Clinical Significance of Risk Factor Analysis in Pancreatic Cancer by Using Supervised Model of Machine Learning

Introduction: Particularly pancreatic cancer, poses a significant global health challenge due to its high mortality rates despite advancements in treatment. Early detection remains crucial as most cases are diagnosed at late stages when surgical intervention is no longer viable. We focused to identify relevant risk factors of pancreatic cancer. Our goal was to determine pertinent risk factors for pancreatic cancer. The best machine learning model was used for risk scoring in pancreatic cancer based on those risk factors and determine their diagnostic value. Methods: We conducted a matched case-control study, retrospectively collecting demographic data and common haemato-logical indicators from all participants. Best model of machine learning among SVM and Logistic regression was chosen to identify risk factors for pancreatic cancer after initial variable selection by dendrogram. Based on these factors, we created a best model for risk scoring in pancreatic cancer and showed higher diagnostic value. Result: 353 cases and 370 controls were finally participated in our study. The discoveries of our machine learning logistic regression with backward elimination showed that Haemoglobin A1c (OR 1.28, 95%CI: 1.08,1.52), Alkaline phosphatase (OR 1.02, 95%CI: 1.01,1.03), CA19-9 (OR 1.01, 95%CI: 1.01,1.01), and Carcinoembryonic antigen (OR 1.41, 95%CI: 1.2,1.66) were related to an expanded risk of PC, while BMI (OR 0.88, 95%CI: 0.81,0.97) were asso-ciated with a diminished risk of PC. Based on these outcomes, the clinical PC for risk scoring was well fitted in the modelled populace, and the score had strong predictive worth with area under receiver operating curve was 0.969 (P < 0.001) which showed higher diagnostic value. Conclusion: HbA1C, ALP, BMI, CA19-9 and CEA levels were associated with the risk of PC. The risk scoring scale (nomogram) might be useful in clinical PC screening as a diagnostic tool by supervised ma-chine learning.