Comparative effectiveness studies of medical products outside of experimental research settings often derive from secondary data sources, such as insurance claims, that include data from many geographic regions and sometimes the entire country. The large source population available through these data sources allows investigators to study uncommon outcomes or exposures. Indeed, such considerations are important for the FDA’s Sentinel Initiative, which aims to create a coordinated national electronic medical product safety surveillance system covering 100 million patients from disparate data sources and thereby geographic regions by mid 2012. A recent publication,1 however, has cast doubt on the validity of this venture owing to potential biases arising from regional variations in medical practice intensity.
