The Predicament of Comparative Effectiveness Research Using Observational Data