Accuracy of CT in predicting malignant potential of cystic pancreatic neoplasms

Background. Cystic lesions of the pancreas are being identified more frequently. Deciding which asymptomatic lesions can be safely followed with serial imaging and which require resection due to malignant potential is an increasingly common question. Current clinical practice is to rely on characteristics of the lesions on CT scan, and additional information from endoscopic ultrasound with fine-needle aspiration (EUS-FNA) and cyst fluid analysis or endoscopic retrograde pancreatography (ERCP) to assess malignant potential. Hypothesis. The malignant potential of pancreatic cystic lesions cannot be accurately predicted by CT scan. Methods. CT scans from 48 patients with cystic lesions of the pancreas were stripped of patient identifiers and retrospectively presented to two expert radiologists. The radiologists recorded specific characteristics of the lesions thought to be important in the differential diagnosis and their opinion of the likely diagnosis. Diagnostic accuracy was assessed by comparing the radiologists' diagnoses to the final pathologic diagnosis after resection. To determine if clinical history, EUS-FNA or ERCP findings improved diagnostic accuracy, medical records were retrospectively reviewed and scored as either supporting or not supporting malignant potential of the lesion. Results. Specific diagnoses based on CT findings alone were correct in an average of 39% of the cases. Even when diagnoses were dichotomized as benign (43%) or potentially malignant (57%, papillary mucinous neoplasms, mucinous cystic neoplasms, cancer), determinations based on CT alone were accurate in an average of 61% of cases. Accuracy rates were 60.4 and 62.5% for the two radiologists, although there was only fair agreement between them (Kappa = 0.28, 95% CI = (0.01 - 0.55), p = 0.05). When all clinical information available was considered together as a single dichotomous indicator of malignant potential, the indicator was accurate in 90% of the cases (Kappa = 0.73, 95% CI = (0.51 - 0.95, p <0.0001)). Conclusion. Specific preoperative diagnosis of pancreatic cystic neoplasms by CT alone is substantially inaccurate. Complementary tests such as EUS-FNA with fluid analysis and ERCP should be recommended to improve diagnosis especially if nonoperative treatment is planned.