Abstract
Objective
In Bangladesh, the unregulated use of antimicrobials in livestock poses a significant threat to the effectiveness of these drugs, yet multi-sectoral data to guide stewardship are scarce. This study investigated veterinary drug use, antibiotic awareness, and risk factors contributing to antimicrobial resistance (AMR) in the Mymensingh region of Bangladesh.
Methods
A cross-sectional survey was conducted among livestock farmers (n = 101) and retail pharmacy sellers (n = 104), complemented by veterinary hospital prescription records (n = 330). Field data were collected and descriptive and analytical statistics were conducted in Python to determine predictors of antibiotic choice. Principal Component Analysis (PCA), Hierarchical Clustering, and network analysis were performed to characterize antibiotic usage patterns and co-prescription structures.
Results
Provider qualification and knowledge of withdrawal periods significantly influenced antibiotic practices in both farm and pharmacy settings. PCA captured major variance components in farm (41.06%), pharmacy (51.89%), and hospital (40.85%) datasets. Hierarchical Clustering demonstrated sector-specific preferences among 27 antibiotic compounds, with Amoxicillin dominating farm use and Tetracycline-class drugs prevalent in hospital prescriptions. Network analysis revealed tightly connected co-prescription clusters, with Amoxicillin and Tetracycline functioning as central nodes.
Conclusions
Significant variation exists in antibiotic awareness and usage across sectors, shaped largely by provider qualification and withdrawal knowledge. The identified predictors and usage networks provide critical evidence for designing targeted antimicrobial stewardship interventions, highlighting the need for stricter pharmacy regulation and enhanced farmer education.
Keywords
Antimicrobial resistance (AMR)
Veterinary drug use
Livestock
Antimicrobial stewardship
One Health
Bangladesh
1. Introduction
Antimicrobial resistance (AMR) has emerged as a major and rapidly escalating global health challenge, threatening the effectiveness of modern medicine and undermining decades of progress in the prevention and treatment of infectious diseases. A recent systematic analysis by the GBD 2021 Antimicrobial Resistance Collaborators, published in The Lancet, reported that in 2021 bacterial AMR was directly responsible for approximately 1.27 million deaths and associated with nearly 4.95 million deaths worldwide. The study further warns that, if effective interventions are not implemented, the global burden of AMR could rise dramatically, with projections suggesting up to 10 million deaths annually by 2050 due to drug-resistant infections [1]. Addressing this challenge requires a One Health approach integrating human, animal, and environmental health [2].
The antimicrobial resistance (AMR) landscape in Bangladesh is characterized by a high prevalence of resistant pathogens, most notably extended-spectrum beta-lactamase (ESBL)-producing Escherichia coli, which has become endemic across both clinical and community settings [3]. This crisis is exacerbated by the pervasive misuse of antibiotics in the poultry and livestock sectors, a practice that facilitates the dissemination of resistance through environmental pathways and the food chain [4]. The human health implications are profound; recent estimates from the Institute for Health Metrics and Evaluation (IHME) indicate that in 2021, the number of deaths associated with AMR among individuals aged 70 and above in Bangladesh reached approximately 41,100 [5]. These findings underscore the urgent need for a One Health approach to mitigate the escalating threat of antimicrobial resistance in the region.
The livestock sector is a major contributor to AMR through routine antibiotic use for growth promotion and disease prevention, facilitating resistant bacteria transmission via food, contact, and the environment [6]. In low- and middle-income countries (LMICs) such as Bangladesh, weak regulation, limited surveillance, and widespread non-prescription access to veterinary drugs exacerbate misuse, including incorrect dosing, incomplete courses, and neglect of withdrawal periods [7].
Bangladesh’s rapidly expanding livestock sector is essential for food security and rural livelihoods but faces governance challenges. Antibiotic distribution often involves a decentralized network of farmers, drug sellers, feed dealers, paravets, and veterinarians, with limited professional oversight in rural areas [8]. Such fragmented systems increase the risk of inappropriate antimicrobial use and hinder effective stewardship, highlighting the need for coordinated multisectoral interventions [[9], [10], [11]].
This study addresses these gaps through a multi-sectoral assessment of veterinary drug practices and antimicrobial resistance (AMR) risk factors in the Mymensingh region of Bangladesh. The study examines antibiotic use patterns, disease management practices, and sources of veterinary advice across farms, pharmacies, and veterinary hospitals to identify key drivers of antimicrobial misuse and inform targeted antimicrobial stewardship strategies.
2. Material and Methods
2.1. Study area
This cross-sectional study assessed veterinary drug practices and antimicrobial resistance (AMR) risk factors across multiple sectors in the Mymensingh region of Bangladesh. Three Upazilas- Ishwarganj, Mymensingh Sadar, and Purbadhala were selected due to their high density of livestock farms and veterinary service providers, representing an important livestock production zone in the country.
2.2. Geographic mapping
A GIS-based map was developed using QGIS v3.28. The map highlights the three selected Upazilas within Mymensingh, with administrative boundary data obtained from the Bangladesh Bureau of Statistics [12] and the Humanitarian Data Exchange [13]. Data were projected using the WGS 84 coordinate system (EPSG:4326).
