Problem definition. The opioid crisis has remained a major public health challenge in the United States for many years. This study develops a data-driven decision support framework to guide policymakers in allocating county-level budgets across multiple expenditure categories in order to address the opioid crisis. Methodology/results. We compile and curate a detailed dataset on fiscal policy and opioid-related outcomes in West Virginia (WV), the state most severely affected by the epidemic. Drawing on this dataset, we identify causal links between county-level budget allocations and two critical outcomes central to the opioid crisis, number of Opioid-Related Deaths (ORDs) and Crime Incidents (CIs). To capture these relationships, we employ Tree Ensembles (TEs) which are trained to predict outcomes as a function of budgetary decisions. We then embed the trained TEs within a Mixed Integer Linear Programming model that produces budget allocation strategies across expenditure categories that maximize the worst-case, risk-averse utility of decision makers across the two outcomes. Our results show that the presented models generate budget allocations that significantly reduce predicted ORDs and CIs across most of the 12 southern WV counties, the area hardest hit by the opioid crisis in the state. For instance, the budget allocations for Cabell County suggest that a 20% reduction in predicted values of both outcomes was possible in 2023 without increasing overall budget. Managerial implications. The proposed framework provides a systematic approach for linking fiscal policy decisions to opioid-crisis related outcomes at the county level. The robust utility optimization framework addresses uncertainty in policymakers’ priorities, increasing the applicability of the results to real-world decision contexts. Our results for 12 southern WV counties suggest that the approach can generate tailored and actionable county-level allocation strategies, supporting policymakers in re-prioritizing expenditures to manage the opioid crisis more effectively.