Cell type identification in leukemia: the level of agreement among four independent diagnostic methods

The level of agreement among four independent diagnostic methods for cell-type identification in leukemia was examined. The four diagnostic methods were routine morphology, electron microscopy, cell surface marker analysis by flow cytometry, and cancer cytogenetics. Sixty-five blood and bone marrow samples from fifty-seven patients were independently investigated by the four methods. It was found that when all cell-type categories were taken into consideration, overall percentages of agreement for pairs of the four diagnostic methods were poor, ranging from 43% to 67%. The kappa statistics, which correct for the agreement expected by chance, were also calculated. These kappa statistics had a range from 0.17 to 0.40 when all cell-type categories were taken into consideration, again indicating poor agreement between the methods pairwise. When each cell type was considered separately, much of the observed agreement was found to be expected on the basis of chance, as indicated by many of the kappa statistics being 0.00. For all categories, overall agreement for all four methods was poor and statistically significant, after correcting for chance (kappa = 0.20, (p = 0.000046). The poor agreement found among the four diagnostic methods indicates the need for further validation studies.