Carcinoid tumours: Predicting the location of the primary neoplasm based on the sites of metastases

Objectives: To predict the primary neuroendocrine tumour of the gastrointestinal tract site based on observed metastatic sites. Methods: We studied data from the radiology database of a single, large cancer centre on 250 patients with pathologically confirmed neuroendocrine tumours. Primary tumour sites and the locations of metastases were collected from pathologic and radiologic reports of all available imaging modalities, such as computed tomography (CT), positron emission tomography (PET/CT), magnetic resonance imaging (MRI) and octreotide scans in the database. A nominal regression model was used to predict primary tumour site using the observed metastatic sites. Regression coefficients that were not statistically significant at the 5 % level were eliminated from the model in a stepwise procedure. Results: Lung and liver metastases were not statistically significant predictors of the location of primary tumours (p = 0.86 and 0.074, respectively); whereas, lymph node, bone, and peritoneal metastases were significant predictors (p < 0.0001, 0.0004, and 0.014, respectively). Conclusions: Metastatic neuroendocrine tumours to the lymph nodes, bone, and peritoneum can be used to predict the primary neuroendocrine site; however, metastases in the lung and liver alone cannot predict the site of the primary tumour site. Key Points: • Imaging helps one to diagnose the location of primary neuroendocrine neoplasm and the associated metastases. • Diffuse metastatic disease often makes identification of the primary difficult. • A prediction model developed may help identification of the primary in this setting. • It may also help identify occult metastases and thereby assist in management.