The determinants of antimicrobial prescribing among hospital doctors in England: a framework to inform tailored stewardship interventions

Abstract

Aim

Little is known about the determinants of antimicrobial prescribing behaviour (APB), how they vary between hospital prescribers or the mechanism by which interventions are effective. Yet, interventions based on a sound theoretical understanding of APB are more likely to be successful in changing outcomes. This study sought to quantify the potential determinants of APB among hospital doctors in south‐west England.

Methods

This multicentre, quantitative study employed a closed answer questionnaire to garner hospital doctors' views on factors influencing their APB. Underlying constructs within the data were identified using exploratory factor analysis and subsequent pairwise comparisons assessed for variance between groups of prescribers.

Results

The questionnaire was completed by 301 doctors across four centres (response rate ≥ 74%) and three key factors were identified: autonomy, guidelines adherence and antibiotic awareness. The internal consistency for the questionnaire scale and for each factor subscale was good (α ≥ 0.7). Subgroup analysis identified significant differences between groups of prescribers: autonomy scores increased with grade until at the specialist trainee level (P ≤ 0.009), foundation doctors scored higher for guidelines adherence than consultants (P = 0.004) and specialist trainees (P = 0.003) and United Kingdom trained doctors scored higher than those trained abroad for antibiotic awareness (P < 0.0005). Scores did not vary significantly between doctors from different centres.

Conclusion

Autonomy, guidelines adherence and antibiotic awareness were identified as important factors relevant to APB, which vary with experience and training. A theoretical framework is offered to facilitate development of more effective, tailored interventions to change APBs.

What is Already Known about this Subject

• Interventions based on a sound theoretical understanding of antimicrobial prescribing behaviour (APB) are more likely to be successful in changing outcomes.

• Qualitative research highlights the importance of social norms, attitudes and beliefs in antimicrobial prescribing and guidelines adherence. However, multicentre studies would confirm the generalizability of these findings.

What this Study Adds

• Autonomy, guidelines adherence and antibiotic awareness were identified as important factors relevant to APB, which vary with experience and training.

• This study extends previous single centre qualitative interview studies to provide more generalizable insights into the perceived determinants of APB among hospital doctors in England.

Introduction

Modern medicine relies on the widespread availability of effective antimicrobials to treat and prevent infection but the emergence of antimicrobial resistant (AMR) organisms threatens their availability for future generations. This situation is recognized as a growing public health and patient safety threat 1, 2, 3, 4. Infections caused by AMR bacteria are associated with increased mortality, longer hospital stays and increased cost 1, 5, 6. Increasingly, we see patients for whom there are no remaining therapeutic options. Multi‐facetted, cross‐sector strategies to combat AMR have been proposed 7, 8, including antimicrobial stewardship, surveillance 1, infection prevention and control and development of new antimicrobials and diagnostics.

‘Antimicrobial stewardship’ pertains to the careful and responsible management of antimicrobial use. Stewardship programmes aim to optimize antimicrobial therapy for individuals, prevent antimicrobial misuse and minimize collateral damage including AMR. Over one‐third of patients in English hospitals are prescribed an antimicrobial at any time 9, yet prescribing is often suboptimal 10 and up to 50% of use may be inappropriate 11. Stewardship strategies have been shown to improve patient outcomes and reduce healthcare costs 6. However, to be effective, they require multidisciplinary engagement across the continuum of care and behavioural change at the individual prescriber level. Unfortunately, evidence to guide stewardship teams is limited and national guidance for acute trusts 12, 13 does not highlight stewardship as a behaviour change issue, underpinned by personal beliefs.

Antimicrobial prescribing is intrinsically complex and one large study which investigated prescribing errors in hospitals found that most prescribing errors are associated with antimicrobial drugs 14. Because infections are ubiquitous across specialities, most are treated by clinicians without expert knowledge of infection management and prescribing is often devolved to junior doctors 15. Non‐expert clinicians often feel a patient's immediate risk outweighs the long term disadvantages of prescribing an antimicrobial 16, 17, which can lead to overuse 18. Implementation research recognizes the importance of understanding the influences on healthcare professionals' behaviour when seeking to change it. However, while systematic reviews of behaviour change interventions have found modest and worthwhile effects, the patterns and mechanisms of effectiveness are often unclear 10, 19.

