3458 - Individualized Pancreatic Cancer Prevention: A Machine Learning Approach Integrating Polygenic Risk Scores for Absolute Risk Prediction in UK Biobank and Community Settings

02:30pm - 03:45pm PT

Hall F

Screen: 19

POSTER

Presenter(s)

Te-Min Ke, MD, PhD - Chi Mei Medical Center, Yungkang City 701 44, Tainan

T. M. Ke¹, A. Lophatananon², and K. R. Muir³; ¹Chimei Medical Center, Tainan, Taiwan, ²2. Division of Population Health, Health Services Research and Primary Care, School of Health Sciences, Faculty of Biology, Medicine and Health, The University of Manchester, Manchester M13 9PT, UK, Manchester, UK, United Kingdom, ³Division of Population Health, Health Services Research and Primary Care, School of Health Sciences, Faculty of Biology, Medicine and Health, The University of Manchester, Manchester M13 9PT, UK., MAnchester, UK, United Kingdom

Purpose/Objective(s):

Pancreatic cancer (PaCa) remains one of the most lethal malignancies, with limited improvements in survival. However, approximately 38% of cases are considered preventable, emphasizing the need for precision prevention. Our study team developed a machine learning-based PaCa risk prediction model in the UK Biobank (UKB), integrating polygenic risk scores (PRS) for better risk stratification. Translating relative risk into absolute risk (AR) is crucial for real-world application. This study aimed to validate a 5- and 10-year AR prediction model in a UK community setting.

Materials/Methods:

We applied our validated UKB-based PaCa risk model. The 5- and 10-year ARs were calculated using: p(a, s, r) = b1 * r / (b1 * r + b2 * (1 - e^(-t * (b1 * r + b2)))), where a, s, r represent age, sex, and odds ratio (OR), b1 denotes age- and sex-specific PaCa hazard rates, and b2 represents competing mortality rates. The model integrates nine risk factors: age, gender, PRS, blood type, smoking, alcohol use, diabetes (DM), pancreatitis, and gallbladder disease. A 2% AR threshold was used for early screening consideration. A web-based AR calculator was developed: https://sunnyteminke.github.io/Pancreatic-Cancer-5-year-and-10-year-Absolute-Risk-Calculator/. The UNIMAN pilot study (EU iHelp project) served as the real-world application. It began on April 20, 2023, with participants providing blood samples for genetic analysis and completing lifestyle and medical history questionnaires. PRS construction was performed using PLINK 1.9, STATA/MP 17, and R (v4.2.1). Ethics approval was granted by the University of Manchester Research Ethics Committee (Ref: 2022-13644-23283).

Results:

A total of 244 participants were recruited; however, only 139 provided complete data. Most were male (93.53%) and aged =50 years. 52.52% had non-O blood types, and 68.35% had low PRS (Q1), while 4% had high PRS (Q4-Q5). 65.31% were never smokers, and 40% exceeded alcohol intake recommendations. DM was present in 7.19%, gallbladder disease in 4%, and pancreatitis in one case. The mean 5-year AR was 0.32% (range: 0.002%–3.643%), and the mean 10-year AR was 1.12% (range: 0.009%–10.573%). Using a 2% AR threshold, only 1.44% (n=2) exceeded 5-year AR, but this increased to 14% (n=23) for 10-year AR. The highest observed 5-year AR was 3.6%, rising to 10.6% at 10 years.

Conclusion:

This study successfully applied an academic PaCa risk model to a UK community, demonstrating feasibility in translating relative risk into clinically meaningful AR values. While the 5-year AR remains low (~0.3%), it substantially increases over 10 years (~1.1%), emphasizing the need for long-term risk assessment. The notable AR rise supports sustained health monitoring, early detection, and targeted preventive strategies. PRS-informed screening and lifestyle interventions may help mitigate future PaCa risk.