Application of parametric method and non-parametric method in estimation of area under ROC curve / 第二军医大学学报

ABSTRACT

Objective: To elucidate and compare the parametric method and non-parametric method in estimation of the area under ROC curve, so as to provide a basis for their application in diagnosis assessment. Methods: The areas under ROC curves were estimated by parametric method of fitting binomial model and by non-parametric method of Mann-Witney statistics. The method was employed in the diagnostic tests of lung cancer. Results: By non-parametric methods, the areas under ROC curves of Cyfra21-1 and CEA were respectively 0.77 and 0.87 in the lung cancer diagnostic tests; by parametric methods, they were 0.78 and 0.87, respectively. It was indicated that when the sample size was large, the values of areas under ROC Curves were similar between parametric method and non-parametric method. Conclusion: Non-parametric method should be used to evaluate the area under ROC curve if the sample size is small, and for large sample size, the parametric method or nonparametric method should be chosen according to the actual situation.