Qualitative research highlights the importance of social norms, attitudes and beliefs in antimicrobial prescribing and guidelines adherence 20. The important role of senior doctors in influencing junior clinicians is widely reported 15, 21, 22. Charani et al. 22 recognized a preference for decision making autonomy in senior doctors, including non‐interference with their peers' prescribing decisions. Clinicians often emulated incorrect prescribing behaviours of fellow clinicians 23. Attitudes towards guidelines vary considerably, from studies that report a lack of confidence in them 24, to accepted non‐compliance 22, accepted compliance 25 and positive evaluations 15. Barriers to optimal antimicrobial use included the tendency of clinicians to perceive AMR as a national problem 23, workflow, role perception, organizational communication and the low priority assigned to antimicrobials 26, concerns about patient outcomes when prescribing narrow‐spectrum therapy and under‐confidence in targeting therapy to culture results 24.

Successful behaviour change initiatives will need to engage with individuals (e.g. beliefs, perceptions, attitudes) and organizations (e.g. workload, systems, priorities). Multicentre quantitative studies of the determinants of antimicrobial prescribing are needed, to enable targeting and tailoring of interventions, for example by grade or specialty. The current study aimed to quantify the perceived determinants of antimicrobial prescribing behaviour (APB) among doctors in English hospitals and explore how these determinants varied between groups of doctors, to inform the design of effective antimicrobial stewardship.

Methods

Setting

The study was conducted at four hospitals in south‐west England, three teaching hospitals including a tertiary referral centre and one district general. The centres were of different sizes, offering varied services, but all had antimicrobial stewardship programmes in place at the time of the study and were recruited via the lead author's existing professional networks. Research Ethics Committee approval was sought but not required. Approval was obtained from each centre's Research and Development department.

Questionnaire development

A literature search, using a variety of search engines (e.g. Medline) alongside a ‘pearl‐growing’ approach whereby papers cited in key publications were sought, was undertaken to identify determinants of APB in secondary care 16, 20, 21, 22, 23, 24, 25, 26, 27, 28 to inform development of items. Models and theories of behaviour 29, 30, 31, good practice recommendations for questionnaire design and other studies' questionnaires and interview schedules were also reviewed. Refinements were made in accordance with feedback from experts and pilot testing with doctors from different specialities and grades. The final iteration included an explanation of study purpose, instructions on completion, demographic questions, 60 short statements (items) relating to antimicrobial prescribing – participants were asked to indicate the extent to which each reflected their own opinion on a Likert scale ((1) strongly disagree, (2) disagree, (3) uncertain, (4) agree, (5) strongly agree) and three free‐text ‘open questions’ to allow participants to elaborate further (see supplementary information for a copy of the questionnaire). Questionnaires were completed using pen and paper. A consent section was not included as the survey was anonymous. Consent was inferred when the doctor completed and returned the questionnaire.

Sampling and recruitment

All qualified doctors working within the study centres during the data collection phase (March 2014–May 2014) were eligible to participate. Site coordinators recruited participants on ward rounds and via pre‐determined meetings, for example departmental/audit meetings and teaching sessions, purposively selected to ensure a good demographic spread. This personal approach was chosen over an e‐mail approach in order to maximize participation and response rate. Doctors were provided with a participant information leaflet before being asked if they would like to participate and it was their choice whether they took part. Care was taken to provide only basic information that would not bias their responses.

Each centre coordinator was asked to complete a summary proforma, identifying how many doctors had been approached to participate and how many had participated, so the response rate could be calculated. Several strategies were adopted to improve the response rate including ensuring the questionnaires had a professional appearance, were easy to read and had a good balance between content and length, pre‐notification of participants (via meeting agendas) and ensuring that participants had dedicated time to complete the questionnaire.

Data analysis

Data were entered into Excel and exported into IBM SPSS Statistics 21 for analysis. Where participants had selected the ‘other’ category for professional characteristics but provided sufficient detail for us to assign a category without making assumptions, we did so (e.g. all surgical specialities in the surgery category). Information provided on the summary proforma was used to calculate a response rate. If data were missing, site coordinators were contacted for their best estimate of the response rate.

