Brigham and Women’s Hospital CERT: Solving the Problem of Alert Fatigue To Minimize Medication Errors

By Shelley Norden Barnes

For the past decade, investigators at the Brigham and Women’s Hospital Center for Education and Research on Therapeutics (BWH-CERT) have helped lead the national effort to incorporate computerized decision support (CDS) tools into electronic health records to reduce medical errors. Their most recent research has focused on the automatic alerts that warn physicians if a medication they prescribe contraindicates a patient’s current drug regimen. The aim of these alerts is to decrease medication errors, prevent adverse drug events, and ultimately to increase patient safety.

The Pervasiveness of Alert Fatigue

At the outset of their medication alert study, the BWH-CERT investigators found only limited published literature on medication-related decision support and how providers respond to warnings. The information they did find, however, was alarming: While medication-related decision support was found to reduce the frequency of preventable adverse drug events and save lives, between 49 to 96 percent of the medication alerts were being over-ridden by physicians.

"The problem with medication alerts is that they’re inundating physicians with too much information," says Shobha Phansalkar, Ph.D., who is spearheading this research. "The profusion of unnecessary alerts frustrates users and is triggering a response known as 'alert fatigue'—the exhaustion that comes when providers spend excessive time and mental energy responding to computer-based medication alerts. As a result, physicians are either ignoring the alerts or simply turning them off, which undermines the purpose of implementing computerized decision support."

Phansalkar and her team seek not only to understand the problem of alert fatigue but also to provide tangible solutions. By improving the content and design of medication alerts they hope to optimize the functionality of CDS tools so that physicians actually use alerts at the point of care to inform decisionmaking.

Phansalkar explained how the BWH-CERT team examined all the information that accompanies alerts, which is the biggest factor contributing to alert over-rides, and found that many clinically insignificant alerts were included in the electronic health record (EHR) and were contributing to the increase in alert volume. She noted, "Providers stopped paying attention to all alerts—even the clinically significant alerts that required action—because they were inundated with so many they chose to ignore them regardless of their severity."

Identifying the Most Severe Drug-Drug Interactions

Intent on creating a scaled-down list of the most critical drug-drug interactions (DDIs) that trigger medication alerts, the team convened a panel of experts that included both academics and representatives of three of the country’s largest medication knowledge-base vendors who supply content to EHR systems.

"Including knowledge-base vendors was critical in that each one has their own criteria for judging the severity of DDIs, meaning that each EHR system may have different alerts," Phansalkar points out. "Our team wanted their input in creating standardized criteria for assessing the most critical DDIs, which could then be used across the board in all EHR systems."

The panel developed stringent exclusion criteria to identify a limited set of medications that should never be coadministered. The result was a list of 15 highly clinically significant drug-drug and drug-allergy interactions that should always generate alerts in EHR programs. Phansalkar is quick to clarify that the list is not meant to be all-inclusive but, rather, a recommended starter set of interactions that should serve as a baseline.

Having established this high-priority list, the team then focused on the clinically insignificant alerts in the medication knowledge bases that were disrupting physicians’ workflow. This resulted in a second set of 33 low-priority DDIs that ideally could be suppressed in EHR systems to reduce the number of alerts and minimize alert fatigue.

Good Design: The Key to Reducing Alert Over-Rides

User acceptance of medication and other alerts could also be affected by their design. Once an alert is deemed significant and accurate, says Phansalkar, the information it contains must be displayed in compliance with human factor principles so that it captures the user’s attention and conveys the appropriate amount of information for the physician to make an informed therapeutic decision. In an earlier CERT project, Phansalkar and her team developed I-MeDeSa, an instrument they are now using to score the various EHRs and then rank these systems on how well they integrate human factor principles into the overall alert presentation and display.

"With this current research, we want to determine if good design correlates with fewer over-rides and if this, in turn, correlates with greater provider satisfaction and acceptance of alerts," Phansalkar explains. "Better alert design should, in practice, improve alert acceptance. So, our goal is not only to satisfy physicians by giving them information they can use but also to see that they are using it to make correct therapeutic decisions for their patients."

To date, little information is available regarding the empirical correlation between better design and higher provider acceptance. While the I-MeDeSa tool will measure design based on a system’s incorporation of human factor principles, the challenge for Phansalkar’s team is to find a way to gauge physician acceptance and satisfaction.

"In thinking about how to measure provider acceptance, the proxy variable that makes the most sense is alert over-ride rates. For example, if there are high over-ride rates for alert programs that are poorly designed based on their I-MeDeSa score, then you can draw correlations between these two variables," she concludes.

While the collective studies of the BWH-CERT team have garnered invitations to present their findings at local and national conferences and incorporation of their findings into vendor products, Phansalkar stresses that there is still much work to be done.

"It’s too soon to know if implementation of our findings has resulted in greater use of alerts by providers, but we hope to be investigating this in the near future as a followup to our current work," she asserts. "The real validation will come when we can measure the impact on providers and on patient outcomes."

The Centers for Education and Research on Therapeutics (CERTs) are a nationwide network of six research centers and a coordinating center that receive core financial support from the Agency for Healthcare Research and Quality. The CERTs conduct research and provide education that will advance the optimal use of drugs, medical devices, and biological products; increase awareness of the benefits and risks of therapeutics; and improve quality while cutting the costs of health care.

For more information about the BWH CERT work on electronic clinical decision support, see the related article: Brigham and Women’s HIT-CERT: A Computerized Approach to Patient Safety.

Abstracts of the studies referenced in this article can be accessed on PubMed:

1. Phansalkar S, Edworthy J, Hellier E, et al. A review of human factors principles for the design and implementation of medication safety alerts in clinical information systems. J Am Med Inform Assoc. 2010 Sep-Oct;17(5):493-501. PMID: 20819851.
2. Zachariah M, Phansalkar S, Seidling HM, et al. Development and preliminary evidence for the validity of an instrument assessing implementation of human-factors principles in medication-related decision-support systems—I-MeDeSA. J Am Med Inform Assoc. 2011 Dec;18 Suppl 1:i62-72. PMID: 21946241.
3. Phansalkar S, Desai AA, Bell D, et al. High-priority drug-drug interactions for use in electronic health records. J Am Med Inform Assoc. 2012 Sep-Oct;19(5):735-43. PMID: 22539083.
4. Phansalkar S, Desai A, Choksi A, et al. Criteria for assessing high-priority drug-drug interactions for clinical decision support in electronic health records. BMC Med Inform Decis Mak. 2013 Jun 13;13(1):65. PMID: 23763856.

Return to Featured CERTs Research