Identification of Confusable Drug Names: A New Approach and Evaluation Methodology

Identification of Confusable Drug Names: A New Approach and Evaluation Methodology Grzegorz Kondrak Department of Computing Science University of Alberta Edmonton, Alberta, Canada, T6G 2E8 kondrak@cs.ualberta.ca Bonnie Dorr Institute for Advanced Computer Studies & Department of Computer Science University of Maryland College Park, 20742, USA bonnie@umiacs.umd.edu Abstract This paper addresses the mitigation of medi- cal errors due to the confusion of sound-alike and look-alike drug names. Our approach involves application of two new methods— one based on orthographic similarity (“look- alike”) and the other based on phonetic sim- ilarity (“sound-alike”). We present a new recall-based evaluation methodology for deter- mining the effectiveness of different similar- ity measures on drug names. We show that the new orthographic measure (BI-SIM) outper- forms other commonly used measures of sim- ilarity on a set containing both look-alike and sound-alike pairs, and that the feature-based phonetic approach (ALINE) outperforms or- thographic approaches on a test set contain- ing solely sound-alike confusion pairs. How- ever, an approach that combines several differ- ent measures achieves the best results on both test sets. 1 Introduction Many hundreds of drugs have names that either look or sound so much alike that doctors, nurses and pharmacists can get them confused, dispens- ing the wrong one in errors that can injure or even kill patients. In the United States alone, an esti- mated 1.3 million people are injured each year from medication errors, such as administering the wrong dose or the wrong drug (Lazarou et al., 1998).1 The U.S. Food and Drug Administration has sought to mitigate this threat by ensuring that proposed drug names that are too similar to pre-existing drug names are not approved (Meadows, 2003). A number of different lexical similarity measures have been applied to the problem of identifying con- fusable drug names. Lambert et al. (1999) tested twenty two distinct methods on a set of drug names extracted from published reports of medication er- rors. The methods included well-known universal measures, such as edit distance and longest common 1For example, a patient needed an injection of Narcan but instead got the drug Norcuron and went into cardiac arrest. subsequence, several variations of measures based on counting common letter � -grams, and measures designed specifically for associating phonetically similar names, such as Soundex and Editex. They identified the normalized edit distance, Editex, and a trigram-based measure as the most accurate. The evaluation methodology of Lambert et al. (1999) involves repeated selection of cut-off thresh- olds in order to compute precision and recall on a test set that contains equal number of positive and negative examples of confusable drug name pairs. However, our own experience with systems for au- tomatic detection of potential drug-name confusions suggests that the usual approach is to examine a fixed number of most similar candidates rather than all candidates with similarity above certain thresh- old. Moreover, the number of non-confusable pairs can be expected to greatly exceed the number of confusable pairs. We present a different method of evaluating the accuracy of a …

Content truncated to 500 words