2.3. Study participants and recruitment
Participants were recruited from three interconnected components of the livestock value chain: livestock farms, retail pharmacies, and veterinary hospitals (Table 1). A multi-stage sampling approach was employed for recruitment. Livestock farmers (n = 101) were recruited through convenient sampling from the selected Upazilas, targeting both commercial and small-scale poultry and cattle operations. Retail pharmacy sellers (n = 104) were identified through local business registries and recruited via on-site visits to veterinary drug outlets. Veterinary hospital data (n = 330) were obtained through a retrospective review of official clinical records from the Upazila Veterinary Hospitals. Inclusion criteria required farmers to have at least one year of experience in livestock management and pharmacy sellers to be actively involved in dispensing veterinary drugs.
Table 1. Multisectoral data collection from farms, pharmacies, and veterinary hospitals.
| Sector | Sample size (n) | Data collection method | Key data points collected |
|---|---|---|---|
| Livestock farms | 101 | Structured questionnaire administered to farmers | Livestock type, disease status, treatment practices, antibiotic usage, and knowledge of drug withdrawal periods. |
| Retail pharmacies | 104 | Structured questionnaire administered to pharmacy sellers | Seller qualification, dispensing practices (prescription vs. non-prescription), most frequently sold antibiotics, and perceived drug effectiveness. |
| Veterinary Hospitals (VH) | 330 (Records) | Retrospective review of hospital records | Demographic details, disease records, and drug prescription profiles. |
2.4. Questionnaire development and validation
A structured questionnaire was developed for this study (Supplementary file 1) through consultation with experts in veterinary medicine, parasitology, and antimicrobial stewardship. The instrument was reviewed to ensure clarity, relevance, and content validity before data collection. It was administered to farmers and pharmacy sellers, while veterinary hospital records were extracted and organized according to the same variables.
2.5. Statistical analysis
All analyses were conducted in Python v3.10 [14] using commonly used scientific computing libraries including pandas, scipy, statsmodels, scikit-learn, matplotlib, seaborn and numpy. A two-tailed p-value threshold of <0.05 was considered statistically significant.
Descriptive statistics summarized antibiotic use, disease prevalence, and provider categories. For categorical comparisons, Chi-square tests [15] were applied when expected counts were ≥5, and Fisher’s Exact Test [16] was used for sparse data.
Spearman’s rank correlation [17] assessed relationships among indicators of antibiotic awareness, misuse, and AMR risk. Cross-tabulations and heatmap visualizations were used to compare antimicrobial practices across farms, pharmacies, and veterinary hospitals.
Logistic regression [18] examined socio-behavioral predictors of withdrawal period awareness. The dependent variable was knowledge of drug withdrawal periods, with education, information access, and years of experience as predictors. Categorical variables were coded using treatment coding with “Below SSC” and “Low/No” as reference groups. Odds ratios (ORs) with 95% confidence intervals (CIs) were reported, and model convergence confirmed stable parameter estimates.
Principal Component Analysis (PCA) [19,20] was applied separately to farm, pharmacy, and veterinary hospital datasets using standardized variables to explore variability in antimicrobial usage. Hierarchical clustering [21] grouped antibiotics according to similarity in usage profiles across sectors. These analyses were exploratory and used to visualize structural patterns rather than infer causality.
Network analysis was used to map relationships among antibiotics, diseases, and sources. Nodes represented antibiotics, diseases, and sectors, while edges denoted co-occurrence. Degree and betweenness centrality were calculated to identify structurally prominent antibiotics within the antimicrobial use network.
2.6. Ethical considerations
The study involved no animal experimentation or identifiable human data. In accordance with Bangladesh Agricultural University policy and national regulations, formal ethical approval was not required. Participation was voluntary, verbal consent was obtained, and all survey responses and hospital records were anonymized.
3. Results
3.1. Antimicrobial use and disease profiles in livestock systems
A total of 535 cases were analyzed across farms (n = 101), veterinary pharmacies (n = 104), and village-level animal health services (n = 330).
At farm level, poultry was the predominant species (60.4%), followed by cattle (29.7%) and goats (6.9%). Major diseases included Mycoplasmosis (14.9%), Newcastle disease (11.9%), Lumpy skin disease (11.9%), and Gumboro (10.9%). Treatments were mainly provided by veterinary surgeons and paravets (each 34.7%). The most frequently used antibiotics were Amoxicillin (31.7%) and Enrofloxacin (20.8%), with 70.3% of farmers reporting adherence to withdrawal periods (Table 2).
