Medical error is the third leading cause of death in the United States, causing nearly a quarter million deaths each year. Clinicians such as physicians, nurses, and pharmacists are shown clinical decision support (CDS) alerts, which are intended to reduce medical errors. However, since these alerts are often irrelevant, clinicians frequently ignore them. This problem is known as alert fatigue.
In this dissertation, I approach the problem in the styles of three disciplines, using three methods. First, I use a human factors style. I conduct a usability test comparing a conventional CDS alert design with an alternative design of my own. Resident physicians participated in a randomized, blinded and controlled study, which simulated patient scenarios, and solicited prescriptions using a mock electronic health record. Two attending physicians evaluated the prescriptions independently. I used ANOVA analysis and found that the residents in the alternative condition wrote more appropriate prescriptions. When shown both designs, residents vastly preferred the alternative design.
Second, I use a medical approach. I conduct a systematic review of the literature, comparing acceptance rates as an outcome on the basis of interactive design and role tailoring—the “fit” of the alert’s contents to the clinician’s role. I find that both interactive design and role tailoring can affect the likelihood that an alert will be accepted.
Third and finally, I use a science and technology studies (STS) style. I combine interviews of hospital clinicians and administrators with a document analysis. I use abductive analysis, and arrive at a novel view of alert fatigue: as a consequence of approaches to patient safety that involve holding healthcare organizations accountable to a growing number of external bodies, including accreditors, payers, government incentive programs, and patient advocacy groups. My findings raise questions regarding what alternative form of accountability might more effectively improve patient safety.
