The pathological TNM classification (AJCC/UICC), based on the assessment of tumor characteristics, provides useful but incomplete prognostic information in colorectal cancer. Numerous studies have highlighted the prognostic value of adaptive immune cells infiltrating the tumor: the in situ density of these cells, measured at the tumor center and the invasive margin, correlates with patient survival, and enrichment in CD8+ cytotoxic T lymphocytes is associated with a favorable prognosis. On this basis, the Immunoscore (IS) was developed, a quantitative digital pathology assay evaluating the immune response, now recognized as a robust immune classifier for estimating the risk of recurrence. The present study examines the reproducibility of visual assessment performed by pathologists on colon tumors stained for CD3+ and CD8+, compared with quantification by the IS.
An international panel of ten expert pathologists evaluated 540 images from 270 randomly selected colon cancer cases. The concordance between the T-score established visually by the pathologists, the corresponding hematoxylin-eosin–stained slides, and the digital IS was analyzed using a two-category classification (high, low) and a three-category classification (high, intermediate, low), before and after a training based on twelve reference cases located at the clinical thresholds.
The results reveal major disagreement. Discordant T-scores between pathologists were reported in more than 92% of cases, and 26% of cases were highly discordant, with the same slide being classified as low, intermediate, or high depending on the observer. Disagreement between the semi-quantitative visual assessment and the reference IS concerned 91% of cases before training and 96% after. Training brought no improvement: 42% of cases had their T-score modified, but concordance indices remained low (Kappa of 0.465 and 0.374). For the 20% of patients located around the decision thresholds, no concordance was observed (all Kappa values below 0.12). The authors attribute this poor reproducibility to the large size of colon tumors, the heterogeneous distribution of CD3+ and CD8+ cells—with an average of 88,000 CD3+ T lymphocytes per slide—and the intrinsically subjective nature of visual estimation.
The authors emphasize the clinical consequences of such misclassification: over- or under-treatment of stage II and III patients, inappropriate surveillance, and exposure to unnecessary toxicity. They conclude that the standardized IS assay outperforms assessment by expert pathologists in the clinical setting and that quantitative digital pathology constitutes a new tool for reproducible and quantitative evaluation of tumor-infiltrating immune cell subtypes.