Table 2. Summary of farm-, pharmacy-, and veterinary hospital-level indicators on livestock types, disease occurrence, treatment sources, and antibiotic use in Mymensingh region.
| Category | Variable | Most frequent responses with frequency (%) |
|---|---|---|
| Farm-level indicators | Upazila | Purbadhala (62.4%), Mymensingh Sadar (23.8%), Ishwarganj (13.9%) |
| Types of animals | Poultry (60.4%), Cattle (29.7%), Goat (6.9%) | |
| Common diseases | Mycoplasmosis (14.9%), Newcastle disease (11.9%), Lumpy skin disease (11.9%), Gumboro (10.9%) | |
| Treatment providers | Veterinary surgeon (34.7%), Paravet (34.7%) | |
| Generic drugs used | Amoxicillin (31.7%), Enrofloxacin (20.8%) | |
| Knowledge of withdrawal | Yes (70.3%), No (20.8%) | |
| Pharmacy-level indicators | Seller qualification | Higher Secondary Certificate (78.9%), Honours (16.4%) |
| Location | Purbadhala (88.5%), Mymensingh Sadar (11.5%) | |
| Prescription practice | Sometimes (91.4%) | |
| Most sold antibiotics | Ceftriaxone (26.0%), Amoxicillin (18.3%) | |
| Record keeping | No (93.3%) | |
| Effective antibiotics | Ceftriaxone (51.0%), Gentamicin (9.6%) | |
| Treatment failure antibiotics | Penicillin (16.4%), Ciprofloxacin (16.4%) | |
| Veterinary hospital indicators | Upazila | Kendua (100%) |
| Types of farms | Cattle (83.6%) | |
| Common diseases treated | Anorexia (22.4%), Worm infestation (21.5%), Dog bite (11.8%), Pneumonia (11.5%), LSD (11.2%) | |
| Prescribed drugs | S.Pben (44.9%), Etracin (36.7%) | |
| Generic drugs used | Sulfadimidine (45.5%), Oxytetracycline (37.6%) | |
| Drug type | Antimicrobial (100%) |
At pharmacy level, most outlets were located in Purbadhala (88.5%). Antibiotics were commonly dispensed without prescriptions (91.4%), and record-keeping was rare (93.3%). Ceftriaxone (26.0%) and Amoxicillin (18.3%) were the most frequently sold antibiotics. Ceftriaxone was perceived as most effective (51.0%), though nearly half of respondents (48.1%) reported treatment failures, particularly with Penicillin and Ciprofloxacin (Table 2).
At veterinary hospital level, cattle were the dominant species treated (83.6%). Common conditions included anorexia (22.4%), worm infestation (21.5%), and pneumonia (11.5%). All cases involved antimicrobial treatment, with Sulfadimidine (45.5%) and Oxytetracycline (37.6%) being most frequently prescribed (Table 2).
3.2. Comparative antimicrobial use in farms and hospitals
Antimicrobial use and disease profiles differed markedly between farms and veterinary hospitals (VH). Oxytetracycline and Sulfadimidine were predominantly used in VH cases, whereas Amoxicillin, Ciprofloxacin, Doxycycline, and Enrofloxacin were more common on farms. Ceftriaxone, Ampicillin, and Erythromycin were also more frequent in VH settings, while Gentamicin, Levofloxacin, and Lincomycin were reported only on farms. Fig. 1
Disease patterns also varied, with VH records showing higher incidences of anorexia, pneumonia, dog bites, and worm infestation, whereas farms reported greater frequencies of Mycoplasmosis, Gumboro, and Newcastle disease. Treatment providers differed substantially: VH cases were managed exclusively by veterinary surgeons, while farm treatments involved paravets, farm owners, and drug sellers (Fig. 2).
3.3. Correlation analysis of awareness, misuse, and antibiotic resistance indicators
In farms, provider qualification showed a moderate negative correlation with drug choice (rs = -0.264, p = 0.008), suggesting that more qualified providers selected less commonly used antibiotics. Associations between provider qualification and withdrawal knowledge (rs = -0.087, p = 0.390) and between drug choice and withdrawal knowledge (rs = -0.167, p = 0.094) were weak and not significant (Table 3).
Table 3. Spearman’s rank correlation matrix (rs) for farm setting indicators.
| Indicator | Treated by | Generic Name of Drugs | Knowledge of Withdrawal Period |
|---|---|---|---|
| Treated by | 1.000 | -0.264* | -0.087 |
| Generic Name of Drugs | -0.264* | 1.000 | -0.167 |
| Knowledge of Withdrawal Period | -0.087 | -0.167 | 1.000 |
*Significant at p < 0.01. P-values: Treated by vs. Generic Name (p = 0.008); Generic Name vs. Withdrawal Period (p = 0.094); Treated by vs. Withdrawal Period (p = 0.390).
In pharmacies, seller qualification was strongly associated with prescription-based dispensing (rs = 0.488, p < 0.001). Qualification also correlated with treatment failure reporting (rs = 0.243, p = 0.013) and perceived antibiotic effectiveness (rs = 0.204, p = 0.038). Prescription-based dispensing was positively associated with perceived drug efficacy (rs = 0.459, p < 0.001). However, the most frequently sold antibiotics showed no significant correlations with other indicators (Table 4).
