Perceptions, views and practices regarding antibiotic prescribing and stewardship among hospital physicians in Jakarta, Indonesia: a questionnaire-based survey

Objectives Antibiotic overuse is one of the main drivers of antimicrobial resistance (AMR), especially in low-income and middle-income countries. This study aimed to understand the perceptions and views towards AMR, antibiotic prescribing practice and antimicrobial stewardship (AMS) among hospital physicians in Jakarta, Indonesia. Design Cross-sectional, self-administered questionnaire-based survey, with descriptive statistics, exploratory factor analysis (EFA) to identify distinct underlying constructs in the dataset, and multivariable linear regression of factor scores to analyse physician subgroups. Setting Six public and private acute-care hospitals in Jakarta in 2019. Participants 1007 of 1896 (53.1% response rate) antibiotic prescribing physicians. Results Physicians acknowledged the significance of AMR and contributing factors, rational antibiotic prescribing, and purpose and usefulness of AMS. However, this conflicted with reported suboptimal local hospital practices, such as room cleaning, hand hygiene and staff education, and views regarding antibiotic decision making. These included insufficiently applying AMS principles and utilising microbiology, lack of confidence in prescribing decisions and defensive prescribing due to pervasive diagnostic uncertainty, fear of patient deterioration or because patients insisted. EFA identified six latent factors (overall Crohnbach’s α=0.85): awareness of AMS activities; awareness of AMS purpose; views regarding rational antibiotic prescribing; confidence in antibiotic prescribing decisions; perception of AMR as a significant problem; and immediate actions to contain AMR. Factor scores differed across hospitals, departments, work experience and medical hierarchy. Conclusions AMS implementation in Indonesian hospitals is challenged by institutional, contextual and diagnostic vulnerabilities, resulting in externalising AMR instead of recognising it as a local problem. Appropriate recognition of the contextual determinants of antibiotic prescribing decision making will be critical to change physicians’ attitudes and develop context-specific AMS interventions.


INTRODUCTION
The global rise in drug-resistant infections is one of the leading threats to public health globally, with increasing rates of morbidity, mortality and escalating healthcare costs. 1 Misuse and overuse of antimicrobial drugs in human healthcare is one of the main drivers 2 3 and also represents a key solution, that is, judicious use of remaining antibiotics. Globally, use of antibiotics remains largely

STRENGTHS AND LIMITATIONS OF THIS STUDY
⇒ The self-developed questionnaire in this study identified a relevant set of attributes through a factor analysis optimisation process, with adequate content, face and construct validity and internal reliability. This study adds important value in the absence of adequately validated instruments regarding antimicrobial resistance and stewardship, with particular applicability for low-income and middle-income countries. ⇒ This study had a large, varied respondent sample and high response rate among physicians at six public and private hospitals in Jakarta, Indonesia, and identified differences between physicians across hospitals, departments, work experience and medical hierarchy, which can guide priority-setting and tailoring of stewardship interventions. ⇒ However, non-participation and the convenient hospital sample could have introduced selection bias, and the data are not necessarily representative for Jakarta or Indonesia. ⇒ Factor analysis is based on using a 'heuristic', which leaves room to more than one interpretation of the same data and cannot identify causality.
Open access unrestrained and poorly governed, with large, unregulated healthcare systems representing an increasingly challenging area for achieving the goal of optimisation. Substantial variations in contributing factors to inappropriate antibiotic prescribing exist across contexts, for example, diagnostic uncertainty, pressure from pharmaceutical industry or patients, 4 5 with the structure and funding of health systems inflecting enactment of optimisation strategies, including antimicrobial stewardship (AMS). 6 Patients infected with drug-resistant bacteria causing common bacterial infections in hospitalised populations are more likely to receive inappropriate antibiotic therapy resulting in higher mortality rates and opportunities for spreading to other patients. 7 8 Hospital AMS programmes aim to control antimicrobial use, and have been associated with reducing hospital-acquired infections, unnecessary healthcare costs and potentially drug-resistant infections. [9][10][11] However, AMS programmes in turn may jar with local constraints and practices and have been shown to have limited traction when attempts to implement occur without adequate understanding of context. 12 The global push to enact effective AMS requires detailed, context-specific data on physicians, given their central role in the complex process of antibiotic prescribing in hospitals, which can inform on how antimicrobial resistance (AMR) is conceived, how current prescribing is rationalised, and how broad AMS principles may be experienced across contexts and nations. 13 14 Few studies to date have been conducted on this topic in low-income and middle-income countries (LMICs), with insufficient evaluation of the psychometric properties of their measurement instruments to examine their suitability to the specific context. 4 5 Indonesia, a diverse lower-middle-income country in Southeast Asia with the world's fourth largest population (275 million), is particularly vulnerable for AMR, ranked among the greatest risers (29th of 76 countries) in antibiotic consumption, estimated at 2.5-fold between 2000 and 2015. 2 The implementation of the National Action Plan for AMR has been hindered due to, among other factors, a limited evidence base of AMR epidemiology, antibiotic utilisation and rational prescribing practices. 15 Weakly enforced antibiotic policies promote inappropriate prescribing and over-the-counter access without a prescription, and AMS is generally in an early stage of implementation. 15 16 Although nationwide representative AMR data are not available, anecdotal evidence suggested high rates of AMR among common Gram-negative bacteria.
To identify context-specific opportunities for AMS interventions, we conducted a questionnaire-based survey among antibiotic-prescribing physicians in hospitals in Jakarta, Indonesia, to evaluate their perceptions of AMR, antibiotic prescribing practice, and views on AMS, and explore differences between physician subgroups. The study also aimed to evaluate the construct validity and psychometric properties of the questionnaire, and explore differences between physician subgroups.

