How can digital innovation support respiratory virus surveillance and monitoring without alienating low-resource settings? Here, Holly Seale, Professor of Infectious Diseases and Social Science at the University of New South Wales, Sydney, explores AI applications, infrastructure gaps, and the future of respiratory virus management.
From your perspective, what are the most important recent advances in AI-enabled respiratory virus detection and management?
One of the most promising applications of AI is in antimicrobial stewardship. Our work in rural Bangladesh has highlighted substantial unnecessary antibiotic use for respiratory infections, largely because treatment is often prescribed empirically when diagnostic testing is unavailable or results are delayed. AI-enabled diagnostics could help distinguish patients more likely to have viral rather than bacterial infections, prioritize confirmatory testing, and support more targeted antibiotic prescribing. Emerging evidence already suggests that AI-assisted stewardship systems can improve prescribing practices and reduce inappropriate antibiotic use.
We're also seeing encouraging research in rural Bangladesh evaluating AI-enabled digital stethoscopes to improve the diagnosis of childhood pneumonia while reducing unnecessary antibiotic prescribing. Importantly, these technologies are not intended to replace clinicians. Instead, they provide frontline healthcare workers with better decision support in settings where diagnostic uncertainty is high and access to conventional testing is limited.
What specific gaps in access or infrastructure are these technologies helping to address?
AI is helping address many of the "last mile" challenges in healthcare, including shortages of clinicians and laboratory professionals, long travel distances, delayed test results, limited transport infrastructure, and gaps in disease surveillance. In low-resource settings, it can extend the capacity of already stretched healthcare teams while helping prioritize scarce molecular testing for the patients most likely to benefit.
How do AI-driven tools need to be adapted to function reliably in low-resource or geographically remote settings?
When considering AI for low-resource or rural settings, the focus should not be on adapting existing tools but on co-designing and developing them with local stakeholders. Building solutions in-country ensures they are tailored to the realities of the setting, including offline or low-bandwidth functionality, low computational requirements, compatibility with basic smartphones or edge devices, and interfaces that can be used easily by non-specialists.
Local validation is equally important. A model developed for one hospital or population may not perform reliably when applied in a different setting with different patient demographics, disease prevalence, equipment, or referral pathways. Sustainable implementation also depends on investing in local expertise, ensuring healthcare professionals are trained not only to use AI systems but also to maintain, evaluate, and improve them over time.
What are the key challenges in validating and regulating AI-based diagnostic tools for respiratory viruses?
One of the biggest challenges is that AI tools often perform well in controlled research settings but less consistently in real-world practice. Respiratory infections are particularly difficult because symptoms overlap, disease patterns vary between populations and seasons, and healthcare infrastructure differs widely. As a result, a model developed in a high-resource hospital may not perform reliably in rural Bangladesh or other low-resource settings. This makes local validation essential, but it is often resource-intensive and overlooked.
Data quality is a persistent issue. How do variability and bias in input data affect the performance of AI models in this space?
Bias can arise from training data, annotation quality, site selection, device differences, patient demographics, and disease prevalence, all of which can affect AI performance. Recent reviews have shown that, if left unaddressed, these biases can reinforce healthcare disparities. Reducing bias therefore requires continuous attention throughout the AI lifecycle – from data collection and model development to validation and clinical implementation.
To what extent can AI help distinguish between respiratory pathogens with overlapping clinical presentations?
Our work is based in rural communities where access to hospitals and trained healthcare professionals is limited or costly. As a result, many people seek care from informal providers, such as village doctors or drug vendors, where access to diagnostic testing is extremely limited. Antibiotics are therefore often prescribed empirically because there is no reliable way to distinguish between viral and bacterial infections.
I don't think AI will provide a complete solution to this problem. These tools can support clinical decision-making, but they should not distract from the broader priority of improving access to trained healthcare providers and diagnostic services. AI should be viewed as part of the solution, not a substitute for strengthening healthcare systems.
Are there risks that increased reliance on AI could distance clinicians and laboratorians from the diagnostic process? How can this be mitigated?
There is sometimes too much emphasis on diagnostic accuracy alone. A tool may perform exceptionally well in validation studies but still fail in practice if it doesn't fit into existing workflows, depends on reliable internet connectivity, or adds to the workload of already overstretched healthcare workers. Particularly in low-resource settings, successful implementation and usability are just as important as algorithm performance.
Looking ahead, what developments in AI or digital diagnostics are most likely to shape respiratory virus management over the next five years?
One of the key lessons from the COVID-19 pandemic is that diagnostics support not only individual patient care but also public health surveillance. AI-enabled surveillance platforms that combine frontline diagnostic data with predictive models could help detect outbreaks earlier and identify changing respiratory virus trends, particularly in regions where traditional surveillance systems are limited or underdeveloped.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
