Clinical significance of risk factor analysis in pancreatic cancer by using supervised model of machine learning

Pancreatic cancer (PC) poses a significant global health challenge due to its aggressive nature, late-stage diagnosis, and high mortality despite advancements in treatment. Early detection remains crucial for timely intervention. This study aimed to identify clinically relevant predictors of pancreatic cancer using a supervised machine learning approach and to develop a risk stratification tool with diagnostic capabilities.

A matched case-control study was conducted retrospectively at the Tenth People's Hospital of Tongji University (2017-2023), involving 353 cases and 370 matched controls. Demographic and hematological data were extracted from medical records. Variables were pre-selected using cluster dendrograms and subsequently refined using logistic regression with backward elimination and Support Vector Machine (SVM) models. A final risk scoring model was developed based on the best-performing model and internally validated.

Key predictors retained in the final logistic regression model included Hemoglobin A1c (HbA1c) (OR 1.28; 95% CI: 1.08-1.52), Alkaline Phosphatase (ALP) (OR 1.02; 95% CI: 1.01-1.03), CA19-9 (OR 1.01; 95% CI: 1.01-1.01), Carcinoembryonic Antigen (CEA) (OR 1.41; 95% CI: 1.20-1.66), and Body Mass Index (BMI) (OR 0.88; 95% CI: 0.81-0.97). The final model demonstrated excellent diagnostic performance (AUC = 0.969, p < 0.001), with high accuracy, sensitivity, and specificity. A nomogram was constructed to facilitate individualized PC risk assessment.

HbA1c, ALP, CA19-9, CEA, and BMI were independently associated with pancreatic cancer. The machine learning-derived risk scoring model demonstrated high predictive accuracy and may serve as a valuable clinical tool for early detection and screening of pancreatic cancer.