Comparison to previous research
Prior research investigated the influence of work experience on antibiotic prescribing decisions with heterogeneous findings. In two international interview studies [
23,
24], physicians reported a perceived influence of work experience on their antibiotic prescribing decision. Furthermore, a Canadian study that evaluated linked administrative health care data found that primary care physicians who were mid-career or more advanced in their career were more likely to prescribe antibiotics than physicians with less work experience [
25]. However, other results indicated no influence of work experience on antibiotic prescribing decisions [
26]. These different findings indicate closer examination in further efforts.
According to the TPB, the significant positive influence of work experience in this study could be interpreted asa well perceived behavioral control and attitude towards a sustainable use of antibiotics, presumably based on many years of professional experience in antibiotic prescribing and accrued knowledge about antibiotics and AMR. This might indicate that with increasing work experience and thus accumulated reference points, physicians' confidence in the ability to perform the desired behavior might rise [
16]. This assumption is supported by prior research that reported a significant lack of knowledge about antibiotics and AMR particularly in the group of physicians with little work experience and potentially paired with uncertainty in decision-making regarding antibiotic prescriptions [
27]. On the other hand, research also shows that younger physicians might more likely be guideline-oriented than more experienced physicians [
28], who may be less current with evidence and guideline-based care than younger physicians [
27]. The effect measured in this study indicates that more experienced physicians perceive participation in a program that uses knowledge-transmitting intervention measures to have an impact on their prescribing decisions. This assumption should be further investigated in future research.
The detected effect could also support the assumption of the project's implementation strategy, which specifically targeted knowledge deficits regarding antibiotics and AMR as well as physicians’ awareness regarding their own prescribing routines. Confirming this assumption, the outcome evaluation of the ARena project observed a significant reduction of antibiotic prescribing in the intervention arms in a pre-post comparison and compared to matched standard care, suggesting that the implementation program had impact on professional practice [
14,
15].
Findings of this study suggest that PCNs were a supportive environment in the ARena project. In the bivariate regression, a positive influence on decision-making in antibiotic prescribing was shown. A qualitative study within the process evaluation of ARena pointed in the same direction [
21]: Physicians described PCNs as supportive with regard to exchange (beliefs, ideas and experiences), management, and the implementation of new routines in practice, which can influence decision-making on antibiotic prescribing [
21,
29]. Research further showed that PCNs can act as effective drivers of innovation and quality improvement, especially when communication and support functioned well within the networks [
21,
30,
31]. Being part of a PCN can thus support the adoption of specific behaviors and beliefs [
13,
21]. With regards to the TPB, PCNs as a contextual factor might have had a strong influence on both social norms and attitudes towards the targeted behavior through social interactions and relationships. The analysis showed that physicians in arm II rated the PCN environment more supportive than physicians in arm I and/or III. This could be due to the implementation strategy since in arm II, medical assistants were to be strongly involved in the intervention’s awareness creating efforts to support physicians and patients alike. This may have been seen as an additional supportive factor and might have had a reinforcing effect on the perception of PCNs as a supportive environment in arm II.
When interpreting results, it should be considered that PCN member physicians might show stronger tendencies to innovation and development than physicians who are not part of a PCN [
13,
21]. The data analyzed in this study were generated in the process evaluation carried out alongside ARena. Therefore, comparisons with standard care (no intervention) could not be performed. Participants might have given socially desirable answers on the survey questionnaires. Potential biases due to the clustered structures in PCNs were accounted for by the MLA where effect sizes marginally differed from the results in the multivariable logistic regression analysis and were statistically unremarkable. However, for further interpretation and discussion of the results, it should be considered that there was a slight cluster effect.
