Title: Applying Natural Language Processing to Extract and Codify Adverse Drug Reaction Data in Medication Labels

Objective To develop an automated method of identifying, extracting and codifying adverse drug reaction (ADR) data contained in the Structured Product Labels (SPLs) for drugs to be studied as part of the Observational Medical Outcomes Partnership (OMOP) project. Background Natural Language Processing (NLP) technology has been used in the medical domain to identify, extract and codify data in biomedical literature and narrative clinical reports. It has also been applied to detect ADRs from narrative clinical reports. We will use NLP to extract ADR data from free text SPLs Method We modified and enhanced an existing NLP program –the Regenstrief EXtraction Tool (REX) for this project to create the Structured Product Label Information Coder and ExtractoR (SPLICER). SPLICER consists of threes main modules: an SPL Parser an ADR extractor, and a MedDRA mapper. It is programmed to first identify ‘raw’ ADR terms contained in the SPL, and then map them to their corresponding MedDRA terms. We also performed an evaluation of SPLICER accuracy using a small sample of SPLs. Results SPLICER processed a total of 80 OMOP drugs that represented 433 drug labels. SPLICER found a total of 40,433 unique ADRs, and achieved a recall, precision and Fmeasure of 96.3, 97.2 and .97 respectively. Conclusion SPLICER accurately identified, extracted, and codified the majority of ADR data contained in SPLs. Its output – a structured ADR database – has value for several potential applications. Friedlin NLP to extract ADR data from SPLs