This study focuses on the predictive identification of patients frequently revisiting the University of Virginia Health System Emergency Department. Identifying these patients can help the Emergency Department formulate strategies that improve patient care and decrease excess Emergency Department utilization. The Health System in particular faces a number of unique challenges in its ongoing mission to reduce extraneous patient revisits. In addition to its status as an academic hospital, it serves a broad geographic region as one of five level-I trauma centers in the Commonwealth of Virginia. In this study we utilized 5 years of data from the University of Virginia Health System data warehouse. These data contain information on 91,297 patients and 196,902 unique encounters, including details on patient demographics, diagnoses and hospital departments visited. From these raw data we engineered features, trained gradient boosted decision trees, and experimented with unsupervised clustering techniques to best approximate 30-day Emergency Department revisit risk at the conclusion of each patient encounter. Our best model for revisit risk resulted in a Receiver Operator Characteristic Area Under the Curve of 0.75. Furthermore, we exhibit the real-time performance of our model as a tool to rank which at-risk patients should receive priority for Emergency Department resources. This test demonstrated a significant improvement over the current allocation of Emergency Department social worker resources with a daily Mean Average Precision of 0.83. The methodologies proposed in this paper exhibit an end-to-end framework to transform raw administrative claims and limited clinical data into predictive models that help the Emergency Department better manage resources and target interventions.
[1]
Jeffrey Dean,et al.
Efficient Estimation of Word Representations in Vector Space
,
2013,
ICLR.
[2]
Dan Chateau,et al.
Frequent users of emergency departments: developing standard definitions and defining prominent risk factors.
,
2012,
Annals of emergency medicine.
[3]
Yong Wang,et al.
Risk Prediction of Emergency Department Revisit 30 Days Post Discharge: A Prospective Study
,
2014,
PloS one.
[4]
Chia-Lun Lo,et al.
Predicting Factors and Risk Stratification for Return Visits to the Emergency Department Within 72 Hours in Pediatric Patients
,
2015,
Pediatric emergency care.
[5]
Wenli Zhang,et al.
Predicting Asthma-Related Emergency Department Visits Using Big Data
,
2015,
IEEE Journal of Biomedical and Health Informatics.
[6]
C. Steiner,et al.
Hospital Readmissions and Multiple Emergency Department Visits, in Selected States, 2006–2007
,
2010
.
[7]
D. Aronsky,et al.
Systematic review of emergency department crowding: causes, effects, and solutions.
,
2008,
Annals of emergency medicine.
[8]
Shaun J. Grannis,et al.
Predicting Emergency Department Visits
,
2016,
CRI.
[9]
Tianqi Chen,et al.
XGBoost: A Scalable Tree Boosting System
,
2016,
KDD.