Antimicrobial resistance (AMR) is a natural phenomenon [1], to which overuse and misuse of antibiotics contribute and augment [1-4]. Globally, antibiotic consumption has increased (eg, by 65% from 2000 to 2015), and 30% to 50% of antibiotic prescriptions were used either inappropriately or unnecessarily [2], further resulting in increased inappropriate use [3,4] and the development of selective pressure on antibiotics [5-7]. Inappropriate prescribing is a key contributing factor to the emergence of AMR [1,8,9] and varies from 62.8% for respiratory tract infections to 78.5% in patients with skin and soft tissue infections [4]. This would strengthen the belief that antibiotics ought to be prescribed and are effective in circumstances when they are not [10]. Physicians’ prescribing behaviors impact not only patient health but also medical expenses and health resources [11]. It is recommended to monitor antibiotic prescribing in hospitals to improve the quality of antibiotic prescribing through education and practice changes [12]. Identifying key behaviors and drivers for the behaviors that may be amenable to change and improve prescribing decisions is an important component of interventions in health care practice to mitigate the burden of AMR [1,10,11]. Antibiotic stewardship programs (ASPs), which are among the most common interventions in health facilities to optimize antibiotic use, are effective, low-cost methods to change behaviors that drive excessive prescribing of antibiotics in health facilities [1].

Human behavior is guided by beliefs about the likely consequence of the behavior (behavioral beliefs), beliefs about the normative expectations of others (normative beliefs), beliefs about the presence of factors that may facilitate or impede the performance of the behavior (control beliefs), shaping attitudes, subjective norms (SNs), and perceived behavioral control (PBC) [13]. It is reported that these behavioral beliefs (attitudes, PBC, and SN) of physicians are predictors of indiscriminate antibiotic prescribing behaviors in hospitals [9,14]; thus, campaigns that address both health service personnel and the general population should take this into account [8]. A high level of knowledge is known to be associated with a more positive attitude and behavioral intention for reducing antibiotic prescriptions and was linked with less complacency, less fear, and less ignorance, although it had indirect effects on intentions to prescribe antibiotics through the attitude of ignorance [14]. On the other hand, perceived higher patient pressure negatively affects attitudes toward the rational use of antibiotics and promotes higher use of antibiotics [15]. Thus, characterizing and designing behavior change interventions based on the behavior change wheel model and theory of planned behavior (TPB) serve as a framework for modeling the antibiotic prescribing behaviors of physicians [13,16,17]. Optimizing antibiotic consumption and reducing the rate of AMR are currently global issues [1,18]. In low- and middle-income countries, the prescribing of antibiotics is highly influenced by inadequate diagnostic facilities, lack of guidelines, difficulty monitoring patient progress, poor intensive care facilities, patient demand for quick relief, perceived patient expectations from past prescriptions, and fear of losing patients to competition [19,20]. This results in high mortality and morbidity due to inadequate regulation, limited access to diagnostic facilities, and antimicrobial over-prescription [21,22]. Based on the behavior change wheel, once a problem is identified and context is considered, functions and policies may be implemented as interventions to understand and change prescribing behavior and improve antibiotic consumption [13,17]. This requires the design and implementation of sustained awareness campaigns to change behaviors and improve health outcomes [9].

In sub-Saharan Africa, physicians still prescribe antibiotics based only on a simple assessment of patients’ symptoms, just as they used to when antibiotics first became commonly used in the 1950s [9], due to a lack of diagnostic and antibiotic susceptibility tests, resulting in up to 95% of antibiotic prescriptions as unnecessary [23]. Prescribing antibiotics requires balancing physician, patient, and facility-related factors [24]. In Ethiopia, antibiotic prescribing in hospitals may account for 52.39% of all prescriptions [25], and one-half of prescribed antibiotics might not be needed [26]. Although behavior change campaigns can be very cost-effective for changing antibiotic prescribing practices, based on identified gaps [9], in Ethiopia, to our knowledge, there have been no studies to model the antibiotic prescribing behavior of physicians other than determining the perceptions of health professionals on AMR and antibiotic use [27,28]. Modeling behavior is needed to help clinical leaders drive ASP and design educational programs to help standardize and improve antibiotic prescribing behaviors in health facilities [29]. Thus, this study assessed the determinants of antibiotic prescribing behavior among physicians serving in outpatient departments (OPDs) in hospitals in northwest Ethiopia using a structural equation modeling (SEM) approach.

Among 200 planned respondents, 185 completed the questionnaires. Thus, the overall response rate was 92.5%.

