Adequately defining tumor cell proportion in tissue samples for molecular testing improves interobserver reproducibility of its assessment.

Gene mutation status assessment of tumors has become standard practice in diagnostic pathology. This is done using samples comprising tumor cells but also non-tumor cells, which may dramatically dilute the proportion of tumor DNA and induce false negative results. Increasing sensitivity of molecular tests presently allows detection of a targeted mutation in a sample with a small percentage of tumor cells, but assessment of tumor cellularity remains essential to adequately interpret the results of molecular tests. Comprehensive tumor cell counting would provide the most reliable approach but is time consuming, and therefore rough global estimations are used, the reliability of which has been questioned in view of their potential clinical impact. The French association for quality assurance in pathology (AFAQAP) conducted two external quality assurance schemes, partly in partnership with the French group of oncology cytogenomics (GFCO). The purpose of the schemes was to (1) evaluate how tumor cellularity is assessed on tissue samples, (2) identify reasons for discrepancies, and (3) provide recommendations for standardization and improvement. Tumor cell percentages in tissue samples of lung and colon cancer were estimated by 40-50 participants, on 10 H&E virtual slides and 20 H&E conventional slides. The average difference between lowest and highest estimated percentage was 66. This was largely due to inadequate definition of cellularity, reflecting confusion between the percentage of tumor cells and the percentage of the area occupied by tumor in the assessed region. The widest range of interobserver variation was observed for samples with dense or scattered lymphocytic infiltrates or with mucinous stroma. Estimations were more accurate in cases with a low percentage of tumor cells. Macrodissection of the most homogeneous area in the tissue reduced inter-laboratory variation. We developed a rating system indicating potential clinical impact of a discrepancy. Fewer discrepancies were clinically relevant since the study was conducted. Although semi-quantitative estimations remain somewhat subjective, their reliability improves when tumor cellularity is adequately defined and heterogeneous tissue samples are macrodissected for molecular analysis.