Development of a Model for Differentiating PDAC from Benign Pancreatic Conditions: A Prospective Case-control Study

Background & aimsPancreatic ductal adenocarcinoma (PDAC) continues to be a devastating disease with late diagnosis and poor overall survival, complicated by clinical presentations similar to benign pancreatic diseases. We aimed to analyse clinical parameters with the goal of developing a prediction model for differentiating suspected PDAC from benign pancreatic conditions. Methods and resultsWe used a prospectively recruited cohort of patients with pancreatic disease (n=762) enrolled at the Barts Pancreas Tissue Bank between January 1, 2008 and September 21, 2021 to perform a case-control study examining the association of PDAC (n=340) with predictor variables including demographics, comorbidities, lifestyle factors, presenting symptoms and commonly performed blood tests. Using a machine learning approach, candidate PDAC risk-prediction algorithms were trained on 75% of the cohort, using a subset of the predictor variables identified from a preliminary observational association study. Models were assessed on the remaining 25%. Multiple imputed datasets were used for both training and validation to accommodate for unknown data. Age (over 55), weight loss in hypertensive patients, recent symptom of jaundice, high serum bilirubin, low serum creatinine, high serum alkaline phosphatase, low lymphocyte count and low serum sodium were the most important features when separating putative PDAC cases from less severe pancreatic conditions. A simple logistic regression model had the best performance with an area under the curve (AUC) of 0.88. Setting a probability threshold of 0.17 guided by the maximum F2 score, a sensitivity of 95.6% was reached in the full cohort which could lead to early detection of around 84% of the PDAC patients. ConclusionThe resultant prediction model significantly outperformed the current UK guidelines for suspected pancreatic cancer referral and could improve detection rates of PDAC in the community. After further work this approach could lead to an easy to understand, utilisable risk score to be applied in the primary and secondary care setting for referring patients to specialist hepato-pancreatico-biliary services.