FACTORS AFFECTING UNIFORMITY IN INTERPRETATION OF PLANAR TL-201 IMAGING IN A MULTICENTER TRIAL

Objectives. This study was designed te, assess factors affecting interobserver agreement in interpretation of planar thallium-201 stress imaging in the Multicenter Study on Silent Myocardial Ischemia (MSSMI). Background. Five hundred fifty-six planar thallium-201 images were interpreted in 24 clinical centers and in a Radionuclide Core Laboratory. The trial's Coordinating and Data Center observed that the participating clinical centers interpreted a significantly greater number of thallium-201 stress studies as abnormal (i.e., myocardial ischemia or scar) than the Core Laboratory, and overall agreement was poor (kappa 0.27). Methods. Agreement in image interpretation between clinical centers and the Radionuclide Core Laboratory was analyzed by kappa statistics. The reproducibility of the Core Laboratory results on 41 randomly selected test studies was excellent (kappa 0.77). In contrast, the reproducibility of interpretation in the clinical centers on their own studies was at best fair (kappa 0.45). It was hypothesized that the poor agreement and reproducibility in the clinical centers were caused by lack of standardization of image display and lack of objective criteria for image interpretation. To test the effect of standardization, 13 clinical investigators interpreted the same 41 test studies using 1) uniform image display, and 2) uniform quantification of images. Results. The agreement in interpretation between clinical investigators and the Radionuclide Core Laboratory improved modestly with uniformity of image display (kappa 0.57) but improved markedly (kappa 0.66) with quantitative circumferential profile analysis. Conclusions. Lack of standardization in image display and lack of objective criteria for interpretation of thallium-201 images are responsible for suboptimal reproducibility and poor interlaboratory agreement in the interpretation of thallium-201 stress imaging. The adoption of a uniformly accepted method for computer quantification of myocardial perfusion images is crucial to improve agreement in interpretation.