Which method is commonly used when there is no gold standard for diagnostic test evaluation?

• A. ROC analysis
• B. Gold standard test
• C. Latent class models
• D. Bayesian priors

Answer: (C)

When there is no gold standard, you can’t observe true disease status to directly measure test accuracy. Latent class models address this by treating the disease state as a hidden variable and using patterns of results from several imperfect tests to infer both how common the disease is and how accurate each test is. The idea is that the probability of each pattern of test outcomes depends on the latent disease status and the tests’ sensitivities and specificities, so you estimate those parameters from the data. This approach works best when you have multiple tests and either make reasonable assumptions about how their results relate to each other (such as partial independence) or use a Bayesian framework with informative priors to help identify the model. ROC analysis, by contrast, relies on comparing outcomes to a known truth, so it can’t be used when no gold standard exists.