Dear allstaters,
in medical diagnostic imaging both the standard care regimen and a non-invasive imaging technique may be applied in each patient. Assume that the outcome variable is binary (e.g., cancer=yes/no). With a gold standard at hand, the computation of confidence intervals of the diagnostic measures sensitivity, specificity, positive and negative predictive values is straightforward for each procedure individually, since these figures are basically binomial proportions. With respect to sensitivity and specificity, approximate CIs for the difference can be calculated by formulae given in Fleiss 2003 (Statistical Methods for Rates and Proportions, 3rd ed., (13.14), (13.15)). With respect to CIs for the difference in predictive values I might apply the delta method (Oehlert GW. A note on the delta method. American Statistician. 1992; 46:27-29).
Are there any suggestions/refernces for the computation of exact confidence intervals for the difference in sensitivity, specificity, predictive values, respectively, having paired observations? Is there an approximate confidence interval for the difference in positive, negative predictive values for paired observations?
Thanks for your help. Many greetings, Oke
Oke Gerke // Department of Statistics // University of Southern Denmark
|