Suitability of the data for factor analysis was confirmed using the Kaiser–Meyer–Olkin Measure of Sampling Adequacy (KMO) and Bartlett's test of sphericity 32. Exploratory factor analysis (EFA) was performed to identify the underlying structure linking some or all of the items (i.e. to determine the number of distinct constructs assessed by the 60 item scale). EFA is particularly well suited to construct identification when there is no a priori expectation about the underlying structure of the data, as was the case. Principal axis factoring (PAF) was selected as the extraction method 32 and cases were excluded listwise (meaning participants with one or more missing values for any of the 60 antibiotic prescribing items were excluded from the EFA). The number of factors to retain was determined using a scree plot. This involves examining a graph of the factor eigenvalues to identify the natural elbow/point of inflection. Factors to the left of the point of infection are retained. Oblique rotation (direct oblimin) was chosen 32 as factors influencing hospital prescribing are unlikely to act in isolation 33 and behaviour is rarely partitioned into neatly packaged units that function independently 34. Rotation assists with identification of a simple structure by maximizing loadings of items onto one factor and minimizing them onto other factors. Oblique rotation is a suitable method when, on theoretical grounds, factors may be related to one another. Factor loadings above 0.4 or more negative than −0.4 were interpreted 32 (factor loadings are a measure of how much each item contributes to the factor – loadings close to –1or 1 indicate the factor strongly affects the item and loadings close to zero indicate the factor has a weak effect on the item).

After the factor solution had been achieved, all negatively correlated items loading onto any factor were reverse coded, for example item 26, loading on to factor one −0.691 became 0.691. A mean score (between 1 and 5) was then calculated for each participant for each factor. Only those who answered every item within each factor were included. Subsequently, subgroup analysis was undertaken to look for differences between factor scores across different groups of doctors using the Kruskal–Wallis test. Following standard conventions, a P value below 0.05 was considered to be significant. Where significant results were found, pairwise comparisons were made using Mann–Whitney tests. Since decisions about thresholds for statistical significance are open to critique, particularly where multiple pairwise comparisons are made, we present the P values in as much detail as possible to allow readers to consider the data against more stringent cut offs should they wish to. For example, a Bonferroni correction would involve taking the ‘usual’ threshold of P = 0.05 and dividing it by the number of subgroup analysis tests (n = 12 in this case), leading to a Bonferroni‐corrected ‘threshold’ of P = 0.004. The reliability of the three factor scale and subscales for each factor were determined using Cronbach's alpha (α).

Results

Participants

Three hundred and one doctors participated in the survey, representing 11% (301/2538) of all doctors employed at the four centres. Since only a subset of doctors at each centre were approached to participate in the research, and the majority of these accepted, the response rate was ≥74% (79%, 74%, >75%, >75% for centres 1 to 4 respectively). Among participants, there was a good spread across the medical hierarchy with consultants (reflecting the top of the hierarchy) representing 31% (n = 92), specialist trainees 14% (n = 44), core trainees 16% (n = 48), through to foundation doctors 38% (n = 114) (the most junior grade) and other doctors 1% (n = 3). There were more males 56% (n = 169) than females 41% (n = 123), with 3% not stating their gender and more doctors from medical specialities 53% (n = 160) than from surgical specialities 25% (n = 75), with 18% (n = 53) stating neither and 4% (n = 13) unspecified. Most participants, 85% (n = 255), undertook medical training within the UK. Only 12% (n = 37) trained abroad and 3% (n = 9) did not answer this question. After excluding participants who had not responded to all 60 prescribing items, 254 remained. The demographics of this group were very similar to those of the 301 whom completed the survey.

Exploratory factor analysis

The KMO was well above the acceptable limit of 0.5 32 at 0.820. Furthermore, Bartlett's test of sphericity was significant (P < 0.0005), supporting the factorability of the correlation matrix (suggesting the data were suitable for EFA). Analysis of the scree plot (Figure 1) showed inflections indicating a strong case for three factors and a reasonable case for five. Both solutions were considered. However, the three factor solution yielded strong factors with a clear theoretical basis explaining 32% of the variance whereas the five factor solution contained three factors that were difficult to interpret, one of which had only two low loading items above 0.4. As factors with fewer than three items are generally weak and unstable 34 and the theoretical basis for a five factor solution was unclear, the three factor solution was retained. Factor loadings from the pattern matrix after rotation are shown in Table 1. The items that clustered on factors one, two and three suggest they represent autonomy, guidelines adherence and antibiotic awareness respectively. The communalities show how much variance in each item is explained i.e. the proportion of an item's variance that is common variance. Twenty items that did not load onto any of the factors above 0.4 and a single item that double loaded were removed (although the participants' responses to all items can be found in the supplementary information accompanying this manuscript). The resultant 39 item scale achieved excellent internal consistency (α = 0.86) and the reliability of each of the subscales (α ≥ 0.7) was good 32. Mean scores for each of the factors are shown in Table 2 and the degree of correlation between factors is illustrated in Table 3.

