Assessment of the Performance of Frozen Section Margin Analysis during Esthesioneuroblastoma Surgery

Introduction: Esthesioneuroblastoma (ENB) is a rare tumor originating from the superior aspect of the nasal cavity and is assumed to arise from the olfactory epithelium. Surgery plays a prominent role in the management of this malignancy, and it is thought that negative margin resections offer a survival advantage. The frozen section procedure represents one of the most commonly used methods for assessing intraoperative margin status. It is therefore important to comprehensively assess the performance of this analysis in the setting of ENB surgery.

Objectives: (1) We assess the performance of the intraoperative frozen section procedure to correctly classify biopsies obtained during ENB surgery into biopsies that contain tumor and those that do not; (2) we define the relationship between post-test probabilities and pretest probabilities from frozen section analysis; and (3) we review all incorrectly classified specimens.

Methods: We searched our institution’s pathology database for patients who had ENB surgery between 1/1/2000 and 7/1/2012. We only included patients who had a definitive diagnosis of ENB prior to surgery and frozen sections obtained during surgery. All frozen sections in this study had corresponding permanent sections available to serve as a gold standard. The pathology reports from these patients were obtained and intraoperative frozen section results and their corresponding permanent results were recorded in a dichotomous fashion—definitive tumor present or definitive tumor absent. Specimens with ambiguous permanent reads were removed from the analysis. Ambiguous frozen section results were coded in a fashion that penalized classification performance. Two authors independently reviewed and verified the data. This database was analyzed to obtain classification statistics. The confidence intervals for classification performance were obtained using the bias correct and accelerated bootstrap sampling method. We used 10,000 samples. Confidence intervals for post-test probability curves were derived using the Taylor series expansion to propagate uncertainty through the appropriate relationships. Finally, we obtained and reviewed the slides from all ambiguous or incorrect reads.

Results: Thirty-four patients were included in the study, and 430 specimens were available for review. Four specimens were discarded, and 12 specimens were reclassified because of ambiguities. Our analysis found the following performance characteristics: sensitivity: 0.89 (0.81, 0.94); specificity: 0.96 (0.94, 0.98); accuracy: 0.94 (0.92, 0.96); likelihood ratio positive: 22.97 (13.75, 41.12); prevalence: 0.21 (0.17, 0.25); positive predictive value: 0.86 (0.78, 0.92); negative predictive value: 0.97 (0.95, 0.98). Histopathologic review revealed that crush artifacts and inadequate specimen size were major sources of incorrect reads.

Conclusion: The negative predictive value for frozen section assessment of ENB is very high, suggesting that the frozen section procedure is an excellent tool for the assessment of intraoperative margins. The relationship between the negative post-test probability and pre-pretest probability suggests that margins should be taken with adequate tissue buffer to keep the pretest probability low to increase classification performance.