AI could help identify coronavirus patients at greatest risk of becoming seriously ill
- Researchers at New York University say their model shows that factors such as age and gender are not the most important indicators of danger
- Increased levels of a liver enzyme, muscle pain and haemoglobin counts were identified as the most important warning signs
Researchers hope that artificial intelligence can help identify which Covid-19 patients will become seriously ill.
An AI algorithm built by New York University researchers found that age, gender, lung image patterns and temperature are not the most useful risk indicators.
Instead, it identified a series of other indicators to predict which newly infected patients would go on to develop severe respiratory disease, according to the study published online March 30 in the journal Computers, Materials & Continua.
In partnership with Chinese doctors, the AI models used data from two hospitals in Wenzhou city, which showed that the three main risk factors were increases in the liver enzyme alanine aminotransferase; myalgias, a type of muscle pain; and rising haemoglobin levels.
The models they built were proven to have 70 to 80 per cent accuracy in predicting severe cases.
The current pandemic is overwhelming health systems around the world and triage tools that help identify those most at risk of severe illnesses may help to allocate resources.
Some patients develop the life-threatening acute respiratory distress syndrome (ARDS) within a matter of hours or days while others show only mild symptoms until they recover.
Ghost in the machine: Chinese scientists create camera that may reveal what goes on in our cells
Although statistics had shown men older than 60 with pre-existing medical conditions tend to be at higher risk, acute respiratory distress and deaths have been seen in all age and gender groups, including people who appear to be young and healthy.
The researchers found age and gender were not strong predictors of whether a patient with initial “flu-like” symptoms would develop acute respiratory distress syndrome later.
Nor were common clinical characteristics like opacities seen in lung images, fever or strong immune responses.
“Our goal was to design and deploy a decision-support tool using AI capabilities – mostly predictive analytics – to flag future clinical coronavirus severity,” said co-author Anasse Bari, a clinical assistant professor in computer science at the Courant Institute of Mathematical Sciences.
“We hope that the tool, when fully developed, will be useful to physicians as they assess which moderately ill patients really need beds and who can safely go home, with hospital resources stretched thin,” he said.
Chinese researchers develop hybrid chip design that holds promise for ‘thinking machines’
In the study, the researchers enrolled 53 patients hospitalised at Wenzhou Central Hospital and Cangnan People’s Hospital, collected their medical history, treatment and clinical outcomes data.
The AI’s predictive analytics technology used machine learning algorithms to determine which factors have the most predictive power for making accurate predictions.
They eventually listed 11 indicators, with alanine aminotransferase, myalgia and haemoglobin as the top three.
“The features that machine learning showed best predicted ARDS were not the indicators a clinician would standardly select, nor were these values grossly abnormal clinically,” the paper noted.
It said that the AI could help doctors to make rapid decisions, but that it needed further validation and refinement with more data and a wider clinical spectrum.
Sign up now and get a 10% discount (original price US$400) off the China AI Report 2020 by SCMP Research. Learn about the AI ambitions of Alibaba, Baidu & JD.com through our in-depth case studies, and explore new applications of AI across industries. The report also includes exclusive access to webinars to interact with C-level executives from leading China AI companies (via live Q&A sessions). Offer valid until 31 May 2020.
