COVID-19 treatment with mechanical ventilation may harm organs beyond the lungs
Judith Somekh, Nir Lotan#, Gur Arieh Yehuda, Ehud Sussman - Department of Information Systems
Machine Learning
Deep learning
Neural Networks
Health
Seed Grant 2020
Mechanical ventilators (MV) assist patients with breathing. MVs are also an invasive treatment method that can cause damage to the lungs. COVID-19 can lead to an acute respiratory distress syndrome (ARDS), a condition that in severe cases requires the intervention of a MV.
In order to evaluate if MVs may cause damage to additional organs besides the lungs, we obtained human gene expression data of six organs: skin, adipose, liver, lung, nerve-tibial and muscle-skeletal. We compared data from patients who were treated with mechanical ventilators and those who were not, for differences in activity levels of genes. We used machine learning methods to identify differences between patients of the two populations; MVs and non-MVs. We built a predictive model to identify if the sample was taken from a MV patient or from a non-MV patient. Our model predicted correctly the type of patient with above 90% accuracy for all tissues. Moreover, we detected higher inflammation and less healthy development in the MV patients’ tissues. This result confirms that when possible, alternatives to MVs should be considered for treatment of patients with major respiratory issues.
This research is supported by a seed grant from the Data Science Research Center (DSRC) at the University of Haifa, Israel. For related studies see our lab web-site.
Mechanical ventilators (MV) assist patients with breathing. MVs are also an invasive treatment method that can cause damage to the lungs. COVID-19 can lead to an acute respiratory distress syndrome (ARDS), a condition that in severe cases requires the intervention of a MV.
In order to evaluate if MVs may cause damage to additional organs besides the lungs, we obtained human gene expression data of six organs: skin, adipose, liver, lung, nerve-tibial and muscle-skeletal. We compared data from patients who were treated with mechanical ventilators and those who were not, for differences in activity levels of genes. We used machine learning methods to identify differences between patients of the two populations; MVs and non-MVs. We built a predictive model to identify if the sample was taken from a MV patient or from a non-MV patient. Our model predicted correctly the type of patient with above 90% accuracy for all tissues. Moreover, we detected higher inflammation and less healthy development in the MV patients’ tissues. This result confirms that when possible, alternatives to MVs should be considered for treatment of patients with major respiratory issues.
This research is supported by a seed grant from the Data Science Research Center (DSRC) at the University of Haifa, Israel. For related studies see our lab web-site.