In a Norwegian hospital setting, an A&F intervention combined with distribution of a pocket version of the national CPG led to a substantial and sustained increased prescribing of appropriate empirical antibiotics. A significant immediate reduction on total treatment duration was transitory and vanished six months post-intervention. With the intervention, we achieved a 10 percentage-point targeted reduction in prescribing of high-dose benzylpenicillin.
Empirical antibiotic prescribing
To halt development and spread of AMR, appropriate prescribing of antibiotics in addition to effective infection control programmes, is important [
7]. Increasing levels of extended-spectrum beta-lactamase producing gram negative bacteria has been observed in Norwegian hospitals the recent years, probably associated with import of resistant strains, either in patients or food, and possibly also increasing domestic use of cephalosporines and fluroquinolones [
7]. Consequently, testing different intervention approaches to reduce use of known drivers for resistance is important. Particularly, use of cephalosporines and fluroquinoles can result in co-selection of resistance to aminoglycosides, and threaten the Norwegian strategy with extensive use of benzylpenicillin and gentamicin [
17]. Cefotaxime is a part of the Norwegian CPG recommendations for severely ill CAP-patients and it can therefore be discussed whether it was correct categorising cephalosporines as inappropriate and tailor a reduction in use. In theory, an intervention to reduce cephalosporines and other inappropriate antibiotics could have resulted in undertreatment. During the feedback session, the ID-specialist highlighted the importance of preferring benzylpenicillin plus gentamicin before cefotaxime, but also emphasised that cefotaxime is an alternative in patients with severe infection and specific complexities. Indicated by CRB-65 score, few patients had a severe infection, and this suggest that cefotaxime might only be necessary in a minority of patients. Altogether, we substantially reduced prescribing of broad-spectrum antibiotics known to promote AMR, and we succeeded in reducing prescribing of these antibiotics with no obvious negative effect on measured clinical outcomes. Consequently, our results indicate that an A&F intervention can have positive effect on empirical antibiotic prescribing and may be expanded to other settings with low level of AMR.
The high degree of maintained empirical prescribing suggests a low level of treatment failure. In other studies, change of empirical prescribing has been reported in 6-31 % of patients [
18‐
20].
The significant higher proportion of patients with penicillin allergy in the pre-intervention phase may theoretically have resulted in a higher prevalence of inappropriate antibiotics prescribing in the pre-intervention phase (i.e. cephalosporins for non-immediate allergy), which again may have biased our positive results post-intervention. To test this, we excluded all patients with penicillin allergy in both phases, which yielded no significant change of the reported mean, trend- or level effect. It is worth noticing that the overall prevalence of penicillin allergy in our study was 12.6 and 5.8 % pre-and post-intervention, which is far higher than the 1 % estimated prevalence of penicillin allergy [
21]. In future, patients erroneously labelled penicillin allergic should be identified as it may restrain patients from more appropriate narrow-spectrum antibiotics [
21].
The lower proportion of AECOPD in the post-intervention phase compared to the pre-intervention phase did not influence our results as empirical prescribing of both conditions is categorized as appropriate.
Total treatment duration
Short-term antibiotic treatment has in randomized controlled trials (RCTs) shown the same efficacy as long-term treatment with regards to clinical-, bacteriological- and radiological success [
22]. Reducing total treatment duration is important, both to increase patient compliance, reduce risk of adverse effects and to reduce AMR development [
23]. For beta-lactams, 3–5 days treatment has been found to be safe [
22,
24]. In a European multicentre study including 14 centres from three countries, mean duration was 8.9 and 11 days for AECOPD and CAP patients, respectively [
9]. Appropriate treatment duration of antibiotics seems to be neglected and interventions are requested [
23,
25].
