Parametric Distribution-free ROC Analysis And Application In Medical Diagnostic Test

The accuracy of diagnostic test can be a common evaluation in medical study. However it is not appropriate to use the most common TPR and FPR as the parameters to reflect the accuracy of the test, which makes ROC regarded as the best indicator of the diagnostic test. In medical study, we need to consider factors that will influence the accuracy of the test, including the covariate as the characteristics of subjects or operating conditions. PDF regression analysis is applied to the clinical trials, which is used to explore the influences brought by the covariates to ROC curve and allow one to evaluate the diagnostic test's accuracy in a more comprehensive way.PDF approach can directly simulate a model of the ROC curve with covariates, which is used to analyze continuous data in diagnostic tests with the influences of the covariates. The model is more flexible. The covariates are continuous or categorical. It can refer to the interaction between covariate andΦ?1(.)and can be distribution-free with respect to the distribution of the test resultsThe basis theory of PDF model and the parametric estimation method were introduced in the paper. Two problems in parametric estimation were studied in the theoretic simulation. It presented the clinical applications. An STATA procedure was applied to the analysis of the tests. The simulation study indicated that the estimates and standard deviation yielded by PDF model were very close to those of other methods. It is an efficient estimation for the parameter of the ROC curve. The PDF model could give us more additional information than other methods. PDF approach required to specify T ,the set of FPRs. The simulation study also indicated that fitting regression model with a small FPRs set can achieve a similar power to maximal set . It is relatively easier to implement and less computationally intensive. Thus 50 equally spaced FPRs, T ={1/51,2/51,…,50/51} is moderate.The results indicated that PDF model can be applied for the evaluation of the ROC curve with covariates and continuous tests. A relatively small number of FPRs can achieve near-maximal statistical efficiency in parametric estimations. PDF model can accurately evaluate some indicators in diagnostic tests.