Misspellings in drug information system queries: characteristics of drug name spelling errors and strategies for their prevention.

INTRODUCTION

Efficient search for and finding drugs is essential for electronic drug information systems which, for their part, are prerequisites for computerized physician order entry systems and clinical decision support with the potential to prevent medication errors. Search failures would be critical: they may delay or even prohibit prescription processes or timely retrieval of vital drug information. We analyzed spelling-correction and error characteristics in drug searches and the suitability of auto-completion as prevention strategy.

METHODS

A blank entry field was presented to the user for unbiased queries in a web-based drug information system containing >105,000 brand names and active ingredients accessible from all 5500 computers of the Heidelberg University Hospital. The system was equipped with an error-tolerant search. Misspelled but found drug names confirmed by users were aligned by dynamic programming algorithms, opposing misspelled and correct names letter by letter. We analyzed the ratios of correctly and incorrectly spelled but found drugs, frequencies of characters, and their position in misspelled search words.

RESULTS

Without error-tolerant search, no results were found in 17.5% of all queries. Users confirmed 31% of all results found with phonetic error-correction support. Sixteen percent of all spelling errors were letters in close proximity to the correct letter on keyboards. On average, 7% of the initial letters in misspelled words contained errors.

CONCLUSION

Drug information systems should be equipped with error-tolerant algorithms to reduce search failures. Drug initial letters are also error-prone, thus auto-completion is not a sufficient error-prevention strategy and needs additional support by error-tolerant algorithms.