Making inferences on treatment effects from real world data: propensity scores, confounding by indication, and other perils for the unwary in observational research | Synapse