A Similarity-based Approach and Evaluation Methodology for Reduction of Drug Name Confusion

This paper facilitates the task of mitigating medical errors due to the confusion of look-alike and sound-alike drug names. Detection of potential confusion is based on both feature-based phonetic comparison (for sound-alike drug names) and orthographic similarity (for look-alike drug names). We present a new recall-based evaluation methodology for determining the effectiveness of different similarity measures on drug names. Using this methodology, we show that a new orthographic measure called BI-SIM outperforms other commonly used measures of similarity on a set containing both look-alike and sound-alike pairs, In addition, we demonstrate that the feature-based phonetic approach outperforms other standard approaches on a test set containing solely look-alike confusion pairs. However, an approach that combines several different approaches achieves the best results on both test sets.