Robust and Adaptive Anticoagulant Control

Abstract

We consider a control theory approach to adaptive dose allocation of anticoagulants, based on an analysis of records of 152 patients on long-term warfarin treatment. We consider a selection of statistical models for the relationship between the dose of drug and subsequent blood clotting speed, measured through the international normalized ratio. Our main focus is on subsequent use of the model in guiding the choice of the next dose adaptively as patient-specific information accrues. We compare a naive long-term approach with a proportional-integral-plus method, with parameters estimated by either linear quadratic optimization or by stochastic resource allocation. We demonstrate advantages of the control approaches in comparison with a naive approach in simulations and through calculation of robust stability margins for the observed data.

An Adaptive Model Predictive Controller to Address the Biovariability in Blood Clotting Response During Therapy With Warfarin

“…These methods were individualized with linear regression approaches; however, in recent years, more advanced techniques have also been investigated, including datadriven algorithms based on multiple linear regression, support vector machines, and artificial neural networks [21]. Moreover, closed-loop approaches were proposed, using deadbeat, proportional integral [29], and proportional integral plus [30] controllers. From these works, the need for a flexible but still robust technique has emerged as one of the critical requirements for automated dosing of warfarin.…”

“…From these works, the need for a flexible but still robust technique has emerged as one of the critical requirements for automated dosing of warfarin. Avery and colleagues suggested the use of MPC as an attractive solution to address this problem, both because of its ability to automatically handle constraints, and because it allows long-term strategies while keeping the flexibility to react to unexpected responses [30]. However, to the best of our knowledge, only two works employed this technique for computer-assisted warfarin administration.…”

Pole assignment control design for time–varying time–delay systems using radial basis functions

“…In part motivated by reference [7], who develop internal multimodel controllers for systems with unknown delays, the present work similarly proposes a Gaussian radial basis function based approach, here using the weighting function to switch between two or more partial models with different time delays. The new algorithm is developed and evaluated via simulation using a non-minimal state space (NMSS) framework, with pole assignment as the design criterion [8][9][10][11].…”

“…However, these are usually for controllers based on neural networks, in which the basis function is an activation function or weighting term [5][6][7]. By contrast, the present article exploits the existing NMSS modelbased design approach [8][9][10][11], adapted here in a novel way so that the input signal is a weighted sum of the partial control inputs for different time delay models. This yields a nonlinear weighted proportional-integral-plus (PIP) control system.…”

“…Each model ( 3) is represented in NMSS form and linear PIP control system design proceeds in the usual manner. Page constraints preclude full discussion of the NMSS/PIP approach here; see [8][9][10][11] and the numerous references therein. For the model ( 3), the non-minimal state vector…”