I have a set of diagnostic accuracy data where about 1800 binary outcomes from radiographs were reported by 3 observers and compared to a gold standard (CT scan).
1. Using diagtest command to compute sensitivity, specificity, PPV, and NPV among subgroups, I encounter 95% CIs of higher than 100% (between 100 and 101). My question is whether this is normal? Does it mean the analysis is not reliable? How should it be reported in a paper, as 100 or 100.60 (for example)?
2. As mentioned above, there are 3 observers in my study. I want to calculate an "overall" diagnostic indices to encompass the answers of all observers. For this purpose can I pool observers' answers and treat each diagnostician as separate case series pretending I have 1800 x 3 sample size and calculate the indices anew?
