Quantitative immunohistochemistry using the CAS 200/486 image analysis system in invasive breast carcinoma: a reproducibility study.

We evaluated the intra- and inter-observer reproducibility of quantitative immunohistochemical (IHC) analyses using the Cell Analysis Systems (CAS) 200/486 image analyzer of Estrogen Receptor (ER), Progesterone Receptor (PR), proliferation-associated nuclear protein (Ki67), HER-2/neu (c-erbB-2) protein over-expression and cathepsin D (CD) in 20 randomly-selected invasive breast carcinomas. Qualitative analysis of IHC Epidermal Growth Factor Receptor (EGF-R) was also assessed in this study for comparative purposes. Duplicate blind assessments by the same observer showed excellent correlations for all quantitative IHC features (P < 0.001; P = 0.004 for neu). However, the immuno-quantitative analyses results between the 3 different operators showed lower correlation coefficient values, thus being less reproducible. This resulted in systematic differences and bias between the observers. This was also clear from the overall agreement between the 3 observers which was 70% for ER, 70% for PR, 56% for Ki67, 79% for c-erbB-2 and 75% for CD. The qualitative visual assessments of EGF-R, expressed as either positive or negative, showed a 75% agreement between observers and 85% intra-observer agreement (comparable to quantitative digital image processing results). The same results were obtained with kappa statistics. A further analysis of the factors causing the lack of reproducibility was performed. For quantitative IHC, segmentation of stored and retrieved digitized images was quite reproducible between and within well-trained observers. However, variation between different fields of vision of one and the same section showed large variations for most cases. Therefore, differences in sampling of fields within a section appeared to be the major cause of lack of reproducibility between observers, although segmentation differences still added slightly to the inter-observer variations. Accordingly, a strict sampling protocol of fields of vision is mandatory to obtain reproducible quantitative IHC results. It is clear from the present study that so-called random (but in fact, at convenience) selection of fields of vision for measurement is not a sufficient guarantee of adequacy of the sampling.