Interobserver accuracy and precision of MKI estimates in neuroblastomas

The Shimada classification of neuroblastomas (NBs) is a frequently used grading system with variably reported prognostic importance. These discrepant reports may be related to the imprecision of the classification scheme as applied by different investigators. An important source of variability is the determination of the MKI, a density sum of mitoses and karyorrhectic cells (MKCs) per 5000 tumor cells. A method for estimating tumor cell density (TCD) and criteria for regarding cells as MKCs is established (Joshi et al.) but there is no independent report of its accuracy and interobserver reproducibility. Our purpose was quantification of the interobserver accuracy and precision in determining the number of MKCs per highpower field (HPF) TCD estimates, and the derived MKI with values of the expert pathologist (Shimida) according to method reported by Joshi. The number of MKC/ HPF, average TCD estimates (total cells per HPF), and the MKI were determined by three pathologists and the expert from the same single representative slide of 31 NBs according to the simplified method for estimating MKI (Mod Path 4:139-146, 1991). The correlation coefficients of three observers with the expert were 0.82, 0.84, and 0.87 for MKCs/HPF; 0.87, 0.34, and 0.63 for TCD; 0.80, 0.87, and 0.88 for MKI. Respective means of three observers and the expert were 46, 23, 47, and 28 for MKCs/HPF; 688*, 603**, 786**, and 681 for TCD; and 311**, 198**, 292**, and 152 for MKI (*P= not significant; **P< .05). The method for estimating TCD and determining MKC/HPF was inaccurate and/or imprecise with resultant high MKI class discordant rates. These findings emphasize the importance of central review for NB grading.