Development of scoring models to predict early pancreatic cancer risk in new-onset diabetes mellitus patients.

Background: Diabetes mellitus is a risk factor of pancreatic cancer (PC). We aimed to develop risk models to predict PC within 1- and 3-year in new onset diabetes mellitus (NODM). Methods: We utilized territory-wide electronic healthcare database to identify NODM patients [fasting glucose ≥7 mmol/L and/or hemoglobin A1c (HbA1c) ≥6.5%] between 2008 and 2017, and observed for three years until PC development, pancreatectomy or death. Outcome of interest was PC at 1- and 3-year. Clinical variables included age, sex, acute pancreatitis history, body mass index (BMI), blood glucose, lipid profile, estimated glomerular filtration rate (eGFR), alanine transaminase (ALT), alkaline phosphatase (ALP), hemoglobin and baseline medication use {metformin, insulin and other anti-diabetic drugs, aspirin, non-steroidal anti-inflammatory drugs, statins, gastroprotective agents [proton pump inhibitors (PPIs) and histamine-2 receptor antagonists (H2RAs)]}. The cohort was randomly split (7:3 ratio) into training and testing sets. Logistic regression was conducted to identify predictors of PC at 1 and 3 years. Results: Among 117,121 NODM patients (mean age: 61.63±12.45 years; male: 51.3%), 189 (0.16%) developed PC within 3-year, with 82 (43.4%) in first year. Predictors in 1- and 3-year risk models included age, sex, acute pancreatitis history, insulin, aspirin, gastroprotective agents, baseline glucose, eGFR, ALT, ALP, and BMI (baseline and change from 1 year prior to baseline). The 1- and 3-year models had area under receiver operating characteristics curve of 0.90 and 0.81, respectively. If specificity was set as 99.9%, number-needed-to-screen (NNS) for one PC case within 1-year was 27. Conclusions: We developed and validated new risk models with high accuracy to predict 1- and 3-year PC risk among NODM patients for targeted screening.