Discussion
There are two important findings from this investigation. First, combining provider and patient education with regular feedback in the form of unblinded normative comparison with local peers results in marked improvement in the inappropriate prescription of antibiotics for ARTI in the primary care setting. Second, almost 90% of providers were shown to be responsive to this method of behaviour change.
Three key components of the intervention likely influenced the outcome. First, provider education was delivered in the form of up-to-date guideline recommendations, educational videos, and pocket guides regarding the evidence-based management of ARTI. Second, providers were assisted with patient education and expectations both prior to, and at the time of, the clinic encounter. Posters placed in patient waiting areas conveying that ‘antibiotics aren’t always the answer’, reset patient expectations prior to their clinical encounter. In the examination room, patient brochures and a virus prescription pad allowed patients to leave with something in hand specifically addressing their viral condition, while providing reinforcement supporting the physician’s decision to avoid antibiotics. Importantly, these messages were delivered under the auspices of the US Department of Health and Human Services, and the CDC, further validating the provider’s recommendation. Third, and most importantly, the use of normative influence by way of biweekly, unblinded reports provided a broader perspective to each providers’ practices.
Although patient education in the form of printed materials used in the waiting and examination rooms was a component of the intervention, this intervention alone has generally been ineffective at modifying patient behaviour and altering patient outcomes.21 However, interventions that use provider and patient education combined with provider insights as to their rate of inappropriate antibiotic prescribing have been successful in reducing rates of inappropriate antibiotic prescribing. Combined with education, Gjelstad et al demonstrated a reduction in antibiotic prescribing (OR 0.72) for ARTI compared with controls following a one-time display of a practitioner’s prescribing profile from the preceding year.22 Meeker et al tested multiple behavioural interventions in primary care practices to evaluate which achieved the greatest efficacy in reducing inappropriate antibiotic prescribing for ARTI. Both accountable justification at the time of prescribing as well as regular reports displaying blinded peer comparison resulted in the lowest rates of inappropriate antibiotic prescribing.15 A clear advantage of peer comparison over accountable justification is that it adds no additional work for the provider during the clinical encounter. Our study was unique in that it provided regular unblinded reports to all PCPs within our health delivery system and was markedly successful at a system level. In such an environment, unblinded peer performance information may facilitate developing a group norm around which individual physicians can anchor their own performance. Although providers may be uncomfortable with the local colleagues seeing their individual performance scores, unblinded data may better support peer-based learning.23
Providers are motivated to administer care that encompasses the current evidence while improving patient well-being. Moreover, individuals often conform to fulfil peer expectations and gain acceptance, particularly if the behaviour aligns with professional standards.12 14 24 The unblinded nature of the reports aided in generating conformity towards an evidence-based goal, in keeping with professional standards and patient’s best interest. In addition, physicians often make different decisions when managing individual patients versus viewing groups of patients from a population-based perspective. Specifically, physicians have been shown to give more weight to a patient’s personal concerns when they consider the patient as an individual and more weight to general criteria of effectiveness when they consider the patient as part of a group or population.25 By viewing biweekly reports of their antibiotic decisions in aggregate, physicians modified their behaviour as a result of this new population-based perspective.26 Interestingly, our study also found that providers who responded to the intervention were significantly younger compared with those who did not. One possible explanation for this is providers may have long-standing patient relationships in which they feel uncomfortable changing practice for fear of a perceived ‘lack of consistency’. Additionally, it is sometimes difficult to change a long-held practice pattern even when new scientific evidence is available.27 Similarly, the CDC recently reported southern, male primary care physicians aged 40 years and older prescribed more antibiotics than specialty-matched peers.28 The barriers to such practice change need to be further evaluated.
It is worth commenting that generating and operationalising this intervention is inexpensive in the context of a comprehensive electronic medical record system. Moreover, continuation of the intervention can take place for any desired duration by creation of automated reports.
Finally, the clinical and financial impact of these changes can be significant. Adverse drug events increase the risk of death, length of hospitalisation, emergency department utilisation and overall cost of care, and antibiotics are the leading cause, representing nearly 24% of adverse drug events present on admission.9 In the second year of the intervention, across all primary care practices, we avoided prescribing an estimated 16 589 unnecessary antibiotics. The estimated savings include approximately $751 664 in antibiotic costs (average wholesale price) and further avoided an estimated 3318 adverse drug events costing an estimated $5 447 796.29–34
Limitations
First, this was a single-centre study, and our results may not be applicable to other practice settings. Second, there were significant baseline differences in the rate of inappropriate antibiotics prescribed, between test and control groups. This may suggest that the test group was more amenable to the intervention than other providers, but the degree of impact in the control group in year 2 suggests otherwise. Third, we did not qualitatively evaluate the thoughts and attitudes of providers who did not respond to the intervention. Further assessment of these outliers may lead to optimisation of similar normative comparison initiatives in the future. Fourth, the intervention consisted of three components together, and we cannot discern the individual impact of each one. Finally, we do not know the durability of the intervention over time, and this will require further study.