Policy4OOD: A Knowledge-Guided World Model for Policy Intervention Simulation against the Opioid Overdose Crisis

TL;DR

Policy4OOD tackles the challenge of evaluating opioid policies by introducing a knowledge-guided world model that blends policy knowledge graphs, spatial spillover modeling, and temporal dynamics. The approach supports forecasting under proposed policies, counterfactual reasoning about alternative decisions, and optimization over intervention portfolios via Monte Carlo Tree Search, all within a unified simulator. A state-level monthly dataset spanning 2019–2024 is constructed, integrating opioid mortality, socio-economic indicators, and structured policy encodings to train the model. Empirical results show improved forecasting accuracy, meaningful counterfactual insights (e.g., effects of policy timing in Tennessee), and practical policy optimization capabilities, highlighting the framework’s potential to support proactive, data-driven public health decision making.

Abstract

The opioid epidemic remains one of the most severe public health crises in the United States, yet evaluating policy interventions before implementation is difficult: multiple policies interact within a dynamic system where targeting one risk pathway may inadvertently amplify another. We argue that effective opioid policy evaluation requires three capabilities -- forecasting future outcomes under current policies, counterfactual reasoning about alternative past decisions, and optimization over candidate interventions -- and propose to unify them through world modeling. We introduce Policy4OOD, a knowledge-guided spatio-temporal world model that addresses three core challenges: what policies prescribe, where effects manifest, and when effects unfold.Policy4OOD jointly encodes policy knowledge graphs, state-level spatial dependencies, and socioeconomic time series into a policy-conditioned Transformer that forecasts future opioid outcomes.Once trained, the world model serves as a simulator: forecasting requires only a forward pass, counterfactual analysis substitutes alternative policy encodings in the historical sequence, and policy optimization employs Monte Carlo Tree Search over the learned simulator. To support this framework, we construct a state-level monthly dataset (2019--2024) integrating opioid mortality, socioeconomic indicators, and structured policy encodings. Experiments demonstrate that spatial dependencies and structured policy knowledge significantly improve forecasting accuracy, validating each architectural component and the potential of world modeling for data-driven public health decision support.

Figures (5)

• Figure 1: (a) The opioid overdose crisis emerges from a complex, dynamic socio-economic system with interacting risk factors and cross-region spillovers. (b) Policy interventions act through multiple, heterogeneous mechanisms, whose effects propagate across intermediate factors and regions. (c) Policy4OOD formulates opioid policy evaluation as a world modeling problem, enabling policy-conditioned forecasting, counterfactual analysis, and policy optimization within a unified framework.
• Figure 2: The schema of the spatial-temporal state graph.
• Figure 3: Overview of Policy4OOD. (a) Policy documents are converted into a policy knowledge graph, together with state-level socioeconomic time series and a cross-state dependency graph. (b) The model jointly encodes policy mechanisms (what), spatial dependencies and spillover effects (where), and temporal dynamics of policy impacts (when), producing a unified policy-conditioned state representation. (c) The learned world model supports forward forecasting, counterfactual reasoning through policy replacement, and policy optimization via MCTS.
• Figure 4: Counterfactual reasoning analysis in Tennessee. (a) Policies used. (b) Predicted overdose death trajectories.
• Figure 5: Case study for policy optimization in Tennessee and Virginia. (a) Candidate policy pool. (b-c) Original and MCTS-selected policies for each state. (d-e) Predicted opioid overdose death trajectories under selected policies.