After his talk at ESCMID Global 2026, detailing the creation of an infection prevention and control (IPC) dashboard to track the spread of healthcare-associated infections, we spoke with Augustine Y. H. Luk, DPhil (PhD) candidate in Clinical Medicine at the University of Oxford, to discuss the challenges in more detail.
At ESCMID Global, what key message were you hoping to convey about the current state of hospital outbreak detection and control?
The key message is that we are not yet making full use of the data already available to help detect outbreaks. We are now working in an increasingly digital healthcare environment. Hospitals routinely collect large amounts of information on electronic health records, including patient movement data, admission and discharge information, and records of interactions with different healthcare facilities. In principle, all of which could support outbreak detection contact tracing and infection prevention.
The challenge is that incorporating too much information can overwhelm systems and staff. The process, therefore, needs to be implemented intelligently, with greater automation to make the data interpretable, scalable, and manageable for healthcare teams.
At present, outbreak detection and contact tracing remain highly manual processes, often relying heavily on spreadsheets and labor-intensive review. Because resource and capacity are limited, investigations tend to focus primarily on high-risk areas, such as ICUs or on pathogens considered especially clinically significant.
Ideally, however, outbreak surveillance and infection control should extend beyond only these specific settings and pathogens to provide broader and more proactive monitoring across healthcare systems. Real-time data systems could help healthcare teams detect potential transmission earlier and respond before outbreaks become more difficult to control.
How are machine learning and data-driven approaches changing the speed and precision of outbreak detection in clinical practice?
Machine learning has the potential to automate much of the pattern recognition involved in outbreak detection and significantly increase speed. However, improvements in precision depend on how the system is implemented and the specific clinical use case.
Many hospitals still rely on relatively simple rule-based approaches, such as flagging case numbers that rise above a 30-day average. While useful, these systems can generate large numbers of false positives, which may increase workload for infection prevention and laboratory teams. They also often fail to account for factors such as seasonality, travel patterns, or population changes that can influence case numbers.
Machine learning approaches, when paired with high-quality data, may help address these limitations by identifying more complex patterns and incorporating a broader range of variables into the analysis. However, more data is not always inherently beneficial. Increased detection is only valuable if the findings can be translated into actionable infection control or public health responses.
What are the most critical data sources – and limitations – that laboratories must address to enable effective real-time surveillance?
In the UK, different NHS trusts use different electronic health record systems, and extracting data from them can be technically difficult, expensive, and often dependent on dedicated IT support. Accessing them in a timely and useable format can be a major barrier.
Ideally, outbreak detection platforms would access these systems in real time, but in practice, strict data governance requirements, legacy infrastructure, and fragmented data systems often prevent this. Delays in accessing patient movement and laboratory data can significantly limit the effectiveness of digital tools for outbreak detection and contact tracing.
Rapid identification of case clusters could allow earlier investigation, targeted sequencing, and faster containment measures. However, if data are only updated every few days or weekly, responses may already be too late.
In many cases, sequencing and laboratory testing capacity already exists within the UK healthcare system. The greater challenge is integrating and accessing clinical data quickly enough to support timely public health action.
How has genomic or genotypic evidence reshaped our understanding of transmission, particularly in high-risk settings such as neonatology wards?
Pathogen genomics offers a high level of specificity when identifying links between cases, and in our pilot study, clinicians and infection control teams placed a high level of trust in genomic evidence. When cases are shown to be genetically related, it provides a strong support for possible transmission.
This can be particularly valuable in high-risk settings, such as neonatal units where patients are vulnerable and consequences of infection can be severe. Sequencing can therefore be a valuable tool for rapidly identifying and containing infections. In these scenarios, laboratories may take an aggressive approach and sequence large numbers of samples to prevent further spread.
However, our ongoing studies also show that genomics alone is often not enough, especially during large outbreaks. In some investigations, we identified clusters involving more than 100 patients. In those situations, genomic relatedness may tell us that cases are connected, but it does not always explain where or how transmission occurred. It may be difficult to determine whether transmission took place within the hospital, originated in the community, or reflected contamination in specific wards, equipment, or clinical areas.
This is where integrating additional data becomes essential. Combining genomic findings with patient movement data, analyzing how each case may have overlapped in space and time while integrating other hospital information can help clarify transmission pathways and provide actionable insight to outbreaks.
What are the practical challenges of integrating sequencing into routine workflows for outbreak investigation?
