Purpose: The failure of light microscopy to predict individual patient survival accurately in pStage I and II colorectal carcinoma can hinder planning postoperative therapy and follow-up. This study was designed and conducted in two parts to assess the influence of relative sensitivity of the light microscope on the pathologist's ability to detect malignant cells in lymph nodes.
Methods: The first part of the study examined the issue of sampling error as a fraction of the number of lymph node sections examined by asking the question, "Does increasing the number of sections (sampling) taken from the block increase tumor cell detection in a lymph node?" Three levels of five sections 4 to 5 microm thick separated by 15 to 20 microm were obtained from each of 494 blocks from 173 cases of pStage I and II colorectal carcinoma. A total of 1,721 lymph nodes were examined. Sections from each level were stained with hematoxylin and eosin and for the expression of cytokeratin. The second part of the study examined the relative sensitivity of the light microscope to detect tumor cells in a lymph node. To simulate lymph nodes, cell blocks were made that contained 10(6) or 10(7) mononuclear cells admixed with increasing numbers of SW480 tumor cells (0, 50, 10(2), 5 x 10(2), 10(3), and 5 x 10(3)). Three pathologists independently examined sections from ten control and ten experimental blocks.
Results: Results from the first part of the study demonstrated cytokeratin-positive cells in 278 lymph nodes from 102 of 172 (59 percent) cases. These cells were identified in the first level in 177 (64 percent) as compared with the second or third level or both in 101 (36 percent) of the lymph nodes. Results from the second part of the study demonstrated an overall sensitivity of light microscopic examination of hematoxylin and eosin-stained sections to be approximately 23 percent, representing tumor cells correctly detected in 7 sections of the 30 sections containing tumor cells. The overall specificity was 87 percent or 26 sections correctly classified as lacking tumor cells of a possible 30. Immunohistochemical staining for cytokeratin expression improved sensitivity of the light microscope to detect tumor cells to 18 of 30 (60 percent) and the specificity to 30 of 30 (100 percent).
Conclusion: This study demonstrates several sources of variability that can induce errors in pathologic staging. These include 1) inadequate section, i.e., sampling, of lymph nodes, 2) use of only hematoxylin and eosin-stained sections, 3) samples with tumor cells below the level of detection sensitivity of the light microscope, and 4) observer variability.