An artificial intelligence-powered system is being used to identify the spread of debilitating illness in communities earlier, with the tool able to detect Melbourne’s thunderstorm asthma outbreak up to nine hours before it was officially reported.
Developed by researchers from CSIRO’s Data61 and UNSW Sydney’s Kirby Institute, the system uses Natural Language Processing and anonymised and publicly available Twitter data to identify symptom keywords on the social media channel.
In the case of the thunderstorm asthma outbreak, terms related to respiratory distress such as ‘coughing’ or ‘breathing’ were assessed.
The severe weather event, which occurred three years ago today, inundated emergency services and hospitals in Melbourne, resulting in over 8,000 hospital admissions by 6pm on November 21st, 2016. The Victorian Department of Health and Human Services attributed nine deaths to the event.
A key challenge in the case of acute disease events is to detect them as soon as possible to assist health agencies to respond swiftly in emergency situations, explains Data61 Postdoctoral Fellow and tool co-creator Dr Aditya Joshi.
“The popularity of social media makes it a valuable source of information for epidemic intelligence,” Dr Joshi said.
“We can draw upon informal sources such as social media data to understand how acute disease events occur, and we can detect when and where an outbreak is likely to occur. This means hospitals and public health agencies can be as prepared as possible.”
The technique combines two fields of artificial intelligence — natural language processing and statistical time series modelling — and a four-step process to ensure the tweets containing the keywords were indeed reports of health conditions and to remove duplicates where an individual might tweet more than once about their condition.
Natural language processing, or NLP, is the ability of a computer program to process human language. The tool uses NLP based on word embeddings, to distinguish between symptoms and unrelated mentions of the keywords.
“Being able to distinguish between tweets that are reports of illness versus tweets which only contain certain keywords is the greatest strength of this tool, because past work in epidemic intelligence has not been able to do this,” said Dr Joshi.
Traditionally, hospitals would report an influx of patients suffering from similar symptoms, however, tweets assessed during this case study focused on early symptoms, such as a cough, which individuals would most likely not consider serious enough to visit a doctor or hospital, but impactful enough to publicly post about on social media.
“This could also be a useful tool for emergency services to better plan their ambulances and emergency staff if they are aware of an impending outbreak.”
Professor Raina MacIntyre, Head of Biosecurity Research Program, Kirby Institute, UNSW Sydney said that this work is a remarkable contribution to public health research.
“Rapid epidemic intelligence systems can give us very early signals of epidemics of other disease events – much sooner than we would otherwise find out,” said Prof MacIntyre.
“Epidemics can grow very rapidly, so the sooner we can detect them, the easier they are to control.”
In future, this system can be used to provide health authorities and the community early warning of a serious and sudden health event, she explained.
“Early detection could significantly improve our capability to mitigate the impact of epidemics.”
The tool can also be used to detect other outbreaks such as Influenza, Ebola and the Zika virus.
More information about thunderstorm asthma and how to stay safe can be found here.
About the research
The paper, Harnessing Tweets for Early Detection of an Acute Disease Event was co-authored by Aditya Joshi , Ross Sparks, James McHugh, Sarvnaz Karimi, Cecile Paris , C. Raina MacIntyre and was published in the Epidemiology Journal.
About the 2016 thunderstorm asthma outbreak in Melbourne