Study design and setting
We conducted a cross-sectional survey between March and August 2019 among all antimicrobial prescribing physicians at six acute-care hospitals in Jakarta, Indonesia, as part of a mixed-method study to identify targets for quality improvement in antibiotic prescribing practices (EXPLAIN study). 17 The participant hospitals were purposively sampled to achieve a representation of the diversity of hospital types in Jakarta, in terms of geographic location, size (small to large), health sector (public and private) and healthcare level (tertiary and secondary). In 2018, a total of 189 hospitals were operational in Jakarta province, of which 59 were public and 130 private, with a total of 15 444 and 15 425 inpatient beds, respectively. 18 The hospital sample was then pragmatically composed based on existing collaborations, site willingness and available study resources, and included two tertiary-care government hospitals and four secondary hospitals, three of which were private hospitals, with between 134 and 853 inpatient beds, situated across all five districts of Jakarta. At the time of the survey, all six hospitals had an AMS programme, although at various stages of AMS implementation, initiated between 2009 and 2018.

Study respondents
All qualified physicians prescribing antibiotics on a regular basis working were eligible for the survey and were approached to participate. The survey intended to include physicians across all clinical departments, work experience and professional hierarchy, to ensure representation of different antibiotic practices and views, perceptions and experiences related to AMR and AMS. 19 We included interns/internship doctors (magang/ dokter internsip; recent graduates working under supervision of a specialist), general practitioners (GPs) (dokter umum; physicians who practice general medicine treating common, non-specialised medical conditions), residents (residen; physicians who are in specialist training), specialist/consultant physicians (dokter spesialis/konsultan; specialised in a particular area of medicine) and other physicians.

Patient and public involvement statement
Patients or the public were not involved in the design, conduct or reporting of the research.

Survey questionnaire
We developed a two-page anonymous, self-administered, paper-based questionnaire, which was easy to complete and based on a conceptual framework that included attributes related to prescribers' perceptions, views and practices. Good practice recommendations for questionnaire design and existing questionnaires in the literature were Open access reviewed and discussed with several experts. The Clinician Pre/Post Perception Survey of the Greater New York Hospital Association United Hospital Fund 20 constituted the initial set of items, supplemented with relevant items from other existing questionnaires. [21][22][23][24][25][26][27] From a preliminary pool, we selected 69 items, of which eight items were worded in the negative to address the acquiescence effect. The instrument was translated from English to Indonesian, and back-translated by an independent translator. The questionnaire was pretested by a convenience panel of 18 physicians (2 GPs, 15 residents and 1 consultant). According to their feedback, we made adjustments to clarify ambiguous items, and remove redundant items. The final version consisted of 40 items and took about 10 min to complete.
The final questionnaire included an explanation of study purpose and completion instructions; 40 short statements (items) to which participants were asked to indicate the extent to which each reflected their own opinion on a 5-point Likert scale, divided into three sections: scope of the AMR problem and key contributors; antibiotic prescribing practices; AMS and respondent sociodemographics (online supplemental appendix).

