Monitoring and forecasting hospitalization rates are of essential significance to public health systems in understanding and managing overall healthcare deliveries and strategizing long-term sustainability. Early-stage prediction of hospitalization rates is crucial to meet the medical needs of numerous patients during emerging epidemic diseases such as COVID-19. Nevertheless, this is a challenging task due to insufficient data and experience. In addition, relevant existing work neglects or fails to exploit the extensive contribution of external factors such as news, policies, and geolocations. In this paper, we demonstrate the significant relationship between hospitalization rates and COVID-19 infection cases. We then adapt a transfer learning architecture with dynamic location-aware sentiment and semantic analysis (TLSS) to a new application scenario: hospitalization rate prediction during COVID-19. This architecture learns and transfers general transmission patterns of existing epidemic diseases to predict hospitalization rates during COVID-19. We combine the learned knowledge with time series features and news sentiment and semantic features in a dynamic propagation process. We conduct extensive experiments to compare the proposed approach with several state-of-the-art machine learning methods with different lead times of ground truth. Our results show that TLSS exhibits outstanding predictive performance for hospitalization rates. Thus, it provides advanced artificial intelligence (AI) techniques for supporting decision-making in healthcare sustainability.

Monitoring and forecasting epidemic diseases are of prime importance to public health organizations and policymakers in taking proper measures and adjusting prevention tactics. Early prediction is especially important to restrict the spread of emerging pandemics such as COVID-19. However, despite increasing research and development for various epidemics, several challenges remain unresolved. On the one hand, early-stage epidemic prediction for emerging new diseases is difficult because of data paucity and lack of experience. On the other hand, many existing studies ignore or fail to leverage the contribution of social factors such as news, geolocations, and climate. Even though some researchers have recognized the profound impact of social features, capturing the dynamic correlation between these features and pandemics requires an extensive understanding of heterogeneous formats of data and mechanisms. In this paper, we design TLSS, a neural transfer learning architecture for learning and transferring general characteristics of existing epidemic diseases to predict a new pandemic. We propose a new feature module to learn the impact of news sentiment and semantic information on epidemic transmission. We then combine this information with historical time-series features to forecast future infection cases in a dynamic propagation process. We compare the proposed model with several state-of-the-art statistics approaches and deep learning methods in epidemic prediction with different lead times of ground truth. We conducted extensive experiments on three stages of COVID-19 development in the United States. Our experiment demonstrates that our approach has strong predictive performance for COVID infection cases, especially with longer lead times.