Drug name confusion errors in the published literature and the USP Medication Errors Reporting Program, 1965-2004: a reconstructed dataset of 724 confused drug name pairs

Background. Medication errors caused by confusion between similar drug names (look-alike/sound-alike, or LASA, errors) have been reported for decades, but the underlying case reports have never been assembled into a public, machine-readable dataset. Between 2002 and 2005 we compiled such a collection from the published literature and from the United States Pharmacopeia (USP) Medication Errors Reporting (MER) Program. It remained in five unreleased Microsoft Access databases. Methods. In 2026 we exported and reconciled the five databases into case-level and pair-level files. We mapped drug names to RxNorm and to WHO Anatomical Therapeutic Chemical (ATC) classes, computed five orthographic similarity measures, and recorded membership on the FDA and ISMP lists of drug names with recommended Tall Man letters. A limited PubMed search added pairs reported from June 2005 to February 2026. We describe the collection and compare the two sources. Results. The dataset contains 709 literature case records (1965-2004), 753 USP reports (1997-2003), and 17 post-2005 records, covering 724 confused drug name pairs. Only 69 of the 708 original pairs appeared in both sources. Published cases were more severe than voluntary reports: among records with a coded outcome, 15.2% of literature records and 1.2% of USP reports were fatal. Similarity measures discriminated reported pairs from random re-pairings of the same names (area under the ROC curve 0.88-0.92), but 30% of pairs had relatively low measured similarity (BI-SIM below 0.40). Of the pairs with at least three records or at least one fatal-coded record, 122 appear on neither Tall Man list. Automated approximate matching of drug names to RxNorm, used in building the dataset, itself produced name-confusion errors of the kind the dataset records; these were found by manual review and removed. Conclusions. This reconstructed collection is a public historical record of drug name confusion through 2004. It shows how little published and voluntarily reported confusions overlap. It can serve as a benchmark for name-similarity methods and as a starting point for review of confusable names.