Antimicrobial resistance in the Northern Territory, Australia: insights from a systems thinking approach to human health

Abstract

Background

Antimicrobial resistance (AMR) is a complex global health threat, contributing to an estimated 4.95 million deaths worldwide with disproportionate impacts across regions. In Australia, the burden varies geographically with up to 50% of infections resistant to antibiotics in northern Australia. This qualitative study engaged stakeholders across One Health sectors to explore the dynamics influencing AMR in the Northern Territory.

Methods

Thirty-five participants across human, animal, environmental and cross-sectoral systems were consulted through interviews and focus groups. Thematic analysis informed the development of a causal loop diagram, grounded in systems thinking, to illustrate complex interactions among factors driving AMR. This paper focuses on key factors and feedback loops influencing AMR in human health.

Results

The causal loop diagram revealed dynamic, interconnected factors driving AMR through reinforcing and balancing feedback loops. Key drivers included antibiotic prescribing and use across human and animal health, as well as antimicrobial residues in the environment. Crucially, broader structural determinants, such as geographical isolation, health system limitations and social inequities, also shaped AMR dynamics. Consequently, reducing antibiotic use alone is unlikely to be sufficient to control AMR. While antimicrobial stewardship and infection prevention and control remain critical, additional leverage points for intervention were identified. These include culturally safe care, healthcare workforce training and retention, improved AMR awareness among professionals and communities, and investment in upstream determinants of health, such as functional housing and access to essential infrastructure.

Conclusions

In regions with a high infectious disease burden, addressing AMR requires a shift from reactive, clinical models to proactive, community-based strategies. Given its systemic complexity, AMR and similar health challenges cannot be addressed by health systems alone or through siloed interventions. Coordinated, multi-sectoral and systems-informed approaches are essential to achieve sustainable impact.

Supplementary Information

The online version contains supplementary material available at 10.1186/s42522-026-00222-7.

Introduction

Antimicrobial resistance (AMR) is a critical global health concern, contributing to approximately 4.95 million deaths worldwide in 2019 [1] with significant regional disparities. By 2050, annual global deaths linked to AMR are projected to reach 8.22 million [2]. In Australia, the burden is disproportionately high in regional and remote areas of northern Australia, where the risk of infectious diseases is three times greater than in urban areas and up to 50% of Staphylococcus aureus infections are methicillin-resistant [3, 4]. In these settings, infectious diseases not only strain healthcare systems but also undermine community wellbeing and socioeconomic stability.

Although AMR is often framed within a clinical context, the emergence and spread of resistant pathogens extend well beyond human health, reflecting a complex network of interconnected drivers across the One Health spectrum [5, 6]. Despite this complexity, AMR is frequently addressed within disciplinary silos, limiting our capacity to understand cross-sectoral dynamics and develop coordinated responses [5]. Gaps also remain in understanding the complex interrelationships and dynamic interactions that underpin AMR across One Health sectors, thereby hindering the development of effective and sustainable interventions [5, 7].

Systems thinking provides a valuable framework for understanding how factors within complex systems interact [8]. It offers analytical tools to characterise system structure, feedback mechanisms and non-linear behaviour, supporting the design of sustainable interventions for complex challenges [9]. This approach is particularly well suited to addressing the multifaceted drivers of AMR across One Health sectors and identifying leverage points for systemic change [5, 10]. Interest in applying systems thinking to AMR has increased in recent years [7]. Examples include systems mapping to develop generic AMR system representations [11], causal loop mapping of AMR across One Health in a German region [12] and in food systems in Europe and Southeast Asia [13, 14], and more recent system dynamics modelling of antimicrobial use and resistance [15]. However, despite this growing interest, the application of systems thinking remains limited, particularly in capturing complex feedback mechanisms across One Health sectors, highlighting the need for further research to support evidence-based decision-making.

Our study applies causal loop modelling, a systems thinking tool, to investigate the interconnected factors driving AMR in the Northern Territory (NT) of Australia. The region experiences a high burden of chronic conditions [16], elevated infectious disease rates [17] and exceedingly high levels of AMR [3, 4], driven by interacting socioeconomic factors, including access to healthcare and social services [18]. By addressing key gaps in understanding AMR dynamics, this research aims to support the development of coordinated, comprehensive and sustainable strategies to mitigate AMR. This paper presents findings focused on the human health dimensions of AMR in the NT and forms part of a larger body of work examining its evolution and transmission across One Health sectors. To our knowledge, this is one of the first studies to apply causal loop modelling to AMR in a human health context. While focused on a regional and remote setting, this approach is transferable to other settings, providing a framework to elucidate AMR drivers and identify leverage points for targeted intervention.

