SAMPLE SIZE DETERMINATION USING ROC ANALYSIS

The paper presents a new method of sample size determination (SSD) based on performance evaluation of systems under study. The method builds upon previous work on Bayesian approach to nonparametric receiver operating characteristics (ROC) analysis with estimation of probability density functions and confidence intervals for parameters of ROC curve. Technical details of the method together with an illustration of its application to a real clinical example is given. The advantages of the method include ability to handle and compare models built using multidimentional, multimodal data, with no need for prior estimation of the effect size and variability. The disadvantages of the approach include high computational requirements, especially when the method of SSD deploys AUC as a measure of systems performance.