Small-scale studies of physician-alerting programs within clinical decision support systems in clinical information systems have long hinted at a widespread problem: physician “alert fatigue” deriving from an overabundance of notifications. These can be drug-drug interaction alerts, drug-allergy interaction alerts, and even drug-disease alerts. Last month, a study by physician researchers at two Boston-based patient care organizations, Beth Israel Deaconess Medical Center (BIDMC) and the Dana-Farber Cancer Institute, is confirming on a broader scale what individual organizations had already guessed at — that physicians are overriding the vast majority of the drug alerts in these systems.
Thomas Isaac, M.D., instructor of medicine in the Division of General Medicine and Primary Care at BIDMC and a project manager at the Center for Health Care Quality Improvement and Safety at Dana-Farber, and Saul Weingart, M.D., Ph.D., vice president for patient safety at Dana-Farber and an internist at BIDMC, authored the study, which was published in the Feb. 9, 2009 issue of the Archives of Internal Medicine. Reviewing the electronic prescriptions and associated medication safety alerts generated by 2,872 clinicians in community-based outpatient physician practices in Massachusetts, New Jersey, and Pennsylvania, Weingart and Isaac found that physicians overrode more than 90 percent of the drug-interaction alerts and 77 percent of the drug-allergy alerts, among 3.5 million prescriptions ordered between Jan. 1 and Sept. 30, 2006. What does this mean in terms of the value of having such alerting programs built into e-prescribing and EMR systems to begin with? Healthcare Informatics Senior Contributing Editor Mark Hagland interviewed Thomas Isaac recently regarding his and Weingart’s findings from that study.
Mark Hagland: Were you surprised by the marquee results of your and Dr. Weingart’s study? Finding that physicians overrode 90 percent of the drug-interaction alerts and 77 percent of the drug-allergy alerts in their e-prescribing systems seems like a pretty big percentage.
Thomas Isaac, M.D.: Actually, we had hypothesized that most drug allergy and drug interaction alerts were being overridden. Previous internal studies had shown similar override rates of around 90 percent in academic medical centers. But we were looking at over 2,500 clinicians across three states in our study, all of them using a commercial e-prescribing product, meaning these outpatient practices had to purchase their e-prescribing technology. I was a little surprised that there was little differentiation among the mild, moderate and severe alert-level categories. I would have thought that for the most severe interactions, doctors would have accepted those alerts more often; and they did, but not by a lot.
MH: Does this say that these alerts really have to be rewritten?
TI: Well, I think it’s a matter of creating alerts that are useful to clinicians. And there is a spread among acceptance rates for alerts. For example, some alerts for drug-drug interactions are overridden virtually 100 percent of the time. The combination of steroids and anti-rheumatic agents — those are generally medications that are often used together for rheumatoid arthritis, and alerts for those were very often overridden.
MH: Then why write one?
TI: Many alerts are written for theoretical reasons. For example, many rheumatic arthritis drugs are often immunosuppressive. And often, you have to use these medications in combination; and there is risk, but doctors are generally aware — rheumatologists are almost certainly aware that there are additive risks involved in adding the drugs together. It’s a little bit hard to say which physicians will know the risks.
MH: Do you believe that alert fatigue may be happening?
TI: Yes, there is definitely evidence of alert fatigue, as shown in other studies. There’s one table in our study where, though we weren’t setting out to examine alert fatigue, there was a suggestion of it. For example, we looked at the volume of prescriptions physicians were writing and what their response was to alerts. And as physicians wrote more prescriptions, they had a higher tendency towards overriding alerts. There could be multiple reasons for this, including enhanced knowledge. But it could also point to alert fatigue.
MH: So, people should be cautioned about this?
TI: I don’t think the message should only be to decrease the number of alerts. Perhaps for the alerts that only speak to very theoretical harm, or nuisance alerts, that are overridden perhaps nearly 100 percent of the time, those probably should be taken away. But physicians should probably be able to turn on or off certain alerts for very common situations among certain specialties or groups of physicians, and individual physicians should probably be given more flexibility as well.
MH: As an industry, are we still a bit behind where we should be with prescribing-oriented clinical decision support?
TI: I think there are some systems that are fairly sophisticated. But I think the vast majority of outpatient physicians who will be purchasing e-prescribing systems like the one we examined will be working with systems similar to the one we studied. There are researchers who have come much further in designing alerts. But I think it is fair to say that the work is still fairly early and hasn’t become mainstream by any means.
MH: Should physicians be more involved in designing the systems?
TI: That definitely would be valuable. Even observing why physicians are overriding specific alerts, and then trying to reduce some of the ‘alert noise,’ would be helpful.
MH: What implications might this study have for the science of alerting more generally?
TI: Probably a couple of things. First, the field is pretty young, and it should be expected that our ‘first crack’ at many of these things would be somewhat crude. But it can be very valuable to observe and study physicians’ responses to the alerts, and get as much physician input as possible in designing these systems, in order to improve and enhance them.
MH: What would your advice be for CIOs as purchasers or co-purchasers of these systems for and with doctors?
TI: It will be important to understand that you’ll have to upgrade these systems every few years, as you would with any other kind of computer system. But overall, these information systems improve patient safety, even in their current iterations.
