(1310) Estimating Excess Stay in Medical-Surgical Units Using Machine Learning

Introduction: Patients who are considered medically fit are often unnecessarily kept in acute care wards due to operational issues or delayed transition of care. This has led to an increase in the average length of stay, especially over the past few years. Furthermore, delayed discharge has been linked to negative impacts on patient outcomes, satisfaction, and staff burnout. We propose the use of a Machine Learning based index to estimate deterioration, defined as the risk of mortality or transfer to a high level of care, for patients in the Medical-Surgical unit. We hypothesize that low-risk patients, identified through this index, are unlikely to require escalation in the level of care and could potentially be discharged early. We then focus on evaluating their median excess stay, their total number of excess days, and the associated costs.

Methods: De-identified data from the Philips eICU (ERI), specifically Med-Surg units, from 2012-2022 (excluding 2020-2021 to reduce COVID bias) was utilized for this study. A Machine Learning model employing XGBoost was developed to forecast deterioration within 24 hours. The model assessment using an area under the receiver operating characteristic curve showed values of 0.79. Low-risk patients were defined as those with continuous deterioration risk values below 0.034 (lowest 5% of the population) for at least 6 hours before discharge. Excess stay was defined as patient hours surpassing the 6-hour buffer before discharge. Stays shorter than 12 hours and longer than 7 days were excluded from the analysis.

Results: Among a population of 38,082 patient stays, 59% were female, with an average age of 57 years and a median length of stay of 43 [23-71] hours. A total of 33,372 patients met the selection criteria, of which 12.6% experienced excess stay, with a median excess of 22 [9-34] hours (49% of their stay). These patients had a total excess stay of 3,833 days. At a conservative rate of $2,000 per day, this extra stay cost an estimated $7.66 million.

Conclusions: By identifying patients with a low risk of deterioration early on, it may be possible to implement strategies for expedited discharge, resulting in cost savings without compromising patient safety.