Predicting Tests Ordered in Hospital Laboratories Using Generalized Linear Modeling

Laboratory services in healthcare systems play a vital role in inpatient care. Most hospital laboratories are facing the challenge to reduce cost and improve service quality. This study focuses on identifying test order patterns in a laboratory for a large urban hospital. The data collected from this facility consists of all tests ordered over a three month time frame and contains test orders for approximately 17,500 patients. Poisson and Negative Binomial regression models are used to determine how well patient characteristics (patient length of stay and the medical units in which patients are placed) will predict the number of tests being ordered. The test order prediction model developed in this study will aid the management and phlebotomists in the hospital laboratory in securing methods to satisfy the test order demand. By implementing the recommendations of this study, hospital laboratories should see significant improvements in phlebotomist productivity and resource utilization, which could result in substantial cost savings.