The Methods Of ROC Analysis And Applications In Medical Research

Using a common, easily interpreted scale, ROC curves present a visual impression for the accuracy of diagnostic systems and display the tradeoffs between sensitivity and specificity for various setting of the decision criterion The area under the ROC curve gives expression to discrimination capacity for two classes of events. ROC analysis is now widely recognized as the best approach for measuring the quality of diagnostic information and diagnostic decisions.Practicability is our aim to pursue After presenting methods, we apply some examples to explain them and some programs to finish the calculation. We used excellent special softwares for ROC analysis and standard statistical packages (such as SAS, SPSS) to deal with data. The specification is given for data input format and how to use these software we presented SAS programs, if a formal procedure or widely distributed program code had not been found.For fitting ROC curves and getting the area under the ROC curve, we applied the parametric methods of binormal models and ordinal regression models, the semiparametric method of Cox proportional hazards model, the nonparametric methods of Hanley, Delong This paper also include estimation of required sample size, evaluation of confounding effects, and estimation of standard error and confidence interval by resampling method for ROC analysis.The main works and results obtained are as follows:1. Binormal models are assumed that normal or abnormal group is normal distribution respectively. At present, the statistical methods of the models are perfect. We call them as the classical parametric methods for ROC analysis. In this paper, univariate or/and bivariate binormal model was established and its parametrers was estimated by the "method of scoring". To calculate the statistical significance of the difference between two ROC curves using any one of three distinct statistical tests: the bivariate Chi-square test, the area z-score test, and the TPR z-score test. The proper binormal ROC model had been applied to the degenerate data. A computer program entitled ROCKIT was used to fit ROC curve to data, such as univariate categorical or continuous data, bivariate independent or correlated data, partially-paired data, multiple readers and replication data and so on.