By Raina MacIntyre, University of New South Wales
SYDNEY, March 25 – Serious infectious diseases have become more frequent in the past decade.
Even before the Covid-19 pandemic, infectious diseases such as tuberculosis, malaria and influenza killed more than 17 million people a year. Covid-19 has claimed at least six million lives (with estimates up to 18 million), and is expected to cost US$12.5 trillion globally by 2024.
Covid-19 could have been a localised epidemic instead of a pandemic if it had been detected and acted upon early.
Using artificial intelligence (AI) to mine vast open-source data can rapidly identify serious epidemics in their early days. AI can scan global news reports and social media for signs of a new illness or unfamiliar disease symptoms in the community.
A team of medical researchers and epidemiologists at the University of New South Wales developed such a system in 2016. EPIWATCH uses machine learning and natural language processing — the ability of a computer program to understand human language as it is spoken and written — to generate epidemic early warning signals.
It then uses the datasets it produces to create a map of upcoming epidemics and provides risk-analysis tools to help health authorities mount an effective response.
A Canadian system called Blue Dot has similar capabilities but is largely for paying private clients. We have vast amounts of open-source data that can provide early warning of epidemics, but this data must be filtered to eliminate irrelevant information and improve the accuracy of forecasts.
AI-driven data filtering can detect signals of serious outbreaks much earlier than traditional public health surveillance, which relies on reporting by doctors, hospitals or laboratories.
A signal for the West African Ebola epidemic in 2014 could have been picked up using Twitter data three months before the WHO declared an epidemic, even though smartphone use in West Africa is lower than the global average.
Using mapping of outbreaks and statistical methods we can automatically generate red flags. While the field is rapidly developing and more systems are coming online, public health has been slow to adopt digital technologies.
The method used by Joshi et al. (doi.org/10.1371/journal.pone.0230322) was able to find Ebola signals in social media data three months before official alerts were issued.
However, attention to epidemic warning systems has increased since the Covid-19 pandemic, with the launch of a Center for Forecasting and Outbreak Analytics in the United States, the announcement of a Pandemic Radar in the United Kingdom, and the establishment of a WHO Hub for Pandemic and Epidemic Intelligence, part of the World Health Organization (WHO) and co-funded by the German government, all in 2021.
Several publicly available web-based applications collect open-source information based on incidents and events. HealthMap collects data on all health conditions, infectious and non-infectious, and detected an alert for a “mystery haemorrhagic fever” nine days before the WHO declared the 2014 Ebola epidemic.
The Global Public Health Intelligence Network, developed by Canadian Public Health and the WHO, provided intelligence drawn from open-source data. Had the network not been defunded in 2019, it could have picked up Covid-19 before it became a global pandemic.
ProMED-mail, a moderated site developed by nonprofit organisation the International Society for Infectious Diseases, receives alerts from health professionals about unusual, severe outbreaks and illnesses.
The site was the first to report the Middle Eastern Respiratory Syndrome coronavirus and Ebola in West Africa.
ProMED-mail has also collaborated with three international health organisations to create a closed global community of vetted public health experts, EpiCore, which gathers information on epidemics.
Systems that do not rely on expert input and moderation can also be valuable. From 2008 to 2015, Google ran Flu Trends as an influenza-forecasting tool.
Metabiotia, a biotech company, has an epidemic tracker that provides a map of outbreaks.
In 2017, the WHO began a system called Epidemic Intelligence from Open Sources, which receives feeds from open sources and other systems such as ProMED-mail.
Other approaches include citizens reporting directly to public health authorities using digital platforms.
To prevent a pandemic, time is of the essence. Even a few days’ warning can make a considerable difference to the long-term outcome.
Epidemics are characterised by exponential growth: 20 cases today could be 80 cases in three days, which could be 320 cases in six days. This rapid growth means health systems are functioning one day and collapsing only days or weeks later.
When large numbers of people are away from work, critical infrastructure is disrupted and supply chains fail.
Governments plan for pandemics and serious epidemics due to their immediate social and economic impacts. Even the relatively small SARS outbreak in 2003 cost the global economy US$54 billion.
There is also the risk of unnatural epidemics due to bioterrorism. The value of an effective early-detection system for serious infectious diseases cannot be overstated.
Raina MacIntyre is Professor of Global Biosecurity, NHMRC Principal Research Fellow and Head of the Biosecurity Programme at the Kirby Institute, University of New South Wales. She leads a research programme in control and prevention of epidemics, pandemics, bioterrorism and emerging infections.
Article courtesy of 360info.