Affiliation: Gillings School of Global Public Health, Department of Biostatistics
The accurate diagnosis of a molecularly-defined subtype of cancer is often a very important step toward its effective prevention and treatment. For the diagnosis of some subtypes of certain cancers, a gold standard with perfect sensitivity and specificity may be unavailable. In those scenarios, the status of the tumor subtype commonly is measured by multiple imperfect diagnostic markers. In many such studies, some subjects are only measured by a subset of diagnostic tests and the missing probabilities may depend on the unknown disease status. In this research, we present novel statistical methods based on an EM algorithm to evaluate incomplete multiple imperfect diagnostic tests under conditional independence and conditional dependence assumptions. We applied the proposed methods to a set of real data from the NCI Colon Cancer Family Registry (C-CFR) on diagnosing microsatellite instability (MSI) for hereditary nonpolyposis colorectal cancer (HNPCC) to estimate diagnostic accuracy (i.e., sensitivities and specificities) and prevalence for 11 biomarker tests. Simulations are conducted to evaluate the small-sample performance of our methods. The advantages and limitations of our methods are discussed. An R package was developed for easy implementation of our methods. Finally, a proposal for future research also was presented.