Data Mining Electronic Health Records for Drug Adverse Events

Harvard Pilgrim Health Care, Inc., Boston MA

Investigators

Abstract

Description (provided by applicant): This application addresses broad challenge area 10: Information Technology for Processing Health Care Data for Research, and the specific challenge topic 10-LM-101: Informatics for post-marketing surveillance. When a new pharmaceutical drug enters the market, there are always questions about its effectiveness and safety profile in comparison with other related drugs. Post-marketing drug safety surveillance is important in order to detect serious adverse events that are too rare to be detected during phase three clinical trials. Such surveillance has traditionally been based on spontaneous adverse event reporting systems but electronic health records from health insurance plans are now increasingly being used instead. In this project we will use two data mining methods, empirical Bayes gamma Poisson shrinkage and the tree-based scan statistic, to search for unexpected acute drug adverse events. We will use a 4.5 million patient electronic health records database from three health insurance plans: Harvard Pilgrim Health Care, Kaiser Permanente Northern California and Kaiser Permanente Colorado. We will assess the safety of the majority of commonly used pharmaceutical drugs. Any signals detected will be evaluated through drug and diagnosis specific logistic regression analyses on the same data, with a variety of covariance adjustments, as well as through temporal scan statistics that looks at the temporal distribution of the time between drug initiation and the adverse event.