Medication safety screening, as I have written about at length previously, has the potential to reduce prescribing errors and therefore to improve patient safety. I have also previously described alert fatigue, which occurs when the signal:noise ratio is so low that providers start to ignore the alerts. Design of clinical decision support (CDS) systems is therefore a critical factor in order to optimize the impact of these systems on patient care.
In a recent article in the International Journal of Medical Informatics, Horsky et al reviewed the literature in this area and formulated some recommmendations. Below are some highlights of their recommendations, from my point of view.
Use severity levels
In the category of reducing excessive alerting, the authors state the importance of tiering alerts by severity level, which is an important attribute of drug interaction databases, such as Wolters Kluwer’s Medi-Span® database. How the severity level is used varies by implementation, but in general, more severe alerts are generally configured as interruptive, whereas less severe alerts may be configured as informative. Interruptive alerts require acknowledgment by the user, whereas informative alerts provide information on the screen but do not require explicit acknowledgment. When a user acknowledgment is required, some systems require the user to provide a reason for overriding the alert. The authors suggest presenting a short list of reasons in a dropdown list.
Ensure correct underlying patient data
Still in the category of reducing excessive alerting, the authors emphasize the importance of having correct patient data in the electronic medical record (EMR). For example, if a patient is listed as having an allergy to penicillin, then allergy alerts will be triggered whenever penicillin or a similar drug is prescribed. If the patient is in fact not allergic to penicillin, the authors state that this allergy should be removed from the patient’s profile.
Use patient context
The authors also talk about using additional patient context as the basis to filter alerts. I previously summarized another paper looking at ways to develop these contextual models. In addition, the authors suggest distinguishing between new orders and renewals of existing orders, since in the latter case, the patient may already be tolerating the drug combination without any problems, so displaying the same alerts every time the medication is renewed contributes to alert fatigue.
Design of alert boxes
The authors provide a detailed table of optimal design attributes for alerts, making recommendations in areas such as color (e.g. reserve red for severe situations), size (e.g. variable), layout (e.g. simple geometry), font (e.g. draw attention to the drug names) and language (e.g. clear and concise). For the content, they recommend providing a link to a detailed explanation. For example, Medi-Span® contains a detailed monograph for each drug-drug interaction.
The authors indicate that physicians may be allowed to turn off individual alerts with certain caveats related to their knowledge and comfort level. For example, Wolters Kluwer’s Medi-Span® Clinical software provides this capability.
In summary, the authors provide a practical set of design recommendations to reduce alert fatigue. The health information technology (HIT) community would be wise to adopt their excellent advice.