What can we learn from the errors in the frozen section diagnosis of pulmonary carcinoid tumors? An evidence-based approach.

The intraoperative diagnosis of pulmonary neuroendocrine tumors can be difficult. Frozen section diagnosis was requested on 87 neuroendocrine tumors including 58 typical carcinoids, 8 atypical carcinoids, 18 large cell neuroendocrine carcinomas, and 3 small cell carcinomas from 2405 patients that underwent frozen section diagnosis at Cedars-Sinai Medical Center from 2002 to 2007. The deferral and error rate for carcinoid tumors was 4.13% and 7.5%, respectively, and resulted in 4 unnecessary lobectomies and 2 second thoracotomies. The most common errors included misdiagnoses as lymphoma, squamous carcinoma or metastasis from breast carcinoma. Thirty one pathologic features were evaluated in the 66 carcinoid tumors and 10 frozen sections each of lymphoma, squamous cell carcinoma, and metastatic breast carcinoma. Seven pathologic features were significant by chi square test at P > .05. Positive likelihood ratios identified 11 pathologic features that were useful for the diagnosis of carcinoid tumor from other neoplasms. The applicability of the 11 pathologic features was tested with a group of pathologists, resulting in significant improvement in diagnostic accuracy as measured by pre and posttests. The value of evidence-based pathology and Bayesian statistics to evaluate complex differential diagnoses in pathology is discussed.