摘要：

Fryback and Thornbury' have pointed out that the efficacy (or effectiveness) of a radiologic procedure can be assessed at any of six levels that range from technical reproducibility to societal impact. Efficacy at the higher levels usually is of greatest interest, but lower-level efficacy is almost always easier to measure. The second level of efficacy, which is called "diagnostic accuracy ," represents the extent to which radiologists' (or physicians') image-based diagnoses agree with patients' actual states of health or disease. Receiver operating characteristic (ROC) analysis is now recognized as the best approach for evaluating diagnostic accuracy,2 primarily because it separates differences in discrimination capacity from differences that are only due to "decision-threshold" effects. In some situations, efficacy at higher levels can be estimated from the results of an ROC analysis by use of collateral data and appropriate assumptions. The purposes of this paper are to indicate the six levels at which efficacy can be evaluated and to describe several recent developments in ROC methodology. The basic principles of ROC analysis have been discussed elsewhere.2'

展开