ROC analysis for the evaluation of continuous biomarkers: Existing tools and new features in SAS® 9.2

Biomarkers have become essential tools for proper diagnosis and treatment of a wide range of illnesses, including cancer, diabetes, and infectious diseases. The growing need for rigorous evaluation of new biomarkers for medical practice has spurred the development and characterization of statistical methods for diagnostic accuracy. The receiver operating characteristic (ROC) curve is the standard analytical tool for evaluating diagnostic tests. However, recent studies suggest widespread use of inappropriate statistical methods for ROC analysis. In addition, while enhancements in SAS 9.2 have greatly simplified basic ROC analysis for SAS users, many common ROC techniques still require extensive additional programming for SAS users. This paper provides an overview of statistical methods for evaluating continuous biomarkers using ROC analysis. For each statistical technique, existing tools available in SAS for performing the task are described, with particular emphasis on methods not addressed in previous SAS papers. The paper also introduces new features for ROC analysis that are now available as a standard component of the LOGISTIC procedure in SAS 9.2. Statistical techniques addressed in the paper include the comparison of the area under the ROC curve (AUC) of two or more biomarkers and the generation of confidence intervals for sensitivity given fixed specificity. Pancreatic cancer data (Wieand et al. 1989) that have been widely used as sample data in statistical literature are used in this paper to illustrate the application of statistical concepts.