Figure 1.

Scree plot showing eigenvalues for the 60 factors

Table 1.

Summary of the exploratory factor analysis (principal axis factoring) results for the antibiotic prescribing questionnaire with oblique (direct oblimin) rotation of the three factor solution (n = 254)

Item	Original item number	Rotated factor loadings*			Communalities**
		Factor 1	Factor 2	Factor 3	Initial	Extracted
I do not usually decide on the antibiotic plan for a patient	26	−0.691	0.156	−0.252	0.693	0.557
I am reluctant to amend the antibiotic prescription of a senior colleague (even if it is not in accordance with the Trust antibiotic guidelines)	45	−0.668	0.072	−0.002	0.618	0.439
I usually prescribe antibiotics on the advice of another doctor	25	−0.637	0.109	−0.123	0.675	0.424
I am reluctant to amend the antibiotic prescription of any colleague (even if it is not in accordance with the Trust antibiotic guidelines)	44	−0.634	0.066	−0.041	0.611	0.400
I am confident making antibiotic prescribing decisions	19	0.629	−0.124	0.214	0.615	0.452
I prefer to be instructed when to stop antibiotic therapy by another doctor	24	−0.625	0.129	−0.056	0.601	0.393
I am usually instructed when to stop/change antibiotic therapy by another doctor	51	−0.616	0.120	−0.171	0.636	0.416
I need more training on antibiotic prescribing	33	−0.604	0.120	0.189	0.533	0.381
I am happy to review (and amend as appropriate) the antibiotic prescriptions of others	21	0.594	0.071	0.108	0.605	0.392
I often advise others on what antibiotic(s) to prescribe	52	0.593	−0.254	0.262	0.564	0.459
I can question the antibiotic prescriptions of senior doctors	50	0.577	−0.051	0.088	0.599	0.342
I am often unsure which antibiotic to prescribe	29	−0.529	−0.049	0.033	0.518	0.288
I will stop antibiotics that others have prescribed in the absence of an appropriate indication	22	0.514	−0.047	0.155	0.484	0.295
Antibiotic prescribing is given the time it needs on ward rounds	37	0.506	−0.054	−0.232	0.523	0.290
My knowledge of antibiotics is sufficient for me to independently prescribe appropriately for the treatment of infection	18	0.497	−0.292	0.185	0.547	0.334
Prudent antibiotic prescribing is a priority for my team	40	0.475	0.141	0.100	0.533	0.283
I can question the antibiotic prescriptions of doctors who are my peers	49	0.425	0.121	0.263	0.530	0.300
When I prescribe antibiotics I consider the potential for the patient to develop a healthcare associated infection	27	0.419	0.142	0.344	0.532	0.358
I often ask for advice when prescribing antibiotics	53	−0.412	0.149	0.012	0.402	0.175
I am often unsure if an antibiotic is necessary	20	−0.412	−0.031	0.120	0.438	0.180
Antibiotic therapy is always reviewed on the ward round	41	0.401	0.093	−0.054	0.454	0.178
When prescribing antibiotics I intend to follow the Trust antibiotic guidelines	10	−0.136	0.700	0.073	0.594	0.494
Following evidence‐based Trust antibiotic guidelines will help optimise treatment outcomes	8	−0.145	0.678	0.189	0.620	0.504
I prescribe antibiotics in accordance with Trust antibiotic prescribing guidelines	35	0.035	0.661	0.062	0.602	0.454
I should not be expected to follow the Trust antibiotic guidelines	11	0.077	−0.590	−0.122	0.541	0.365
I have confidence in the Trust antibiotic guidelines	9	0.006	0.579	0.221	0.611	0.404
I am supposed to follow the Trust antibiotic guidelines when prescribing antibiotics	16	−0.057	0.559	0.248	0.549	0.386
I like to choose freely which antibiotic to prescribe	14	0.219	−0.551	−0.034	0.558	0.322
I may decide not to follow the Trust antibiotic guidelines	36	0.376	−0.539	0.306	0.579	0.464
Trust antibiotic guidelines make it easy to prescribe antibiotics prudently	15	0.113	0.521	0.172	0.482	0.346
It takes more effort to follow the Trust antibiotic guidelines (than to not)	17	0.013	−0.463	0.096	0.418	0.216
Trust antibiotic guidelines are often not suitable for the patient I am prescribing antibiotics	13	−0.029	−0.460	0.059	0.459	0.216
I know how to access the Trust antibiotic prescribing guidelines	7	0.071	0.411	0.087	0.482	0.195
I intend to prescribe antibiotics prudently	32	0.142	0.358	0.509	0.636	0.459
Antibiotic resistance is a problem in England	3	0.007	0.067	0.497	0.567	0.257
Antibiotics are overused in this hospital	2	−0.138	−0.042	0.494	0.525	0.253
Antibiotics are overused in England	1	−0.051	0.136	0.485	0.565	0.261
Antibiotics are a precious resource that I must use responsibly	59	0.016	0.308	0.415	0.511	0.288
It is my responsibility to optimize the antibiotic therapy of the patients I treat	31	0.278	0.285	0.409	0.568	0.381
Eigenvalues		9.40	6.38	3.46
% of variance explained		15.67	10.63	5.73