Materials and methods

Causal loop modelling

This study developed a causal loop diagram (CLD), a qualitative modelling tool that visualises the feedback structures underlying complex problems—a crucial step in the system dynamics modelling process [19]. Drawing on stakeholder input, the study aimed to foster a shared understanding of AMR evolution and spread by exploring system-wide dynamics influencing AMR across the One Health spectrum and identifying potential leverage points for interventions.

Causal loop models or CLDs consist of variables connected by arrows that indicate causal relationships. If an increase (or decrease) in variable A (e.g., infections) leads to a corresponding increase (or decrease) in variable B (e.g., antibiotic use), a ‘+’ symbol is used to indicate a positive relationship. Conversely, if an increase in A (e.g., infection control) leads to a decrease in B (e.g., infections), a ‘-’ symbol represents a negative relationship [9]. CLDs feature two types of feedback loops that influence system dynamics or behaviour over time: reinforcing (R) or positive loops, which amplify change, and balancing (B) or negative loops, which resist change and promote system stability or equilibrium [9]. Delays (//) indicate time lags between causes and effects, often contributing to trade-offs between short- and long-term outcomes, and may lead to unintended consequences [19]. CLDs also enable the identification of leverage points—specific areas within a system where small shifts can produce significant, lasting impact. Leverage points enable fundamental, long-term actions or interventions that address underlying system drivers, rather than short-term actions targeting symptoms alone [9].

Data collection and analysis

Data for CLD development were collected through stakeholder interviews and focus groups. The study followed the Consolidated Criteria for Reporting Qualitative Research checklist [20] (Supplementary Table S1), and a one-page project overview (Supplementary File S1) was shared with participants at the time of invitation. All participants provided written informed consent prior to participation or verbal consent at the start of recorded interviews or focus groups. Participant recruitment took place between 23 August 2023 and 20 December 2024.

A systematic approach to stakeholder identification, adapted from Schiller, Winters [21], was used to develop a stakeholder map comprising key categories and groups associated with AMR. Participants were initially recruited through the investigators’ professional networks using purposive sampling [22]. At the end of each interview, participants were invited to share recruitment materials or refer other potential participants, consistent with snowball sampling [23].

Semi-structured interviews, widely recognised as effective for system dynamics model formulation [19], were conducted to explore root causes, causal relationships and feedback processes driving AMR and identify potential leverage points. Interview questions were tailored to participants’ professional background (examples provided in Supplementary File S2). Interview data informed the development of a preliminary CLD, which was subsequently presented in focus groups for discussion and validation. Interview participants willing to contribute further were invited to participate in focus groups and refer other potential participants. Focus group discussions with One Health experts validated causal relationships identified in interviews and revealed additional variables and connections. Leverage points identified from interview data with the greatest potential to address AMR were discussed during focus groups.

Interview recordings were professionally transcribed and then analysed using MaxQDA software. An inductive coding approach [24] was applied to identify themes representing direct and indirect factors driving AMR emerging from the dataset. Similar codes were grouped into the same categories and iteratively refined until dominant thematic patterns were identified [25]. For example, the codes scabies and impetigo were grouped under skin infections, together with trachoma, secondary infections and recurrent infections, later consolidated into the theme exposure to infections and reinfections.

Following thematic analysis, the five-step approach described by Baugh Littlejohns, Baum [26] was adapted to determine the polarity (positive or negative) of relationships between dominant themes. For instance, overcrowding was identified as having a positive relationship with exposure to infections and a negative relationship with food safety. Causal relationships were translated into word-and-arrow diagrams using Stella Architect software (Stella), with feedback loops automatically identified where reciprocal relationships existed. These were assembled into an initial CLD, reviewed and verified by the research team to resolve unclear polarity and ambiguous feedback structures.

Details of the study design, model building approach, and the initial CLD were shared with participants ahead of focus groups and revisited during discussions. Participants validated variable names and causal relationships, and additional relationships were incorporated where consensus was reached. Participants’ views were largely aligned across focus groups, and no disagreement requiring adjudication was observed. As a result, the validated CLD reflects a shared and consistent representation of participant perspectives. Focus group recordings were transcribed and analysed using the same approach as interview data to cross-check variables and relationships. The CLD was subsequently refined and finalised by the study team.

Results

A total of 35 participants were included, comprising 29 participants across 30 interviews (one participant was interviewed twice, before and after model development) and 10 participants in two online focus groups. Interviews lasted approximately 60 minutes each, while focus groups ran for around 120 minutes each.

Participants represented a diverse mix of professions, including clinicians (n = 15; e.g., physicians, rural medical practitioners, pharmacists and veterinarians); policymakers (n = 12; in health, housing, electricity and water services); researchers (n = 4); and other representatives (n = 4; from the private sector, including a pharmacy manager and a food systems consultant, and from a national policy advisory body for Aboriginal community health). Participants were categorised into One Health domains—human (n = 17), animal (n = 11), environment and plant (n = 5), and cross-sector (n = 2)—as well as by professional background (Fig. 1). Some participants have expertise spanning multiple domains or hold dual roles relevant to AMR. Specifically, five of six physicians also have backgrounds in chronic or infectious disease research; two veterinarians have experience in animal health research; and six policymakers working in animal health were also trained as veterinarians or researchers.

