Detection and correction of errors in real-time locating system data

Installing real-time locating systems (RTLS) in health clinics promises to provide hard data on operations, including wait times and resource utilization. However, interpreting the data is made more difficult by its own inaccuracies and by the complexities of a multi-service clinic. We must find ways of processing erroneous “jumps” in the RTLS signal and identifying true movement from noisy data. The objective of this study is to investigate factors that lead to longer waits for infusion. We also aim to provide a framework for transforming raw RTLS data that contain many errors into more reliable data that can be utilized to provide accurate measures of clinic conditions. We conduct a case study at a university hospital cancer center in Virginia. First we develop rules to detect and eliminate erroneous signals in the RTLS data. These errors can be caused by fluctuations in the wireless signal that cause the system to misidentify a patient's location. Then, using the RTLS data, we develop pattern mining and clustering models to identify typical patient behaviors associated with specific activities, focusing on patient waiting times. Finally, we demonstrate how this cleaned data can provide clinics with accurate assessments of clinic operations, including patient wait times, and patterns of patient flow.