^* Rotated factor loadings: a measure of how much each item contributes to the factor. Loadings close to −1 or 1 indicate that the factor strongly affects the item and loadings close to zero indicate that the factor has a weak effect on the variable. Factor loadings over 0.4 appear in bold.

^** Communalities: show how much variance in each item is explained i.e. the proportion of an item's variance that is common variance.

Table 2.

Descriptive statistics for the three factors derived from the questionnaire data

Factor	Number of items	Mean score	SD	Cronbach's alpha (α)
Factor 1: Autonomy	21	3.44 (n = 275)	0.55	0.90
Factor 2: Guidelines adherence	12	4.12 (n = 291)	0.45	0.83
Factor 3: Antibiotic awareness	6	4.08 (n = 289)	0.46	0.70

Table 3.

Factor correlation matrix

Factor	Factor 1	Factor 2	Factor 3
Factor 1: Autonomy	1.000
Factor 2: Guidelines adherence	0.135	1.000
Factor 3: Antibiotic awareness	0.076	0.072	1.000

Extraction method: Principal axis factoring. Rotation method: direct oblimin with Kaiser normalization.

Subgroup analysis

Significant differences were identified across the medical hierarchy with regards to autonomy (χ²(3, n = 273) = 113.089, P < 0.0005) and guidelines adherence (χ²(3, n = 288) = 12.451, P = 0.006) but not antibiotic awareness (χ²(3, n = 286) = 4.862, P = 0.182). Pairwise comparisons revealed the nature of these differences. For autonomy, foundation doctors scored significantly lower when compared with consultants ([Mann–Whitney test statistic (U) = 807; z‐zcore (Z) = −9.343], P < 0.0005), specialist trainees ([U = 604.5; Z = −6.880], P < 0.0005) and core trainees ([U = 958; Z = −5.883], P < 0.0005). Core trainees scored significantly lower than consultants ([U = 1030.5; Z = −4.322], P < 0.0005) and specialist trainees ([U = 687; Z = −2.616], P = 0.009) and consultants' and specialist trainees' scores did not significantly differ ([U = 1523.5; Z = −1.146], P = 0.254). For guidelines adherence, foundation doctors scored significantly higher than consultants ([U = 3685.5; Z = −2.911], P = 0.004) and specialist trainees ([U = 1617; Z = −2.980], P = 0.003). In considering the threshold for P values, it is worth noting that only the pairwise comparison of core trainees and specialist trainees in relation to autonomy would change from being statistically significant to not being statistically significant if a Bonferroni correction was applied. Statistically significant differences were not seen between other grades. Doctors who completed their medical training in the UK achieved significantly higher antibiotic awareness scores than those who trained abroad (P < 0.0005 [U = 2276.5; Z = −4.867]). However, no differences were found with regards to autonomy and guidelines adherence. There were no significant differences between the centres with regards to any of the three factors (autonomy: H(3) = 3.95, P = 0.267, guidelines adherence: H(3) = 1.50, P = 0.682, antibiotic awareness: H(3) = 3.159, P = 0.368). Specialities and genders were not compared as they were not matched with respect to grade.