Fig. 1.

Study participants across One health sectors. DMO, duty medical officer; RMP, rural medical practitioner. Number in brackets indicates the number of participants in each One health domain or professional category

Summary of findings

Analysis of interview and focus group data identified seven major themes associated with AMR in human health in the NT, which, together with additional factors, form the core components of the causal loop model (Table 1). Associated quotes supporting these themes are provided in Supplementary Tables S2 and S3. As coding and analysis progressed, relationships between factors were defined and subsequently translated into Stella to represent their interactions within the human AMR system.

Table 1.

Major themes and other model factors

Theme	Other major model factors*
Antibiotic prescribing and use	Use of antibiotics with adverse reactions Increased environmental pathogens Severe infections and sepsis Need for broad-spectrum antibiotics Antibiotic shortage
Patient adherence to antibiotic treatment	Trusting relationships Engage with healthcare services User-friendly antibiotics
Antimicrobial stewardship	Updatable clinical decision support tool
Healthcare infrastructure and culturally safe care	Over-capacity hospitals Nursing home shortage Premature/self-discharge Culturally safe care Delayed care seeking Disease severity at presentation Treatment effectiveness Disease complications Admission/Readmission Hospital penalisation Hospital budget Cultural/language barriers Staff training, retention and support Staff turnover
Overcrowding and housing infrastructure	Overcrowding Functioning housing Hygiene conditions Food safety Poverty Exposure to infections and reinfections Housing funding Functioning housing
Access to health infrastructure	Climate change
Engagement with healthcare services	Health literacy Disease normalisation Other priorities

*Several themes share the same model factors – details are provided in Supplementary Tables S2–S3

Stakeholder input from interviews and focus groups identified antibiotic prescribing and use in both human and animal health, as well as antibiotic residues in the environment, as key drivers of AMR evolution and spread in the NT, consistent with global evidence. However, the dynamics underpinning AMR extend beyond clinical settings, reflecting the broader complexity of the AMR system.

A series of reinforcing and balancing feedback loops was identified, illustrating how interactions among system factors can either amplify or mitigate AMR emergence and spread over time. A generic global One Health AMR diagram was created to briefly represent interconnected human, animal and environment systems (Supplementary Fig. S1 and Table S4), highlighting that AMR cannot be effectively managed when addressed in isolation. The generic diagram reflected established global One Health AMR dynamics and, while not intended to generate novel insights, provided a useful foundation for facilitating discussion with participants, particularly those unfamiliar with AMR or systems thinking. In this paper, interactions between factors driving AMR in human health are examined through the healthcare and community sub-models. Details of causal relationships and feedback loops are provided in Supplementary Files S3–S6.

The healthcare sub-model

The human healthcare sub-model reflects participant perspectives on factors driving AMR and their interactions across hospital and primary healthcare settings in the NT. The model comprises 59 variables and 96 causal connections (Fig. 2 and Supplementary File S3), generating 506 reinforcing and 101 balancing feedback loops (Supplementary File S4).

Fig. 2.

AMR evolution and spread in the healthcare sub-model. →(+) and →(-), positive and negative relationships; ABX, antibiotics; AMR, antimicrobial resistance; AMS, antimicrobial stewardship; R and B, reinforcing and balancing loops; //, delay; green colour, leverage points for AMR intervention

Reinforcing loop R1 (Fig. 2 or Table 2) illustrates how increased reliance on antibiotics elevates selective pressure, leading to treatment failures, rising infection incidence, and continued antibiotic demand, thereby reinforcing AMR risk. One clinician described scenarios where first-line antimicrobial therapy failed due to resistance, necessitating the use of last-resort agents such as colistin to manage severe infections (loop B1). While effective in some instances, such antibiotics are often associated with serious side effects (Supplementary Table S2, quote 1.1). In certain cases, this may include nephrotoxicity, resulting in acute kidney injury [27] and requiring interventions such as dialysis (loop B2).

Table 2.

