Limitations of postmarketing surveillance in the analysis of risk of pregnancy loss associated with maternal mefloquine exposure.