黑料吃瓜群网

A new experimental AI tool, developed by (NYU)* could predict which COVID-19 patients will go on to contract acute respiratory distress syndrome (ARDS), before any severe symptoms emerge.

ARDS, a life-threatening complication of COVID-19 which causes fluid to build up in the lungs, affects around 40% of those hospitalised with severe cases of the novel coronavirus.

Until now there has been little way of predicting the likelihood of ARDS among newly diagnosed COVID-19 patients. In fact, many have described a COVID-19 diagnosis as a 鈥渓ottery鈥� 鈥� you either get severe respiratory symptoms or you don鈥檛.

鈥淲e were looking to develop a tool to reliably identify who will develop ARDS when they first present," explained co-author Dr Megan Coffee, a clinical assistant professor of infectious diseases and immunology.

The challenge for the researchers was that, although most ARDS patients are male, most men do not develop ARDS. In addition, risk factors like late age and compromised immunity have been shown to predict death as a result of ARDS, but not reliably indicate who will develop the disease or further complications.

Given the 50% overall mortality rate of ARDS, acceptance of this seemingly 鈥榣ottery-like鈥� scenario could increasingly put lives in danger, as ambulance services and ICU wards stretch thinner towards the COVID-19 peak.

In contrast, this new AI tool could reverse this trend, by identifying high-risk patients early. In turn, it could help doctors decide which patients should be given a bed versus sent home; meaning that resources are better deployed and more lives saved.

The study analysed laboratory, demographic and radiological findings from 53 patients who had a coronavirus diagnosis and only mild symptoms. The researchers designed complex computer models and fed this data into them.

Through designing and tracking a series of 鈥榙ecision trees鈥� they were able to test the efficacy of the AI at making predictions, based on these various datasets. They found that the more data the AI consumed, the more 鈥榢nowledgeable鈥� it became, and the more accurate its predictions were.

Contrary to expectations, seemingly obvious risk factors 鈥� like ground glass opacities (lung image patterns), fever and an over-compensating immune response 鈥� were not reliable indicators of a later ARDS diagnosis.

Instead, reported myalgia, haemoglobin levels and levels of alanine aminotransferase (ALT) 鈥� a liver enzyme 鈥� were highly predictive of severe respiratory illness, with more than 80% accuracy.

鈥淭he AI can support physicians in their clinical decision-making,鈥� said fellow co-author Dr Anasse Bari, a clinical assistant professor of Computer Science at NYU.

鈥淲hile we still need to validate this tool on large datasets before deploying it, the approach holds the promise as another AI tool to assist clinicians in identifying which patients might develop severe cases and need immediate care.鈥�

*The tool was developed by researchers at the NYU Grossman School of Medicine and NYU Computer Science at the Courant Institute.

Image credit:聽漏stock.adobe.com/au/Halfpoint