Evaluation of signal detection methods for use in prospective post licensure medical product safety surveillance

The Sentinel Initiative aims to develop a national, electronic network to link data on 100 million patients from multiple existing health care data systems by 2012 in order to conduct post-licensur e safety monitoring of FDA-regulated medical products. A critical component of this Initiative is determining what statistical methods can bes t be employed within this framework to accurately, robustly, and flexibly detect safety problems. In this report, we review existing signal detection methods for possible use in the Sentinel Initiative. We statistically and clinicall y describe each method, summarize prior and current uses 0 f each method, and assess what is kn own about each method's robustness across different data sources and each method's flexibility to accommodate different signaling threshold levels and different types of adverse event outcomes. Recommendations for pursuing the most promising signal detection methods for use in the Sentinel Initiative are given.