Discussion

This study provides a process evaluation of a multicomponent Antimicrobial Stewardship intervention that successfully influenced physicians’ prescribing behavior, leading to a 26.6% relative reduction in DOT with broad-spectrum antimicrobials in hospitalized CAP patients. Prospective audit and feedback, clinical lessons, and engagement of local opinion leaders were considered the most effective intervention components, while the e-learning and poster were perceived the least effective and, therefore, potentially redundant components.

Opinion leaders, clinical lessons and prospective audit and feedback have been identified as key components for influencing prescribing behavior. ^{Reference Davey, Marwick and Scott11,Reference Dellit, Owens and McGowan15,Reference Grol, Bosch, Hulscher, Eccles and Wensing18,Reference Vaughn, Krein and Hersh19} However, the opinion leader component in this study was perceived as only moderately effective. To improve this, interviewees recommended selecting opinion leaders from each relevant specialty, with regular meetings to evaluate performance and address barriers to guideline adherence. All participants identified clinical lessons among the most effective components, despite only 31% of the target audience attending the first lesson. Based on our findings we recommend lessons to be scheduled during existing educational sessions, include all relevant medical specialties, and supervisors actively encouraging residents to attend. Prospective audit and feedback were considered both effective and well appreciated by the participants, though it required 15 to 60 minutes per day. The use of algorithms to select patients for feedback was recommended, as this could reduce time required for audit. Pocket cards were not identified as either the most or least effective component, and were scored comparable for overall appraisal as clinical lessons, but were considered easy to implement. Their use varied across hospitals, with residents using them in teaching hospitals, and both residents and specialists in the general hospital. The e-learning content was appreciated, but due to an overload of competing e-learning invitations, it received lower scores compared to clinical lessons. Finally, none of the interviewees recalled the poster, and it remained unclear whether, or for how long, it was displayed.

Detailed reporting of performed Antimicrobial Stewardship interventions and conducting process evaluations are essential for enabling reproducibility and facilitating comparison of outcomes across studies. ^{Reference Davey, Marwick and Scott11,Reference Hulscher and Prins14,Reference Rzewuska, Charani, Clarkson, Davey and Duncan20} Generalizability of ASP studies largely depends on the quality of implementation of the different intervention components. Additionally, qualitative studies focusing on the barriers and facilitators of ASP components are needed to address gaps in understanding factors that influence prescribing behavior and to identify the most effective strategies for future ASP efforts. ^{Reference Abdel Hadi, Eltayeb and Al Balushi21,Reference Morris, Calderwood and Fridkin22} Although the intervention study was conducted between 2015 and 2017, this process evaluation conducted in 2019 provides valuable insights into how the components were implemented, how they were received by stakeholders, and how they might be improved. Despite technological advances, adherence to CAP guidelines remains suboptimal, ^{Reference Shanmuganathan, Kornum, Bruun and Mygind23–Reference Yoon, Nolan, Humphrey, Duffy, Thomas and Ritchie25} and our findings remain relevant to effective Antimicrobial Stewardship.

The multicomponent ASP intervention was applied as a one-size-fits-all approach, but reductions in broad-spectrum antimicrobial DOT varied across hospitals. Previous studies have suggested that locally adapted ASPs may be more effective than standardized approaches. ^{Reference Hulscher and Prins14,Reference Hartman, van de Pol and Heltveit-Olsen26} Designing the intervention as a toolkit and allowing hospitals to tailor components to their local context, could enhance its effectiveness. For example, although the e-learning content was considered valuable, it was perceived as one of the least effective components. In hospitals already offering numerous e-learning opportunities, and where low engagement with the module was anticipated, this component could reasonably have been omitted with efforts redirected towards more effective components.

A known barrier to implementing prospective audit and feedback is its time-consuming nature. ^{Reference Barlam, Cosgrove and Abbo27} Although not explicitly mentioned in the interviews, possibly because some interviewees were no longer directly involved in the audit, prior meetings with PIs and researchers suggest that the time required for auditing was negatively perceived. One hospital introduced an algorithm to automate auditing and continued to carry out prospective audit and feedback outside of study participation. Automating the identification of CAP patients with broad-spectrum antimicrobials could substantially reduce the time and effort required for auditing, thereby improving workflow continuity for feedback providers. Automation may facilitate more timely and direct feedback to prescribing physicians and support the guidance of appropriate empirical therapy through integration with electronic healthcare systems. ^{Reference Wilde, Song and Moore28}

Various facilitators and barriers may have influenced the impact of the ASP intervention in the participating hospitals. In hospital 1, the low activity of opinion leaders may have limited the potential effect. Nevertheless, half of the target audience attended the clinical lessons, and all provided feedback was incorporated. The ASP had a moderate effect only in hospital 2, despite already high baseline guideline adherence, which might be attributed to the highly active opinion leaders and a 52.6% e-learning completion rate. In hospital 3, the high intervention effect might be explained by short communication lines between physicians, combined with a high recommendation acceptance rate and nearly half of the target audience attending the clinical lessons. Finally, in hospital 4, both the influence of a nearby academic hospital and internal conflicting opinions may have hindered the intervention, as reflected in a low feedback acceptance rate of 41.8%.

A limitation of this study is that not all data necessary for a full process evaluation were collected. Specifically, the exact time required to prepare clinical lessons, exposure to follow-up lessons, use of pocket cards and poster, and the actual performance of local opinion leaders were not recorded. Nevertheless, the available data provide sufficient detail to understand the caveats and key drivers of a successful ASP and facilitate its implementation in other settings. Although interviews were conducted two years after the trial, all participating hospitals were actively using intervention components at the time, which may have minimized the risk of recall bias. Only four hospitals were interviewed and inclusion of all hospitals might have provided additional contextual insights. However, saturation was reached after eleven interviews and similar components were perceived as most effective, irrespective of hospital-level outcomes. Therefore, these data provide a comprehensive overview of the perceived component effectiveness.

To conclude, hospitals aiming to reduce broad-spectrum antimicrobial use in CAP patients may consider to conduct daily prospective audit and feedback, implement regular clinical lessons, distribute pocket cards, and appoint actively engaged local opinion leaders across relevant departments. Additionally, it is recommended to periodically report prescribing performance within the hospitals, compare performance between departments, and establish performance evaluation structures across hospitals.