The FDA requires manufacturers to conduct postmarketing studies when there are suspected safety issues arising from product use. Randomized clinical trials for assessing safety are often not feasible, and more importantly not ethical, to conduct. Alternatively, observational cohort studies of large healthcare databases have been used to fulfill postmarketing requirements. There are many sources of bias and issues of confounding that make drawing statistical inference difficult when assessing product safety with such non-randomized study designs. As a result, there has been increased use of propensity scores (PS) which attempt to account for these issues. It is not clear, however, the degree to which confounding effects are reduced with PS methods and whether it is necessary that statistical analyses account for matching when PS matching is used. 1. What are some key considerations when designing observational database studies that use PS methods? 2. What methods can be used to measure the degree by which confounding effects are reduced by PS? 3. Is it necessary to account for matching in PS matched data analysis, for example, stratified analysis by matching identifier?