Respondent recruitment
The hospital management provided the total number of prescribing physicians for each department. The questionnaires were delivered to the head of each unit who then distributed the survey to all eligible staff. The study coordinator kept a record of numbers of physicians approached and participated. On survey completion, respondents could enter a raffle to win one of three gift cards in each hospital (US$14 each); there were no other incentives for participation.

Statistical analysis
The percentage of respondents selecting each answer choice was calculated using the total number of responses as the denominator. For an exploratory factor analysis (EFA), a common lower bound for sample size is 10 cases per variable, suggesting a minimum sample size of 400; to allow for meaningful subgroup comparisons and minimise selection bias, we targeted a>50% response rate and a sample size of >1000 across the six hospitals.
We performed EFA to identify underlying distinct constructs, using factor, pcf command in Stata with orthogonal (varimax) rotation. For this analysis, the eight items worded in the negative were reverse-coded, and missing data for categorical variables were treated as a separate category. The Kaiser-Meyer-Olkin (KMO) was calculated to ensure EFA requirements were met. Each item was assigned to a certain factor based on the highest absolute factor loading of the rotated solution. To determine the optimal factor solution, we used the Kaiser criterion, scree plot of eigenvalues, and Horn's parallel analysis (paran package in Stata). Parallel analysis adjusts the original eigenvalues for sampling error induced collinearity among the variables to arrive at adjusted eigenvalues. Analogous to the Kaiser criterion, only factors with adjusted eigenvalues >1 were retained. The scree plot of eigenvalues and the scree plot of parallel analysis were visually inspected and factors were retained above the inflection. To interpret the factor solutions, we examined the factor loadings, and assigned each item to a certain factor based on the highest absolute factor loading (minimum loading 0.30), and then produced an umbrella label for each factor that best characterised the joint meaning of all the variables associated with it. If a certain item did not fit with the factor structure (eg, the results were not interpretable or a factor comprised less than three items), the analysis was rerun without the pertinent item(s). After the optimal factor solution had been achieved, we used the regression scoring method to compute the factor scores based on the factor loadings, and Cronbach's α coefficient to test internal reliability. Using the factor scores as the dependent variable, we developed six (one for each factor score) multivariable linear regression models to assess associations with key physician characteristics ('subgroups') as the following five independent variables, that is, hospital sector and care level, grouped departments, work experience and medical hierarchy. Given the hierarchical structure of the data (respondents clustered within hospitals) we applied multilevel (two-level) modelling to control for possible clustering within hospitals. We also adjusted for possible confounders. Values of p<0.05 were considered statistically significant. All analyses were done with Stata/IC v16.1 (StataCorp).

Respondent characteristics
All 1896 antibiotic prescribing physicians at the six hospitals were approached, and 1007 (53.1%) participated in the survey. Table 1 summarises the participants' key characteristics. Online supplemental table S1 summarises the response rates. Figure 1 and online supplemental table S2 summarise the responses to all 40 items. Open access according to hospital protocol after discharge of a patient with a multidrug-resistant organism (Q7); 56.9% (570/1002) thought adherence to hand-hygiene protocols to be excellent (Q8); and 22.5% (226/1005) felt that their hospital does not provide adequate staff education regarding multidrug-resistant organisms (Q9).

Exploratory factor analysis
The KMO was 0.8773 overall and >0.5 for all items, suggesting the data were suitable for EFA. Analysis of the scree plot (online supplemental figure S1) indicated a case for four factors, whereas the parallel analysis (online supplemental figure S2) indicated a case for seven factors. The four-factor solution yielded strong factors but explained only 39.9% of the variance and lacked a theoretical basis for one factor. The seven-factor solution contained one factor with only three items that was difficult to interpret; two of these items (Q9 and Q10) did not load well with any factor in various alternative factor solutions and were removed. Therefore, a six-factor model with a clear theoretical basis based on 38  Internal reliability was excellent for the overall 38-item scale (α=0.85) and factor 1 (α=0.8734) and 2 (α=0.8334), good for factor 3, 4 and 5 (α=0.70 each) and acceptable for factor 6 (α=0.57). Table 4 summarises the results of the subgroup analysis.

Hospitals
Statistically significant differences were identified between hospitals for awareness of AMS purposes (factor 2), views regarding rational antibiotic prescribing (factor 3), perception of AMR as a significant problem (factor 5) and immediate actions to contain AMR (factor 6), but not for awareness of AMS activities (factor 1) and confidence in antibiotic prescribing decisions (factor 4). None of the factor scores differed between prescribers in public vs private, or secondary vs tertiary hospitals.

