When should we believe nonrandomized studies of comparative effectiveness?

Abstract

The demand for data from randomized comparative-effectiveness trials will always outstrip supply. Given their susceptibility to bias, several factors should be considered when examining nonrandomized comparative-effectiveness studies. These include comparability of treatments, magnitude of difference observed, sufficient attention to the underlying biology, examination of relationships supporting the main findings, whether the study includes only new users of the study treatments, whether the study end point is validly recorded, and replication of results.