Multifaceted prospective interventions by antimicrobial stewardship teams have been shown effective in order to reduce length of antibiotic treatment. Avdic et al. showed that a team comprising an ID-specialist and a clinical pharmacist recommended reduction of duration of antibiotic treatment in 59 % of patients by applying an algorithm suggesting the appropriate duration. The mean treatment duration was reduced from 10 to 7 days [
26]. Lesprit et al. showed a reduction in mean antibiotic treatment duration from 10 to 7 days in a multicentre RCT in surgical and medical departments, where an ID-specialist performed systematic reviews of antibiotic prescribing at day 1 and day 3–4 in the intervention group. No negative effects on clinical outcomes were observed [
27]. In a study by Murray et al., they recommended antibiotic duration based on a severity score, automatic stop dates at time of initiating empirical prescribing and pharmacist feedback to prescribers. Overall, the mean duration of treatment was reduced from 8.3 to 6.8 days, with a subsequent reduction of gastrointestinal adverse effects [
28]. A retrospective A&F probably requires fewer recourses compared to prospective A&Fs, and we therefore found it important to test this approach
. In our study, we achieved a smaller reduction in total antibiotic duration compared with studies based on more comprehensive intervention strategies. In future, prospective interventions involving a multidisciplinary approach should be tested in a Norwegian setting.
Dose of benzylpenicillin
In Norway, there have been concerns about an unjustified increase in prescribing of high-dose benzylpenicillin over the recent years [
29]. High doses are associated with increased overall consumption of antibiotics, which may increase risk of adverse effects and AMR development [
30‐
33]. Benzylpenicillin possesses time dependent killing, and time above the minimum inhibitory concentration (T > MIC) should cover about 50 % of the dosing interval. Peak efficacy is reached at about 5 times above MIC [
34]. Based on a national high proportion of susceptible
S.pneumoniae (MIC ≤ 0.06 mg/L), most patient will benefit 1.2 g x 4 (i.e. low dose) [
7]. For
H.influenzae a clinical breakpoint for benzylpenicillin has not been defined. Time-kill experiments suggested that 3.0 g x 4 (i.e. high dose) can be recommended in
H.influenzae without resistance mechanisms [
35]. However, as
H.influenzae rarely results in an invasive infection and is expected in a minor proportion of patients, the Norwegian CPG have chosen to recommend low-dose benzylpenicillin as empirical treatment in non-severe patients [
6]. It is relevant to notice that in case of identified
H.influenzae, a change to high-dose benzylpenicillin (when susceptible) or to an alternative antibiotic may be warranted. In severely ill CAP-patients, 3.0 g x 4 is recommended due to possible altered volume of distribution and protein binding. In our study population, only 6 and 11 % (pre- and post-intervention) of the CAP-patients had a severe infection according to CRB-65. Accordingly, we suspect there is a potential for improvement beyond 10-percentage point decline in high-dose benzylpenicillin achieved in this study. However, this must be further explored.
Behaviour change
At UNN Tromsø, the local infection-, microbiology- and infection control team have focused upon the importance of appropriate empirical antibiotic prescribing over several years. This continuous focus may have lowered the odds of our A&F intervention to increase appropriate empirical antibiotic prescribing. Opposite, there has been little focus on treatment duration and dose, where we observed a lower effect of our intervention. Avdic et al
. assessed prescribers’ knowledge and practice related to antibiotic prescribing of hospitalized CAP-patients. They observed that most prescribers were uncomfortable with short treatment duration, and suspect that prescribers are not up to date on studies supporting short-time treatment [
26]. This is supported in our study by comments received during feedback session. Physicians stated that possible explanations on the discrepancies between audit results and CPG recommendations included that they were not up to date on all CPG recommendations and that prescribing often were based on traditions (i.e. unaware that doxycycline is not recommended for AECOPD and unaware of recommended treatment duration). Qualitative studies on behaviour are important in order to identify barriers and tailor interventions.
