Safe automation of interfering medical treatments via control barrier functions and reachability analysis: a fluid resuscitation-sedation-vasopressor infusion case study.

Abstract

Despite advances made toward the automation of medical treatments, existing work has predominantly focused on the automation of individual medical treatments while overlooking interferences between them. In our prior work, we developed an automation architecture to administer interfering medical treatments with safety, which (i) monitors internal physiological state of a patient using an extended Kalman filter, (ii) mediates medical treatments to bound the estimated internal state within a prescribed safe regime based on control barrier functions, and (iii) treats the patient to a prespecified treatment target through a number of intermediate targets derived from reachability analysis. The goal of this work was to investigate the scalability of this automation architecture in more complex and challenging medical treatment scenarios with a number of conflicts. Using a critical care resuscitation scenario including fluid resuscitation and intravenous sedative-vasopressor infusion, we examined if our automation architecture can achieve treatment goals while ascertaining the safety of internal state in a large number of diverse in silico patients. The results suggested that (i) the extended Kalman filter could continuously monitor a patient's internal physiological state, (ii) the control barrier functions could mediate interfering medical treatments and protect patients against unsafe internal physiological state, and (iii) the reachability analysis could treat a patient as closely as possible to a treatment target while ensuring the safety of the patient's internal physiological state, all despite complex and entangled conflicts between them. Our automation architecture may provide a viable means to autonomously de-conflict interfering medical treatments.