Sample size for a randomized paired screen positive design

Abstract

The design of epidemiologic studies for the validation of screening tests necessitates accurate sample size calculations to allow for the estimation of relative measures of accuracy within a specified level of precision and with desired level of confidence. It is imperative to consider both the true positive rate (TPR) and the false positive rate (FPR) when comparing the accuracies of the screening tests since the TPR quantifies the key benefit of screening while the FPR quantifies the disadvantage of screening. We derive the sample size formulae for a novel Randomized Paired Screen Positive (RPSP) design. The formulae are derived under the assumptions that the tests are (1) conditionally independent and (2) dependent. The point estimates obtained correspond to those of the unpaired screen positive (USP) and paired screen positive (PSP) designs, respectively. Thus, this study acts as a guide to those investigators wishing to compare the performance of a new screening test relative to a standard one but are in dilemma about which design to use and which method of sample size computation should be adopted. The sample size depends on the standard deviation, the non-central parameter and the TPR of the new screening test. Further, the sampling distribution follows the non-central t-distribution