Mining electronic health records for adverse drug effects using regression based methods

The identification of post-marketed adverse drug events (ADEs) is paramount to health care. Spontaneous reporting systems (SRS) are currently the mainstay in pharmacovigilance. Recently, electronic health records (EHRs) have emerged as a promising and effective complementary resource to SRS, as they contain a more complete record of the patient, and do not suffer from the reporting biases inherent to SRS. However, mining EHRs for potential ADEs, which typically involves identification of statistical associations between drugs and medical conditions, introduced several other challenges, the main one being the necessity for statistical techniques that account for confounding. The objective of this paper is to present and demonstrate the feasibility of a method based on regression techniques, which is designed for automated large scale mining of ADEs in EHR narratives. To the best of our knowledge this is a first of its kind approach that combines both the use of EHR data, and regression based methods in order to address confounding and identify potential ADEs. Two separate experiments are conducted. The results, which are validated by clinical subject matter experts, demonstrate great promise, as well as highlight additional challenges.