Cautions in the use of antecedents as surrogates for confounders

When lacking information on confounding variables, epidemiologists have used surrogates which are antecedents of both the exposures and confounders of interest. The usefulness of this strategy is explored in a series of scenarios for a prospective epidemiologic study wherein risk ratios relating antecedent to confounder, antecedent to exposure, and confounder to exposure were varied. Antecedent-adjusted, confounder-adjusted, and crude risk ratios were calculated and compared. The antecedent-adjusted risk ratio was useful, that is, was closer to the confounder-adjusted risk ratio than was the crude risk ratio, in 1,067 (49%) of 2,187 scenarios. The antecedent-adjusted risk ratio, the crude risk ratio, and the risk ratio relating confounder to exposure together predicted the usefulness of the antecedent (or any variable) as a confounder proxy. The antecedent was useful in 97% of scenarios wherein: 1) the antecedent-adjusted risk ratio was less than the crude risk ratio, and the risk ratio relating confounder to exposure was greater than 1.0, or 2) the antecedent-adjusted risk ratio was greater than the crude risk ratio, and the risk ratio relating confounder to exposure was less than 1.0. In the remaining scenarios, it was useful only 5% of the time.