Anasse Bari and Megan Coffee
Blending Infectious Disease Knowledge and Computer Science to Outsmart the Virus
By Wook Kim
One potential breakthrough in the fight against COVID-19 is a diagnostic tool that could help frontline doctors quickly determine which newly infected patients need immediate and specialized care.
Bridging the worlds of medicine and artificial intelligence (AI), this prototype represents the joint efforts of the Grossman School of Medicine’s Megan Coffee, a clinical assistant professor of medicine in the Division of Infectious Diseases and Immunology, and the Courant Institute of Mathematical Sciences’ Anasse Bari, a clinical assistant professor of computer science. The timing of their collaboration, initially focused on tackling Ebola, was fortuitous: the pair had begun pooling their knowledge just before the first COVID-19 cases were reported. Pivoting their efforts, they reached out to two hospitals in China (and later a growing number in the United States).
Their decision-support tool, Coffee says, uses “basic information that doctors collect when a patient first comes in with COVID” and “can predict who may go on to have ARDS [acute respiratory distress syndrome], which is . . . one of the more severe manifestations of COVID.” In addition to tipping off doctors to start patients on potentially lifesaving treatments, these predictions also aim to help overstretched hospitals determine how best to allocate medicines and critical bed space.
“The tool is based on a technology called predictive analytics,” explains Bari. “It looks at experience and historical clinical data and tries to learn patterns—basically, what are the combinations that we see on a patient’s laboratory values that could lead to a severe case.” Building the tool was done in conjunction with Coffee’s team and meant testing combinations of ever-improving data sets. Because its algorithms have a learning component, this AI tool is continuously becoming smarter.
The results from their initial study surprised them. Key indicators of COVID-19—fever, particular patterns in lung images, and strong immune responses—were not reliable predictors of later severe respiratory disease. “We found that the values that most stood out were body aches, elevated hemoglobin, and inflammation of the liver,” says Coffee. The tool can predict the onset of ARDS with up to 80 percent accuracy, and their teams continue to collect and analyze more data sets. But Bari already counts their efforts as a model for other such endeavors. “One of the reasons for our success is because we have an international multidisciplinary team,” he says. “A collaboration between experts in AI and infectious diseases, right there on the field where all this started.”