Background: Computerized physician order entry systems are known to improve patient safety in acute-care hospitals. However, as clinicians frequently override drug interaction and allergy alerts, their value in ambulatory care remains uncertain.
Objective: The purpose of the study was to examine whether ambulatory care clinicians were more likely to accept drug-drug interaction alerts that an expert panel judged to be of high clinical value.
Study design: We convened an expert panel to examine drug-drug interaction alerts generated by 2872 clinicians in Massachusetts, Pennsylvania and New Jersey who used a common electronic prescribing system between 1 January 2006 and 30 September 2006. We selected 120 representative drug interaction alerts from the most commonly encountered class-class interactions.
Measurements: The expert panel rated each alert based on the following categories: (i) strength of the scientific evidence; (ii) probability that the interaction would result in an adverse drug event (ADE); (iii) severity of typical and most serious ADEs; (iv) the likelihood that a clinician could act on the information; and (v) the overall value of the alert to the average primary care clinician. We then used multivariate regression techniques to examine the relationship between the alert acceptance rate and the expert panel's mean rating of each category.
Results: The decision of clinicians to accept drug interaction alerts increased (relative to a baseline alert acceptance rate of 8.8%) by 2.7% (95% CI 0.4, 5.1) for interactions that panelists judged would result in an ADE, by 2.3% (95% CI 0.9, 3.7) when primary care providers (PCPs) lacked prior knowledge about the information presented in the alert, and by 3.3% (95% CI 0.9, 5.8) when the PCP could readily act on the information provided in the alert.
Conclusion: The value of electronic drug interaction alerts is influenced heavily by clinicians' judgements about the clinical value of the alert. Expert judgement should be taken into account when developing electronic decision support.