Table 4. Spearman’s rank correlation matrix (rs) for pharmacy setting indicators.
| Indicator | Qualification | Prescription | Most Sold AB | Works Well | Treatment Failure |
|---|---|---|---|---|---|
| Qualification of Pharmacy Seller | 1.000 | 0.488*** | -0.033 | 0.204* | 0.243** |
| Sell with or without Prescription | 0.488*** | 1.000 | 0.052 | 0.459*** | 0.143 |
| Most Sold Antibiotics | -0.033 | 0.052 | 1.000 | 0.076 | -0.043 |
| Which Antibiotic Works Well | 0.204* | 0.459*** | 0.076 | 1.000 | -0.090 |
| Treatment Failure with Antibiotics | 0.243** | 0.143 | -0.043 | -0.090 | 1.000 |
Significance: *** p < 0.001, ** p < 0.01, * p < 0.05. P-values: Qualification vs. Prescription (p =1.51 × 10-7); Qualification vs. Works Well (p = 0.038); Qualification vs. Treatment Failure (p = 0.013); Prescription vs. Works Well (p =1.52 × 10-6). All other correlations were non-significant (p > 0.05).
3.4. Cross-tabulation of drug type by source
Cross-tabulation revealed substantial differences in antimicrobial use across farms, pharmacies, and veterinary hospitals. Veterinary hospitals contributed the largest number of records (n = 330), compared with pharmacies (n = 104) and farms (n = 101), reflecting differences in reporting intensity and treatment protocols.
Professional qualification appeared to influence prescribing and dispensing practices; however, weak correlations suggest that commercial and practical factors may play a dominant role in antimicrobial use patterns (Fig. 3). Figure 4 further illustrates variations in disease distribution and antimicrobial use across sources.
3.5. Determinants of withdrawal awareness
Multivariable logistic regression showed that none of the evaluated socio-behavioral factors, including education level, information access, or farming experience, were significantly associated with farmers’ awareness of antimicrobial withdrawal periods (p > 0.05). For example, education above the Secondary School Certificate level had an odds ratio of 1.324 (95% CI: 0.101-17.331), but the association was not statistically significant (p = 0.831). Wide confidence intervals indicate substantial variability in the estimates (Table 5).
Table 5. Logistic regression results for predictors of withdrawal awareness.
| Predictor | Coefficient | Std. Error | z-score | P>|z| | Odds Ratio (OR) | 95% CI Lower | 95% CI Upper |
|---|---|---|---|---|---|---|---|
| Intercept | 0.4788 | 0.457 | 1.047 | 0.295 | 1.614 | 0.658 | 3.955 |
| Education: Above SSC vs Below SSC | 0.2807 | 1.312 | 0.214 | 0.831 | 1.324 | 0.101 | 17.331 |
| Education: SSC vs Below SSC | -0.5656 | 0.719 | -0.786 | 0.432 | 0.568 | 0.140 | 2.303 |
| Info Access: Moderate/High vs Low/No | 0.8676 | 0.756 | 1.148 | 0.251 | 2.381 | 0.541 | 10.474 |
| Experience (Years) | 0.0592 | 0.108 | 0.549 | 0.583 | 1.061 | 0.859 | 1.309 |
3.6. Hierarchical clustering of antibiotic usage
Hierarchical clustering identified three main groups of antibiotics based on sectoral frequency distributions (Fig. 5): i) antibiotics concentrated in veterinary hospitals, ii) widely used agents across farms and pharmacies, and iii) low-frequency antibiotics with limited presence across sectors. Short branch distances indicated similar cross-sector usage patterns, whereas longer branches reflected divergence in sectoral concentration.
3.7. PCA-based clustering of antibiotic usage profiles
PCA revealed distinct variance structures across sectors. In farms, PC1 (27.43%) captured variance associated with treatment provider and drug usage variables, while PC2 (13.63%) reflected livestock-type and location variation (Fig. 6. A). In pharmacies, PC1 (33.57%) was primarily associated with dispensing practices and documentation variables, whereas PC2 (18.32%) reflected perceived treatment outcomes (Fig. B6). In veterinary hospitals, PC1 (22.62%) captured drug selection variability, and PC2 (18.23%) reflected administrative characteristics (Fig. 6. C).
Scatterplots of PC1 and PC2 demonstrated visible separation of subgroups within sectors, indicating heterogeneity in antimicrobial usage profiles. Farms and pharmacies displayed more defined clustering, while veterinary hospitals exhibited broader dispersion, suggesting greater variability in institutional prescribing patterns. These separations represent statistical differentiation in covariance structure rather than evidence of causal drivers of antimicrobial choice.
3.8. Factor structure of antimicrobial practices
Exploratory factor analysis identified latent structures in antimicrobial practices.
In farms, ten factors were retained (eigenvalue >1). The first factor represented veterinary surgeon-led treatment (loading = 0.935), while the second reflected owner-led treatment and tetracycline use. Other factors captured patterns associated with fluoroquinolones, oxytetracycline, and sulfonamides (Fig. 7).
In veterinary hospitals, a more complex structure was observed, with 27 factors exceeding the eigenvalue threshold. The first three factors explained substantial variance and reflected clusters related to broad-spectrum antibiotic use, tetracycline versus sulfonamide usage, and disease-treatment associations (Fig. 7).
