Retrospective detection of potential medication errors involving drugs with similar names

To estimate frequencies of potential errors involving similarly named drugs using a retrospective claims database and measure the association between frequencies of potential errors and two measures of drug name similarity, edit distance (minimum number of insertions, substitutions, or deletions of characters required to change a given word into another target word) and normalized edit distance (proportion of letters that must be changed to commute one word to another, and ranges from 0 to 1, with 0 indicating identical words, and 1 indicating a pair of words with no common letters).

Retrospective database analysis.

Idaho Medicaid claims data from 1993 to 2000.

Not applicable.

Potential errors were detected using adjacent claims generated by dispensing of one drug followed by dispensing of the other drug with a similar name. In all, four potential error criteria were developed: two for detecting potential refill errors and two for detecting potential initial errors. A total of 10 drug pairs were randomly selected from the Idaho Medicaid claims database for each value of edit distance, which ranged from 1 to 30 (n = 300).

Frequencies of potential medication errors in claims sequences for initial and refill claims, edit distance, and normalized edit distance.

Of 300 drug pairs studied, 106 (35.33%) were involved in at least one potential error. A total of 1,138 dispensing episodes satisfied the criteria for potential errors. Frequencies of potential errors per drug pair were negatively associated with edit distance (r = -0.133, P < .05) and normalized edit distance (r = -0.226, P < .01). Frequencies of potential initial errors also were negatively associated with edit distance (r = -0.126, P < .05) and normalized edit distance (r = -0.222, P < .01). Potential refill errors also had negative association with edit distance (r = -0.134, P < .05) and normalized edit distance (r = -0.226, P < .01).

Error criteria were successfully applied to a retrospective claims database to detect potential initial and refill errors that involved similarly named drugs.