Over the past few years, for several classes of drugs, formal meta-analyses have been used by the FDA to inform changes to class labels (e.g., risk of suicidality with anti-epileptic drugs and with SSRI’s in adolescents and young adults). In some instances, however, meta-analyses have been considered insufficient to address specific safety concerns (e.g., mortality risk with use of erythropoietin-stimulating agents in the oncology setting). In presenting these examples, the emphasis will be on the principles that need to be considered in conducting such meta-analyses, such as adjudication of endpoint events, broad vs. narrow definitions of events, statistical methods for rare events, and investigating possible treatment effect modifiers.