Key feedback loops in the healthcare sub-model

Loop number	Loop structure
Reinforcing loops
R1	ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) Infectious disease burden →(+) ABX prescribing and use
R2	Exposure to infections and reinfections →(+) Infectious disease burden →(+) Over-capacity hospitals →(+) Exposure to infections and reinfections
R3	Over-capacity hospitals →(+) Premature/self-discharge →(+) Admission/Readmission → (+) Over-capacity hospitals
R4	ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) Need for new ABX →(+) New ABX available →(+) ABX prescribing and use
R5	Culturally safe care →(-) Delayed care seeking →(+) Disease severity at presentation →(-) Treatment effectiveness →(-) Disease complications →(+) Admission/Readmission →(+) Hospital penalisation →(-) Hospital budget →(+) Culturally safe care
R6	Culturally safe care →(-) Premature/self-discharge →(+) Admission/Readmission →(+) Hospital penalisation →(-) Hospital budget →(+) Staff training, retention and support →(+) Culturally safe care
R7	Over-capacity hospitals →(+) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence →(+) Morbidity and mortality →(+) Healthcare costs →(-) Healthcare budget →(+) Expanded hospital infrastructure →(-) Over-capacity hospitals
R8	Over-capacity hospitals →(+) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence →(+) Morbidity and mortality →(+) Healthcare costs →(-) Healthcare budget →(-) Nursing home shortage →(+) Over-capacity hospitals
R9	Trusting relationship →(+) Adherence to ABX treatment →(-) Need for broad-spectrum ABX →(+) ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) Use of ABX with adverse reactions →(+) Morbidity and mortality →(+) Healthcare costs →(-) Healthcare budget →(+) Culturally safe care →(+) Trusting relationship
R10	Trusting relationship →(+) Engage with healthcare services →(+) Adherence to ABX treatment →(+) Appropriate ABX prescribing and use →(-) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence →(+) Morbidity and mortality →(+) Healthcare costs →(-) Healthcare budget →(+) Culturally safe care →(+) Trusting relationship
R11	Healthcare budget →(+) Priorities for AMS →(+) Updatable clinical decision support tool →(+) Effective AMS →(+) Appropriate ABX prescribing and use →(-) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence →(+) Morbidity and mortality →(+) Healthcare costs →(-) Healthcare budget
Balancing loops
B1	ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) Use of ABX with adverse reactions →(-) Infectious disease burden →(+) ABX prescribing and use
B2	AMR awareness →(+) Appropriate ABX prescribing and use →(-) Effect of ABX in increasing AMR →(+) Use of ABX with adverse reactions →(+) Morbidity and mortality →(+) Public attention →(+) AMR awareness
B3	ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence→(+) Morbidity and mortality →(+) Public attention →(+) AMR awareness →(+) Infection prevention and control -→(-) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use
B4	AMR awareness →(+) Priorities for AMS →(+) Updatable clinical decision support tool →(+) Effective AMS →(+) Appropriate ABX prescribing and use →(-) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence→(+) Morbidity and mortality →(+) Public attention →(+) AMR awareness

→(+) and →(-), positive and negative relationships; ABX, antibiotics; AMR: antimicrobial resistance; AMS: antimicrobial stewardship; R and B, reinforcing and balancing loops

Key themes identified include antibiotic prescribing and use, patient adherence to antibiotic treatment, antimicrobial stewardship, and healthcare infrastructure and culturally safe care

Antibiotic prescribing and use

High antibiotic use in the NT is driven by interrelated clinical, environmental and structural factors. A high burden of severe infections and sepsis, compounded by significant prevalence of chronic conditions such as diabetes, kidney and cardiovascular disease [28, 29], predisposes individuals to recurrent infections (loop R2).

Broad-spectrum antibiotics are commonly used to manage or prevent secondary bacterial infections (Table S2, quotes 1.2–1.4), including those associated with viral respiratory infections and scabies-related skin disease, and to reduce sepsis-related mortality. One participant noted: “Scabies is really high up here, especially in remote communities… one of the most common causes of getting recurrent strep infections and that’s sort of one of the big drivers of acute rheumatic fever” (Participant (P) 10). This practice may, in part, reflect a preventive strategy to reduce life-threatening complications, including rheumatic heart disease. As another clinician explained: “Our antibiotics are often broader … Unfortunately, the outcome is that we probably drive AMR” (P1).

Participants also linked climate variability and extreme weather events to greater exposure to environmental pathogens, particularly during the wet season. These conditions increase infection risk, especially among individuals with underlying health conditions. “In wet season, we get melioidosis, people get nasty necrotising infections from the water. So we do have a lot of broad-spectrum antibiotic pressure. We have to do that because otherwise people will die” (P1; Table S2, quote 1.5). Persistent burdens of chronic diseases and infections such as melioidosis [30, 31] drive frequent use of broad-spectrum antibiotics. In remote areas, prophylactic prescribing was commonly reported due to limited diagnostic capacity, shortages of narrow-spectrum antimicrobials, and challenges with treatment adherence (Table S2, quotes 1.6–1.8).

Patient adherence to antibiotic treatment

Patient adherence is influenced by access to and engagement with healthcare services, both of which are underpinned by trusting relationships (Table S2, quote 2.1) with healthcare providers (loops R9-10). “Having a trusting relationship is the absolute key requirement for attending healthcare” (P7; Table S2, quotes 2.2).

