Applying the Levenshtein Distance Algorithm to Detect Look-alike Drug Names

Abstract

Objective: To apply the Levenshtein distance algorithm (LDA) as a tool for detecting look-alike drug names in the Krabi hospital’s drug list for fiscal year 2024. Methods: This experimental study applied the LDA to detect look-alike drug names, focusing solely on drug names and their strength shown on the drug labels on containers dispensed to patients. The study compared the outcomes of similarity assessment of 150 pairs of drug names by using the LDA at threshold levels of 50, 60, and 70 and those by 24 participants who were staff of the pharmacy department at Krabi hospital. Evaluation of the algorithm's performance at each threshold level was based on F-scores and the proportion of false negatives in assessment of drug name pairs, i.e., judged as not being look-alike pairs by the LDA, but judged as being look-alike pairs by the participants. Results: Results of the LDA application at thresholds levels of 50 and 60 to assess the similarity of drug names were not statistically significantly different from those assessed by the participants (P > 0.05). The LDA at a threshold level of 50 showed the highest performance with an F-score of 0.80 and the lowest number of false negative outcomes of assessment (12 drug name pairs). Conclusion: The LDA at a threshold level of 50 can be used as a tool for detecting look-alike drug names. Its detection of look-alike drug names on the drug labels was consistent to those assessed by human. The study provides information for selecting drug name pairs to be included in the hospital's look-alike drug names list. The list could be distributed for alerting health professionals on the risky drug name pairs that may cause medication errors, and for using in the risk management plans for patient drug safety accordingly.