Cross-sectional studies of medical tests for diagnosing presence or absence of disease are often beset by numerous potential confounders, e.g., site, treatment, age, medical history, conditions of testing, etc. By definition, a confounder of the association of test result with disease status is correlated with both variables. Logistic regression is a standard method for adjusting the odds ratio between test result and disease status for confounders. However, when the number of covariate levels grows with sample size, the logistic regression-based maximum likelihood estimator of the odds ratio adjusted for the covariates is well known to be asymptotically biased. Alternatively, estimators of the adjusted odds ratio that are valid even when data are sparse within covariate levels can be obtained using conditional logistic regression or the model-free Cochran-Mantel-Haenszel (CMH) method. In this talk, I’ll review CMH-like estimators of quantities that are typically of more interest in diagnostic studies than the odds ratio, namely, risk difference, relative risk, and diagnostic likelihood ratios. I’ll illustrate these estimators with the concrete examples of diagnostic tests and their potential confounders.