A Q&A About Alert Fatigue
Computerized order entry has popped up in a lot of recent conversations I’ve had about HITECH incentives. Not surprisingly, hospitals ramping up order entry systems are having a devil of a time avoiding programming alert fatigue into their CPOEs.
Following is a Q&A session on the topic I had with Linda Peitzman, M.D., who heads the clinical solutions group at Wolters Kluwer Health. Linda previously was a practicing physician and medical director at HealthSystem Minnesota/Park Nicollet Clinic.
Is there an alert(s) for a specific condition or medication that frequently crop up as a problem when it comes to alert fatigue?
Studies have shown that drug-drug alerts related to contraindications like allergies cause a lot of fatigue because they are very common and they tend to be implemented in the least specific ways.
But the real problem isn’t with certain medication s or conditions; the real problem lies with alerts that are not specific enough or that are inappropriate. For example, if an alert about the potential for an allergic reaction fires every time aspirin is ordered, whether or not the patient is allergic, that is problematic.
Also problematic are alerts that are not designed to fit within clinician workflows. Some clinicians prefer it when alerts only fire for certain conditions or in certain clinical settings, such as the ICU. When clinician preferences are ignored, alerts will be overridden and frustration will grow until it spills over to all alerts. This ultimately impacts the effectiveness of the entire clinical decision support system.
Have you identified a threshold where alerts become an obstacle rather a benefit to physicians?
If alerts fire at inappropriate times or are too vague or inaccurate, they become obstacles that will affect all clinicians, including those who would otherwise have benefited from them. It also depends upon how CDS tools, including alerts and order sets, are implemented within the facility’s EMR or CPOE system. Different systems have different interfaces that impact different parts of the clinical workflow. It is very institution- and alert-specific.
What have you found is the best approach to tackling alert fatigue from a technology perspective?
First, it’s important to understand that alerts are just one CDS tool. There might be other, more appropriate, ones that can effectively impact clinician behavior without causing fatigue, such as order sets. So most important is ensuring that the “right” CDS tools are being utilized.
However, once the decision is made that alerts are the way to go, there are five things that can be done to reduce or eliminate fatigue. First is to choose those areas where alerts really can have a significant patient and organizational impact. If 99% of MI patients are given aspirin, you probably don’t need an alert to remind clinicians to do so. Save alerts for areas where there are higher error rates or greater potential for adverse outcomes. When fewer alerts are triggered, and those that are fired are appropriate and impactful, you have less fatigue.
Second, make sure alerts are as specific as possible. This does depend somewhat on the technology and underlying clinical content, but if you can be very specific with the alert so it only fires when a set of criteria, such as a patient’s gender, age, condition, allergies and medication, are met or only in certain clinical areas, it helps ensure that the alert is appreciated and helpful rather than an obstruction.
Third, ensure that alerts can to be customized to match specific organizational needs. Hospitals should be able to modify alerts over time. Sometimes, technologies and content don’t allow that customization and modification as much as they should.
Fourth, involve clinicians in establishing the alerts. If you start firing alerts and the clinician really isn’t aware of the decision-making behind that alert, it can become an obstruction. If clinicians understand why and how an alert is established, the system will work better.
Finally, you have to monitor and continually refine alerts. It goes beyond determining how often they’re ignored or overridden to include how they are affecting care over time. What feedback are you getting? How can you measure or solicit input on unintended consequences? Someone needs to be in charge of this.
Have you found that alert fatigue is most common right after CPOE is implemented, or is it something that builds up over time?
Alerts aren’t always only implemented within CPOE system, but if we’re talking specifically about CPOE then it is completely dependent upon how they are implemented. Some prefer to implement CPOE first and then add alerts later or carefully implement only a few very specific alerts first. These approaches tend to work well in the long-term and keep alert fatigue to a minimum. If you immediately turn on all possible alerts at the same time CPOE is implemented, they are annoying from the start and that tends not to change over time. But again, it differs from one CPOE system to the next.
What's the worst real-world example of alert fatigue you've encountered?
It’s really more of a “worst type of alert” problem, because there are all kinds of alerts and some are more appropriate in certain situations than others. “Uninterruptive” alerts pop up in the background and don’t require any action on the clinician’s part, while “interruptive” alerts force some sort of response before you can move on. To me, the worst possible type of alert is an interruptive alert that is inaccurate or non-specific, because it needlessly takes time away from the clinician. It stops the workflow and can cause problems because care can’t be completed.
Beyond that, there are anecdotal examples that are just silly, like multiple alerts that pop up for every possible interaction, including food, whenever a particular drug is ordered. This forces clinicians to click through every single sequence, every single time. I’m not sure that it’s dangerous to the patient, but it can be very dangerous for the computer monitor.
How can providers in the implementation stage of CPOE best avoid programming too many alerts into the software?
By implementing alerts slowly over time rather than all at once and ensuring that those alerts are as meaningful and specific as possible. This is accomplished by using a drug database that is capable of targeting alerts to very specific patient conditions. It’s also important that the CPOE system and the medication CDS system both allow modifications to be made to alerts based on clinical and workflow parameters. It doesn’t do any good if the EMR allows customization but the underlying drug database doesn’t have the corresponding ability to say when an alert should be fired. Finally, it helps to understand that alerts are only one tool in the CDS toolkit. Often times, there are other decision support tools, such as order sets, that are equally effective but without the risk of alert fatigue.