A deep learning-based prognostic prediction model for distal cholangiocarcinoma incorporating the metabolism-inflammation marker monocyte-to-high-density lipoprotein cholesterol ratio

Abstract

Background: Inflammatory responses and lipid metabolism play a pivotal role in tumor initiation and progression, significantly impacting the prognosis of patients with malignant tumors. This study aims to investigate the prognostic relevance of the monocyte-to-high-density lipoprotein cholesterol ratio (MHR)— a novel inflammation-metabolism biomarker—in patients with distal cholangiocarcinoma (dCCA), leveraging deep learning-based analytical approaches. Methods: Clinicopathological records of dCCA patients managed at The First Affiliated Hospital of Bengbu Medical University (Bengbu, China) between January 2011 and July 2023 were retrospectively reviewed. Receiver operating characteristic (ROC) analysis was performed and the area under the curve (AUC) used to quantify the ability of MHR to predict outcomes. Associations were evaluated using Cox proportional hazards regression in both univariable and multivariable forms. Predictors that remained in the multivariable model were compared with the highest-importance features from the random forest ranking; the intersecting variables were incorporated to construct a survival-prediction nomogram. Results: One hundred and eighty-eight patients with dCCA following radical pancreaticoduodenectomy (PD) were enrolled. The area under AUC for MHR in predicting 1-year postoperative survival was 0.651 [95% confidence interval (CI): 0.5538–0.7485], with an optimal cutoff value of 0.74. Patients were divided into a high MHR group (MHR?>0.74, n=82) and a low MHR group (MHR ≤0.74, n=106) based on this cutoff value. The median disease-free survival (DFS) time were 42 months and 18 months, respectively (P=0.002) whereas median overall survival (OS) times for the low and high MHR groups were 36 months and 17 months, respectively (P<0.001). Multivariate analyses combined with random forest analysis and least absolute shrinkage and selection operator (LASSO) regression identified that preoperative MHR, carbohydrate antigen 19-9 (CA19-9), lymph node metastases, portal system invasion and tumor differentiation were independent predictors of postoperative mortality. Conclusions: By combining a readout of systemic inflammation with HDL-related lipid status, MHR emerges as an informative prognostic index in dCCA. In parallel, CA19-9 concentration, nodal involvement, portal system invasion, and histologic differentiation are each independently associated with survival. Integrating these variables within our deep-learning-based prognostic model enables earlier risk triage and more targeted postoperative management, with potential to improve clinical outcomes.