Agreement Between Diagnostic Tests

The delta method [36, s.26] obtains variance in the percentage of compliance by category In diagnostic precision simulations, sample design provided efficiency gains for the estimation of specificity and APP (Table 6). Of the sample fractions observed, 670/2331-28.7% of women with the gold standard were identified, but specific variances and APP were inflated by about 2 times compared to the gold standard test by all women. This is due to the fact that the design of the sample is more obvious to the few women who have developed CIN2. For the final design, which also applies to all women HPV (670-102) /2331-33.1%) differences in specificity and APP are less than 10% of the variance in all women subjected to the gold standard. This is also due to the fact that this project would represent an average of 34.2 of the 38.5 CIN2 estimated in the full study. However, the accumulation of the sample for CIN2 has little effect on the variances in sensitivity and NPV. Instead, the sample design, which doubles the negative sample percentage for the three screening tests, provides efficiency gains in estimating sensitivity and NPV (but not for specificity and APP). These results illustrate the general result [30] that a cost-effective estimate of specificity (or APP) with two-phase constructions requires an overcut of positive test results (which tend to record most of the true positives of the gold standard), but for sensitivity (or NPV), an overcuting of test results requires. The overall agreement will always be somewhere between the positive percentage agreement and the negative percentage agreement. There are many ways to describe the accuracy of the diagnosis. Appropriate indicators include estimates of sensitivity and specificity pairs, probability ratio of positive and negative pairs of results, and ROC (operational receptor characteristic) analysis, and confidence intervals. See the latest edition of the CLSI EP12-A and GP10-A guidelines; texts by Lang and Secic (1997), Pepe (2003), Zhou et al.

(2002); References in these texts; and the bibliography at the end of this document. To interpret these measures, we recommend that you provide the definition of conflict of interest, the standard of reference, the population of use and a description of the population studied. We find that if there are no sampling strata, the APA and APA estimates that ignore the sample are identical to those responsible for sampling.