Deciding when to intervene: a Markov decision process approach

The aim of this paper is to point out the difference between static and dyn amic approaches to choosing the optimal time for intervention. The paper de monstrates that classical approaches, such as decision trees and influence diagrams, hardly cope with dynamic problems: they cannot simulate all the r eal-world strategies and consequently can only calculate suboptimal solutio ns. A dynamic formalism based on Markov decision processes (MPPs) is then p roposed and applied to a medical problem: the prophylactic surgery in mild hereditary spherocytosis. The paper compares the proposed approach with a s tatic approach on the same medical problem. The policy provided by the dyna mic approach achieved significant gain over the static policy by delaying t he intervention time in some categories of patients. The calculations are c arried out with DT-Planner, a graphical decision aid specifically built for dealing with dynamic decision processes. (C) 2000 Elsevier Science ireland Ltd. All rights reserved.