POWER_ROC: Stata module to compute power and sample-size analysis for receiver operating characteristic (ROC) analysis

The power roc package is helpful for planning studies that use receiver operating characteristic (ROC) curves to evaluate the accuracy of diagnostic tests. An advantage of ROC curves for evaluating the accuracy of diagnostic tests is that it incorporates both sensitivity (the probability of a test detecting disease when the patient actually has the disease) and specificity (the probability that the test is negative when the patient is actually disease-free), into a single measure of accuracy. power roc is loosely based upon the code used in the SAS® macro ROCPOWER (Zepp 1995). power oneroc computes sample size or power for a one-sample receiver operating characteristic (ROC) analysis. Variance functions are computed using either the method described in Obuchowski, Lieber and Wians (2004) for continuous and ordinal data assuming a binormal distibution (the default), or the method described by Hanley and McNeil (1982) for continuous data which is based on the Mann-Whitney version of the rank-sum test. power tworoc computes sample size or power for a two-sample receiver operating characteristic (ROC) analysis. When there is no correlation corr(0) between the two alternative AUCs (auc1 and auc2), power tworoc computes sample size (power) for an independent two-sample test. When the correlation does not equal 0, power tworoc computes sample size (power) for a paired two-sample test. As with power oneroc, variance functions are computed using either the method described in Obuchowski, Lieber and Wians (2004) for continuous and ordinal data assuming a binormal distribution (the default), or the method described by Hanley and McNeil (1982) for only continuous data which is based on the Mann-Whitney version of the rank-sum test. Sample size can be computed given power, and power can be computed given sample size. Results can be displayed in a table (default) and on a graph ([PSS-2] power, graph).