Machine learning algorithm pipeline for endothelial damage detection and adverse outcome prediction.

Endothelial cells form the lining of the blood vessels of the entire vascular system, from the heart to the smallest capillary, regulating vascular tone, immune response and exchange of materials in and out the blood stream among others.

Endothelial damage has been observed in the early stages of most cardiovascular diseases, atherosclerosis or in patients with iflammatory and infectious diseases (e.g. COVID-19, septic shock).

Typically endothelial damage is measured by means of blood test analysis and provocative tests either invasive, performed by using pharmacological agents, or non-invasive, such as flow mediated dilation that on the other hand, suffer from high operator-independency and no-automatization.

This proposal revolves around the design, development and validation of a supervised machine learning algorithm (ML) to evaluate endothelial damage and predict adverse outcome in critically ill patients in the ICU. The ML pipeline will use as input data the one from the Horizon 2020 project VASCOVID clinical validation.

These data comprises of physiologically relevant variables that can be measured non-invasively with a completely automatized platform.

This smart platform combines diffuse optics and an automatized tourniquet for performing a reactive test on peripheral muscle (thenar muscle).

By means of this device it is possible to access in an accurate and robust way information about early impairment in perfusion, metabolic rate of oxygen consumption, and microvascular functionality and tissue capability of locally regulate the blood flow.

These variables bring an important physiological insight concerning the interpretability of machine learning algorithm from the clinical community who does not fully trust this approaches.