Optimal region-specific social distancing strategies in a complex multi-patch model through reinforcement learning

Although non-pharmaceutical interventions such as social distancing have proven effective in curbing outbreaks, they also carry economic consequences. This poses a dilemma for policymakers striving to find a balance between disease control and economic burden. This delicate balance varies regionally, influenced by non-epidemiological factors such as population movements, socio-demographic characteristics, and the intricacies of social distancing policies. These factors interact in intricate ways, shaping the transmission dynamics of COVID-19. To address this complexity, we propose an innovative approach utilizing deep reinforcement learning (RL). This method assists in tailoring intervention policies for diverse regions, taking into account their unique dynamics. We incorporate South Korea’s social distancing policies and their economic impact into a RL framework with a multi-region epidemic model, offering a comprehensive solution. We integrate official mobility data and GDP specific to each region, employing the proximity policy optimization algorithm to determine the most appropriate region-specific social distancing policy. The algorithm’s reward function considers both outbreak control and economic impacts, providing policymakers with the flexibility to fine-tune the balance between these two factors according to their preferences. This adjustment can be performed across three distinct cost scenarios: High, Base, and Low-cost scenarios. In scenarios with High-costs, social distancing measures are aimed at regions with extensive connectivity and higher transmission rates. When costs are moderate, policies center around the period of peak prevalence, illustrating adaptable strategies in areas characterized by high transmission rates, budget limitations, and population mobility. In situations with Low-costs, these measures encompass most regions, excluding those with low transmission rates. The study’s results support focused interventions in specific regions to balance outbreak control and economic impact mitigation.