UKHSA using AI to identify sources of food poisoning

The UK Health Security Agency (UKHSA) is exploring how AI could be used to analyse online restaurant reviews and identify sources of food poisoning.

In a study, UKHSA researchers assessed how effective various large language models (LLMS) were at identifying potential symptoms of gastrointestinal (GI) illness from “thousands” of online restaurant reviews.

They rated each LLM’s ability to spot symptoms like abdominal pain, diarrhoea and vomiting and what food guests had eaten, with the hope that the approach could eventually be used to target investigations into foodborne illness outbreaks.

This would provide more information on rates of GI illness which are not captured by current systems, UKSHA said, as well as help identify possible sources of food poisoning outbreaks.

Steven Riley, chief data officer at UKHSA, said: “We are constantly looking for new and effective ways to enhance our disease surveillance.

“Using AI in this way could soon help us identify the likely source of more foodborne illness outbreaks, in combination with traditional epidemiological methods, to prevent more people becoming sick.”

The initiative forms part of UKHSA’s evaluation of AI to perform different tasks within public health.

While the agency has previously explored how AI could aid in identifying illness outbreaks from restaurant reviews, its most recent study involved analysing a more detailed list of terms and language that could potentially help identify their sources.

More than three thousand reviews were manually annotated by epidemiologists after being collected and filtered.

Reviews were then filtered for those containing a comprehensive list of possible GI-related keywords, which were reviewed for relevant symptoms.

More general symptoms, such as headache, fever and respiratory symptoms were not included, due to the fact these are not specific to GI illness, UKSHA explained.

One of the key challenges identified by researchers was access to real-time data.

While the LLMs enabled the team to gather general information on the type of food restaurant guests had eaten and which may be linked to illness, determining whether any specific ingredients or other factors were involved was difficult, UKHSA said.

Variations in spelling, the use of slang and people misattributing their illness to a specific meal were also identified as issues.

Riley said: “Further work is needed before we adopt these methods into our routine approach to tackling foodborne illness outbreaks.”

Meanwhile, in March 2025, NHS England announced it would roll out AI software by Cera that can be used detect the symptoms of winter illnesses like Covid, flu, RSV, and norovirus.