Assessment and control of confounding in trauma research

Confounding is a central problem in nonexperimental trauma research. In this article, the authors review recent advances in the theoretical conception of confounding, with an emphasis on the counterfactual definition of confounding. The strengths and limitations of different techniques for controlling confounding are discussed. Special attention is given to propensity scores, a method that has been used widely in the health sciences, but only rarely in trauma research, in the last several years. The article is written for researchers who use data analytic methods to control confounding and for clinicians who read original research articles to inform their clinical practice. Guidelines to assess the appropriate use of multiple regression models and propensity scores are offered.