AI-powered blood test set to predict survival prospects of Covid-19 patients

Researchers have developed a way to predict if a person will survive after developing Covid-19 – and it involves taking a drop of blood. Found to be 96 per cent accurate, the method could help doctors decide who to prioritise for the most intensive treatments. Artificial intelligence has been used to develop the technique, which involves tests for a number of proteins in the blood. One of the study’s senior authors, Dr Florian Kurth, of Charity-University Medicine Berlin, said at present it was “difficult” for doctors to assess whether an individual patient will deteriorate or die from Covid-19. It was, he said, “especially challenging” to work out the probable outcome among patients who currently have similar levels of the disease. Based on proteomics – the large-scale study of proteins – the new method involves looking at levels of 14 proteins in a drop of a patient’s blood. “The proteomic risk prediction was far better than the prognosis derived from established risk-assessment scores, which are routinely used in clinical care,” Dr Kurth said. Machine learning, a form of AI, was used to find out which out of more than 300 proteins in the blood could predict how a patient’s disease will develop. It identified 14 proteins “that showed trajectories different between survivors and non-survivors”, in the words of the new study. This latest application of AI comes after many other initiatives to employ the technology to combat the pandemic and heralds what is expected to be its greater use in medicine as a whole. Other examples include an AI-based method developed by New York University Abu Dhabi researchers that uses information from, for example, chest X-rays to help forecast a Covid-19 patient’s outcome over the next 96 hours. The technology has also been used to design coronavirus vaccines and to identify which drugs out of thousands being used to treat other conditions might be effective at combating Covid-19. AI has also been used by scientists to combine data on Covid-19 patients from many hospitals, which helps to make research more reliable. In the new study the researchers said most proteins that could work out a patient’s prognosis were involved in clotting or were part of the immune system called the complement cascade. The clotting system is known to be important, the researchers said, because a “substantial proportion” of patients severely ill with Covid-19 develop thromboembolic events, which involve a blood clot that has been dislodged from one part of the circulatory system causing a blockage in another blood vessel. “Our study demonstrates that plasma proteomics can give rise to prognostic predictors substantially outperforming current prognostic markers in intensive care,” the researchers wrote in Plos Digital Health. They found that the method accurately predicted the outcome for 18 out of 19 patients who were seriously ill with Covid-19 but went on to survive, and five out of five patients who died. John Oxford, an emeritus professor of virology at Queen Mary University of London and co-author of the textbook Human Virology , said better ways of knowing whether a person is at risk could be helpful in deciding, for example, whether to give antiviral drugs. “It would be excellent if there were any testing system that could harden up prognosis and give real data on how serious this could be for that particular individual,” said Prof Oxford, who is not connected to the new study. The new technique has not been made available for use in hospitals but Dr Kurth said the researchers were planning to find out if it could be used by standard testing laboratories, which could lead to its wider adoption. “We also want to assess the method further in larger cohorts of patients and possibly also in other diseases,” he said.

AI-powered blood test set to predict survival prospects of Covid-19 patients

New method could help medics to develop treatment plans accordingly