The Difficulty of Detecting Occult Metastases in Patients with Potentially Resectable Pancreatic Cancer: Development and External Validation of a Preoperative Prediction Model

Abstract

Occult metastases are detected in 10-15% of patients during exploratory laparotomy for pancreatic cancer. This study developed and externally validated a model to predict occult metastases in patients with potentially resectable pancreatic cancer. Model development was performed within the Dutch Pancreatic Cancer Audit, including all patients operated for pancreatic cancer (January 2013-December 2017). Multivariable logistic regression analysis based on the Akaike Information Criteria was performed with intraoperative pathologically proven metastases as the outcome. The model was externally validated with a cohort from the University Hospital of Verona (January 2013-December 2017). For model development, 2262 patients were included of whom 235 (10%) had occult metastases, located in the liver ( n = 143, 61%), peritoneum ( n = 73, 31%), or both ( n = 19, 8%). The model included age (OR 1.02, 95% CI 1.00-1.03), BMI (OR 0.96, 95% CI 0.93-0.99), preoperative nutritional support (OR 1.73, 95% CI 1.01-2.74), tumor diameter (OR 1.60, 95% CI 1.04-2.45), tumor composition (solid vs. cystic) (OR 2.33, 95% CI 1.20-4.35), and indeterminate lesions on preoperative imaging (OR 4.01, 95% CI 2.16-7.43). External validation showed poor discrimination with a C-statistic of 0.56. Although some predictor variables were significantly associated with occult metastases, the model performed insufficiently at external validation.