Strengths, limitations and lessons learned
ITS design is the “the strongest quasi-experimental design to evaluate longitudinal effects of time-delimited interventions” [
10]. The advantage with ITS, compared to using before-after studies only, is that the trend pre-intervention is accounted for, sustainment of the effect is explored and graphical presentation facilitate interpretation of results [
36]. Applying a before-after analysis would have deprived us the opportunity to reveal the non-sustainable effect on treatment duration. In a Cochrane review, it is recommended to include 12 data points both pre-and post-intervention in order to adequately evaluate seasonality, and to assess the immediate effect and sustainability of the intervention [
10]. Moreover, increasing number of observations per data point reduces variance and provides more stable estimates. Pre-intervention, the trend is non-significantly decreasing for both empirical antibiotic prescribing and total treatment duration, which we cannot rule out is due to variance (fewer patients admitted during summer months), seasonality or that prescribing is affected by other factors such as deputies at work during summer months. Altogether, there exist some limitation on internal validity in our study. However, the ITS seemed sensitive to detect change in level and trend. Potential bias in form of competing interventions cannot be ruled out without a control to establish our hypothesis. Data collection was not affected by the intervention and outcome variables were objectively assessed.
Another strength in our study is that the physicians at the study department were not informed about the pre-intervention audit before we presented the audit results. Consequently, Hawthorne effects can be ruled out. With regard to the influence on prescribing results, we retrospectively acknowledge that our results could have been even better if we had defined explicit targets together with the department physicians (e.g. “the target is to shortening the mean total treatment length to 7 days”). In addition, providing the feedback both in oral and written format more frequently, as recommended in a recent Cochrane review, could have been beneficial [
11]. Electronic prescribing and medication charts are not implemented in the majority of Norwegian hospitals. Consequently, electronic audit and surveillance is currently impossible and limits the possibility for feeding back results as the study progresses. As a consequence of only providing the feedback once and only orally, we could not reach all physicians by our intervention. Despite this, we believe that for instance junior physicians in the emergency department (ED) anyway has been exposed to advices based on our intervention, as they frequently turn to the physicians at the respiratory medicine department when handling CAP and AECOPD patients. These ED-physicians were not invited to the feedback session. In addition, junior staff at the clinical departments rotates every 4–6 months. Consequently, some that are active during the pre- and post-intervention phase may not have been present at the feedback session. However, we emphasize the recommendation from literature that interventions are recommended to target senior physicians, and not junior physicians, as the seniors act as supervisors for the less-experienced physicians [
37,
38]. Moreover, lack of adherence to CPGs is more pronounced among seniors compared to juniors [
37].
It is relevant to notice that in this study an ID-specialist had a central role in the intervention team. In Scandinavian countries the ID-specialist is perceived as important for antibiotic prescribing [
37]. This is in contrast to countries/settings with a more hierarchical work system, where it has been found that clinical leaders and senior physicians overrule the ID-specialists’ advice on antibiotic prescribing [
38]. Altogether, it is a limitation that we only provided the feedback once, but a possible negative effect of this is probably reduced by high attendance of senior physicians at the feedback (four out of five attended) and the active participation of the ID-specialist.
We emphasise that CPG recommendations does not address all patient scenarios, and complete adherence to CPG is probably not desirable. In this study, examples of justified non-adherence are use of cefotaxime in patients with severe infection and specific complexities, empirical antibiotic prescribing in patients with penicillin allergy, microbiological findings and delayed clinical response. We have not collected data on this, but it should be considered for future studies.
Our study has other limitations. First, our study has limited external validity. Optimally the intervention should have been performed in multiple hospitals. Second, economic and microbiological measures where not assessed in this study, and should if possible, be included in future studies. Third, proportion of intensive care units admissions pre- and post-intervention is also an alternative outcome measure that should be collected to describe severity of disease in the patient population. Fourth, CRB-65, which is a well-validated severity-scoring tool, was not documented in patient records and we therefore calculated the severity based on information at admission. Fifth, we are unable to distinguish the effect of our A&F intervention from the effect of the pocket version of the national CPG that was distributed to all departments at the hospital during same month as our intervention. Preceding the distribution of this pocket version, the national CPG was published online 2013 [
6]. One the other hand, as literature indicate that distribution of educational material alone is ineffective, we do not believe that the effect of this pocket version may have been substantial in itself [
39].