Few doctors responded to the open questions. However, when they did, they often cited guidelines, Clostridium difficile infection and medical microbiologists as important influences on their antimicrobial prescribing and stated that they would like more antibiotic prescribing teaching.

Theoretical model of antimicrobial prescribing behaviour

A theoretical model of APB in English hospitals is proposed (Figure 2‐B) which highlights factors underpinning APB, identified via the EFA and links them to key components of any behaviour 31. In this context, autonomy has been mapped onto capability as items in this factor relate to perceived competence to prescribe and review antimicrobial therapy independently. Antibiotic awareness has been mapped onto motivation as these items relate to doctors' awareness of AMR and antibiotic overuse and their perception that they must prescribe judiciously. Lastly, guidelines adherence has been mapped onto opportunity as following local guidelines enables doctors, particularly junior doctors and non‐infection specialists, to prescribe appropriately for the treatment of infection when they may not have the knowledge base to support this complex activity independently.

Figure 2.

Theoretical framework of antimicrobial prescribing behaviour. Parker & Mattick (2016) this paper

Discussion

This study aimed to quantify the perceived determinants of APB among doctors in English hospitals. A questionnaire was used to garner doctors' views on the variables influencing their antimicrobial prescribing and EFA revealed three underlying factors, autonomy, guidelines adherence and antibiotic awareness. These key perceived determinants were incorporated into a theoretical behavioural model which proposes links between the factors and behavioural components 31 and could be used to underpin the development of more effective stewardship interventions.

Doctors scored items within the guidelines adherence factor highly, suggesting a general acceptance of local guidelines, consistent with a study by Cortoos et al. 25. Foundation doctors' guidelines adherence scores were significantly higher than other grades. This reflects their positive attitude towards guidelines, preference for following them 15, 35 and perhaps also a lack of capability to prescribe antibiotics without the support of guidelines, or to recognize when patients fall ‘outside’ the scope of the empirical guidance.

Doctors also scored items in the antibiotic awareness factor highly, demonstrating a good level of appreciation for the problems of AMR, antibiotic overuse and personal responsibility for prescribing antibiotics prudently, consistent with a recent single centre American study 36. UK trained graduates scored higher than their non‐UK trained counterparts, suggesting overseas graduates may benefit from tailored stewardship training, although this finding should be interpreted cautiously as the numbers trained abroad were small. Antimicrobial awareness scores did not vary significantly by grade.

Participants scored lowest on the autonomy factor but with the highest level of inter‐participant variation. Autonomy scores significantly increased moving up the medical hierarchy, emphasizing the likely influence of seniors as decision‐makers 21, 22, 24, 25. However, decision‐making autonomy and the medical hierarchy may have less positive aspects, including divergence from evidence‐based guidelines and creating challenges for more junior doctors 15, 22. This study indicates that, as autonomy grows, prescribers are more confident to review and amend the prescriptions of others. Non‐interference seems to represent the antithesis of (decision‐making) autonomy, although the rationale may be bound in social etiquette 22, 37 and/or capability. Notably, high levels of autonomy and guidelines adherence are not mutually exclusive. However, the ‘buy‐in’ of senior clinicians when developing local guidelines is likely to be of key importance to support adherence 22. Then, autonomous professionals can choose to adhere to guidelines as their perceived best course of action and potentially dichotomous organizational and senior expectations will be aligned, lessening confusion for juniors who often feel ‘stuck‐in‐the‐middle’ 15.

Strengths and limitations

As with all studies, there are strengths and limitations to this research. The strengths include the rigorous questionnaire development process (literature review, expert consultation and piloting), the large sample size across multiple sites, high response rate and good participation across medical grades. The study involves theory‐driven choices made during the EFA process 38 and makes a contribution to theory through the theoretical model offered. It is important to note that the four data collection sites were in south‐west England, with medical trainees sometimes rotating between settings, so the barriers may be less challenging than between English hospitals in different regions.