In the NT, high staff turnover, partly driven by geographic isolation, has increased reliance on fly-in fly-out healthcare workers, disrupting continuity of care and formation of trust [32].

Treatment adherence may be improved through more ‘user-friendly’ antibiotic formulations, such as easy-to-swallow tablets or less frequent dosing regimens (Table S2, quote 2.3). In turn, adherence can influence antibiotic prescribing decisions (loops R9-10), when convenience may sometimes be prioritised over targeted treatment, potentially undermining antimicrobial stewardship and increasing AMR risk.

Antimicrobial stewardship

Antimicrobial stewardship (AMS) programs aim to optimise antimicrobial use, improve patient outcomes and reduce AMR risk. In the NT, limitations in AMS implementation [33] are linked to inadequate prioritisation, potentially driven by limited awareness of AMR severity (loop B4, Table 2) and constrained healthcare budgets (loop R11).

Clinicians emphasised the importance of timely access to local AMR surveillance data and point-of-care testing to guide effective AMS. Additionally, access to up-to-date treatment guidelines, integrated AMS education, and interoperable electronic health records across hospital and primary care were identified as essential. A critical gap was the absence of a contemporary, updatable clinical decision support tool. “We really need good antimicrobial stewardship to do clinical decision support … to rationalise quickly and choose narrow antibiotics according to what pathogen is growing” (P1; Table S2, quote 3.1). “If there was something that could tell me the AMR rates in a particular community, therefore guide me on the best choice of antibiotics, then I would absolutely follow it. The reality is that at the moment I’m guided by a paper book” (P21; Table S2, quote 3.2).

Healthcare infrastructure and culturally safe care

Participants identified systemic healthcare pressures as key contributors to infections, reinfections and AMR escalation. Overcrowded hospitals, driven by the high burden of infections and chronic diseases, ageing infrastructure, limited single-occupancy rooms, and delays in aged care admission were described as drivers of increased transmission (loops R2, R7 and R8, Table 2). “People share four-bedded rooms … There’s over 100 people at the moment waiting at hospitals for a nursing home bed” (P1; Table S2, quotes 4.1 and 4.3).

High rates of self-discharge [34, 35], partly linked to cultural safety issues, poor communication, extended wait times or family responsibilities, exacerbate readmissions, hospital strain and budget (loops R3 and R6) (Table S2, quotes 4.4–4.7). “So Northern Territory has the highest national rates of self-discharge. That’s at 12% of all hospital admissions which is way higher than the national average. And that’s because hospitals are too cold and culturally unsafe. The doctors and nursing staff don’t speak the language. There’s not enough uptake of interpreters in the health system. They don’t know what’s going on. It’s terrifying” (P7). Readmission following premature or self-discharge can subsequently reduce hospital revenue, which is often tied to inpatient activity. “When you’re readmitted, the hospital gets financially penalised each time … and it doesn’t matter if it’s self-discharged. And it affects our budget” (P1). This further limits resources available for culturally safe care (loops R5-6), which involves ensuring environments that are free from assault, challenge or denial of Aboriginal and Torres Strait Islander peoples’ identity and needs, and that support shared respect and self-determination [36].

Insufficient staff training, retention and support may also hinder culturally safe care (Table S2, quotes 4.6–4.10). Loops R5 and R6 describe how these deficits contribute to delayed care seeking and increased disease severity at presentation. “People just have different priorities … Late presentations are a problem” (P14). This, in turn, reduces treatment effectiveness, increasing disease complications and repeat readmissions.

Participants emphasised the need for targeted staff training, especially comprehensive orientation for fly-in fly-out workers, to improve understanding of the NT context, alongside robust retention and support strategies (Table S2, quotes 4.11–4.13). These, in turn, may foster trust, strengthen cultural safety, enhance patient engagement and improve treatment adherence.

Key leverage points

Despite numerous reinforcing loops driving AMR in the NT (Fig. 2 or File S4), participants identified several leverage points (highlighted in green in the CLD) capable of shifting this dominant long-term trend. These include appropriate antibiotic prescribing and use (loops B4 or R10), supported by effective AMS implementation (loop B4 or R11), increased AMR awareness among healthcare providers and communities (loops B2-4), improved culturally safe care (loops R5-6, R9-10), and increased investment in staff training, retention and support (loop R6). Participants also highlighted the need for strengthening infection prevention and control (loop B3).

Importantly, findings indicate that reducing the upstream burden of infectious diseases is critical to alleviating healthcare pressure and limiting antibiotic demand, underscoring the importance of preventive efforts. These community-level strategies are explored further in the following community sub-model.

The community sub-model

The community sub-model characterises AMR dynamics outside formal healthcare settings. While partially overlapping with the healthcare sub-model, it highlights multidirectional pathways of AMR transmission between community and clinical environments (Fig. 3). The sub-model contains 42 variables and 79 connections (Fig. 3 and Supplementary File S5), generating 40 reinforcing and one balancing feedback loops (Supplementary File S6).