3.9. Network structure of antibiotic use across diseases and service provider
Network analysis examined relationships among antibiotics, diseases, and service sources (farms, pharmacies, and veterinary hospitals) (Fig. 8). The network contained 89 nodes and 174 edges, representing co-occurrence patterns among 34 antibiotics, 51 diseases, and three service settings.
Amoxicillin and Oxytetracycline showed the highest connectivity, linking multiple diseases across sectors, while Ciprofloxacin also displayed notable associations, particularly with respiratory and gastrointestinal conditions. Antibiotics such as Erythromycin, Penicillin, and Sulfonamides occupied intermediate positions, connecting smaller clusters of diseases and providers.
Source-level patterns indicated more diffuse connectivity in farms and pharmacies, suggesting broader antibiotic usage, whereas veterinary hospital prescriptions were more clustered, particularly around tetracycline and sulfonamide antibiotics. Overall, the network topology indicates that a limited number of antibiotics dominate treatment practices across diseases and service settings.
The network represents co-occurrence patterns in antimicrobial usage rather than biological resistance mechanisms.
4. Discussion
This multi-sectoral study reveals a complex and fragmented antimicrobial ecosystem in Bangladesh’s Mymensingh region, where high usage of Critically Important Antimicrobials (CIAs), significant stewardship gaps, and systemic surveillance biases collectively threaten One Health efforts to contain AMR.
The epidemiological landscape is dominated by preventable diseases like Mycoplasmosis (14.85%), Newcastle Disease (11.88%), and Lumpy Skin Disease (11.88%), linked to intensive farming, poor biosecurity, and inadequate vaccination [22,23]. This high disease burden occurs within a fragmented treatment landscape. While veterinary surgeons and paravets each provide 34.65% of treatments, a concerning 30.69% of interventions are managed by farm owners and local drug dealers [24,25]. This unsupervised access fuels a heavy reliance on CIAs, with amoxicillin (31.68%), enrofloxacin (20.79%), and ciprofloxacin (9.9%) dominating farm use, and ceftriaxone (25.96%) being prevalent in pharmacies [26,27]. The routine, often prophylactic use of these last-resort drugs for common conditions critically erodes the future veterinary therapeutic arsenal, posing a direct threat to both animal and human health. This is exacerbated by pharmacy practices, where 91.35% of antibiotics are dispensed without prescription and 48.08% of sellers report treatment failures [27]. Concurrently, veterinary hospitals rely on empirical, broad-spectrum treatments with sulfadimidine (45.45%) and oxytetracycline (37.58%) due to limited diagnostic infrastructure [28,29].
Our comparative analysis uncovered fundamental contextual disparities between farm and veterinary hospital (VH) settings. VHs predominantly manage complex, referred cases (e.g., anorexia, dog bites), while farms grapple with endemic production diseases (e.g., Mycoplasmosis, Gumboro) [30]. This epidemiological divide, coupled with a pronounced source-level bias where VHs contributed 330 records versus 104 from pharmacies and 101 from farms, highlights a critical flaw in current surveillance. National AMR systems relying solely on institutional VH data will systematically underestimate the resistance drivers circulating in the broader livestock population, necessitating integrated One Health surveillance that captures data from all parallel veterinary ecosystems [31].
The behavioural drivers of AMR also varied significantly by sector. In farms, qualified providers were significantly more likely to avoid commonly misused antibiotics (rs = -0.264, p = 0.008), yet their qualification did not translate to better knowledge of withdrawal periods [32]. This indicates that training improves drug selection but fails to address food safety compliance. In contrast, pharmacy-level data revealed that seller qualification was strongly correlated with prescription-based dispensing (rs = 0.488, p < 0.001) and awareness of treatment failures [33]. This provides a clear mandate for policy: instead of unenforceable blanket bans, a tiered licensing system should be implemented where trained paravets or certified sellers can dispense a limited list of lower-tier antibiotics, while all CIAs remain strictly veterinarian-prescribed. This approach leverages the existing workforce to maintain animal health access while enforcing stewardship.
The lack of statistical significance across all examined predictors (p > 0.05) suggests that withdrawal awareness may be influenced by factors beyond traditional socio-demographic variables. While the odds ratio for higher education (Above SSC vs. Below SSC: OR = 1.324) hinted at a positive trend, the wide confidence interval (95% CI: 0.101-17.331) underscores the high variability and lack of precision in this association. This aligns with recent observations in peri-urban farming systems where knowledge of antimicrobial resistance and withdrawal periods remains low regardless of formal education levels [34]. The low Pseudo R-squared (0.02871) and non-significant LLR p-value (0.4737) further confirm that the current model captures only a negligible fraction of the variance in awareness. This discrepancy may be attributed to “knowledge-action gaps,” where farmers may possess general information but lack the specific, technical understanding required to implement withdrawal periods effectively [34]. Alternatively, information access categorized as “Moderate/High” may still lack the quality or specificity needed to impact awareness, as drug sellers and feed vendors often the primary sources of information for smallholders, may prioritize sales over public health safety [35].
