Knowledge Augmented AI for Precision Surveillance and Control of Hospital-Acquired Infections

Knowledge Augmented AI for Precision Surveillance and Control of Hospital- Acquired Infections Healthcare-Associated Infection causing Pathogens (HAICPs) due to organisms, such as Escherichia coli and Staphylococcus aureus, as well as due to multi-drug resistant organisms (MDROs), such as Clostridioides difficile, Carbapenem-Resistant Enterobacterales (CRE), and methicillin-resistant S. aureus (MRSA), pose a heavy health burden, especially in healthcare facilities, leading to increased cost of surveillance, response, control, and treatment. They also spread within the community, nursing homes and long-term care facilities, and are becoming increasingly resistant to antibiotics, making them a significant public health threat. Reducing the risk of HAICPs within healthcare facilities remains a very challenging problem. As in the case of other clinical problems, machine learning methods using electronic health record (EHR) data have been used quite extensively to predict patients at-risk of HAICPs. However, there remain significant gaps in the robustness and external validity of prior models, and these haven’t made it into regular clinical practice, and into general Clinical Decision Support (CDS) systems. Surveillance and response during infection control also haven’t been optimized using AI and machine learning techniques, which presents a large burden for many healthcare facilities, especially smaller facilities, and those with limited resources and staffing. We propose to develop a novel approach for HAICP risk prediction, surveillance and infection control prob- lems, using a combination of AI and machine learning, human factors and systems science techniques as part of a highly multi-disciplinary collaboration. Our project consists of the following tasks: (1) joint risk prediction of multiple HAICP clinical tasks using a combination of deep learning and agent based models, on patient data from the EHR enhanced with data on potential transmissions by health care workers (HCWs) and important clinical features drawn from provider notes using large language models (LLMs), (2) implementation of these methods in a federated manner across hospitals, (3) optimization of testing and surveillance policies, and (4) development of an Infection Prevention Operations Dashboard (IPOD). Our team consists of nationally known clinicians (infection preventionists, infectious disease experts), epidemiologists, and computer scientists, hu- man factors and systems engineers, and patient safety experts from the University of Virginia (UVA), Johns Hopkins University (JHU), Georgia Tech and Virginia Tech. Our project is innovative in (1) its focus on both HAICP clinical risk prediction, as well as surveil- lance and response, which will aid in HAI infection control efforts, and (2) knowledge augmentation of deep learning and LLM methods using transmission models and clinical insights, and (3) an Infection Prevention Operations Dashboard that incorporates these components, and that will be developed us- ing a human-centered design approach, and scales to six diverse hospitals through federation across the UVA and Johns Hopkins (JH) healthcare systems.