Online Control for Adaptive Tapering of Medications

TL;DR

This paper develops an adaptive, model-free control protocol for tapering medications that maintains well-being while reducing doses, outperforming non-adaptive methods in simulations.

Contribution

It introduces a robust integral control-based protocol for medication tapering modeled as an online optimization problem, requiring only approximate dose response knowledge.

Findings

Outperforms non-adaptive methods in simulations

Maintains well-being during tapering process

Requires no system identification

Abstract

We investigate adaptive protocols for the elimination or reduction of the use of medications or addictive substances. We formalize this problem as online optimization, minimizing the cumulative dose subject to constraints on well-being. We adapt a model of addiction from the psychology literature and show how it can be described by a class of linear time-invariant systems. For such systems, the optimal policy amounts to taking the smallest dose that maintains well-being. We derive a simple protocol based on integral control that requires no system identification, only needing approximate knowledge of the instantaneous dose response. This protocol is robust to model misspecification and is able to maintain an individual's well-being during the tapering process. Numerical experiments demonstrate that the adaptive protocol outperforms non-adaptive methods in terms of both maintenance of well-being and rate of dose reduction.