Professional hierarchy
For awareness of AMS activities (factor 1), consultants, GPs and residents scored higher than interns. For awareness of AMS purposes (factor 2), consultants scored lower than interns and residents. For views regarding rational antibiotic prescribing (factor 3), consultants scored higher than GPs. For confidence in antibiotic prescribing  Overall 47.4% Table shows the results of the exploratory factor analysis (principal axis factoring) with orthogonal varimax rotation of the six-factor solution using the factor, pcf command in Stata. 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 0 indicate that the factor has a weak effect on the item. Item #9 and 10 were excluded from the analysis as explained in the Results section. Uniqueness: shows the proportion of the item's variance that is not explained by the factors Open access decisions (factor 4), consultants scored lower than residents, whereas for immediate actions to contain AMR (factor 6), consultants scored higher than residents. No differences were identified for perception of AMR as a significant problem (factor 5).

Departments
For awareness of AMS activities (factor 1) and purpose (factor 2), physicians in surgery and medicine scored higher than the acute specialties, whereas for awareness of AMS activities (factor 1), medicine scored higher than surgery. For views regarding rational antibiotic prescribing (factor 3), surgery scored higher than acute specialties. For confidence in antibiotic prescribing decisions (factor 4), medicine scored lower than surgery and other specialties. For perception of AMR as a significant problem (factor 5), surgery scored lower than medicine. For immediate actions to contain AMR (factor 6), surgery scored higher than medicine and other specialties and the acute specialties scored higher than surgery and medicine.

Work experience
Physicians with little (0-5 years) work experience scored lower than more experienced colleagues, for awareness of AMS purpose (factor 2); for perception of AMR as a significant problem (factor 5); and for immediate actions to contain AMR (factor 6). No differences were identified for factors 1, 3 and 4.

