Agreement in human interpretation of analog thallium myocardial perfusion images

To assess the agreement of human interpretation of analog thallium myocardial perfusion images, four experienced interpreters evaluated 100 images on two occasions using a form designed to limit reader variability. A high intraobserver agreement (agreement by same observer at separate times) of 89--93% was found when films were interpreted as normal or abnormal (a dichotomous decision). Interobserver agreement for a majority grouping of observers (three or four) was 75% for an abnormal and 68% for a normal interpretation. However, agreement ranged from 11--79% when interpreters were asked to read the anatomic location of defects. Posterior and lateral wall defects were interpreted with the least amount of agreement. These results indicate that caution must be taken when interpreting defect location. Using a scale of 1--10 to grade the severity of a defect, correlations of 0.82--0.86 were found when reading defects in the lateral and anterior projections. Higher correlations, from 0.86--0.94, were found in left anterior oblique views. Use of reporting forms with specific criteria, multiple observers at one occasion, and/or computer processing may improve agreement. A brief review of the agreement of cardiology testing procedures is also presented.