Alert Fatigue

28 Dec 2008 - 4:16pm
5 years ago
2 replies
778 reads
Michael Kordek
2008

Dear Professors
Hello, and thanks for any help ahead of time.
I am a Medical Informatics student at OHSU and right now I am in a research
methods class at my institution: Geisinger.
For My project I wanted to investigate Alert Fatigue, suffering, myself,
from the condition.
So far most of what I can document comes from high-risk industries like
nuclear power, aerospace and the military. There is very little data in the med
informatics literature.
As I begin the topic though, I can find very little nomenclature defined
either.
Is there taxonomy of alerts? And as I start, I really have not found a
definition of an alert either.
Is there an official medical informatics definition of ‘alert’? I propose
the definition below, would you please comment on it?
Alert: “unsolicited 1 signal 2 of [possible/potential/definite] 3 error
condition”
explanation:
(1) a self posted reminder, like ‘3 days until tax day’ would not be an
alert for my purposes.
(2) Any kind of signal or communication, it could be a vibrating watch, and
it does not have to be interruptive or require an action to continue.
(3) Many alerts are before the fact, and in medinfo most are trivial or
false alarms.
I would appreciate any help in pointing me to medical informatics articles
on experiments in alert fatigue. There is a lot of data about alerts, and
about overrides, but not a lot about the human behavior and what makes an alert
productive and what makes an alert cause other alerts to fail. I have found
nothing about how many alerts the mind can take before it zones out. And even
the articles that mention how many alerts were fired in a system and
overriden, there is nothing I have found yet about how many alerts per encounter, and
how many alerts per doc per day.
I have found a few articles that look at the fatigue effect itself. A 2008
preprint on overrides vs. number of alerts and a 2000 article that talks about
a drag effect of one alert on overall compliance. These are two of the
articles that I found that talk about alert fatigue:

Tiering Drug-Drug Interaction Alerts by Severity Increases Compliance Rates
http://www.jamia.org/cgi/reprint/M2808v1

Improving Allergy Alerting in a Computerized Physician Order Entry System
http://www.pubmedcentral.nih.gov/picrender.fcgi?artid=2243998&blobtype=pdf

There are 3 things that I would like to do in my project:
1: define an alert, and come up with taxonomy of alerts, there are some
textbooks that mention this.
2: present a simple count of alerts by taxonomy and by per doctor per day.
So far all I have found is number of alerts in a system over a month or a
year, nothing from the doctor point of view or using the doctor in the
denominator.
3: start to define the fatigue factor itself. This is based on a hypothesis
that the more alerts the more overrides and the less time an alert is on
screen.

A simple approach to the third step would be percent of overrides morning
vs. afternoon, Monday vs. Friday as in the sample attached graph called
overrides. By pairing Monday morning with Monday afternoon etcetera, we can show
more significance, and by controlling for rates within each doctor
morning/afternoon//Monday-Friday we can get even more significance.

Any advice on the next step and other articles would be very welcome.

Thank you
Michael Kordek
**************One site keeps you connected to all your email: AOL Mail,
Gmail, and Yahoo Mail. Try it now.
(http://www.aol.com/?optin=new-dp&icid=aolcom40vanity&ncid=emlcntaolcom00000025)

Comments

31 Dec 2008 - 2:46pm
Marc Resnick
2006

I would recommend looking at the Human Factors Literature. The field
has many researchers who specialize in various kinds of warnings,
including alerts.

You can start at the HFES web site which has archives for a variety
of their publications http://www.hfes.org/Publications/

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Posted from the new ixda.org
http://www.ixda.org/discuss?post=36778

31 Dec 2008 - 6:17pm
Bruce Esrig
2006

The model for how the alerts are generated, automatically filtered, and
distributed will be an essential part of the analysis.

If all alerts reach a human, then there's no way to win. Alerts increase
until all the humans that you can afford to notify are exhausted.

The key is to look at the events that are generating alerts and figure out
how to winnow the alerts so the significant events (that you can afford to
respond to) get a response from a qualified responder (who actually has the
time, attention, and resources required to respond effectively).

Best wishes,

Bruce Esrig
Madison, NJ

On Wed, Dec 31, 2008 at 2:46 PM, marc resnick <resnickm at fiu.edu> wrote:

> I would recommend looking at the Human Factors Literature. The field
> has many researchers who specialize in various kinds of warnings,
> including alerts.
>
> You can start at the HFES web site which has archives for a variety
> of their publications http://www.hfes.org/Publications/
>
>
>
>
> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
> Posted from the new ixda.org
> http://www.ixda.org/discuss?post=36778
>
>
> ________________________________________________________________
> Welcome to the Interaction Design Association (IxDA)!
> To post to this list ....... discuss at ixda.org
> Unsubscribe ................ http://www.ixda.org/unsubscribe
> List Guidelines ............ http://www.ixda.org/guidelines
> List Help .................. http://www.ixda.org/help
>
>

Syndicate content Get the feed