Improving Hospital Efficiency: Predicting Post-Acute Care Facility Placement Using Machine Learning and Patient Mobility Scores from the Electronic Medical Record

点击翻译按钮获取中文摘要

Project Summary / Abstract Annually, approximately 8 million people are discharged from an acute care hospital to a post-acute care facility, accounting for >20% of all hospital discharges and >40% of all Medicare discharges. Post-acute care facilities frequently provide rehabilitation services for patients experiencing functional limitations after acute illness who cannot return home safely. Clinicians in acute care hospitals often fail to recognize hospital- acquired functional limitations until after resolution of acute medical/surgical issues. This failure delays hospital discharge and the start of rehabilitation in a post-acute care facility, which can exacerbate hospital-associated functional limitations. Patients' mobility status, one component of physical function, is an important factor in determining the requirement for a post-acute care facility. Simple, validated tools for routinely evaluating patient mobility are increasingly common in acute hospitals but are not routinely used to predict the need for discharge to a post-acute care facility. One such tool, the Activity Measure for Post-Acute Care Inpatient Mobility Short Form (AM-PAC IMSF), is a validated and reliable mobility measure for patients in acute care hospitals. The AM-PAC IMSF is used, as part of routine clinical care throughout hospitalization, for all patients in our acute care hospital. In a pilot study, we demonstrated that lower AM-PAC IMSF scores at hospital admission were strongly associated with post-acute care facility placement. Our goal is to expand upon our preliminary work to develop a formal model to predict which patients are likely to require post- acute care facility placement. Such prediction would be invaluable for improving the discharge planning process and expediting receipt of rehabilitation services at a post-acute care facility. Our overall objective is to demonstrate that prediction models, leveraging `big data' from electronic medical records, can help optimize the hospital discharge process. Thus, we propose the following Aims: 1) To determine if baseline patient mobility status, measured by the AM-PAC IMSF within 48 hours of hospital admission, is predictive of hospital discharge to specific levels of post-acute care; and 2) To develop a dynamic prediction model, using both the hospital admission AM-PAC IMSF score and the subsequent trajectory of daily scores after hospital admission, to predict hospital discharge to specific levels of post-acute care. This proposed research addresses the AHRQ priority of improved efficiency and quality of healthcare delivery via improving the hospital discharge process, with associated improvement in patient outcomes.