Background: Computerized drug-drug interaction (DDI) screening is widely used to identify potentially harmful drug combinations in the inpatient and outpatient setting.
Objective: To evaluate the performance of drug-drug interaction (DDI) screening software in identifying select clinically significant DDIs in pharmacy computer systems in community and hospital pharmacies.
Methods: Ten community pharmacies and 10 hospital pharmacies in the Tucson metropolitan area were invited to participate in the study in 2004. To test the performance of each of the systems used by the pharmacies, 25 medications were used to create 6 mock patient profiles containing 37 drug-drug pairs, 16 of which are clinically meaningful DDIs that pose a potential risk to patient safety. Each profile was entered into the computer pharmacy system, and the system response in terms of the presence or absence of a DDI alert was recorded for each drug pair. The percentage of correct responses and the sensitivity, specificity, positive predictive value, and negative predictive value of each system to correctly classify each drug pair as a DDI or not was calculated. Summary statistics of these measures were calculated separately for community and hospital pharmacies.
Results: Eight community pharmacies and 5 hospital pharmacies in the Tucson metropolitan area agreed to participate in the study. The median sensitivity and median specificity for community pharmacies was 0.88 (range, 0.81-0.94) and 0.91 (range, 0.67-1.00), respectively. For hospital pharmacies, the median sensitivity and median specificity was 0.38 (range, 0.15-0.94) and 0.95 (range, 0.81-0.95), respectively.
Conclusion: Based on this convenience sample of 8 community pharmacies and 5 hospital pharmacies in 1 metropolitan area, the performance of community pharmacy computer systems in screening DDIs appears to have improved over the last several years compared with research published previously in 2001. However, significant variation remains in the performance of hospital pharmacy computer systems, even among systems manufactured by the same vendor. Future research should focus on improving the performance of these systems in accurately and precisely identifying DDIs with a high probability of resulting in true potential adverse effects on clinical outcomes and creating a low .noise. ratio associated with false-positive alerts.