Fig. 3.

AMR evolution and spread in the community sub-model. →(+) and →(-), positive and negative relationships; ABX, antibiotics; AMR, antimicrobial resistance; R and B, reinforcing and balancing loops; //, delay; green colour, leverage points for AMR intervention

Key themes identified include overcrowding and housing infrastructure, access to health infrastructure, and engagement with healthcare services.

Overcrowding and housing infrastructure

Participants identified household overcrowding as a critical structural determinant influencing both limited access to functional housing and increased exposure to infectious disease. This, in turn, contributes to recurrent infections and elevated AMR risk. Overcrowded and poorly functioning housing, often compounded by poverty and socioeconomic disadvantage, undermines hygiene practices and food security (loops R1-3), thereby increasing susceptibility to infections [37] (Supplementary Table S3, quotes 5.1–5.3).

Persistent shortages result in high household occupancy with individuals sharing limited living and sleeping space. One participant noted: “There are not enough houses. It’s common to hear of places with more than 10 people per house” (P24). Housing design was frequently described as misaligned with cultural practices and living norms in remote communities. In overcrowded conditions, housing functionality is further compromised (Fig. 3 or Table 3), placing additional strain on essential amenities. These challenges are particularly pronounced in remote areas, where poverty and social disadvantage are more prevalent (Table S3, quotes 5.4–5.7).

Table 3.

Key feedback loops in the community sub-model

Loop number	Loop structure
Reinforcing loops
R1	Overcrowding →(-) Functioning housing →(-) Overcrowding
R2	ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence→(+) Societal costs →(+) Poverty →(+) Overcrowding →(+) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use
R3	ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence→(+) Societal costs →(+) Poverty →(-) Functioning housing →(+) Food safety →(-) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use
R4	Access to health infrastructure →(-) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence→(+) Societal costs →(+) Poverty →(-) Access to health infrastructure
R5	Culturally safe care →(+) Trusting relationship →(+) Engage with healthcare services →(-) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence →(+) Morbidity and mortality →(+) Healthcare costs →(-) Healthcare budget →(+) Culturally safe care
R6	ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence →(+) Morbidity and mortality →(+) Healthcare costs →(-) Healthcare budget →(+) Health promotion →(+) Health literacy →(-) Disease normalisation →(-) Engage with healthcare services →(-) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use
Balancing loops
B1	AMR awareness →(+) Infection prevention and control →(-) Exposure to infections and reinfections →(+) Infectious disease burden →(+) ABX prescribing and use →(+) Effect of ABX in increasing AMR →(+) AMR emergence and prevalence→(+) Morbidity and mortality →(+) Public attention →(+) AMR awareness

→(+) and →(-), positive and negative relationships; ABX, antibiotics; AMR, antimicrobial resistance; R and B, reinforcing and balancing loops

While timely housing maintenance remains challenging, participants highlighted programs such as Healthy Homes (Grealy et al., 2025) as important steps toward strengthening local capacity for repair and upkeep. However, competing priorities and limited funding were reported as ongoing barriers. “It really ought to be a priority, but other things get prioritised … We built a lot of new houses, but we’re not providing people with living skills to look after them” (P24; Table S3, quote 5.8).

Access to health infrastructure

Beyond built environment conditions, access to essential health-supporting infrastructure, including safe drinking water and reliable electricity, was identified as a key determinant of community health in the NT (loop R4). Bore water is the primary source for many remote communities; however, some areas have reported chemical and physical exceedances of national drinking water guidelines.

Rising temperatures and extreme weather conditions in the NT exacerbate challenges in maintaining water infrastructure durability and functionality (Table S3, quotes 6.1–6.2). As a result, infrastructure degradation increases the likelihood of microbial contamination and chemical pollutants, compromising water safety and quality.

Poor water quality and limited access to basic utilities, combined with overcrowding, were perceived to contribute to high rates of skin and soft tissue infections. Such infections, if left untreated, can progress to severe sequelae, such as sepsis, rheumatic fever and rheumatic heart disease, which remain highly prevalent in remote communities [38, 39].

Engagement with healthcare services

Participants identified multiple factors underlying low healthcare engagement. While geographic isolation presents a physical barrier, individuals may remain reluctant to seek care even when services are accessible. Consistent with findings in the healthcare sub-model, this reluctance was frequently linked to limited trust in the healthcare system and the absence of culturally safe care (loop R5) (Table S3, quotes 7.1–7.3).

Additional barriers include cultural and language differences, low health literacy and the normalisation of disease, which together reduce the perceived urgency of seeking medical care (loop R6). Limited health literacy can impair individuals’ ability to understand, navigate and effectively engage with healthcare, thereby reducing service uptake. The normalisation of conditions such as impetigo and scabies in remote communities [40], often compounded by stigma, was reported to delay care-seeking behaviour and reduce recognition of illness severity among both patients and healthcare providers. Other contributing factors include competing priorities, such as family or carer responsibilities or participation in ‘sorry business’ (Table S3, quotes 7.4–7.9).

