Mining Adverse Drug Reactions from Electronic Health Records

Over 2 million serious side effects, including 100,000 deaths, occur due to adverse drug reactions (ADR) every year in the US. Though various NGOs monitor ADRs through self reporting systems, earlier detection can be achieved using patient electronic health record (EHR) data available at many medical facilities. This paper presents an algorithm which allow existing ADR detection methods, which were developed for spontaneous reporting systems, to be applied directly to the longitudinal EHR data, as well as a new ADR detection method specifically for this type of data. Preliminary results show that the new method outperforms existing methods on EHR datasets. Future work on the method will extend it to detecting potential cause-effect relationships between events in other types of longitudinal data, handling multiple cause and effect items, and automatically selecting surveillance windows.