My top 5 papers in priority pathogens and AI from the past year

Had the pleasure of collaborating with Phil for a “Top 10 papers in priority pathogens and AI” at the IPS conference this year. We’re doing a Journal Club this coming Wednesday to cover the papers (you can register for that here). Last week, Phil posted about his top 5, so here are mine…

1. AI for SSI surveillance

The paper by Abbas et al. caught my attention because it explores whether widely available large language models (LLMs) can help automate SSI surveillance, which is one of the most time-consuming activities in IPC. Semi-automated (or fully automated) surveillance for SSI is not a new thing. In fact, many years ago, we managed to embed a semi-automated SSI surveillance system into our electronic systems. However, the platform was clunky and limited in functionality, so it didn’t really take off. The approach explored by Abbas et al. has more potential to transform our surveillance activities. The study was a small retrospective pilot study, in which five LLMs were compared with manual NHSN-based SSI surveillance in 40 abdominal surgery patients. All models performed pretty well, achieving accuracies of 90-95% and showing strong agreement with infection preventionist review. The study is limited by its small sample size, single-centre design and reliance on documentation quality, but it provides an intriguing proof of concept.

2. Patient acceptability of AI-supported surgical wound monitoring

Tanner et al. explored the acceptability of AI-supported surgical wound monitoring among patients and NHS staff. There are now loads of studies exploring the technical abilities of AI for IPC (and medicine in general) but far fewer that evaluate the views of staff and patients around the use of this sort of approach. The authors conducted a multi-centre qualitative interview study involving 20 cardiac surgery patients and 10 NHS staff who participated in an AI-enabled wound surveillance programme. Their interviews revealed strong enthusiasm from both groups, with participants highlighting the potential for earlier SSI detection, improved access to care, greater reassurance, efficiency and scalability. At the same time, respondents raised sensible concerns around data security, over-reliance on AI and the possible loss of human interaction. The study was relatively small, involved only two cardiac centres and, as a feasibility study, cannot tell us whether widespread implementation would be equally successful. However, the overall positive view of patients and staff was encouraging.

3. Air disinfection and respiratory infection in a care home setting

My next choice was the AFRI-c cluster randomised trial by Hay et al., which evaluated whether stand-alone HEPA filters could reduce winter respiratory infections in care homes.  Air disinfection technologies have attracted enormous attention since the COVID-19 pandemic, and so I was really keen to see the outcome of this one! The study was a large pragmatic cluster-randomised trial involving 91 care homes across England, comparing the use of HEPA filtration in communal areas and bedrooms against standard IPC measures over a winter season. The results were disappointing: HEPA filtration did not reduce respiratory infection rates, outbreaks, antibiotic use or staff absenteeism. The study was conducted during a period when enhanced IPC measures were already widely established, which may have limited any additional benefit from filtration. This paper goes to show that high-quality negative studies are just as important as positive ones!

4. C. auris on the rise in Turkey

The next came from Bali et al., reporting a national surveillance study of Candidozyma auris infections across Turkey. I selected this study because C. auris continues to be one of the most concerning emerging healthcare pathogens worldwide, combining multidrug resistance, environmental persistence and outbreak potential. We have been managing an outbreak of C. auris in London – in fact, one of Phil’s choices was a case-control study that we undertook showing very limited clinical impact from C. auris in our setting. This study from Turkey provides a rather stark contrast! Using data from Turkey’s mandatory HCAI surveillance system, the authors analysed all laboratory-confirmed infections reported between 2023 and 2025. More than 2,800 infections were identified, the incidence increased more than tenfold over the study period, almost all infections were bloodstream infections (the majority of which were central line-associated bloodstream infections), and mortality approached 50%. Whilst this mortality is not directly attributable to C. auris, it’s a striking figure. Given the rapid increase in prevalence, I thought that this would be a simple case of ascertainment bias through a voluntary national surveillance programme. But this is not the case – this was a mandatory surveillance programme with no chance in methodology over the study period. As with any surveillance dataset, there were limitations. The study focused on infections rather than colonisation, lacked susceptibility data and had substantial missing outcome information for some patients. Even so, the overall message is that C. auris is undergoing a sharp rise and showing its clinical teeth in Turkey.

5. Enhanced AMS and HCAI

My final selection was the stepped-wedge cluster intervention study by Classen et al., examining the impact of enhanced IPC and antimicrobial stewardship on HCAI. I chose this paper because it provides some interesting findings around the relative impact of IPC and AMS-related interventions. The study involved 15 departments across five German university hospitals and sequentially introduced enhanced IPC and AMS interventions. The most important finding, perhaps not surprisingly, was a significant reduction in Clostridioides difficile infection following implementation of the combined programme. Interestingly, the intervention did not significantly reduce VRE or third-generation cephalosporin-resistant Enterobacterales bloodstream infections. The study was conducted in hospitals that already had mature IPC and AMS programmes, baseline infection rates were low, and the follow-up period may have been insufficient to detect changes in relatively uncommon bloodstream infections. Despite these limitations, the study reinforces the need for a tight AMS programme to reduce the risk of C. difficile.

Summary

Taken together, these five papers illustrate the breadth of modern IPC. A key part of our role in a modern IPC service is to evaluate emerging technologies such as AI, interpret genomic and surveillance data, respond to the threat of multidrug-resistant organisms, understand environmental reservoirs of infection, and integrate AMS into our core prevention strategies. As always, no single study provides all the answers, but collectively they offer a valuable snapshot of where our field is heading and the questions we need to keep asking.