Advanced analytics provided a strategic roadmap for targeted interventions. Hierarchical clustering identified distinct “intervention units,” such as a cluster of veterinary-dominant antibiotics (Sulfadimidine, Oxytetracycline) for VH-specific guideline audits, and a cluster of broad-spectrum agents (Ciprofloxacin, Doxycycline) shared by farms and pharmacies for cross-sectoral supply chain regulation [36,37]. Principal Component Analysis further illuminated operational heterogeneity within each sector, revealing distinct farm and pharmacy archetypes defined by management practices and compliance levels, confirming that these sectors are not monoliths and require nuanced interventions [[38], [39], [40]]
The latent drivers of drug use, uncovered through Exploratory Factor Analysis, contrasted sharply between settings. The farm dataset exhibited a simpler structure centred on treatment authority (owner vs. professional), while the VH dataset revealed a complex latent space where drug choices were tightly linked to specific disease syndromes [41]. This fundamental difference between community-level autonomy and clinical complexity necessitates tailored stewardship strategies.
Most critically, the network analysis reveals a highly concentrated antimicrobial usage landscape, where a few broad-spectrum antibiotics notably Amoxicillin and Oxytetracycline, function as structural hubs. Their high connectivity across numerous diseases and institutional settings signifies their role as the dominant drivers of antimicrobial exposure in this ecosystem. This finding is critical, as the structural centrality of these agents suggests they are not just frequently used but are also the primary conduits of selective pressure across the entire network [42]. Furthermore, the analysis identified distinct, sector-specific prescribing patterns. The diffuse connectivity within farm and pharmacy networks points toward broader, less specialized usage, whereas the clustered patterns in veterinary hospitals suggest more protocol-driven therapeutic choices. This structural divergence underscores that effective antimicrobial stewardship (AMS) must be tailored to the specific context of each sector rather than applying a uniform strategy [43]. Therefore, stewardship efforts such as mandating sensitivity testing or promoting alternatives focused on these hub drugs would have a disproportionately high impact on slowing AMR emergence across the entire system, a core objective of the One Health approach [44].
In summary, the fundamental challenge extends beyond the mere reliance on CIAs to the pervasive irrationality of antibiotic choice, driven by a fragmented provider network and weak regulation. Our data-driven identification of sector-specific usage clusters and system-wide hub antibiotics provides a clear, evidence-based roadmap for prioritizing stewardship interventions where they will have the greatest impact on preserving antimicrobial efficacy for both animal and human health.
5. Limitation
The multivariate (PCA, clustering, factor analysis) and network-based approaches employed in this study were exploratory and descriptive in nature. These methods identify covariance structures, similarity groupings, and connectivity patterns within cross-sectional antimicrobial usage data. They do not establish causal pathways, transmission directionality, resistance gene dissemination, or temporal dynamics.
Because microbiological susceptibility testing, genomic resistance profiling, and longitudinal surveillance were not incorporated, structural prominence within the network should not be equated with confirmed AMR spread. Future research integrating phenotypic and genotypic resistance data will be necessary to evaluate how usage connectivity relates to resistance evolution and dissemination.
6. Conclusion
In conclusion, the fundamental challenge in Bangladesh’s livestock sector is not merely the reliance on CIAs, but the pervasive irrationality and poor predictability of antibiotic choice, driven by a fragmented provider network and weak regulation. Our data-driven identification of sector-specific usage clusters and system-wide hub antibiotics provides a clear, evidence-based roadmap for prioritizing stewardship where it will have the greatest impact. Mitigating AMR risks requires integrated One Health surveillance and coordinated, context-specific interventions that address the unique drivers in each sector of this fragmented ecosystem.