One major challenge is cost. Although sequencing has become cheaper, rapid testing during outbreaks can remain expensive, particularly when batch processing is not possible.
Another challenge is the need for specialist expertise to interpret genomic data and determine which findings are clinically meaningful. Establishing reliable thresholds for relatedness requires large validation studies and detailed epidemiologic analysis.
Speed is also critical. If sequencing results are not available quickly enough to guide infection prevention and control measures, they become more useful for retrospective analysis than real-time outbreak management. For sequencing to change routine practice, it needs to be embedded into workflows in a way that produces timely, interpretable, and actionable outputs.
What makes an IPC dashboard truly effective in tracking and interrupting transmission?
The way I see it, an effective IPC dashboard removes much of the manual workload involved in contact tracing while aiding interpretability. Traditionally, infection prevention teams have to manually review patient records to track where patients have been, who they interacted with, which staff cared for them, and even which equipment or facilities they used. This is extremely time consuming and resource intensive, especially because most infection prevention and control teams are relatively small.
A dashboard can automate much of this data collection and rapidly presents the most relevant information in one place. It also helps address another major problem in healthcare systems: fragmented data sources. Patient movement records, laboratory results, GP data, and other clinical information are often stored across separate systems that do not communicate efficiently with one another. Integrating these data streams into a single real-time platform could help teams to identify transmission risks and investigate more cases more quickly.
Speed is particularly important because traditional investigations can take days, by which time infections may already have spread further. In a real-time model, if a patient tests positive for an organism such as Clostridioides difficile, the system could immediately flag the case, identify relevant contacts, and support rapid containment measures or targeted testing.
The approach also reflects a broader focus on pandemic preparedness following COVID-19 and earlier outbreaks such as SARS. One key lesson from these events was that hospitals can become major sites of transmission, making rapid identification and containment critical for preventing wider spread into the community.
Where do you see AI adding the most value beyond detection?
AI is particularly useful for pattern recognition and can process huge volumes of data far faster than humans. Even as an early warning or alert system, it could significantly improve infection surveillance.
The challenge, however, is ensuring that the information remains actionable and does not overwhelm healthcare teams with excessive alerts or unnecessary information that may increase workload rather than reduce it.
Beyond detection, there is also potential for prediction. In our study, we constructed patient contact networks as graphs, which can potentially be used to train AI systems to predict which patients are most likely to become infected next. Factors such as duration of contact, shared healthcare teams, and patient vulnerability to infection could all contribute to these predictions.
This could help laboratories and infection prevention teams prioritize resources more effectively. For example, if sequencing capacity is limited, predictive models may help identify which patients are most likely to yield clinically important findings and, therefore, should be tested first.
What safeguards are needed to ensure these tools are trustworthy, interpretable, and usable in laboratory and clinical settings?
Any tool that influences clinical decision-making moves closer to being considered a medical device, which means it requires strict regulation and rigorous validation.
For the dashboard itself, the risk is lower because it mainly reorganizes and presents existing clinical and laboratory data in a more accessible way. The underlying information is completely traceable and verifiable by clinicians. Even so, the system still needs to be reliable, secure, and carefully tested in real workflows.
The ethical and regulatory challenges become more significant when AI moves beyond detection into prediction. For example, if an AI model predicts that a patient is likely to have an infection and recommends isolation, incorrect predictions could have important clinical, operational, and personal consequences.
To address this, safeguards need to include rigorous validation studies, compliance with medical device standards, prospective clinical testing, and continuous human oversight. Qualitative studies examining how clinicians interact with these systems are also important to ensure that unintended consequences or workflow issues are identified early. Ultimately, these tools should support clinical judgment, not replace it.
What one change would most improve how laboratories and healthcare systems anticipate and prevent outbreaks?
One important step forward would be improving access to healthcare data, particularly through real-time integration between hospital systems. Direct and secure access to electronic health records and laboratory systems could make outbreak detection tools far more effective and responsive.
I also think the healthcare impact of outbreak detection is sometimes underestimated because attention tends to focus on only the highest-risk pathogens or patient groups. However, infections such as resistant E. coli outbreaks or catheter-associated infections still place a major burden on healthcare systems in terms of cost, workload, and patient outcomes, even if they do not always make headlines.
Implementing broader surveillance systems does require investment in sequencing, software infrastructure, and specialist staff to interpret and act on the data. More health economics studies demonstrating the long-term value and cost savings of early outbreak detection could help healthcare leaders see these systems as worthwhile investments.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
