Abstract
Certain symptoms are of diagnostic value because they reflect the progress of a multistage disease. To analyze such cases, a dynamic symptom/disease model is constructed. The model is seen to be consistent with observations about skewed, dependent symptom distributions for certain diseases. The uniformly most powerful unbiased test of hypotheses about the disease factor leads to a linear discriminant function. Prior knowledge of the symptom/factor relationship can be used to correct the procedure for nuisance effects or lacking such information, the model can be structured so that the test is independent of the nuisance factors.