Machine Learning-Based Classification of Depression Using Inflammatory Biomarkers in Pancreatic Cancer Patients.

Abstract

Inflammation is a common mediator of pancreatic cancer and depression. This study investigated the predictive value and clinical associations of inflammatory markers and depression in cancer patients using machine learning (ML) and statistical modeling. Pancreatic cancer patients (n = 328; mean age, 65 years; majority with stage IV disease) were assessed using the Patient Health Questionnaire-9 (PHQ-9; depression defined as PHQ-9 ≥ 10). Clinically significant depression was present in 35% of subjects at baseline, and the rate declined at follow-up. Four ML models (logistic regression, random forest, support vector machine, and extreme gradient boosting; XGBoost) were trained using routinely collected clinical data and showed comparable performances with moderate but consistent discriminative capacity (AUC: 0.70-0.72). Permutation importance analysis revealed C-reactive protein (CRP), neutrophil-to-lymphocyte ratio (NLR), and albumin as key predictors of depression. Generalized estimating equations further confirmed that elevated CRP (OR = 1.32, p = 0.001) and NLR (OR = 1.55, p = 0.001) were independently associated with depression. These findings suggest that inflammatory markers can not only help to identify patients at risk for depression but also underscore the linkage between inflammation and depression. ML models incorporating these markers may therefore support early detection and intervention in pancreatic cancer care.