Abstract
AUTHORS' REPLY: We appreciate Horn and Hansten's interest in our recent publication. Their expertise regarding drug-drug interaction information is well known. We look forward to evaluating the article that describes their efforts to customize a drug-drug interaction clinical decision support system. 1
The current drug-drug interaction alert system is broken. The “out-of-the-box” commercial drug-drug interaction knowledge bases need to be improved. As implied in a recent commentary in The Annals, we spend substantial money for this information, but many question its value. 2
We have no choice but to customize. Horn and Hansten caution that turning off drug-drug interaction alerts by sections (we assume this means by severity) is risky for institutions and patients. However, many institutions have no other option. The “out-of-the-box” alerts in commercial knowledge bases, regardless of severity level, may be worse than no alerts. With many irrelevant alerts, important alerts are obscured; however, it creates a false sense of security and could increase liability. We need solid evidence to show the best methods for optimizing alerts. Which approaches best prevent patient harm? Is there evidence that turning off alerts by categories is worse than other methods, like using the “out-of-the-box” alerts or customized approach using an explicit methodology? Also, the degree of customization varies among clinical decision support systems, adding an additional layer of complexity.
Hiring experts to customize alerts or develop filters to improve upon commercial databases may be an option for institutions that can devote the labor or extra money for this purpose. But why should we have to do this? What is the purpose of severity levels, except to make decisions by category? Shouldn't the system be fixed, rather than using a band-aid approach? We are long overdue for standards for drug interaction alerts that would make expensive consultants and customization unnecessary.
Most alerts occur during order entry or verification. Alerts should occur only when an action is needed. There should be very few alerts that require the order entry or verification process to stop for an intervention (ie, change to a different drug, reduce a dose, order a laboratory test). Information that will help clinicians better monitor the patient should be provided in a rounding document or other report that allows the healthcare professional to do the suggested monitoring at the appropriate time (eg, serotonin syndrome progress note that describes symptoms of this condition). Alerts and monitoring recommendations would ideally be based on the professional's role; for instance, different alerts and rounding documents should be provided to prescribers, pharmacists, nurses, and other health-care professionals.
Although some customization may be needed, clearly defined standards for drug interaction alerts should decrease this need over time. Can we really afford to review 15,000 drug-drug interactions at each institution? Often, the institution assumes the maintenance of the drug-drug interactions after the customization process. It may be unreasonable to maintain all needed changes. And why are we paying for the knowledge base in the first place?
