A method of developing a gene expression profile indicative of the presence or stage of a selected a disease, disorder or genetic pathology in a mammalian subject employs penalized discriminant analysis with recursive feature elimination. A method of diagnosing a cancer in a mammalian subject includes the steps of examining a sample containing the subject's immune cells and detecting a variance in the expression of a statistically significant number of genes, e.g., at least 10 non-tumor genes from those same genes in a characteristic disease or healthy gene expression profile. A significant variance in expression of these genes when compared to a gene expression profile, preferably an average gene expression profile of a normal control, or significant similarities to an average gene profile of subjects with cancer, correlates with a specific type of cancer and/or location of tumor.