To Predict or Not to Predict: The Case of the Emergency Department

Coupling healthcare datasets with advanced statistical methods has the potential to improve the efficiency and quality of healthcare dramatically. However, data used for predictive decision making in healthcare delivery has significant variable costs. We offer a novel and generalizable framework that helps reduce the costs associated with the use of data for healthcare analytics while maintaining high predictive accuracy. We utilize this approach in the emergency department (ED) context and specifically in predicting whether a patient will be admitted to an interior hospital unit or discharged from the ED. Data used to generate this prediction are available at different times, introducing a time‐cost associated with different data used in the prediction. We focus on minimizing the time‐cost of prediction without sacrificing accuracy. Using our approach, we are able to reduce the time‐cost of prediction by more than two‐thirds and significantly reduce the need to use privacy‐sensitive features. Yet, we still maintain high accuracy of prediction that is comparable to standard approaches which do not reduce data costs (area under the ROC curve of 0.86). Our work has significant potential value to healthcare entities and contributes to a growing stream of work on how to realize the value of healthcare analytics efforts.

• 저널Production and Operations Management · 31(2) · 799–818
• 토픽Machine Learning in Healthcare · Artificial Intelligence
• DOI10.1111/poms.13580
• 저자Sriram Somanchi, Idris Adjerid, Ralph Gross