Perceptions, views and practices regarding antibiotic prescribing and stewardship among hospital physicians in Jakarta, Indonesia

Objectives: Antibiotic overuse is one of the main drivers of antimicrobial resistance (AMR), especially in low and middle-income countries. This study aimed to gain an understanding of perceptions, views, and practices regarding AMR, antibiotic prescribing, and 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 general hospitals in Jakarta in 2019. Participants: 1007 of 1896 (53.1% response rate) antibiotic prescribing physicians. Results: EFA identified six latent factors (overall Crohnbach alpha=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. 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. Physicians 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 health care is one of the main drivers 2,3 and also represents a key solution, i.e. judicious use of remaining antibiotics. Globally, use of antibiotics remains largely unrestrained and poorly governed, with large, unregulated healthcare systems representing an increasingly challenging area for achieving the goal of optimization. Substantial variations in contributing factors to inappropriate antibiotic prescribing exist across contexts, e.g. diagnostic uncertainty, pressure from pharmaceutical industry or patients 4,5 , with the structure and funding of health systems inflecting enactment of optimization strategies, including antimicrobial stewardship (AMS) 6 .
AMS programs aim to control antimicrobial use, and have been associated with reducing hospital-acquired infections, unnecessary healthcare costs, and potentially drug-resistant infections [7][8][9] . 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 10 .
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 AMR is conceived, how current prescribing is rationalised, and how broad AMS principles may be experienced across contexts and nations 11,12 . Few studies to date have been conducted on this topic in low and middleincome countries (LMIC), with insufficient evaluation of the psychometric properties of their measurement instruments to examine their suitability to the specific context 4,5 . Indonesia, a diverse middle-income country in Southeast Asia with the world's fourth largest population (275 million), is particularly vulnerable for AMR, driven by dense urban populations combined with rising antibiotic consumption 13 , a decentralised health system 14 , and weakly enforced antibiotic policies 15 , hence promoting inappropriate prescribing and over-thecounter access without a prescription. Despite progress in government policies, AMS is generally in an early stage of implementation 15,16 .
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, accounts of antibiotic prescribing, and views on AMS. We performed exploratory factor analysis (EFA) to evaluate the construct . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; https://doi.org/10.1101/2021.09.05.21263144 doi: medRxiv preprint validity and psychometric properties of the questionnaire, identify distinct underlying constructs in the data, 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 public and private general 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 hospitals included two tertiary-care government hospitals and four secondary hospitals, three of which were private hospitals. At the time of the survey, all six hospitals had an AMS programme, albeit at an early stage of implementation.

Ethical considerations
The research ethics committee of the Faculty of Medicine University of Indonesia (1364/UN2.F1/ETIK/2018) and the Oxford Tropical Research Ethics Committee (559-18) approved the study, with additional permission from hospital management. As the survey was anonymous, participant consent was inferred when the doctor completed and returned the questionnaire, as explained in the survey introduction.

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 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; https://doi.org/10.1101/2021.09.05.21263144 doi: medRxiv preprint address the acquiescence effect. The instrument was translated from English to Indonesian, and back-translated by an independent translator. The questionnaire was pre-tested by a convenience panel of 18 physicians (2 GPs, 15 residents, and 1 consultant), and adjustments were made in accordance with their feedback, reducing the number of items to 40. The final version took about 10 minutes 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 3 sections: scope of the AMR problem and key contributors; antibiotic prescribing practices; AMS; and respondent socio-demographics (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. Upon 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.

Ethical approval
The research ethics committee of the Faculty of Medicine University of Indonesia (1364/UN2.F1/ETIK/2018) and the Oxford Tropical Research Ethics Committee (559-18) approved the study, with additional permission from hospital management. As the survey was anonymous, consent was inferred when the participant completed and returned the questionnaire.

Statistical analysis
The percentage of respondents selecting each answer choice was calculated using the total number of responses as the denominator. For an 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 minimize 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 reversecoded, 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 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; https://doi.org/10.1101/2021.09.05.21263144 doi: medRxiv preprint solution, and we then assigned an umbrella term to each factor. After the optimal factor solution had been achieved, we calculated factor scores using the regression scoring method, and Cronbach's α coefficient to test internal reliability. Using the factor scores (dependent variable), multivariable mixed-effects linear regression was used to analyze physician subgroups (independent variables i.e., hospital sector and care level, grouped departments, work experience and medical hierarchy), while adjusting for possible clustering within hospitals as well as relevant confounders. P-values<0.05 were considered statistically significant. All analyses were done with Stata/IC Version 16.1 (StataCorp, College Station, TX, US).
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; https://doi.org/10.1101/2021.09.05.21263144 doi: medRxiv preprint

Respondent characteristics
All 1896 antibiotic prescribing physicians at the six hospitals were approached, and 1007 (53.1%) participated in the survey. Table 1 summarizes the participants' key characteristics. Table S1 summarizes the response rates.

