REVIEW OF PRESCRIPTION MODIFICATIONS: A NOVEL STRATEGY TO ASSESS PHYSICIAN RESPONSE TO MEDICATION ALERTS

Purpose: High volume of irrelevant alerts may lead to “alert fatigue”, whereby physicians largely ignore alerts generated by Clinical Decision support (CDS) that have the potential to prevent unsafe drug administration and ordering.  Prior studies have assessed alert fatigue by assessing physicians’ overriding of alerts.  Unlike these studies, we sought to assess the effects of physicians’ response to alerts by determining whether prescriptions were modified or canceled in response to alerts, or whether the physician chose to override the alerts by proceeding with the prescription without making modifications.

Method: We conducted a retrospective, cross-sectional study of 38 primary care physicians’ usage of the same electronic medical record system with e-prescribing implemented by two regional health information organizations (RHIO’s) in New York over a one year period.  We analyzed audit logs generated by the system which record medications prescribed, alerts generated and actions taken by the physician. We compared the drug name, dose and directions of the initial prescription to the final signed prescription to determine whether a modification was made.

Result: The primary care physicians participating in the study were largely comprised of family practitioners (45%), pediatricians (18%) or internal medicine (37%) physicians. A majority (84%) had been in practice for atleast 10 years.  Physicians wrote 199 (SD 514) prescriptions on average during the study.  Excluding refills, 7732 unique new prescription attempts were made, of which 3394 (43.9%) generated a total of 4823 alerts (1.42 alerts/ prescription with alert).  Of the 3394 prescriptions with alerts, 2921 (86.1%) were completed without any modifications (overridden), 279 (8.2%) were canceled or aborted, and 473 (13.9%) were modified.  Alert acceptance rates varied by type of alert: Duplicate Medication alert (41.6%), Drug Interaction Alert (30.4%) and Elderly Patient Alert (19.6%) had the highest acceptance rates of 23.8%, 18.2% and 22.2% respectively.

Conclusion: We found that only 14% of prescriptions were modified in response to alerts, and another 8% were aborted in response to alerts.  Examining changes made to a prescription in response to an alert may help better elucidate the impact of alerts on actual prescribing behavior, and could be used in conjunction with traditional methods of analyzing alert response that focus on overrides of alerts. The information can then be used to better calibrate CDS systems.