AMR is a global crisis [27,29-32], calling for urgent action to resolve it. One of the important strategies for combating AMR is improving antibiotic prescribing practices by determining factors that affect physicians’ prescribing behaviors [10,11,14]. Physicians’ antibiotic prescribing practices are influenced by physician-related factors (knowledge, expertise, specific prescription patterns, time constraints, and communication with patients), patient-related factors (preferences, expectations, knowledge, culture, economic status, and previous experience), and health system–related factors (guidelines, policies, regulations, and financial incentives) [24]. Thus, the TPB in its original or modified (MTPB) form provides a theoretical framework to identify the determinants of physicians’ antibiotic prescribing behaviors [14,15,33]. This study, using SEM, confirmed the theoretical framework for physicians’ antibiotic prescribing behaviors with some modifications. The MTPB model revealed that physician knowledge was associated with PBC and SN (P<.001). However, attitudes, SNs, and PBC did not influence intentions around prescriptions and perceived prescribing behaviors of physicians. Overall, the study revealed that physicians’ perceived antibiotic prescribing behaviors were not affected by intentions to reduce and prescribe antibiotics.

A qualitative study in Ethiopia uncovered that “junior physicians (interns and residents) are more likely to prescribe broad-spectrum antibiotics, and they further speculate these practices are driven by poor knowledge” [27]. In this, 121 (65.4%) of the 185 physicians had good knowledge (answered 5 questions out of 11), although the respondents answered only 44.13% of the total questions about antibiotic prescriptions correctly. This was relatively low compared with the 55% to 86% for physicians in hospitals in China [14,15], Lao People's Democratic Republic, Democratic Republic of Congo, and Peru [34-36]. In this study, MTPB knowledge was directly linked with the PBC and SNs of physicians (P<.05). This was different than the findings of a study in China that reported a link between high knowledge and positive attitudes toward the rational use of antibiotics [14,15] but similar to a study that reported a link between high knowledge and decreased SNs to prescribe antibiotics [14]. This study revealed a lack of indirect links between knowledge and antibiotic prescribing behavior, which was similar to a study in China that reported a lack of a link between knowledge and antibiotic prescribing practices [14,15]. Knowledge helps physicians weigh the treatment options and increases the accuracy of risk perception; thus, physicians with different professional titles (such as resident physicians and general practitioners) and length of practice may have different attitudes toward antibiotics, based on findings in previous studies [37,38]. Although training can contribute to knowledge acquisition, this study revealed a lack of difference in knowledge between those who completed training on antibiotic prescribing and AMR and those who did not.

This study also revealed a lack of a link between attitudes, SNs, and PBC to prescribe antibiotics. This was different from a report in China that confirmed that “intentions to prescribe antibiotics are predicted by the attitudes, subjective norms, and sense of behavioral control of the prescribers” [33], although another study reported a lack of relationship with intentions to prescribe antibiotics [14,15]. It was reported that attitudes, SNs, and PBC were predictors of antibiotic prescribing behaviors [11,39]. However, this study did not show indirect relationships between attitudes, SNs, and PBC and prescribers’ perceived antibiotic prescribing behaviors.

Of the patients seen in the OPDs weekly, 54.8% were estimated to receive at least one antibiotic, and of the 1850 estimated patients who presented with a URTI, 916 (49.5%) were estimated to be prescribed at least one antibiotic. In Ethiopia, antibiotic prescribing ranges from 56.0% in primary health care facilities [40] to 73.7% in inpatient wards in the national referral hospital [41]. The majority of the prescriptions (32.9%-39.3%) were for respiratory tract infections, although about 54.2% of all antibiotic prescriptions might not be needed [27]. This could explain the high rates of perceived antibiotic prescriptions for outpatient visits and patients with URTIs in this study. In ambulatory care facilities in Tanzania, 95% to 96.3% of presenting cases were receiving at least one antibiotic [42], which is higher than the estimated prescriptions among weekly visits and patients with URTIs in this study. Another study in the same hospital reported 66.9% of patients were treated with or prescribed at least one antibiotic among weekly visits [28], which was higher than the perceived weekly prescriptions and prescriptions for URTIs in this study. A study in China reported that physicians prescribed antibiotics to an estimated 40% of patients with URTIs [14], and actual antibiotic prescribing behavior was 44.3% [33], both of which are lower than the rates in this study.

In this study, intention to reduce antibiotic use and intention to prescribe antibiotics were not linked with perceived antibiotic prescribing behaviors. Similar findings from China supported the limited role of intentions on antibiotic prescribing behavior [33]. Intraphysician prescription variability is affected by the availability of clinical guidelines, experience, peer prescribing practices, pharmaceutical pressure, time pressure, financial considerations, individual practice patterns, practice volume, and relationships with patients [24]. Differences in these factors might explain the discrepancy in the findings from this study and those from a study in China [33], which reported that a positive attitude toward antibiotics resulted in a higher intention to prescribe antibiotics. The difference might be due to differences in the type of patients, clinical practice, and availability of structural and process controls in antibiotic prescribing. Two studies, one in Eritrea and one in Ethiopia, confirmed this by reporting that patient age (<18 years), gender (male), and the number of drugs in a prescription (≥2) were associated with the prescription of antibiotics [43,44]. Thus, physicians in different health systems may be subjected to different working environments and social pressures, which could affect their overall intention to prescribe and prescribing behavior.

