Saul Weingart, MD, PhD, senior author on the medication safety study

Computer-based systems that allow clinicians to prescribe drugs electronically are designed to automatically warn of potential medication errors, but a new study reveals clinicians often override the alerts and rely instead on their own judgment.

The study, led by investigators at Dana-Farber Cancer Institute and Beth Israel Deaconess Medical Center (BIDMC), suggests that most clinicians find the current medication alerts more of an annoyance than a valuable tool. The authors conclude that if electronic prescribing is to effectively enhance patient safety, significant improvements are necessary. The study's findings appear in the Feb. 9 issue of the Archives of Internal Medicine.

"Electronic prescribing clearly will improve medication safety, but its full benefit will not be realized without the development and integration of high-quality decision support systems to help clinicians better manage medication safety alerts," says the study's senior author, Saul Weingart, MD, PhD, vice president for patient safety at Dana-Farber and an internist at BIDMC.

The researchers reviewed the electronic prescriptions and associated medication safety alerts generated by 2,872 clinicians at community-based outpatient practices in Massachusetts, New Jersey, and Pennsylvania to learn how clinicians responded to the alerts.

The clinicians submitted 3.5 million electronic prescriptions between Jan. 1 and Sept. 30, 2006. Approximately one in 15 prescription orders, or 6.6 percent, produced an alert for a drug interaction or a drug allergy. The vast majority of the 233,537 alerts (98.6 percent) were for a potential interaction with a drug a patient already takes.

Clinicians overrode more than 90 percent of the drug interaction alerts and 77 percent of the drug allergy alerts. Even when a drug interaction alert was rated with high severity, clinicians typically dismissed those for medications commonly used in combination to treat specific diseases. They also were less likely to accept an alert if the patient had previously been treated with the medication.

The high override rate of all alerts, the researchers contend, suggests that the utility of electronic medication alerts is inadequate, adding that for some clinicians, most alerts "may be more of a nuisance than an asset."

"The sheer volume of alerts generated by electronic prescribing systems stands to limit the safety benefits," says Thomas Isaac, MD, MBA, MPH, of BIDMC and Dana-Farber and the paper's first author. "Too many alerts are generated for unlikely events, which could lead to alert fatigue. Better decision support programs will generate more pertinent alerts, making electronic prescribing more effective and safer."

Although the study analyzed orders generated on only one electronic prescribing system, PocketScript, the researchers say their observations are relevant to other systems because the alerts they reviewed were typical and were generated by a commercial database, Cerner Multum, used by other electronic prescribing systems.

Based on these findings, Weingart and his colleagues offer several recommendations to improve medication safety alerts, including reclassifying severity of alerts, especially those that are frequently overridden; providing an option for clinicians to suppress alerts for medications a patient already has received; and customizing the alerts for a clinician's specialty. The research team identified a list of potentially dangerous drug interactions based on those alerts that most often changed the clinicians' decision to prescribe. This list is available at www.dana-farber.org/electronic-medication-safety.

"We need to find a way to help clinicians to separate the proverbial wheat from the chaff," says Weingart. "Until then, electronic prescribing systems stand to fall far short of their promise to enhance patient safety and to generate greater efficiencies and cost savings."

In addition to Weingart and Isaac, the paper's other authors are Joel Weissman, PhD, Executive Office of Health and Human Services, Commonwealth of Massachusetts; Roger Davis, ScD, BIDMC; Daniel Sands, MD, MPH, BIDMC and Cisco Systems, San Jose, Calif.; Michael Massagli, PhD, PatientsLikeMe, Inc., Cambridge, Mass.; and Adrienne Cyrulik, MPH, Blue Cross Blue Shield of Massachusetts, Boston.

The research was supported by a grant from the Physicians' Foundation for Health Systems Excellence, Boston.