Few reference studies were found which also address general conditions of medical practices. For example, some studies showed that communication and the organisational model have an influence on antibiotic prescribing [
26,
32,
33]. However, the results of the bivariate regression in this study show that specific contextual factors such as structural conditions, environment of existing processes, or externally defined general conditions had an influence on participation in the project and influenced decision-making on antibiotic prescribing. Due to the mean score variables used in this analysis, a differentiated consideration of the domains is not possible. Nevertheless, the results indicate that such general conditions may have an effect regarding antibiotic prescribing decisions and might be a facilitating or inhibiting factor in an implementation process. As these factors are highly individualized at physician level, practice level, or health system level, it is difficult to compare different study samples at all levels or generalize results. Further research is needed regarding such contextual factors to analyze their impact in detail. In the TPB, Ajzen [
16] described the influence of similar conditions such as legal requirements, availability of resources, or cooperations on so-called actual behavioral control. Behavioral control and motivation to behave can both significantly influence the intention and, consequently, decision-making [
16].
While one study identified a significant influence of practice area on antibiotic prescribing [
22], this study found no influence. Furthermore, a study showed that physicians who treated a high number of patients were more likely to prescribe antibiotics [
25]. This cannot be confirmed by this study either. In total, it can be seen that general conditions of the medical practice, such as the practice size or practice area, as well as specific conditions as described above, were hardly considered or mentioned in research before [
34].
No effects could be attributed to the affiliation to the intervention arm. This could be due to the fact that all groups largely received the same intervention components and that the effects of the additional intervention components in arms II and III, in addition to a small number of cases, were not statistically strong enough. Also, in this analysis, only the assignment to the intervention arm was used as a predictor. More detailed analyses on the uptake of the individual intervention components and their possible influence on physicians’ perception of antibiotic prescribing are therefore needed.
Overall, previous studies which explored contextual influencing factors were mainly qualitative. Statistical correlations in the specific context of decision-making regarding the prescription of antibiotics were only examined to a limited extent. This study therefore contributes to the identification of statistical correlations regarding this topic as a first of its kind. As described in the introduction, the description and differentiation of the concept of context is complex. In the CICI, context is differentiated into seven domains [
10]. In this study, the predictive factors were classified into all contextual domains related to individual factors and the intervention arm allocation in order to examine their influence and relationships in terms of the CICI. Due to the number of cases, only a few numbers of factors could be analyzed. The large number of contextual factors given the small number of cases can be best considered indicative of relationships and influences only. However, investigating to which extend they jointly influenced decision-making in antibiotic prescribing was limited. Large-scale studies are needed to sufficiently investigate a diverse environment around physicians’ practices in intervention studies. By doing so, effects between context and CMCI could be identified and contextual factors in the sense of the CICI could be covered as broadly as possible.
Strengths and limitations
Several strengths and limitations must be considered when interpreting the results of this study. One strength of this study was the application of the STROBE-Statement which granted methods and results to be transparent and comprehensible.
The applied questionnaire was not validated but theory-based in conceptualization. However, it can be assumed that all questions were well understood since the number of missing values was small and they were completely at random. Another strength of the study was the high response rate. Nevertheless, the number of cases was statistically too small to conduct a comprehensive analysis of all possible contextual factors identified in theory [
17]. The inclusion of predictive variables was thus limited in multivariable regression analysis and only significant factors of the bivariate regression analysis were included. Given the relatively small sample size and a potential lack of statistical power, the number of included factors in the multivariable regression analysis might have reduced the degrees of freedom so that effects were not detected. The study focused on the physicians’ perceived impact on decision-making regarding antibiotic prescribing through their participation in the ARena project. This is a self-reported impact, which may contain an over- as well as an underestimation of individual performance. In addition, it was not possible to link the results of the process evaluation to the actual prescribing rate from the intervention study. Therefore, no statement can be made about the influence of the contextual factors on actual prescribing behavior.
Finally, the generalizability of the results of this study is limited. On the one hand, the participants’ age and work experience might limit the transferability of results to younger primary care physicians. On the other hand, the study setting in PCNs must be considered when transferring results. The nested data cannot exclude the possibility that socially desirable responses were given. A potential selection bias might be present in participants' pre-existing motivation to reduce antibiotic use. Furthermore, the contextual factors used in this study are specific to the German health care system. Thus, the results cannot simply be transferred to practitioners who do not participate in a PCN as well as to non-German health systems.