I-Corps: Translation Potential of Generative Artificial Intelligence (AI) and Natural Language Processing (NLP) Methods to Quantify Network Flow Models

This I-Corps project is based on the development of software technology to help physicians avoid unnecessary escalation of patients to higher levels of care. Currently, hospital beds are often occupied by patients who could be cared for at a lower-level facility, causing these hospitals to turn away patients who truly need beds. Other patients have extended stays, sometimes multiple days, in an area of the hospital that is not ideal for them, such as in the emergency department.

This incompatibility between needs and available space is causing significant revenue loss and harming patients. This solution is designed to provide decision support on where and when each potential patient could receive care and to make limited hospital capacity available for patients who need it most. In addition, the technology identifies space in highly specialized hospitals for patients who require specialized care.

This technology may also enable more efficient use of limited resources in hospitals and healthcare systems encompassing multiple hospitals, improving patient outcomes.

This I-Corps project utilizes experiential learning coupled with a first-hand investigation of the industry ecosystem to assess the translation potential of a software technology that uses artificial intelligence (AI) to enable decision support and learning for placement of patients in hospitals. Current clinical decision support and prediction technology focuses on acuity or first-come, first-served bed placement, and lacks a learning system and outcomes considerations.

This technology uses generative AI and natural language methods to quantify systems where the parameters are multidimensional and continuously changing, such as network flow models. The software uses surfacing and quantifying variation in agent decisions allocating constrained resources for inflow demand. The technology continually updates a surfacing algorithm using feedback from top performing agents.

In addition, decision support is designed to customize to a specific healthcare system for admissions and other key flow decision points. Initial technical results from small scale testing of a partly manual model have demonstrated a reduction in the number of unnecessary patient escalations. The goal is to enable effective use of bed capacity, providing value by reducing time and resources, and improving patient outcomes.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.