CRII: SCH: Towards Smart Patient Flow Management: Real-time Inpatient Length of Stay Modeling and Prediction

Chapman University, Orange CA

Investigators

Abstract

Patient length of stay has been used as an essential criterion for the effective planning and management of hospital resources. Prolonged stay increases patients’ risk of hospital-acquired infections and disrupts patient flow and access to high-quality healthcare services. As such, a model that can reliably predict the length of stay for a specific patient is desirable to mitigate the prolonged stay and guide personalized decision-making. However, the length of stay can be affected by a multitude of factors and can vary based on different patients’ conditions and disease progression. The complex and dynamic nature of massive clinical data, not to mention the presence of a large portion of missing and censored values in the healthcare data, poses significant challenges for efficient modeling and dynamic prediction. This project aims to offer an integrated solution by establishing a pipeline consisting of advanced statistical modeling, monitoring, and deep learning techniques based on patient information collected from heterogeneous medical systems over time. The success of the project will catalyze a transition from a traditional standard-driven discharge scheduling service to a data-driven proactive scheduling paradigm. The success of the project will alleviate the hospital’s pressure on resource allocation and improve patient flow and, more importantly, pandemic preparedness. The project can provide opportunities for research-based interdisciplinary training of undergraduate and graduate students in health informatics, statistics, and artificial intelligence from diverse backgrounds, including women and underrepresented minorities. This project will address the critical challenges of healthcare data analysis, i.e., heterogeneity, multi-modality, and data sparsity. Conventional data-driven methods have been predominantly focused on identifying the factors that strongly influence the length of stay as opposed to predicting the length of stay itself. Moreover, the existing approaches failed to address the inherent uncertainty and were incapable of incorporating different data modalities and dynamic prediction. The project proposes a personalized framework by integration of advanced tensor fusion and time-to-event modeling techniques towards smart patient flow management, which ultimately allows for faster achievement of health outcomes and reduction of hospitalized costs. The proposed intelligent framework will advance the state-of-art research of real-time data fusion and personalized prognosis in the following aspects: (1) brings the data fusion and length of stay prediction into a unified framework; (2) facilitates personalized length of stay prediction in a real-time manner; (3) naturally has the capability to incorporate uncertainties in the decision-making process to provide a confident and intelligent scheduling service. Although the methodology is proposed for the patient length of stay prediction, it does not depend on any restrictive assumptions of domain knowledge and specific disease and thus can potentially be applied to a broad range of events predictions. 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.