These interacting factors limit healthcare engagement and delay initiation of appropriate treatment. The interplay between low health literacy and disease normalisation also contributes to high rates of self-discharge from healthcare facilities.

Key leverage points

The reinforcing loops identified in the community sub-model (Fig. 3 or File S6) illustrate how interacting structural and social factors drive persistent infections and reinfections. The broader societal costs associated with AMR (e.g., prolonged hospitalisation and productivity losses) further exacerbate poverty, overcrowding and chronic disease (loops R2–R4).

Breaking this cycle requires targeted action at key leverage points, including increasing access to functional housing and essential utilities (loop R1), enhancing culturally safe care and trust in healthcare services (loop R5), and sustaining community-based health promotion (loop R6). Together with clinical interventions described in Sect. “The healthcare sub-model”, these community-level strategies highlight the interdependence of community and healthcare environments and the need for coordinated, system-wide action to reduce AMR transmission across both sub-systems.

Discussion

Drawing on input from stakeholders across One Health sectors, we developed a causal loop model to conceptualise the direct and indirect factors driving the emergence and spread of AMR in a high burden region of Australia. By focusing on regional and remote settings, our findings demonstrate how complex interrelationships among clinical, social and structural factors shape the dynamics of infectious diseases and resistance. These findings underscore that addressing AMR through isolated interventions is insufficient and unlikely to result in sustained reductions in disease burden, highlighting the importance of systems-based approaches. For example, despite widespread antibiotic use, infections such as recurrent scabies persist in the region (Table S2, quote 1.9), reflecting a broader failure to address upstream determinants of health. Our model shows that reliance on downstream clinical responses is itself a symptom of an overwhelmed system, repeatedly emerging when these foundational drivers remain unaddressed.

We found that the prevailing narrative attributing the high burden of AMR in the NT primarily to high volumes of broad-spectrum antibiotic use needs critical re-evaluation. Given the disproportionately high burden of infectious diseases, AMR risk is compounded by systemic challenges, including healthcare delivery and access in geographically remote settings with high workforce turnover [41], substandard housing and infrastructure [37], and socio-economic disadvantage [18]. In this context, antibiotic use is often clinically appropriate and essential to reduce morbidity and mortality. What may be externally perceived as antibiotic overuse, in fact, reflects necessary responses to prevent serious infections and adverse outcomes.

This interpretation aligns with previous research reporting high levels of appropriate antibiotic prescribing (over 85%) in accordance with clinical guidelines across regional and remote primary care settings in northern Australia, including the NT [42]. Moreover, what remains underexplored is the role of antimicrobial-resistant strains and resistance genes circulating within both healthcare and community environments as key contributors to AMR. Consequently, reducing antibiotic use alone is unlikely to sufficiently control AMR [43]. While the development of new antibiotics remains important, reliance on pharmaceutical innovation alone will continue to perpetuate cycles in which AMR predictably re-emerges. Without addressing underlying systemic drivers, such advances will only provide temporary relief.

Our systems thinking approach identified key leverage points for high-impact interventions capable of producing sustainable change while minimising unintended consequences across healthcare and community settings in the NT. These include interoperable AMS and infection control programs; culturally safe care; workforce training, retention and support; AMR awareness and health promotion; and improved access to functional housing and health-enabling infrastructure.

As illustrated in our model, while new antibiotics may offer short-term benefits, they address only the symptoms rather than the root causes of AMR. While vaccine development and deployment depend on external factors, other preventive measures, particularly enhanced infection prevention and control, are more likely to deliver sustainable reductions in infectious disease burden and subsequent antibiotic demand. Participants consistently emphasised the importance of addressing infections at their source as a key strategy to mitigate AMR risk, necessitating a shift from reactive clinical responses to proactive, community-based interventions.

Continued investment in functional housing, community-led maintenance programs, and access to health hardware is essential to reduce overcrowding and its downstream impacts. Improvements in cultural safety and continuity of care, including greater investment in staff training, retention and support [44], alongside effective communication [35], AMR awareness and health promotion, are critical for building trust, enhancing healthcare engagement and supporting treatment adherence. Although such investments require resources, they have the potential to significantly reduce long-term infection and AMR burden, thereby alleviating pressures on healthcare systems and society at large.

Strengths and limitations

The CLD developed in this study provides insights beyond qualitative data and literature synthesis by explicitly revealing feedback mechanisms, cross‑system interactions, and unintended consequences that shape AMR dynamics in this setting. In particular, the model showed how clinically appropriate antibiotic use is reinforced by upstream drivers, creating self‑perpetuating cycles that are not apparent when factors are considered in isolation.

