Adaptive control of anticoagulation during hemodialysis

Abstract

Anticoagulation by heparin administration is the primary method employed to prevent clotting in the extracorporeal circuit during hemodialysis [1,2]. The heparin infusion regimen usually consists of a bolus loading dose followed by either additional boluses or continuous infusion into the arterial line of the dialyzer. The level of anticoagulation may be monitored by measuring the activated clotting time (ACT) for a sample of whole blood drawn from the extracorporeal circuit. In patients not at risk of systemic anticoagulation, heparin is usually administered to achieve and maintain an elevated ACT that is typically 150% of the baseline level. A series of ACT measurements obtained during the first few hemodialysis treatments a patient receives may be used in a mathematical modeling approach to determine the average heparin requirement for the individual patient [2]. Mathematical modeling of heparin requirements is particularly important when patients are at increased risk of complications from systemic heparinization. In these instances, anticoagulation is more closely monitored, and the heparin dose is decreased to keep the ACT at 125% of the baseline [3]. However, applying the mathematical modeling approach to determine the heparin requirements requires laborious calculations, or the use of nomograms, and manual recording of the heparin dosage and its timing [4]. Since the patient's heparin requirements may change over time, they should be reassessed monthly and at any time untoward clotting or hemorrhage episodes occur. For hemodialysis in acute situations, there are no past heparin dosage requirements available for guidance; it is difficult to apply the mathematical modeling approach during acute hemodialysis to maintain precise control of anticoagulation. This paper describes the use of a computer-controlled system that uses sparse ACT measurements as the basis for automatically adjusting the heparin infusion rate to provide a target level of anticoagulation. The patient receives an infusion regimen that would achieve and maintain a desired ACT in a mathematical model of the patient response to heparin infusion. The infusion regimen is delivered by means of a computer-controlled syringe pump. A menu-driven program allows the operator to enter patient data, verify parameters such as the syringe size and concentration of heparin in the syringe, and direct the operation of the adaptive control system. Before heparin administration, the patient's baseline ACT is measured, and a setpoint is determined and entered into the system. The height, weight, and sex of the patient are entered into the computer system for use in determining the initial model parameter estimates. After data entry, the operator selects a menu entry to begin drug infusion. The system then delivers the infusion regimen computed to achieve the target response in the patient model. When an ACT measurement is to be used for adaptive control, the operator selects the appropriate menu entry and enters the ACT measurement and the time at which the corresponding blood sample was drawn into the system. The system then automatically estimates the parameters for the individual patient, adapts the model parameters to the new estimates, and adjusts the infusion rate as required to move the ACT to the target level. If the measured ACT is below the target, heparin is infused rapidly to raise the model-based ACT prediction to the target level, and then a continuous infusion is begun at a rate that would maintain the target response for the model. If the measured ACT is above the target, the infusion rate is set to zero until the predicted ACT falls to the target, and then a continuous infusion is begun at a rate that would maintain the target response for the model. The model parameters are estimated using a Bayesian method in which a new parameter estimate is calculated by weighing the confidence in the ACT measurement against the confidence in the current estimate of the model parameters [5, 6]. Bayesian estimation is especially useful in situations where model parameters are estimated based on sparse measurements of patient response [7].