Workshop on guidelines to the epidemiology of weak associations. Confounding.

The magnitude of confounding is examined in nine case studies of two "weak" relationships: between artificial sweeteners and bladder cancer, and between oral contraceptives and cervical dysplasia. Confounding had little or no influence on the results of any published study. The responsible epidemiologist must always consider the possibility of confounding, no less when associations are weak than when they are strong. Identification of potentially confounding variables is an integral part of good epidemiologic practice. Rarely, however, does confounding itself, especially from unidentified sources, live up to its reputation for introducing seriously spurious associations. An investigator is more likely to be led astray by undetected biases than by pure confounding.