On the distortion of risk estimates in multiple exposure level case-control studies

The impact of error in assessing exposure to hypothesized pathogens on distortion of the appearance of relative risk is considered in the case-control study design. We assume there is a true dose-response relationship which links exposure to disease status. We also assume that multiple categories are used to classify levels of exposure, and that the errors which occur are those between adjacent exposure categories. The findings suggest that, if the pattern of errors is random, even though they are numerous, the resulting bias in risk will not be severe enough to mask a substantial risk increment. A large and consistent understatement of exposure by cases, and/or consistent overstatement of exposure by controls, can effectively mask even a strong pattern of risk enhancement. To produce the appearance of risk where there is none would require a substantial error in the opposite direction among the cases and controls.