Artéfacts de monitoring et recherche dans les grandes bases de données: ce que vous ignorez pourrait vous nuire

Abstract

During the last two decades we have witnessed an exponential growth in information technology. Although anesthesia information management systems (AIMS) have been around for more than 30 years, the transition from paper records to electronic medical records has been slow. However, recent data demonstrate a rapid rise in adoption of electronic records at academic medical centres. As AIMS have grown in popularity, so too has research that uses automated electronic data at its epicentre. Retrospective observational studies are no longer a rarity in peerreviewed journals. In fact, it is often difficult to find a leading anesthesiology journal without this type of clinical research. In this issue, Kool et al. attempt to examine the prevalence of artifacts in physiologic data stored by AIMS. This single-centre, prospective observational study of 86 patients at the University Medical Center Utrecht is an important step in the maturation of clinical research using AIMS data. The aim of the study was to determine the reliability of data collected from an AIMS. All patient monitoring and data recordings contain artifacts, but historically their extent and the parameter-specific variation in AIMS data have been understudied. Kool et al. used independent study personnel (another anesthesiology provider) to monitor each case and note any artifactual data displayed on the physiologic monitor or in the AIMS. The authors focused on five physiologic parameters: pulse oximetry oxygen saturation, electrocardiography (ECG)derived heart rate, ECG-derived ST segments, intermittent noninvasive blood pressure, and continuous invasive blood pressure. Whereas the heart rate, oxygen saturation, and noninvasive blood pressure remained ‘‘relatively’’ free of artifacts, more than 5% of ST segments and invasive blood pressure values displayed on the monitor and stored in the AIMS database were deemed artifacts. These data are among the first to evaluate systematically the prevalence of artifact data during physiologic monitoring and in AIMS data. We applaud Kool et al. for publishing work that is long overdue. The relatively high frequency of ST-segment and invasive blood pressure artifact data reminds the scientific community that all analyses using monitoring data must account for artifacts. Many of today’s manuscripts using automated physiologic data interfaces for analyses do disclose how artifact data are defined, identified, and managed. For example, it is common to perform median functions over relatively lengthy periods, such as 5 or 10 min, to define epochs of physiologic derangement. In addition, ‘‘invalid’’ values may be removed prior to median function, such as invasive systolic blood pressures [ 280 mmHg or invasive blood pressures without a reasonable pulse pressure between systolic and diastolic readings. Mature observational analyses using automated physiologic data often report such artifact management in their published manuscripts. However, there are neither standards for artifact management nor standards for how to report it. The current work by Kool et al. has significant implications to all consumers or creators of observational studies using physiologic data. It is an important first step toward improved transparency in data acquisition and management. Their findings demonstrate that every publication should have a detailed section on materials and methods, M. Martin, MD S. Kheterpal, MD Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, MI, USA