Optimizing Wearable Devices in Personalized Opioid Use Disorder Treatments Under Budget Constraint

Abstract

Wearable devices have the potential to improve the effectiveness of personalized treatments for opioid use disorder (OUD) by measuring patient responses to different treatment regimens in real time. A variety of wearable devices (such as Fitbit, Garmin, EmbracePlus) with different features, sensitivities, and costs are available. Whether such devices are cost-effective to incorporate into treatments for OUD, which are under budget constraints already, and if so, how they should be used, are critical questions that we aim to address. To help answer these questions, we build a finite-horizon, non-stationary constrained partially observable Markov decision process (CPOMDP) to model patients' health states and transitions, potential interventions, treatment effects, and costs. Specifically, we explicitly account for budget constraints in our POMDP formulation to assist decision making. To facilitate solving our models, we propose a novel reformulation that identifies all optimal solutions lying on the original formulation's solution convex hull. We then demonstrate that our reformulation can be solved using a binary search in conjunction with an exact POMDP algorithm. Leveraging these elements and parameters estimated from publicly available data and past literature, we conduct a numerical study to  evaluate the value of different wearables in OUD treatments. We consider various levels of budget, wearable cost and precision, patient treatment adherence (TA), patient treatment dynamics, and two outcome metrics: the quality adjusted life days (QALD) and overdose and OUD related deaths (nOD). Our findings indicate that wearables are particularly valuable when patients exhibit differential responses to treatments across the population. Furthermore, this value is high for patients with low or moderate TA. The value is high for patients with low or moderate TA at medium budgets, and high for high TA patients at low budgets. Outside of these settings, the marginal benefit of wearables appears low relative to their cost at this time.

Keywords: POMDP, CPOMDP, Budget Constraint, Opioid Use Disorder, Personalized Treatment