Abstract

Background/Objectives: Early detection of pancreatic cancer (PC) remains challenging due to the lack of effective screening tools. We aimed to develop a clinically applicable model to identify individuals at high risk of PC using routine health check-up data. Methods: In a 1:4 matched case–control study (111 cases and 439 controls) using data collected between 2010 and 2018, PC cases were defined as individuals diagnosed with PC within 2 years of a health check-up. A model was developed using conditional logistic regression and validated in an independent cohort of 52,043 individuals (40 PC cases; 2019–2021). Results: Five risk factors were identified: carbohydrate antigen 19-9 ≥ 39 U/mL, hemoglobin A1c ≥ 6.5%, alkaline phosphatase > 110 IU/L, weight loss ≥ 5%, and dyspepsia. A 12-point risk scoring system was constructed, with an area under the curve of 0.784 in the development and 0.841 in the validation cohort. At a threshold of ≥5, the high-risk group had a significantly higher 2-year incidence of PC than the low-risk group (2.56% vs. 0.047%; p < 0.001), a 54.76-fold risk enrichment, with a negative predictive value of 99.95% and a number-needed-to-follow of 39. Conclusions: This simple, validated model effectively stratifies short-term PC risk using routine data, potentially facilitating targeted surveillance in the general population.

IPC Classification

Keywords