To the Editor.
—Those of us long devoted to investigating and improving the quality of bedside diagnosis1 always welcome articles like that of Dr Heckerling and colleagues2 who evaluated precordial percussion and palpation using ultrafast computed tomography as a standard. Although appropriate statistical methods, particularly the receiver operator characteristic and the κ value, were used, one wonders whether the investigative method was appropriate for a fair evaluation of the techniques studied. My principal concern is this: was there a sufficient number of bedside investigators, especially since the second investigator evaluated an undefined subset of only 14 patients, compared with 103 patients by the principal investigator?Even with good concordance on their common 14 patients as indicated by the Landis-Koch observer agreement scale, should there not have been at least three observers, each blindly evaluating the entire series? Observer variability, even among recognized experts (analyzed elsewhere),3 can be
Spodick DH. Accuracy of Precordial Percussion and Palpation. JAMA. 1994;271(17):1318-1319. doi:10.1001/jama.1994.03510410030019