Discriminant model for cytologic distinction of large cell neuroendocrine carcinoma from small cell carcinoma of the lung

Rira Hoshi,Noriyuki Furuta,Takeshi Horai,Yuichi Ishikawa,Satoshi Miyata,Yukitoshi Satoh

DOI: https://doi.org/10.1097/JTO.0b013e3181ce3cdd

Abstract:Background: To establish cytologic criteria for pulmonary large cell neuroendocrine carcinoma (LCNEC), we developed and evaluated a discriminant model for cytologic differential diagnosis between LCNEC and small cell lung carcinoma (SCLC). Methods: Aspiration cytologic and/or imprint smears from 29 LCNEC cases were reviewed in comparison with 26 SCLC cases. We selected the following parameters for assessment: background, cellular arrangement, cell clusters, cell cohesion, arrangements, cell dimensions areas, the presence of cytoplasm and/or prominent nucleoli, nuclear features, mitosis, naked nuclei, and nuclear streaking. To demonstrate the utility of differences in frequencies of cytologic parameters for LCNECs and SCLCs, a discriminant model was developed and evaluated. Results: Among the cytologic parameters investigated, large clusters (consisting of >or=60 tumor cells) with tight cohesion and small tumor cells (showing <or=120 microm) without prominent nucleoli on each case were particular focuses of attention, because statistically significant differences with good power were evident between the LCNEC and SCLC groups for their frequencies (p < 0.0001). On the basis of variation in plotted location on scatter plots, a discriminant model for LCNEC and SCLC was made and evaluated by logistic discriminant analysis. Sensitivity, specificity, and accuracy were all 100%. With leave-one-out cross validation, the predicted error rate of the discriminant model for new cases was 0.00545. Conclusion: Our model based on the cytologic features of large cell clusters with tight cohesion and of small tumor cells without prominent nucleoli should be a useful aid for distinction between LCNECs and SCLCs.