Conference proceeding
End-to-End Risk-Aware Reinforcement Learning to Detect Asymptomatic Cases in Healthcare Facilities
2024 IEEE 12th International Conference on Healthcare Informatics (ICHI), pp.83-92
06/03/2024
DOI: 10.1109/ICHI61247.2024.00019
Abstract
This paper studies the problem of detecting asymptomatic cases in epidemic outbreaks within healthcare facilities. Asymptomatic cases pose a significant obstacle in our fight against epidemic outbreaks as they drive latent infection spread, are challenging to surveil, and are hard to intervene against. Detecting asymptomatic cases is challenging for numerous reasons, including lack of data except for large-scale serological surveys, poor generalization from symptomatic cases, and bias towards symptomatic cases in existing epidemiological datasets. Prior works fail in accurately detecting asymptomatic cases as they ignore individual risk factors, ignore the infection transmission pathways, or fail to integrate the two in a principled manner. Here, we formulate the asymptomatic case detection problem over a temporal network as a Prize Collecting Steiner tree with learnable latent prizes, where the latent prizes correspond to individual risks and the edge weights represent the cost of infection transmission. We translate the problem into an equivalent bi-level reinforcement learning problem and propose a deep Q-learning algorithm to tackle the problem. To demonstrate the efficacy of our proposed approach, we conduct extensive experiments on real-world networks derived from healthcare facilities. Our experiments over simulated healthcare-associated outbreaks of Clostridioides difficile infection (CDI) and COVID-19 reveal that the proposed approach has significant advantages over all the state-of-the-art baselines. Our approach outperforms the closest competitor by up to 29.62 % in detecting asymptomatic cases, which leads to more accurate predictions of symptomatic cases. In our experiments, we also demonstrate that accurately detecting asymptomatic cases leads to more accurate prediction of symptomatic cases. Finally, our case study in real CDI outbreak reveals that the asymptomatic cases detected by our approach are indeed high risk cases.
Details
- Title: Subtitle
- End-to-End Risk-Aware Reinforcement Learning to Detect Asymptomatic Cases in Healthcare Facilities
- Creators
- Yongjian Zhong - University of IowaWeiyu Huang - University of Iowa,Department of Computer Science,Iowa City,USABijaya Adhikari - University of Iowa
- Resource Type
- Conference proceeding
- Publication Details
- 2024 IEEE 12th International Conference on Healthcare Informatics (ICHI), pp.83-92
- Publisher
- IEEE
- DOI
- 10.1109/ICHI61247.2024.00019
- ISSN
- 2575-2634
- eISSN
- 2575-2634
- Grant note
- University of IowaCDC MInD Healthcare group: U01-CK000594
This work was partially supported by the CDC MInD Healthcare group under cooperative agreement U01-CK000594 and its associated COVID-19 supplemental funding and a startup fund from the University of Iowa. The authors acknowledge feedback from other University of Iowa CompEpi group members.
- Language
- English
- Date published
- 06/03/2024
- Academic Unit
- Computer Science
- Record Identifier
- 9984699521502771
Metrics
2 Record Views