Methodology to automatically detect abnormal values of vital parameters in anesthesia time-series

Methods for automatic detection of abnormal vital signs occurring during anesthesia are poorly documented in the literature.Existing methods are not reproducible between databases.We have developed a data model and an algorithm that allow automatic detection of abnormal values of vital parameters occurring during anesthesia.Predefined thresholds and various parameters such as time between measurements and time spent outside predefined thresholds provide adaptability to various clinical situations, e.g. hypotension occurring after start of anesthesia.The relation between occurrence of abnormal values of vital parameters and mortality and length of stay may then be studied on a large and automated scale. Abnormal values of vital parameters such as hypotension or tachycardia may occur during anesthesia and may be detected by analyzing time-series data collected during the procedure by the Anesthesia Information Management System. When crossed with other data from the Hospital Information System, abnormal values of vital parameters have been linked with postoperative morbidity and mortality. However, methods for the automatic detection of these events are poorly documented in the literature and differ between studies, making it difficult to reproduce results. In this paper, we propose a methodology for the automatic detection of abnormal values of vital parameters. This methodology uses an algorithm allowing the configuration of threshold values for any vital parameters as well as the management of missing data. Four examples illustrate the application of the algorithm, after which it is applied to three vital signs (heart rate, SpO2, and mean arterial pressure) to all 2014 anesthetic records at our institution.