|
Abstract:
|
Tumor grade is an important prognosticator in survival of cancer patients. Tumors are graded into three categories: well, moderate and poor under a complex system. However, the reproducibility is poor. With at least three readings, the prevalence and classification rates are estimable up to a permutation of these three categories under a latent class model (Dawid, 1979). Various structural models were imposed in the past to identify the categories when instrumental variables are available. Here a new method is proposed to identify the categories and classification rates when an auxiliary variable is known to be positively or negatively associated with the latent variable, e.g., the true tumor grade. This method is illustrated by analysis of tumor grade reading data from a joint study of the National Surgical Adjuvant Breast and Bowel Project (NSABP) and the Genomic Health Inc.
|
