Congestion reduction in the Emergency Department of Massachusetts General Hospital

Abstract

The MGH Emergency Department (ED) and General Medicine Floor currently experience heavy patient volume and rising patient wait times, despite recent capacity expansions. While several projects have been piloted to divert patients towards alternative care paths, MGH management wants to better understand what types of patients are being admitted to the hospital and what features are deterministic of patient admission. This thesis addresses this information gap by using binary logistic regression models to assess predictive and significant patient features for admission. Our analysis uses both patient demographic information and decision point data gathered in the Emergency Department of patient visits. On out-of-sample data, our predictive model achieves an area under the receiver operating characteristic of 0.82, and we conclude that the predictive features for admission are within good clinical practice. Further analysis of patient care suggests that provision of IV antibiotics in the outpatient setting could reduce MGH admissions by approximately 307 bed-days per year, with additional possible reductions in excess of 1,000 beddays for different provisions of care. We also assess the outpatient usage of MGH patients and conclude that 75 percent of cellulitis, pneumonia and urinary tract infection patients are not seeing a clinician in the outpatient setting prior to ED presentation. This analysis indicates that more proactive management of these patients could prevent both their visit to the ED and potentially their admission. We demonstrate that statistical methods based on real time patient data. can contribute to effective healthcare planning and operations.