Abstract
Objective Clinical decision supports (CDS) in electronic medication order systems identify alerts for clinicians. However, CDS may cause alert fatigue, which is the tendency for clinicians to ignore prompts presented by CDS due to excessive numbers and/or their perceived limited clinical significance. Alert fatigue may increase the risk of missing clinically relevant alerts. At North York General Hospital, pharmacists managed over 50% of all medication CDS alerts amounting to approximately 60 alerts per day per pharmacist with an override rate of over 90% indicating a high likelihood of alert fatigue. Thus, we attempted to reduce pharmacists’ alert fatigue utilizing customizable filters.
Methods Optimizing medication CDS has traditionally centered around turning on or off alerts, changing alert severity levels or clinician role tailoring. These strategies can be labor and time-intensive requiring clinicians from different specialties to review hundreds of individual alerts. As such, this study pursued the use of customizable, context-based filters to reduce unnecessary alerts. Utilizing data from the EHR vendor’s visual analytics dashboard and guided by pharmacists’ feedback, three customizable filters were applied. First, a filter to suppress alerts for medications that are ordered by the same prescriber during one session was implemented. Second, a filter to reduce alerts for medications that are commonly ordered both as scheduled and as needed was applied. Finally, customization was done on how long discontinued medications are eligible for alert checking by the medication CDS system. Data was collected 1 month prior to and 3 months after implementation for a duration of one month each. Alerts data was taken from the analytics dashboard. Pharmacists’ perceptions of alert fatigue were collected using a voluntary online survey. Adverse medication events data was obtained from the hospital’s incident reporting tool.
Results Comparing before and after implementation, total alerts decreased by 48.4% for pharmacists. In practice, this represented a reduction from 59.7 to 27.1 medication CDS alerts per day per pharmacist. However, pharmacists’ alert override rate was minimally changed from 98.1% to 97.3%. Fourteen (78%) of the 18 pharmacists surveyed felt there was an overall decrease in unnecessary alerts while 67% perceived they were able to spend more time on reviewing meaningful alerts post-implementation. Compared to pre-implementation, pharmacists reported a minor reduction in the percentage of alerts they deemed unnecessary or inappropriate from 66.8% to 59.3%. However, 78% still remarked that there was room for improvement in the CDS alerting system. The number of adverse medication incidents were similar between the periods before and after implementation. No incidents were found to be a result of the new customized contextual filters.
Conclusion The use of customizable filters may be a viable alternate approach to reducing alert volume without needing to completely turn off specific alerts or changing alert severity. Pharmacists’ perceptions of alert fatigue appeared to improve modestly post implementation. Comparison of medication incidents before and after implementation did not show an increase in medication errors. However, override rates remain elevated and pharmacists felt that further improvements could still be made to the medication CDS system.