Funding
None
Ethical approval
Not required
Acronyms
| AMR: | Antimicrobial resistance |
|---|---|
| CIAs: | Critically Important Antimicrobials |
| CI: | Confidence interval |
| EFA: | Exploratory Factor Analysis |
| FMD: | Foot-and-mouth disease |
| LMICs: | Low- and Middle-Income Countries |
| LSD: | Lumpy skin disease |
| NMDS: | Non-metric multidimensional scaling |
| OR: | Odds Ratio |
| PCA: | Principal Component Analysis |
| VHs: | Veterinary hospitals |
| WHO: | World Health Organization |
Declaration of competing interest
None declared
Appendix. Supplementary materials
Download: Download Word document (35KB)
References
- [1]M. Naghavi, S.E. Vollset, K.S. Ikuta, et al.Global burden of bacterial antimicrobial resistance 1990-2021: a systematic analysis with forecasts to 2050Lancet, 404 (10459) (2024), pp. 1199-1226View in ScopusGoogle Scholar
- [2]WHO Regional Office for Europe. Action against antimicrobial resistance requires a One Health approach. 2024. https://www.who.int/europe/publications/i/item/WHO-EURO-2024-9510-49282-73655.Google Scholar
- [3]F. Chowdhury, S. Mah-E-Muneer, S. Bollinger, et al.Prevalence of colonization with antibiotic-resistant organisms in hospitalized and community individuals in Bangladesh, a phenotypic analysis: findings from the Antibiotic Resistance in Communities and Hospitals (ARCH) studyClin Infect Dis, 77 (Suppl 1) (2023), pp. S118-S124CrossrefView in ScopusGoogle Scholar
- [4]M. Samtiya, K.R. Matthews, T. Dhewa, AK. PuniyaAntimicrobial resistance in the food chain: trends, mechanisms, pathways, and possible regulation strategiesFoods, 11 (19) (2022), p. 2966CrossrefView in ScopusGoogle Scholar
- [5]IHMEThe burden of antimicrobial resistance (AMR) in BangladeshInstitute for Health Metrics and Evaluation (2023)https://www.healthdata.org/sites/default/files/2023-09/Bangladesh.pdfGoogle Scholar
- [6]E. Enshaie, S. Nigam, S. Patel, V. RaiLivestock Antibiotics Use and Antimicrobial ResistanceAntibiotics., 14 (2025), p. 621CrossrefView in ScopusGoogle Scholar
- [7]World Health OrganizationAction against antimicrobial resistance requires a One Health approach (No. WHO/EURO: 2024-9510-49282-73655)World Health Organization. Regional Office for Europe (2024)Google Scholar
- [8]A. Matheou, A. Abousetta, A.P. Pascoe, et al.Antibiotic Use in Livestock Farming: A Driver of Multidrug Resistance?Microorganisms, 13 (4) (2025), p. 779CrossrefView in ScopusGoogle Scholar
- [9]M.S. Sarker, S.C. Nath, I. Ahmed, et al.Knowledge, attitude and practice towards antibiotic use and resistance among the veterinarians in BangladeshPlos one, 19 (8) (2024), Article e0308324CrossrefView in ScopusGoogle Scholar
- [10]A.A. Masud, E.K. Rousham, M.A. Islam, et al.Drivers of antibiotic use in poultry production in Bangladesh: Dependencies and dynamics of a patron-client relationshipFront Vet Sci, 7 (2020), p. 78View in ScopusGoogle Scholar
- [11]S.M. Ahmed, N. Naher, SNBK Tune, BZ. IslamThe Implementation of National Action Plan (NAP) on Antimicrobial Resistance (AMR) in Bangladesh: Challenges and Lessons Learned from a Cross-Sectional Qualitative StudyAntibiotics, 11 (5) (2022), p. 690CrossrefView in ScopusGoogle Scholar
- [12]Bangladesh Bureau of StatisticsGeospatial Data PortalBBS, Dhaka (2023)cited 2025 Nov 10https://bbs.gov.bdGoogle Scholar
- [13]Humanitarian Data ExchangeBangladesh Administrative BoundariesHDX (2023)cited 2025 Nov 10Available fromhttps://data.humdata.orgGoogle Scholar
- [14]Python Software Foundation. (n.d.). Python Language Reference. Retrieved from https://www.python.orgGoogle Scholar
- [15]KX. PearsonOn the criterion that a given system of deviations from the probable in the case of a correlated system of variables is such that it can be reasonably supposed to have arisen from random samplingThe London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 50 (302) (1900), pp. 157-175CrossrefGoogle Scholar
- [16]RA. FisherOn the Interpretation of χ2 from Contingency Tables, and the Calculation of PJ R Stat Soc, 85 (1) (1922), pp. 87-94CrossrefGoogle Scholar
- [17]Spearman C. The proof and measurement of association between two things. 1961. https://doi.org/10.1037/11491-005.Google Scholar
- [18]RE. WrightLogistic regressionL.G. Grimm, P.R. Yarnold (Eds.), Reading and understanding multivariate statistics, American Psychological Association (1995), pp. 217-244CrossrefGoogle Scholar
- [19]K.LIII PearsonOn lines and planes of closest fit to systems of points in spaceThe London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 2 (11) (1901), pp. 559-572, 10.1080/14786440109462720Google Scholar
- [20]H. HotellingAnalysis of a complex of statistical variables into principal componentsJ Edu Psychol, 24 (6) (1933), pp. 417-441CrossrefView in ScopusGoogle Scholar