Six main limitations were identified: 1) the self‐reported nature of the data is inherently vulnerable to recall bias, selective reporting, and/or participant error. These aspects were mitigated by the anonymous and voluntary nature of participation, the rigorous consultation process and the large sample size, but it is possible that participants cannot know certain influences on their prescribing. It is also not known how respondents interpreted ‘uncertain’ at the mid‐point on the Likert scale, 2) the voluntary participation in the study means that those more interested in antimicrobial stewardship issues and/or who were easier to access are possibly over‐represented in the sample. We are reassured to note that, when comparing our sample to national data, the proportion of female participants and non‐UK graduates is largely as expected (http://www.gmc‐uk.org/Chapter_1_SOMEP_2015.pdf_63501394.pdf), although foundation doctors and trainees are overrepresented because, in practice, there are more doctors on the specialist register (consultants) than foundation doctors and hospital trainees combined, 3) coordinators at the remote study sites did not always provide a precise denominator for the number of participants approached, meaning that we could not calculate an overall response rate, beyond saying that it was >74%, 4) the factors identified in the data were reliant on sufficient relevant items being present in the questionnaire, thus some smaller categories of influencing factors may have been omitted, 5) the three factors identified through EFA accounted for 32% of the variance, so a large proportion of variance remains unexplained, reflecting the complex nature of APB and 6) our application of a behaviour change framework to empirical data is offered as an example to promote discussion at the theory‐practice interface and to conceptualise recommendations for practice and future research but it only reflects one framework and one possible interpretation and should be considered in that light.

Recommendations for practice and future research

Hospital doctors should be viewed as a heterogeneous population and interventions should be tailored accordingly. Guidelines provide essential support for juniors and should be readily available. However, stewardship teams must engage stakeholders to improve alignment between the organization's guidelines and senior clinicians' preferences. Consultants and specialist registrars are usually proactive decision‐makers suggesting senior input is important to support timely antimicrobial review and de‐escalation in accordance with best practice 11, 39. Antimicrobial prescriptions should be reviewed daily by senior clinicians, who should share their decision making rationale with the juniors 15, to facilitate their development. Doctors who prescribe antimicrobials are likely to benefit from tailored antimicrobial teaching. For juniors this should focus on guidelines adherence and when to deviate, whereas consultant interactions should focus on local resistance epidemiology and aligning the evidence‐base, hospital guidelines and consultant preferences (Figure 2).

Previous research found greater differences between specialities/wards than between hospitals 15, suggesting successful interventions in one setting may be transferrable to similar wards in different hospitals. The finding that factor scores did not vary significantly between hospitals supports this view and stewardship teams may benefit from working collaboratively, to develop tailored interventions for implementation across multiple centres. The theoretical framework could be a starting point to facilitate this work (Figure 2), an under underutilized approach. Stewardship teams would need to define their target behaviour and group, consider what needs to change and which key factors should be targeted with intervention(s) to bring about the desired change 40. For example, when seeking to improve guidelines adherence among junior doctors it would be crucial to consider the factors impacting on their APB. Senior clinicians' preferences are likely to impact on juniors' capability/autonomy with regards guidelines adherence, so not only must juniors be aware of guidelines, and capable of using them, but they must have the support of seniors to comply.

Future users might improve the scale's parsimony by reducing the number of items per factor provided an acceptable level of reliability is retained. Further research is needed to explore whether additional perceived determinants of APB, for example risk aversion, could be quantified with the inclusion of additional items in the questionnaire; to explore how individuals' factor scores vary longitudinally; to assess for differences between doctors from different specialities; and to explore the impact of the wider multi‐disciplinary team.

In conclusion, hospital doctors perceive autonomy, guidelines adherence and antibiotic awareness to be important determinants of APB, which vary with experience and training. A theoretical framework is offered to facilitate the development of more effective, tailored stewardship interventions.

Competing Interests

All authors have completed the Unified Competing Interest form at http://www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare no support from any organization for the submitted work, no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years and no other relationships or activities that could appear to have influenced the submitted work.

We would like to gratefully acknowledge the important contributions of Dr Robert Tilley, Mr Antony Zorzi and Dr Tom Lewis who acted as site co‐ordinators, collecting survey data, Ms Esmita Charani for advice on questionnaire design, Dr Obioha Ukoumunne and Dr Roy Powell for statistical support and all the doctors who reviewed or completed the questionnaire. Furthermore, we thank the Imperial College London Infection Management Committee who provided feedback on the study design.

Supporting information

Table S1 Table showing how doctors responded to questionnaire items (n = 301)*

Supplementary Table

Parker, H. M. , and Mattick, K. (2016) The determinants of antimicrobial prescribing among hospital doctors in England: a framework to inform tailored stewardship interventions. Br J Clin Pharmacol, 82: 431–440. doi: 10.1111/bcp.12953.