Pain management is an international health issue. The Eugene McDermott Center for Pain Management at the University of Texas Southwestern MedicalCenter at Dallas conducts a two-stage interdisciplinary pain management program that considers a wide variety of treatments. Prior to treatment (stage 1), an evaluation records the patient's pain characteristics, medical history and related health parameters. A treatment regime is then determined. At the midpoint of their program (stage 2), an evaluation is conducted to determine if an adjustment in the treatment should be made. A final evaluation is conducted at the end of the program to assess final outcomes.

The structure of this decision-making process uses dynamic programming (DP) to generate adaptive treatment strategies for this two-stage program. Our stochastic DP formulation considers the expected final outcomes when determining treatment. An approximate DP solution method is employed in which state transition models are constructed empirically based on data from the pain management program, and the future value function is approximated using state space discretization based on a Latin hypercube. The state transition probabilistically models how a patient's pain characteristics change from stage 1 to stage 2. The optimization seeks to minimize pain while penalizing excessive.