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 ( Figure S1) indicated a case for four factors, whereas the parallel analysis ( 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 items was deemed most suitable, explaining 47.4% of the variance, with KMO 0.8802 overall and >0.5 for each item ( Table 2). The six latent factors are ( Table 3, see Table S2 for details): 1) Awareness of AMS activities; 2) Awareness of AMS purpose; 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. 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). Figure 1 and Table S3 summarise the responses to all 40 items. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  Table 4 summarizes the results of the subgroup analysis.

Physician 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 versus private, or secondary versus 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 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. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; https://doi.org/10.1101/2021.09.05.21263144 doi: medRxiv preprint

Discussion
This survey assessed the perceptions, views and practices regarding AMR, antibiotic prescribing and AMS among over 1000 physicians in Indonesian hospitals. Through an exploratory factor analysis we identified six distinct constructs in the dataset, i.e., 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 outline a series of dynamics around AMR and AMS in the Indonesian context. Spanning issues around visibility (diagnostics) 26 , awareness (education) 27 and institutional form (governance) 28 , the survey results tease out many of the core issues illustrated in other settings 4,5 , but in turn, illustrate a specific mix of variables at play, i.e. uncertainty, risk, and lack of sense of responsibility. For instance, only about one-third of physicians recognised that antibiotic overuse was an issue at their own hospital, many physicians were hesitant to stop antibiotics that others had prescribed in the absence of an appropriate indication, and 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. The study findings expand on our recently published paper on the patterns and quality indicators of antibiotic prescribing in the same hospitals, which identified several priority areas for stewardship 17 .
The most significant factor in guiding the future agenda in Indonesia around effective AMS implementation, is the perceived "externality of AMR" as a problem 29 . That is, physicians acknowledge its significance but do not take ownership or responsibility, thus reflecting a production of AMR as an externality, e.g. 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 reproduces the perception of AMR as a "problem of elsewhere". This feeds a lack of engaging with AMS, as it is not recognised 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 somewhat offset (in the short term) problems around room cleaning, hand hygiene and staff education. In this way, AMR as an externality and the vulnerabilities of the institution, offer an environment conducive to the ongoing over-use of antimicrobials 30,31 . The higher incidence of hospital-acquired infections in LMIC than in high-income countries could further promote defensive prescribing as a way to compensate for substandard IPC practices 32 . 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 33 . is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; https://doi.org/10.1101/2021.09.05.21263144 doi: medRxiv preprint 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" (e.g. lack of data on bacterial susceptibility patterns) and that they bear final patient responsibility, introducing the fear of losing a patient or legal consequences 34 , 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 includes AMS, as opposed to late-career physicians 35 . 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 36,37 .
The acute specialties (including emergency, 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 38 . 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, based on our self-developed questionnaire, we were able to identify a relevant set of attributes through a factor analysis optimization process, with adequate content, face and construct validity and internal reliability. In the absence of adequately validated instruments regarding AMR and AMS 39 , this study adds important value to the field, with particular relevance and applicability for LMIC. 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 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; https://doi.org/10.1101/2021.09.05.21263144 doi: medRxiv preprint convenient 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 self-report has potential for social desirability bias. Third, 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 12 .
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 10, 2021. ; 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. Uniqueness: shows the proportion of the item's variance that is not explained by the factors. Item #9 and 10 were excluded from the analysis as explained in Results.  Abbreviations: GP, general practitioner. The table summarizes the results of the multivariable mixed-effect linear regression models to assess the associations between independent variables and each of the factor scores (dependent variable), while adjusting for the possible interdependence of observations clustered within hospitals as well as confounders (sex, AMS training). * This model did not include hospital as an independent variable due to collinearity.  Table S3.
Data are reported as n (%). Doctor type was missing for 8 respondents, across a hospitals (02 (2), 03 (3), and 05 (3)); b sector (public 8, private 0); c level of care (secondary 3, tertiary 5); and d departments (obstetrics/gynaecology 2; ENT 1; paediatrics 1; multiple units 1; dental/oral surgery 2; missing dept 1). e Invitations to participate in the survey were sent out according to primary work units; the discrepancies reported for the acute and other departments result from the fact that n=177 respondents who worked across multiple units (mostly GPs and internship doctors) indicated a different primary department on their questionnaire.  Table 3 Item #9 and 10 were excluded from the analysis, as explained in Results.  Figure 1. Parallel analysis adjusted the original eigenvalues for sampling error-induced collinearity among the variables to arrive at the adjusted eigenvalues.