By integrating stakeholder perspectives into a single system representation, the model enabled identification of where and how interventions in non‑clinical domains (e.g. housing, culturally safe care, workforce stability) could influence downstream antibiotic demand and resistance risk. These system‑level feedback mechanisms and leverage points would not have been apparent through interviews or literature synthesis alone. Importantly, the modelling process also supported stakeholders in shifting from linear problem framing to systems‑based reflection, facilitating shared understanding and targeted discussion of intervention priorities.

The CLD provides a comprehensive systems-level view of interconnected factors influencing AMR in human health; however, it does not capture all possible relationships. This limitation reflects both the scope of the study and the inherent complexity of AMR systems. The sample size was relatively small, which may introduce a degree of subjectivity into the model. While perspectives from stakeholders working in health, housing and utilities provided valuable population-level insights, future research would benefit from deeper community engagement to elicit individual-level lived experiences. In addition, although participants appeared to engage positively with the approach as a way of understanding AMR as a systems‑based problem, the impact of the modelling process on stakeholder engagement was not formally evaluated, as this was not an explicit aim of the study.

Another limitation lies in the challenge of prioritising the many pressing issues associated with AMR in the NT. While stakeholders identified a wide range of needs, our analysis focused on leverage points highlighted during focus groups as having the greatest potential to shift the current trajectory of the AMR system.

We acknowledge that potential biases may exist, and our ongoing work aims to test and refine these hypotheses using quantitative system dynamics simulation informed by real-world data. This next phase will allow further model refinement and support the design and evaluation of effective policy and management strategies.

The CLD was shared in detail primarily with participants who opted for continued involvement in focus groups. While this facilitated rich validation and systems-level discussion, broader dissemination may have generated additional perspectives.

Conclusions

This study draws on stakeholder perspectives to explore the interrelated factors shaping the evolution and spread of AMR with a focus on human health. These insights were synthesised into a visual model, a causal loop diagram, representing a systems-based hypothesis of the key drivers underpinning AMR. While many contributing factors have been identified in previous work, our model highlights how these elements interact and collectively reinforce the dynamic complexity of the AMR system. The causal loop model provides a valuable means of identifying root causes, hidden drivers and feedback mechanisms while capturing a holistic picture of infectious disease and AMR burden. By focusing on human health dynamics, this approach addressed a gap in the literature and is transferrable to broader contexts, particularly in low-resource settings. Importantly, the approach enables stakeholders to adopt a systems thinking mindset when engaging with complex public health challenges and provides a platform to identify opportunities for targeted interventions to inform decision-making.

As part of the broader project, we are also investigating the direct and indirect drivers of AMR in animal and environmental systems, including wildlife, and their interactions with human health. Understanding these cross-sectoral dynamics is critical yet remains underrepresented in AMR research [45–47].

The next phase of the study involves the development of a quantitative system dynamics simulation model, enabling the design and testing of alternative strategy scenarios and assessment of their potential impact on reducing AMR risk. This will support evidence-informed decision-making and prioritisation of interventions.

Overall, our findings reinforce that addressing AMR cannot rely on a single system, particularly healthcare or antimicrobial stewardship in isolation. In settings characterised by high infectious disease burden, coupled with healthcare system constraints and social inequities, achieving sustained reductions in AMR requires addressing upstream determinants of health and shifting towards preventive, community-based interventions. There is no ‘silver bullet’ to tackle the multifaceted, systemic challenges posed by AMR or similar public health issues. Meaningful and sustainable progress will require coordinated, multi-sectoral and systems-informed approaches integrating collective perspectives across multiple domains, such as healthcare, community and the environment.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

ABX

Antibiotics

AMR

Antimicrobial resistance

AMS

Antimicrobial stewardship

CLD

Causal loop diagram

NT

Northern Territory

Author contributions

YP and TW conceptualised the study and conducted the investigation. YP designed the methodology, analysed the data, carried out the formal analysis, developed the model and wrote the manuscript text. YP, TW and SJ validated the results and reviewed and approved the final manuscript.

Funding

Open access funding provided by CSIRO Library Services. YP was funded by a CSIRO Early Research Career Postdoctoral Fellowship from the Commonwealth Scientific Industrial and Research Organisation (R-19391-01).

Data availability

All data generated or analysed during this study are included in this published article and its supplementary information files.

Declarations

Ethics approval and consent to participate

Ethical approval including the study design protocol was obtained from the Human Research Ethics Committee of the Commonwealth Scientific and Industrial Research Organisation (2023_028_HREC), as well as from the Northern Territory Health and Menzies School of Health Research (2023-4668). Written or verbal informed consent was obtained from all participants prior to conducting interviews and focus groups.

Consent for publication

Not applicable

Competing interests

The authors declare no competing interests.