Principal component analysis. A factor analytical technique for the determination of interobserver variation in...

The insufficient interobserver reproducibility limits the practical use of histomorphological tumor grading in daily routine pathology. In this study, the reasons for the quite low rates of agreement between different observers have been investigated by the application of a factor analytical technique, i.e. principal component analysis with varimax rotation, on a tumor grading system. Grading results from 44 cases of G1 and G2 giant cell tumors of the bone (GCT), graded by three different observers according to the five criteria of Jaffe, were taken as an example. It could be proven that the single criteria were used in an observer-specific way. Two criteria, for example, which are scored highly correlated by one observer, may be used independently by another. The resulting observer-specific different recognition patterns may provide an explanation for their quite different grading results, which were identical in only 48.6% to 54.1% (mean: 50.9%) of the cases. No correlation of GCT grading with recurrence was found in 31 cases which had been treated by intralesional surgery.