DISCUSSION
This survey, among over 1000 physicians in Indonesian hospitals, assessed their perceptions and views regarding the AMR problem and its key contributors, antibiotic prescribing practices and AMS. Starting from those three broad themes, we used EFA to identify distinct underlying constructs in the dataset, that is, (1) awareness of AMS activities; (2) awareness of AMS purposes; (3) views regarding rational antibiotic prescribing; (4) confidence in antibiotic prescribing decisions; (5) perception of AMR as a significant problem and (6) immediate actions to contain AMR. The survey findings thus outline a series of dynamics around AMR and AMS in the Indonesian context. Spanning issues around awareness (education) 28 (factors 1, 2, 5), visibility (diagnostics) 29 (factor 3, 4 and 6), and institutional form (governance) 30 (factors 3 and 6), the survey results tease out many of the core issues illustrated in other settings, 4 5 but in turn, illustrate that several specific factors play a role, that is, uncertainty, risk and lack of sense of responsibility.
In this respect, an important finding was that only few physicians recognised that antibiotic overuse was an issue at their own hospital, that many physicians were hesitant to stop antibiotics without an appropriate indication that others had prescribed, and that many felt that antibiotic restrictions impaired their ability to provide good patient care. Lack of confidence in prescribing decisions and defensive prescribing were common due to diagnostic uncertainty, fear of patient deterioration or complications, or because patients or their relatives insisted. These views and perceptions concur with the observed overuse of broad-spectrum antibiotics, the underuse of diagnostic stewardship to guide antibiotic decisions and poor guideline compliance of antibiotic prescribing in the same hospitals, described in our recent paper, which identified several priority areas for stewardship. 17 Based on the survey data, we considered that an important obstacle that hinders effective AMS implementation in this context could be linked to perceived 'externality of AMR' as a problem. 31 That is, physicians acknowledge its significance but do not take ownership or responsibility, thus reflecting a production of AMR as an externality, for example, a result of irrational use elsewhere in communities or other hospitals. The lack of systematic surveillance of AMR and antibiotic use and the underutilisation of bacterial cultures, recognised by many of the respondents, also reinforces the perception of AMR as a 'problem of elsewhere'. This feeds a lack of engaging with AMS, as physicians do not recognise it as a value-add for an already stretched institutional context, and in turn provides the context for continued defensive prescribing 'to be on the safe side'. Moreover, defensive prescribing practices may somewhat offset (in the short term) problems around room cleaning, hand hygiene and    Open access staff education. In this way, AMR as an externality and the vulnerabilities of the institution, offer an environment conducive to the ongoing overuse of antimicrobials. 32 33 The higher incidence of hospital-acquired infections in LMICs than in high-income countries could further promote defensive prescribing as a way to compensate for substandard IPC practices. 34 All in all, this supports the notion that physicians tend to prioritise managing immediate clinical risks, reputation and concordance with peer practice, vis-à-vis the long-term population consequences of AMR. 35 The identified differences between hospitals regarding awareness of AMS purposes, views regarding rational antibiotic prescribing, perception of AMR as a significant problem and immediate actions to contain AMR could not be attributed or related to the pertinent hospital sector (public vs private) or healthcare level (secondary vs tertiary), but they were found to be associated with several differences between the physician subgroups. First of all, work experience and medical hierarchy were found to influence the awareness of AMS purpose, AMR as a significant problem, and immediate actions to contain AMR. Interestingly, compared with junior physicians, specialists/consultants expressed lower confidence to make antibiotic decisions in uncertain situations while showing higher confidence in actions to contain AMR. Possible explanations include that specialists/consultants have a better recognition of the 'unknowns' (eg, lack of data on bacterial susceptibility patterns) and that they bear final patient responsibility, introducing the fear of losing a patient or legal consequences, 36 whereas taking actions to curb AMR can be a remedy to compensate their fear. Conversely, residents' higher confidence in antibiotic prescribing may also relate to their contemporary medical training, which promotes the use of evidencebased prescribing guidelines, as opposed to late-career physicians. 37 GPs had low scores on views regarding rational antibiotic prescribing compared with consultants which could reflect the GPs' limited responsibility in the antibiotic decision-making hierarchy, possibly leading to a lack of positive attitude towards guidelines and preference for complying with them. 38 39 Furthermore, the acute specialties (including emergency, Intensive Care Unit (ICU) and anaesthesiology) had lower awareness of AMS activities and purposes, compared with surgery or medicine, but scored higher for immediate actions to contain AMR. Compared with surgeons, physicians in medicine had a greater awareness of AMS activities and recognition of AMR as a significant problem, but they had lower confidence in antibiotic prescribing decisions and immediate actions to contain AMR. These observations are in line with a UK study that found that emergency physicians experienced pressure for immediate action out of fear of losing a patient and a lack of ownership of antibiotic decision making due to the patient transitioning to inpatient care, that medical doctors adopted a more policy-informed, interdisciplinary approach and that senior surgeons left complex antibiotic decisions to junior staff, resulting in potential defensive and inappropriate antibiotic use. 19 Variations in the social norms, values and behaviours between specialties should inform what is the best approach to antibiotic decision making.
Our study had several strengths and weaknesses. First, we were able to identify a relevant set of attributes through a factor analysis optimisation process, with adequate content, face and construct validity and internal reliability. In the absence of adequately validated instruments regarding AMR and AMS, 40 this study adds important value to the field, with particular relevance and applicability for LMICs. Nonetheless, further questionnaire validation steps (such as criterion-related validity) are necessary to achieve a fully valid and reliable instrument. Second, the study had a large, varied respondent sample and high response rate. However, non-participation and the purposive and limited hospital sample could have introduced selection bias, and the data are not necessarily representative for Jakarta or Indonesia at large. The authenticity of the answers was maximised by protecting the respondents' identities, although reliance on selfreport has potential for social desirability bias. Third, this study focused on hospital settings. It is, however, acknowledged that the problem of AMR extends beyond hospital settings to the wider community through the primary healthcare system as well as informal providers, which warrants further study in the Indonesian context. 41 Fourth, factor analysis is based on using a 'heuristic', which leaves room to more than one interpretation of the same data and cannot identify causality.

CONCLUSION
AMS implementation in Indonesian hospitals is likely highly dependent on institutional, contextual and diagnostic vulnerabilities. These may result in the problem of AMR being externalised, instead of recognised as a local hospital problem. Current AMS strategies may be insufficiently successful in promoting prudent antibiotic use, due to lack of systematic engagement with and feedback to prescribers, aimed at building confidence in antibiotic decision making and ownership of the AMR problem. Appropriate recognition of the contextual and social determinants of antibiotic prescribing decision making, including hospital factors, dynamics in medical hierarchy and experience, among others, will be critical to design context-specific AMS interventions that are adopted by healthcare professionals and successfully influence behaviours. 14