New Machine Learning Tool Can Help Predict COVID-19 Death Risk

Researchers from Yale University developed the multiscale PHATE machine learning tool that provides a detailed analysis of millions of immune cells and information regarding which type could lead to COVID-19 mortality.

During the COVID-19 pandemic, the researchers discovered information about the novel coronavirus and how it affects the human body. They found that specific immune system cells, such as T cells and antibody-producing B cells, are usually protective against pathogens that cause COVID-19. However, the Multiscale PHATE tool, which utilizes an algorithm called PHATE, found that certain activity in these cells can lead to death in some cases.

The tool analyzed 163 patients, a total of 55 million blood cells, all infected with COVID-19 at the Yale-New Haven Hospital.

The main findings were that T cells are generally protective against COVID-19, while two white blood cells known as granulocytes and monocytes were more likely to link to death.

But, on a more granular level, researchers found that when clustered with immune system cells IL-17 and IFNG, another T cell known as TH17 could lead to mortality.

“Machine learning algorithms typically focus on a single resolution view of the data, ignoring information that can be found in other more focused views,” said Manik Kuchroo, a doctoral candidate at Yale School of Medicine who helped develop the technology and is co-lead author of the paper, in the press release. “For this reason, we created Multiscale PHATE which allows users to zoom in and focus on specific subsets of their data to perform more detailed analysis.”

With regard to predicting death risk, the Multiscale PHATE tool had an accuracy level of 83 percent, researchers concluded.

Use of machine learning to analyze diseases such as COVID-19 is on the rise.

For example, a recent study from Infectious Diseases of Poverty used various machine-learning techniques to examine how they predicted COVID-19 infection risk. The researchers concluded that one of the three tested techniques could accurately predict when a person is infected with the deadly disease.  

Another recent study from the Association for Professionals in Infection Control explained how researchers used machine-learning algorithms to predict which patients will develop Clostridiodes difficle infections within a hospital. Similar to the above study, one provided a higher level of accuracy of the various methods they tested.

Machine learning use has also been proven effective within surgical procedures. A 2021 study described how machine learning can help monitor patients under anesthesia, providing doctors with better insight into the amount of a drug a patient may need.