An RCT Simulation Study on Performance and Accuracy of Inexact Matching Algorithms for Patient Identity in Ambulatory Care Settings

Patient misidentification has been identified as a primary cause of electronic medical record (EMR) related patient harm. One potential function at which such errors can occur is the user interface, during patient look-up. In this study we apply a within participant randomized control trial (RCT) simulation to assess the impact of a variety of inexact matching techniques on the rate of patient misidentification. We compare a string edit distance algorithm (Jaro-Winkler) and a phonetic algorithm (Metaphone) against a commonly used leading sequence match algorithm baseline. The algorithms were evaluated based on response distribution to the stimulus and on processing time. Our participant group of 24 consisted primarily of students. Our results indicate that the baseline algorithm performs significantly better than either intervention algorithms on the basis of processing time, but on the basis of correctness it performs more poorly than the interventions when the stimulus is in the searchable set. The primary outcome of our study is the recommendation to only employ the interventions in clinical settings where patients´ who´s names are sought are present rather than novel (e.g., Community clinics and inpatient settings). Where patients are typically novel (e.g., Walk in clinics and emergency rooms), we recommend the use of the baseline algorithm instead.