Examination of medication clinical decision support using Bayes’ theorem

Drug–drug interaction (DDI) alerts are the most common computerized clinical decision support alerts encountered by ordering clinicians.1 However, DDI alerts are unpopular and overridden a large percentage of the time, and, as measured by published override rates and anecdotal reports from clinicians, not much progress has been made in improving clinician acceptance of DDI alerts (Figure 1).2–6 Override rates are a crude measure but remain a reasonable proxy for physician acceptance. Efforts to improve acceptance of DDI alerts have included reducing the number of drug pairs that trigger them,7 alerting only for the most serious DDIs, applying principles of alert theory and usability to the alerts,8 and developing frameworks to evaluate clinical decision support systems.9 What else might we do to increase their acceptance? ... It may help to think of a DDI alert as a test—a test that is run each time a new medication is ordered. The intent of the test is to highlight the risk of a harmful drug interaction so that the order can be reconsidered, special monitoring planned, or other action taken.