With insights from ESCMID Global 2025, Gerry Hughes explores how artificial intelligence is supporting the management of infectious diseases, as well as the associated risks and future directions.
Artificial intelligence (AI) is reshaping infectious disease care, from the diagnostic laboratory to antimicrobial stewardship at the bedside. At ESCMID Global 2025, there was widespread agreement that AI’s greatest value lies in its ability to support, not supplant, expert clinical judgment.
Researchers and healthcare professionals showcased practical applications where AI can help to address early diagnosis, outbreak detection, antimicrobial prescribing and information dissemination.
AI roles across the infectious disease diagnostic pathway
AI is driving tangible improvements in diagnostic and documentation workflows, potentially enhancing speed, accuracy and scalability in both laboratory and clinical settings.
Presenting work from Monash University in Melbourne, Australia, at ESCMID Global, Dr Nenad Macesic and colleagues explored how AI could extract more value from matrix-assisted laser desorption/ionisation time-of-flight (MALDI-TOF) mass spectrometry. Although widely used for bacterial species identification, this technology has previously not been reliable for strain typing.
The Monash team trained a neural network using nearly 2,000 clinical bacterial isolates using MALDI-TOF and whole genome sequencing, which is the gold standard analysis tool.
For common pathogens like E. coli, S. aureus, and E. faecium, the AI model achieved strong accuracy, particularly when paired with high-resolution genomic data. While further work is required, the findings demonstrated the potential application of AI-augmented MALDI-TOF analysis of clinical bacterial strains in practice.
Complementing this, a European collaborative led by researchers at the University of Oldenburg in Germany, presented CarbaDetector – a free web-based research tool that uses machine learning to predict carbapenemase-producing Enterobacterales from basic disk-diffusion results.
In external testing, it showed 97.8% sensitivity, with higher specificity (56.6%) than the reference antibiogram. The researchers suggested that CarbaDetector may help reduce unnecessary confirmatory testing in clinical labs.
A third project presented by Justin Xu, a PhD candidate in machine learning at the University of Oxford’s Big Data Institute, UK, demonstrated how AI can support the interpretation of chest X-ray reports. His team used GPT-4 to convert unstructured radiology text into structured formats, enabling further classification by a mathematical model.
Using large publicly available datasets, the model achieved high performance in identifying infectious disease-related findings. Future work on this approach is set to apply similar AI methods to infectious disease consultation notes, aiming to improve diagnosis and documentation while reducing the costs of manual data annotation.
Smarter stewardship through decision support
Professor Shruti Gohil, associate medical director, hospital epidemiology at the University of California, Irvine in the US, discussed results from the INSPIRE antimicrobial stewardship trials. These studies evaluated how context-aware clinical prompts integrated into electronic healthcare systems could improve empiric antibiotic choices.
Unlike traditional alerts, the prompts were iteratively developed over a one-year period to ensure user acceptability and to reduce known challenges such as alert fatigue. They used local microbiological data and patient-specific risks to tailor advice. Importantly, they did not require clinicians to respond with a simple ‘yes’ or ‘no’ but provided actionable recommendations and tracked prescriber actions arising from the alerts.
Publications arising from these studies, for pneumonia and skin and soft tissue infections, reported optimised extended-spectrum antibiotic selection without an increase in adverse events. Professor Gohil noted the importance of accompanying the prompt rollout with an educational campaign to support adoption.
Generative AI and infectious disease knowledge
Generative AI represents a transformative force in medicine, comparable in impact to the introduction of electricity or the internet, Professor José Ramón Paño-Pardo, attending physician at the University Clinical Hospital of Zaragoza in Spain said during his ESCMID Global session.
While not a substitute for clinical judgment, he continued, large language models (LLMs) like ChatGPT can enhance diagnostic reasoning and accelerate knowledge dissemination. He brought attention to the results of a recent randomised trial demonstrating the impact of LLMs on diagnostic decision-making.
In that study, 50 physicians in family, internal and emergency medicine found that access to a LLM did not significantly improve diagnostic reasoning compared with conventional resources alone.
Physicians using the LLM performed similarly to those relying solely on standard tools. Interestingly, the LLM outperformed physician groups when assessed independently, suggesting that while the technology shows promise, further development is needed to effectively support AI-mediated clinical reasoning in practice
Professor Paño-Pardo cautioned that generative AI tools, while capable of synthesising information and streamlining documentation, must be subject to rigorous clinical oversight. Their outputs are not inherently reliable and should support rather than supplant the judgement of infectious disease specialists, he said.
ChatGPT as writing assistant, not author
Recent European guidance on the responsible use of generative AI urges the research community to apply such tools with transparency and caution – particularly in writing and manuscript development. The guidance advises against using AI in sensitive tasks such as peer review, and calls for robust oversight, training and a clear commitment to research integrity.
Professor Erlangga Yusuf, a consultant in medical microbiology at the Erasmus Medical Centre in Rotterdam, Netherlands, echoed these recommendations in using generative AI to support scientific writing.
During his ESCMID Global session, he highlighted that while these tools struggle with technical tasks like data visualisation or systematic literature reviews, they excel at improving grammar, clarity and flow.
He reported successfully using ChatGPT to rephrase manuscript sections, align writing with journal guidelines and refine logical progression of concepts and ideas. However, he cautioned researchers to stay actively involved in the writing process, ensuring outputs remain accurate and meaningful.
Key limitations of partnering with LLMs in academic writing include hallucinated references, a lack of true comprehension and potential data privacy risks, he warned. Professor Yusuf encouraged researchers to treat LLMs as collaborators rather than authors and maintain rigorous oversight of AI-generated content.
Conclusion
This year’s ESCMID Global stimulated a deeper appreciation of how AI can be responsibly integrated into infection services. Whether in the form of predictive modelling, structured prompts or support with documentation, the technology is here – and the likelihood is that its role in research and healthcare will only expand.
However, one particular message was abundantly clear: AI is not a shortcut. It demands validation, oversight and transparency. When implemented with rigour and accompanied by education and governance, AI can enhance clinical excellence.
For infection specialists, now is the time to engage with these tools, shape their evolution and ensure they serve the needs of both patients and professionals alike.
