Background: Histopathologic, prognosis-related grading of malignancy by means of morphologic examination in transitional cell carcinomas of the urinary bladder (TCC) may be subject to observer variation, resulting in a reduced level of reproducibility. This may confound comparisons of treatment results. Using objective, unbiased stereologic techniques and ordinary histomorphometry, such problems may be solved.
Experimental design: A study of 110 patients with papillary or solid transitional cell carcinomas of the urinary bladder in stage Ta through T4 was carried out, addressing reproducibility of both qualitative and quantitative grading methods. Grading of malignancy was performed by one observer in Japan (using the World Health Organization scheme), and by two observers in Denmark (using the Bergkvist system). A "translation" between the systems, grade for grade, and kappa statistics were used in evaluating the reproducibility. Unbiased estimates of nuclear mean volume, nuclear mean profile area, nuclear volume fraction, nuclear profile density index, and mitotic profile density index were obtained twice in 55 of the studied cases by one observer in Japan and one in Denmark, using a random, systematic sampling scheme.
Results: The results were compared by bivariate correlation analyses and Kendall's tau. The international interobserver reproducibility of qualitative gradings was rather poor (kappa = 0.51), especially for grade 2 tumors (kappa = 0.28). Likewise, the interobserver agreement on the Bergkvist scheme was poor (kappa = 0.43). On the other hand was the interobserver agreement on invasion high (kappa = 0.75). The intraobserver reproducibility of the quantitative histopathologic variables was excellent in both Japan and Denmark for estimates of nuclear mean volume (r = 0.93), for nuclear mean profile area (0.78 < r < 0.83), and for nuclear profile density index (0.85 < r < 0.89), whereas the reproducibility for nuclear volume fraction was somewhat poorer (0.68 < r < 0.64). The slopes of the correlation lines were not significantly different from unity. Estimates of mitotic profile density index also showed acceptable intraobserver reproducibility (Kendall's tau > 0.53).
Conclusions: The international, interobserver reproducibility of the quantitative estimators yielded similar results for all histopathologic variables investigated, except for nuclear volume fraction (r = 0.54). This can probably be related to the manual design of the sampling scheme and may be solved by introducing a motorized object stage in the systematic selection of fields of vision for quantitative measurements. However, the nuclear mean size estimators are unaffected by such sampling variability. The results obtained in this study stress the need for objective, quantitative histopathologic techniques substituting qualitative, subjective methods in prognosis-related grading of malignancy.