Unfortunately, Scientific American doesn't give any clues as to what study this information comes from, and my attempts to find it have all come up blank (anyone else have more luck?)

Of the two studies I can find on inter-rater reliability of coronary angiography, one shows greater than 98%, and the other shows high 90s. No doubt these are completely different types of coronary scans with completely different criteria for success than whatever Eddy was doing, but given the lack of pointers to the original study it's impossible to say exactly what's going on.