Multi-drug therapy design for HIV-1 infection using nonlinear model predictive control

Since anti-retroviral drugs for HIV-1 therapy have also undesirable side effects there is a trade-off in there dosage. This motivates the solution of the dynamic optimization problem that consists in minimizing the amount of drug administered to the patient while achieving the therapeutical objective of keeping the viral load bellow a specified value. Since the development of nonlinear state-space models for the HIV-1 infection, several works approached this problem using control algorithms. This work proposes a nonlinear model predictive control algorithm embedding nonlinear multirate state estimation with an extended Kalman filter. It is shown that the use of a nonlinear model yields a significant improvement in performance. The effect of the parameters that configure the controller is studied by showing their impact on performance. It is shown that the weights of the cost function can be used to select the relative amount of the different drugs administered, a fact that may be used to optimize toxicity.