COST-EFFICIENT STUDY DESIGNS FOR RELATIVE RISK MODELING WITH COVARIATE MEASUREMENT ERROR

Covariate measurement error and misclassification problems are pervasive in biomedical research. Much useful work has been published concerning the correction of point and interval estimates of relative risk for these errors, but virtually none of it has been used by biomedical investigators. It is suggested that a primary reason for this lack of use is the lack of necessary data, namely small substudies which replicate or validate the mismeasured variables to provide information on the measurement error or misclassification process. Although a small number of papers have been published which investigate design optimality under this main study/validation study framework, almost none provide explicit sample size formulas. In addition, the definition of optimality used by most investigators is not of the greatest interest for the applied investigator, and alternative criteria are discussed. The existing papers are reviewed, the connection between main study/validation study and two-stage designs is restated, and future directions for research are proposed.