Hierarchical adaptive and supervisory control of continuous venovenous hemofiltration

Continuous venovenous hemofiltration (CVVH) is a lifesaving renal replacement therapy used in clinical intensive care settings. Patients undergoing a life threatening illness often develop kidney failure. CVVH performs the blood filtering process while the kidneys recover. Hemofiltration devices which are currently available operate through open-loop fluid flows induced by peristaltic pumps, and possess precisions of /spl plusmn/30 ml of fluid removed per hour. This precision is inappropriate for a neonate, who can have a total blood volume of only 150 ml, and a 15 ml blood volume reduction could cause hypovolemic shock. Because of the patient risk involved in using such equipment, most medical centers measure the fluid removed from the patient on an hourly basis. This measurement step leads to a significant increase in nursing care and therefore increases the cost of the therapy. Presented in the paper is an intelligent CVVH system that deals with accuracy and cost issues while improving the care delivered by timely responding to changes in a patient's cardiovascular state. The intelligence aspect of the system is achieved through a hierarchical control architecture for CVVH. The control methodology uses a direct adaptive control scheme for the peristaltic pumps and a supervisory control algorithm for high-level decisions on the safe operation of the system. Adaptive control of the pumps results in improved accuracy of fluid flow, while the supervisory controller provides greater autonomy which reduces the burden on clinical personnel. The performance of the proposed hierarchical controller is illustrated by experiments on a hemofiltration machine using a simulated patient. These experimental simulations showed that the measured cumulative ultrafiltration after one hour of simulation was only 1.7 ml different from the target value, which is well within the acceptable medical range.