Likelihood-based confidence intervals for the risk difference of two-sample binary data with a fallible classifier and a gold standard

摘要

We develop likelihood-based confidence intervals for risk difference in two-sample misdassified binary data. Such data consist of two studies. The first study is the main study where individuals are classified by an inexpensive fallible classifier which may misclas-sify. The second study is a validation substudy where individuals are classified by using both the fallible classifier and an expensive gold standard which classifies perfectly. We propose and examine three likelihood-based confidence interval methods and conclude that the modified Wald method applied to small-number adjusted new data performs well and has nominal coverage probabilities.