An Antimicrobial Prescription Surveillance System That Learns from Experience

Inappropriate prescribing of antimicrobials is a major clinical concern that affects as many as 50 percent of prescriptions. One of the difficulties of antimicrobial prescribing lies in the necessity to sequentially adjust the treatment of a patient as new clinical data become available. The lack of specialized health-care resources and the overwhelming amount of information to process make manual surveillance unsustainable. To solve this problem, we have developed and deployed an automated antimicrobial prescription surveillance system that assists hospital pharmacists in identifying and reporting inappropriate prescriptions. Since its deployment, the system has improved antimicrobial prescribing and decreased antimicrobial use. However, the highly sensitive knowledge base used by the system leads to many false alerts. As a remedy, we are developing a machine-learning algorithm that combines instance-based learning and rule induction techniques to discover new rules for detecting inappropriate prescriptions from previous false alerts. In this article, we describe the system, point to results and lessons learned so far, and provide insight into the machine-learning capability.