Confounding is an important source of bias in nonexperimental studies, arising when the effect of an exposure on the occurrence of an outcome is distorted by the effect of some other factor. In nonexperimental studies of patients with CKD or who are on chronic dialysis, confounding is a significant concern owing to the high burden of comorbid disease, extent of required clinical management, and high frequency of adverse clinical events in this patient population. Confounding can be addressed in both the design stage (restriction, accurate measurement of confounders) and analysis stage (stratification, multivariable adjustment, propensity scores, marginal structural models, instrumental variable) of a study. Time-dependent confounding and confounding by indication are 2 special cases of confounding that can arise in studies of treatment effects and may require more sophisticated analytic techniques to adequately address. The availability and expanded use of large health care databases have ensured greater precision and have now placed the focus on validity. Addressing the major threats to validity, such as confounding, should be a first-order concern.
© 2012 National Kidney Foundation, Inc. Published by Elsevier Inc. All rights reserved.