Food-Associated Multidrug-Resistant Pathogens in Oncology Patients: An AI-Assisted Literature Mining Study Through a One Health Lens

Background: AMR in food-associated bacterial pathogens threatens immunocompromised patients through food chain transmission. Conventional reviews lack scalability; AI-assisted pipelines offer a reproducible alternative for systematic AMR surveillance synthesis.

Objectives: To characterise resistance prevalence among food-associated pathogens in oncology-related infections using a multi-agent AI literature mining pipeline.

Methods: A three-stage agentic AI pipeline was developed. A GPT-based conversational agent received natural language queries and autonomously generated optimised PubMed search strings using validated MeSH terms. A retrieval agent then fetched abstracts via the NCBI Entrez API in batch mode. A structured extraction agent (Llama-3, Groq API) processed each abstract to extract pathogen identity, resistance phenotypes, resistance genes, cancer type, infection site, resistance rates, and study design. Data were aggregated for frequency analysis and pathogen–cancer co-occurrence mapping across 2015–2025.

Results: One thousand peer-reviewed articles were analysed. All identified pathogens are recognised foodborne or food-associated organisms. E. coli showed resistance to penicillins (81.84%), cotrimoxazole (65.79%), and monobactams (61.61%). K. pneumoniae exhibited near-total penicillin resistance (98.99%). A. baumannii demonstrated an XDR profile: carbapenems (82.58%), third-generation cephalosporins (84.10%), fluoroquinolones (80.37%), and cotrimoxazole (75.77%). Enterobacter spp. and E. faecium showed penicillin resistance of 91.77% and 90.64%, respectively. P. aeruginosa resistance to third-generation cephalosporins reached 49.41%. S. aureus exhibited macrolide resistance (55.63%) and MRSA prevalence (45.29%), concordant with livestock-associated MRSA documented in the food chain.

Conclusions: Resistance profiles identified in oncology patients mirror patterns reported in veterinary and agri-food surveillance, supporting a food-to-patient transmission hypothesis. This multi-agent AI pipeline provides a scalable, reproducible tool for cross-sector AMR intelligence aligned with One Health priorities.