Automatic detection of drug interaction mismatches in package inserts

The US Code of Federal Regulation (21 CFR 207) mandates that pharmaceutical manufacturers submit their FDA-approved drug information as medication Package Insert (PI). PI should provide comprehensive, current, and accurate information about the medical use of a drug. However, PI narratives are cumbersome for healthcare providers to navigate and are therefore rarely used by them. We are developing Paracelsus, a tool for automatically extracting structured drug information from PI. Paracelsus has the potential to allow healthcare providers to benefit from the rich drug information in PI for patient care. In this study, we report the development and evaluation of Paracelsus on drug-drug interactions. We show that Paracelsus performs with a high accuracy, discovering interactions not covered by other medical compendia, and in addition automatically detecting inconsistencies in the package inserts.