In general, this study uncovered a lack of volitional control among physicians when prescribing antibiotics. This indicates the complex nature of antibiotic prescribing practices, which are influenced by various factors [24]. Thus, a campaign is needed to reduce over-prescriptions of antibiotics using a systems approach to addressing gaps in the knowledge and attitudes of prescribers [33]. Introducing educational programs and training on antibiotic prescribing practices and antibiotic resistance, preparing targeted guidelines to address gaps in antibiotic prescriptions for URTIs, and involving specialists will help address the gaps in antibiotic prescribing [24,33,45]. Providing antibiotics as a universal therapy due to gaps in knowledge and skills and financial or reputational incentives on the one hand and a lack of antibiotic and poor facility regulations, the absence of a regulatory framework, and poor implementation of existing policies on the other hand might be drivers of inappropriate antibiotic prescriptions [46]. Thus, disease-specific prescribing guidelines like those for URTIs can facilitate the translation of intentions into practices, since the inability to make a clear diagnosis and over-prescription of antibiotics may be linked with physicians’ clinical capacity rather than behavioral control intentions [24]. Despite this, self-regulation, outcome expectation, and anticipation of possible barriers may still have a considerable effect on prescribing behaviors and will help reduce inappropriate antibiotic use, since inappropriate prescribing of antibiotics in ambulatory care is known to be linked to current knowledge on antibiotics, ASPs, and AMR among prescribers [42].

Therefore, ASPs must have fiduciary responsibility for all health care institutions across the continuum of care [47]. A comprehensive approach through a hospital policy on the rational use of antibiotics is essential to developing and implementing an evidence-based antibiotic use policy and standard treatment guidelines for common infectious diseases, improving antimicrobial prescribing through educational and administrative means, and monitoring and providing feedback regarding antibiotic resistance, all of which are strategic approaches [48]. Knowledge of determinants that influence antibiotic prescription behavior is essential for the successful implementation of antimicrobial stewardship interventions [49]. In Ethiopia, at present, there are limited or no national or coordinated legislative or regulatory mandates designed to optimize the use of antimicrobial therapy through ASPs [6,26]. This research on behavioral determinants may have a substantial impact on designing policies on antibiotic prescribing behaviors and implementing effective, efficient, and evidence-based interventions. It urges strengthening efforts to improve prevention and control efforts for infectious diseases, including the adoption of ASPs in all health care facilities. Research is also needed to define the optimal elements and goals of ASPs in different health care settings; expand educational efforts on ASP; devise novel mechanisms to prevent the over-prescription of antibiotics; and implement rapid, point-of-care diagnostic tests that would enable appropriate prescription and care.

Overall, this study will help understand prescribing behaviors by applying the TPB model and will be helpful for regulating prescribing behaviors, improving clinical management, promoting physician-patient communication, and establishing a harmonious physician-patient relationship to improve rational prescribing behaviors in delivering high-quality medical services. Policymakers should also consider multiple scenarios rather than merely concentrating on creating awareness. The model can hopefully be incorporated as part of multilevel interventions designed to decrease irrational prescriptions for actual patients. Furthermore, it might help initiate a nationwide survey on factors affecting antibiotic prescribing behaviors, and more research is needed to explore the views of other stakeholders on antibiotic use. Although this study presents opportunities for future studies in the country, it does have its limitations. The sample size was relatively small, and self-administered questionnaires may not provide the possibility for respondents to verify their answers, resulting in socially desirable answers. Due to a lack of records for antibiotic prescriptions, it was not possible to determine the actual antibiotic prescribing behaviors of physicians. Another limitation might be that it did not include all comprehensive physician-related factors (expertise, knowledge, specific prescription patterns, time constraints, and communication with patients), patient-related factors (knowledge, preferences, expectations, culture, economic status, and previous experience), and health system–related factors (financial incentives, guidelines, policies, and regulations) that influence antibiotic prescribing practices. Thus, further studies using TPB with a large number of physicians are warranted.

In conclusion, there was a high level of estimated antibiotic prescriptions for URTIs and weekly outpatient visits. However, the perceived behaviors around antibiotic prescription were not affected by the intention to prescribe antibiotics or the intention to reduce antibiotic use. Although the physicians had a good level of knowledge about antibiotics, antibiotic resistance, and prescriptions, which were linked with the attitudes and SNs of physicians, intentions to reduce use and prescribe antibiotics were not significantly associated with attitudes, SNs, or PBC. This may show the complex nature of antibiotic prescriptions, which cannot be justified simply by intentions and behaviors of physicians, as determined based on TPB.