Exploring the Nexus of Inflammation, Depression and Pancreatic Cancer Through Machine Learning (Preprint)

Abstract

Depression is a prevalent comorbidity in patients with pancreatic cancer, negatively impacting their quality of life, treatment adherence, and survival. Inflammation is a key biological process that may link pancreatic cancer and depression, with inflammatory markers like C-reactive protein (CRP) and neutrophil-lymphocyte ratio (NLR) playing significant roles. Machine learning (ML) provides a novel approach to predicting depression based on these biomarkers, potentially enabling early intervention and improved patient outcomes.

This study aimed to investigate the clinical associations between depression, inflammation, and pancreatic cancer, and to utilize machine learning to predict depression using biomarker levels and clinical data.

A prospective cohort study was conducted between May 2021 and November 2023, including 328 patients diagnosed with pancreatic cancer. Depressive symptoms were measured using the Patient Health Questionnaire-9 (PHQ-9). Generalized estimating equations (GEE) were used to explore relationships between depression and inflammatory markers, with significant variables from univariate analyses included in multivariate models. Various ML algorithms, including Random Forest, K-Nearest Neighbors (KNN), and Extreme Gradient Boosting (XGBoost), were employed to predict depression, evaluated using classification metrics such as accuracy, precision, recall, and F1-score.

The cohort had a mean age of 65 years, with most patients diagnosed at stage IV. Clinically significant depression (PHQ-9 ≥10) was observed in 35% of patients at baseline, decreasing over time. Univariate analyses indicated associations between depression and factors such as lack of surgical resection and metastatic disease; however, these were not significant in multivariate models. Only log-transformed CRP and NLR remained significant inflammatory markers in multivariate analyses. Among the ML models, XGBoost achieved the highest performance, with an accuracy of 81%, precision of 84%, recall of 91%, and an F1-score of 86%. CRP, NLR, and platelet-lymphocyte ratio (PLR) emerged as the strongest predictors of depression.

Depression is prevalent among pancreatic cancer patients and is likely linked to inflammation. Machine learning models, particularly XGBoost, effectively predicted depression using inflammatory markers. Integrating ML-based predictions into mental health care within cancer treatment protocols may improve patient outcomes. Future research should focus on refining these predictive models and exploring their clinical implementation for early depression detection and intervention in pancreatic cancer care.