SpreadIt News | Digital Newspaper

New study where machine learning model predicts Covid survivability using blood tests

A single blood sample from a severely ill COVID-19 patient was utilized in a new study for analysis by a machine learning model that analyses blood plasma proteins to predict survival.

The research was published in the ‘PLOS Digital Health Journal.’

Advertisement

Healthcare systems throughout the globe are struggling to handle a large number of critically sick COVID-19 patients who require specialized medical treatment, particularly if they are classified as being at high risk. Clinically accepted risk assessments in critical care medicine, such as the SOFA or APACHE II, have relatively little predictive value for COVID-19.

Researchers examined the levels of 321 proteins in blood samples obtained at 349-time points from 50 severely sick COVID-19 patients being treated in two different health care facilities in Germany and Austria in the current study. To discover connections between the measured proteins and patient survival, a machine learning technique was applied.

Advertisement

The group lost fifteen patients; the average duration from admission to death was 28 days. The median length of stay in the hospital for those who survived was 63 days. The researchers identified 14 proteins that altered in different directions over time in patients who survived intensive care vs. those who did not survive.

The researchers then created a machine learning model to predict survival based on a single time-point measurement of key proteins, which they evaluated on an independent validation cohort of 24 severely sick COVID-10 patients. In this cohort, the model had good predictive power, correctly predicting the outcome for 18 of 19 patients who survived and 5 of 5 patients who died (AUROC = 1.0, P = 0.000047).

Advertisement

The researchers concluded that blood protein tests if validated in larger cohorts, might be beneficial in both identifying patients with the highest mortality risk and determining whether a specific treatment affects an individual patient’s expected trajectory.

Advertisement