Confidence limits for risk differences and risk ratios adjusted for estimated sensitivity and specificity

In this paper, a procedure to construct approximate confidence intervals (CIs) of risk differences and risk ratios taking into account the sensitivity and specificity of a diagnostic test, and also taking into account the size of the samples from which sensitivity and specificity were estimated is proposed. The new method combines the originally proposed method of Rogan and Gladen, and the general method described by Zou and Donner for calculating the CI of difference between two effective measures based on their individual CI limits. The performance of the new method will be evaluated by calculating the exact coverage of the resulting CIs. More methods are known to account for sensitivity and specificity considered to be constant, however, when the sensitivity and specificity of a test were estimated within studies with small sample sizes then ignoring their variability may lead to biased estimates. This will be shown through real examples.