[Behavioral] Alert!

Sometimes practice changes permanently based on single studies of few participants. The use of behavioral alerts provides an example.

We have all experienced alert fatigue. Most alerts are generated as one interacts with the medical record (e.g., when entering a new order). But one type of alert pops up as a patient’s electronic medical record is entered: the behavioral alert.

This form of electronic alert first appeared in the 1980s at the Portland Oregon Veterans Administration Medical Center as a response to a series of violent incidents. After studying their experience, David Drummond, Landy Sparr, and Geoffrey Gordon shared their findings in the Journal of the American Medical Association (aka JAMA).

Here is how the system was described in the 1989 article:

If there was a pattern of seriously disruptive behavior and a consensus among committee members, a recurrent computer entry—a flag—was added to the patient's automated database within the medical center-wide computerized scheduling system. The flag was designed to alert medical staff to the patient's violence potential. When a patient who has been flagged is checked in by a clerk at a computer terminal or is scheduled for an appointment, an advisory note on the computer monitor appears, and a subtle audio signal is emitted. Clerks are trained to print these warnings and bring them to clinicians' attention before the patient is seen. The flag comprises a brief directive, eg, “Patient should be searched for weapons,” “Hospital police should be asked to stand by until released by examining clinician,” and, usually, a reference to a dated medical progress note that describes past difficulties with the patient.

And here’s what the authors found:

Drummond DJ, Sparr LF, Gordon GH. Hospital Violence Reduction Among High Risk Patients. JAMA 1989;261(17):2531–4.

You might notice a dramatic reduction in outpatient incidents after introducing flagging (44 in 226 (19%) visits before versus 3 in 137 (2.2%) visits after). This appears remarkable, but your enthusiasm should be tempered by a reminder that this was a pre/post-study. Notably, there were 40% fewer visits in the period after flagging was implemented. There is no way to know how those who continued to be seen were different from those who didn’t back back. It is not crazy to imagine that these groups were different. It is also not hard to imagine that those who stayed away were also those who had been responsible for behavior incidents before flagging was implemented.

In the intervening years, behavioral alerts became standard. I have seen them at all the hospitals in which I have worked. But I also click through them without the acknowledgment they are intended to promote.

More importantly, as Zoë Kopp, Irina Kryzhanovskaya, and Susannah Cornes note in an excellent Things We Do For No Reason article on the topic, “there is little evidence that [behavioral alerts] actually reduce incidents of violence. Instead, they negatively label patients already at high risk for stigma and have significant potential to bias provider care.”

And yet they persist. They began with a small intervention published in a prestigious journal. That lent them greater support than the data suggested. And there have been no good studies published in the interim to validate these early findings. They are so ingrained that I am not optimistic that we’ll see such a study any time soon.