Engineering Inpatient Discharges: Disposition Prediction and Day-of-Discharge Planning

Inpatient discharge planning is a critical decision point in patient care, with implications for the efficiency of the inpatient unit as well as other units of the acute care hospital. Inefficient discharge planning can cause patient boarding (waiting for beds) in the upstream units. While this is a poignant and well-known problem in healthcare, very little quantitative research exists that proposes approaches to alleviate it. To address this issue, we apply Systems Engineering methods with focus on three key challenges in inpatient discharge planning. First, to aid inpatient care providers in predicting discharge disposition (home vs. non-home) within 24-hours of a patient being admitted, we develop an early-warning prediction tool. This tool is derived from a multivariable logistic regression model built using data from a general medicine unit at a VA hospital. The tool is expected to aid the inpatient staff in proactively classifying non-home discharges from home in an effort to initiate early discharge planning and avoid non-medically related discharge delays. Second, to improve hospital bed flow and reduce upstream patient boarding, we propose a novel discharge target strategy, n-by-T, for an inpatient unit\u27s planning of daily discharges. A stochastic simulation model developed in collaboration with a trauma unit at a local hospital predicted that this strategy could offer significant advancement in discharge completion time and reduction in upstream boarding; these findings were later validated via a pilot at the unit. Consistent findings via an extension to a neurology unit at another hospital suggest potential generalizability of this strategy. Third, to assist ancillary service providers on inpatient units in sequencing their daily patient workflow, we propose a novel approach to construct implementable and robust strategies. We develop a scenario-specific mixed-integer programming model to derive optimal sequences that minimize average upstream patient boarding under due date constraints. We then design a simulated annealing based metaheuristic to derive a single sequencing strategy that is promising across all scenario-specific optimal sequences for a given system configuration. An experimental evaluation of our approach suggests that our proposed strategies outperform several realistic strategies on boarding time. In summary, our research proposes easy-to-understand and implementable strategies derived from optimization and data analytics based methodologies to aid effective and efficient planning of discharges and improve patient flow through the hospital