- [21]JH. WardHierarchical Grouping to Optimize an Objective FunctionJ Am Sta. Assoc, 58 (301) (1963), pp. 236-244View in ScopusGoogle Scholar
- [22]M.K. Hassan, M.H. Kabir, M. Abdullah-Al-Hasan, S. Sultana, M.S.I. Khokon, SML. KabirPrevalence of poultry diseases in Gazipur district of BangladeshAsian J Med Biol Res, 2 (1) (2016), pp. 107-112CrossrefGoogle Scholar
- [23]C. LanataStudies of food hygiene and diarrhoeal diseaseInt J Environ Health Res, 13 (sup1) (2003), pp. S175-S183View in ScopusGoogle Scholar
- [24]FAO. Veterinary Paraprofessional Competency Framework Toolkit. Rome, 2024.Google Scholar
- [25]M.C. Sartelli, T. Hardcastle, F. Catena, et al.Antibiotic Use in Low and Middle-Income Countries and the Challenges of Antimicrobial Resistance in SurgeryAntibiotics, 9 (8) (2020), p. 497CrossrefGoogle Scholar
- [26]E. Enshaie, S. Nigam, S. Patel, V. RaiLivestock antibiotics use and antimicrobial resistanceAntibiotics, 14 (6) (2025), p. 621CrossrefView in ScopusGoogle Scholar
- [27]M.A. Islam, Z. Akhtar, M.Z. Hassan, et al.Pattern of antibiotic dispensing at pharmacies according to the WHO access, watch, reserve (AWaRe) classification in BangladeshAntibiotics, 11 (2) (2022), p. 247Google Scholar
- [28]M.A. Kalam, M.S. Rahman, M.A. Alim, et al.Knowledge, attitudes, and common practices of livestock and poultry veterinary practitioners regarding antimicrobial use and resistance in BangladeshAntibiotics, 11 (1) (2022), p. 80CrossrefView in ScopusGoogle Scholar
- [29]BDVetsBDVets Online DatabaseVeterinary drugs in Bangladesh (2023)https://bdvets.com/veterinary-drugsGoogle Scholar
- [30]N.N. Farhana, R. KhatonClinicopathological investigation of Newcastle disease in commercial poultry farms at Bogura district of BangladeshAsian-Australas J Biosci Biotechnol, 9 (3) (2024), pp. 45-55CrossrefGoogle Scholar
- [31]H. Kim, J.S. Park, D. Kim, et al.Standardization of an antimicrobial resistance surveillance network through data managementFront Cell Infect Microbiol, 14 (2024), Article 1411145View in ScopusGoogle Scholar
- [32]Food and Agriculture Organization of the United NationsAntimicrobial resistance and residues in animal productsFAO, Rome (2019)Google Scholar
- [33]World Health OrganizationGlobal antimicrobial resistance and use surveillance system (GLASS) report 2022WHO, Geneva (2022)Google Scholar
- [34]A. Buckel, C. Kankya, M.A. Caudell, et al.Minding the knowledge-action gap: Results from a mixed-methods study of antimicrobial use among dairy farmers in central UgandaPLoS One, 21 (1) (2026), Article e0339969, 10.1371/journal.pone.0339969View in ScopusGoogle Scholar
- [35]M.T. Hossain, K. Rafiq, M.Z. Islam, et al.A Survey on Knowledge, Attitude, and Practices of Large-Animal Farmers towards Antimicrobial Use, Resistance, and Residues in Mymensingh Division of BangladeshAntibiotics, 11 (4) (2022), p. 442CrossrefView in ScopusGoogle Scholar
- [36]T.F. Landers, B. Cohen, T.E. Wittum, EL. LarsonA review of antibiotic use in food animals: perspective, policy, and potentialPublic Health Rep, 127 (1) (2012), pp. 4-22CrossrefView in ScopusGoogle Scholar
- [37]A.H. Holmes, L.S.P. Moore, A. Sundsfjord, et al.Understanding the mechanisms and drivers of antimicrobial resistanceLancet, 387 (10014) (2016), pp. 176-187View PDFView articleView in ScopusGoogle Scholar
- [38]H. Abdi, LJ. WilliamsPrincipal component analysisWiley Interdiscip Rev Comput Stat, 2 (4) (2010), pp. 433-459CrossrefView in ScopusGoogle Scholar
- [39]I.I. Otaigbe, CJ. ElikwuDrivers of inappropriate antibiotic use in low- and middle-income countriesJAC-Antimicrob Resist, 5 (3) (2023), p. dlad062View in ScopusGoogle Scholar
- [40]N. De Briyne, J. Atkinson, L. Pokludová, S.P. Borriello, S. PriceFactors influencing antibiotic prescribing habits and use of sensitivity testing amongst veterinarians in EuropeVet Rec, 173 (19) (2013), p. 475CrossrefView in ScopusGoogle Scholar
- [41]K. Loosli, F. Nasuwa, M. Melubo, et al.Exploring drivers of self-treatment with antibiotics in three agricultural communities of northern TanzaniaAntimicrob Resist Infect Control, 13 (2024), p. 94View in ScopusGoogle Scholar
- [42]A.C. Tompson, L. Manderson, CIR. ChandlerUnderstanding antibiotic use: Practices, structures and networksJAC-Antimicrobial Resistance, 3 (4) (2021), p. dlab150View in ScopusGoogle Scholar
- [43]S. Moya, R. Hibbard, G. Asenjo, et al.Capturing the complexity of veterinarians’ antibiotic prescribing practices in the livestock sector: A meta‑ethnography across contextsJAC‑Antimicrobial Resistance, 6 (6) (2024), p. dlae177View in ScopusGoogle Scholar
- [44]B.J.A. Berendsen, M.G. Pikkemaat, H. Yassin, HJ. van der Fels-KlerxThe role of antibiotics in the spread of resistance genes in the environmentEnviron Pollut, 204 (2015), pp. 289-295Google Scholar