease, abdominal pain to check it was not diverticulitis, to rule out acute pancreatitis, and knee pain to rule out septic arthritis.) Participants most often discussed using POC-CRPTs to convince patients they did not need an antibiotic, although some did use it when uncertain: 'Our patients will demand antibiotics and so we found the testing extremely, extremely useful for that because once you could give them the result and say, look, antibiotics really won't be useful, they seem to accept that more than the explanation. I think we're pretty confident that our prescribing did go down.' (E1, GP) Two practices (C, F) mentioned that the SureScreen lateral flow POC-CRPT was more practical than Afinion during home visits or when they wanted to avoid leaving their consultation room.
In summary, results indicated that practice J had engaged with implementation as intended and had chosen to use all three AMS strategies in a way that worked for their practice (Box 1), practices A, E, F, and G had engaged partially with implementation (focusing on the physical resources of POC-CRPTs and leaflets), and the remaining practices had engaged very little.

DISCUSSION Summary
Prescribing data indicated no evidence of change overall, or by antibiotic type. Practice J did not appear to reduce their prescribing, although their prescribing remained steady as opposed to other practices where prescribing rose. Engagement with intervention materials differed substantially between champions, with website engagement being poor. Lack of time and competing priorities in general practice were frequently cited as reasons for low engagement.
Champions in five practices initiated changes to adopt AMS strategies, most often the POC-CRPT, which was used most frequently when allocated to one person.

Strengths and limitations
The study emulated a real-life scenario with an intervention delivered remotely with minimal interaction between practices and researchers. Complementary AMS strategies were available to prescribers with choice to use all or some. This allowed us to assess how interventions were received outside of a trial setting. High-prescribing practices were selected to represent practices who had likely not previously engaged with AMS initiatives. Use of intervention materials could have been monitored more closely by visiting practices, although this may have influenced behaviour.

Comparison with existing literature
The discrepancy between participant reports of improved prescribing and actual prescribing rates may be explained by the use of routine data and inability, within these data, to identify antibiotic use for specific indications. Previous research has shown positive effects of AMS strategies on antibiotic prescribing for RTIs, but not on antibiotic prescribing overall; this may apply to this study. 22 Researchers have called for better diagnostic coding. 23 Data also indicated that communication across teams was poor, with intervention materials often only supporting the prescribing decisions of one clinician.
Five practices had implemented at least one AMS strategy, indicating that initial adoption of strategies in highprescribing practices is possible, although further optimisation is clearly required to improve prescribing. Previous research has emphasised the importance of champions; 22,24 however, implementation was often conducted as a one-off brief activity where decisions caused minimal disruption to existing ways of working, mainly providing additional resources to be used as desired. In some practices champions had adopted strategies themselves, but not informed others, as seen elsewhere. 25

Box 1. A summary of engagement reported by participants from practice J Engagement
• The antibiotic champion was a GP partner with time allocated to administrative tasks that could be spent on study activities. They were reported to be enthusiastic and effective at getting colleagues engaged with the study. They encouraged a team approach and set the precedent that everyone was expected to contribute.
• The champion looked at the antibiotic optimisation website closely and used the presentation slides provided to run the practice meeting. The practice meeting covered all three antimicrobial stewardship strategies and attendees discussed how each would work in their practice.
• Practice J was the only practice not to use the Affinion point-of-care C-reactive protein test (POC-CRPT) as they deemed it impractical to use one machine. The practice team made decisions to put SureScreen POC-CRPTs and leaflets in each consultation room so clinicians would have them to hand. The group also decided on how they would consistently issue delayed prescriptions.
• In interviews, both the champion and another member of staff displayed detailed knowledge of the antibiotic optimisation website. Practice J was the only practice where interview participants acknowledged the specific communication strategies (as detailed in the website) to discuss no antibiotic decisions and delayed prescriptions.
Champions appeared to see their role as time limited and predominantly focused on raising awareness, rather than encouraging active engagement over time, contrasting with previous work on champions as catalysts for dissemination of innovation. 26 The physical resources were used most by participants, highlighting the importance of environmental cues. These not only acted as reminders of AMS but also provided new tools, readily available at the time of the prescribing decision, to allow clinicians to approach consultations differently (behavioural substitution). 27 Access to POC-CRPTs had been a particular motivation to join the study and practices were primed to receive this strategy. However, despite evidence that POC-CRPTs were used, how they were used appeared suboptimal. The antibiotic optimisation website specified that POC-CRPTs were most valuable when there was diagnostic uncertainty or when considering prescribing (scenarios used in trials). 6,7 It was encouraging to see that most CRP results were low (<20). However, like previous research, participants reported that POC-CRPT was most often used to convince patients of a no antibiotic decision, so low test results were perhaps unsurprising. 28,29 Although this may have reduced prescriptions by avoiding prescribers 'giving in' to patients, POC-CRP testing was not used to its full potential and such use is arguably an expensive form of communication, particularly if it lengthens consultations. 7,30 It is also important to recognise that clinicians may overestimate patient expectations for antibiotics and, as such, use of a POC-CRPT to 'convince' patients that an antibiotic is not required may be misplaced at times. [30][31][32] Participants also reported using POC-CRPTs for a range of presentations, indicating mission creep, again seen in previous work. 28 Lack of engagement with the website meant prescribers did not know how AMS strategies could benefit them and their patients. The communication training ('Finding the right words') had been named to appeal to prescribers, who report difficulties in discussing prescribing decisions. 11,30,33 Despite being approved by clinicians during intervention development, this content was misperceived as something that prescribers already do; however, from observational work it is known this is not done consistently. 34 Delayed prescriptions were viewed in the same vein. As a result, prescribers did not see a discrepancy between their current behaviour and the desired behaviour.

Implications for practice
Champions are needed until new ways of working become ingrained. Champions are often self-selecting and internally motivated to undertake additional activities. This role is hard to replicate in high-prescribing practices where there are competing priorities without additional resource. 23,35 There may therefore be benefit to having champions outside the practice team, and, if so, the authors of this study would advocate that they should have easy and regular access to prescribers to be able to review and give feedback on adoption and use. The champion's role should be defined as a longer-term position and appropriately supported.
Champions in the current study did not engage all prescribers in their practices, which is likely to be a continuous challenge with increases in part-time working and staff turnover. Rather than training all prescribers to be fluent in all AMS strategies, it may be more feasible to triage patients with acute infections to specific individuals or teams who are trained and supported to use a breadth of AMS strategies. Such teams may utilise nurses and allied healthcare professionals, and incorporate continued professional development activities.
Interventions that are delivered remotely and passively meet challenges in how they are received in primary care. Physical resources delivered in the current study were given most attention as additions to the environment. In contrast, the website was overlooked. Online interventions are likely to be better received if incorporated into continued professional development programmes and existing electronic systems; however, this needs to be in line with existing workflows to avoid adding burden. In-person training is likely required, either with champions or practice teams, to ensure introductory messages about AMS strategies are received. An example is the TARGET 'Train the Trainers' scheme. 36 Such training allows opportunity to address preconceptions about strategies and specify how they can be used for greatest benefit.
Introduction of POC-CRPTs runs the risk of it being used in practice to support communication rather than reduce diagnostic uncertainty. A general practice consultation may therefore not be the best environment for such diagnostics. NHS England is encouraging POC testing in community pharmacies; and tests to