Similarity Mapping of National Drug Code Formulary Systems Between Nations

Background: A fundamental problem in international health care research is comparing national prescribing practices and drug formularies between nations that use inconsistent drug naming, dosages, and packaging sizes. However, there is currently a lack of quality mapping between different ontology systems. This study aims to evaluate the efficiency of automated mapping by drug names using similarity measure algorithms. The motivating example was cross-mapping between the British National Formulary (BNF) and the Anatomical Therapeutic Chemical (ATC) drug classification systems. Method: We used edit-based, token-based, and hybrid similarity algorithms to match the drug’s active pharmaceutical ingredients between classification systems. Results: : A variety of similarity measures were also compared. We found a substantial trade-off between optimizing matching and false positives for the cross-mapping of drug datasets. Conclusion: In general, edit-based algorithms with a pre-processing step to remove stop words had a consistently good performance by recall, precision, and F1 scoring.