A pathomics-integrated multimodal model to evaluate chemoimmunotherapy efficacy in unresectable intrahepatic...

Chemoimmunotherapy has emerged as the first-line therapy for unresectable intrahepatic cholangiocarcinoma (ICC). However, durable clinical responses are observed in less than 30% of patients, necessitating biomarkers of survival benefit. Thus, the aim of this study was to develop and validate a pathomics-based prognostic signature for patients with ICC receiving chemoimmunotherapy.

This multicenter study included patients with ICC with biopsy samples. Pathomics features were extracted from digital H&E-stained images, and the pathomics signature for ICC (PS-ICC) was developed by machine learning (ML). SHapley Additive exPlanations provided algorithmic explanation, and The Cancer Genome Atlas database supported biological interpretation. We explored the potential of PS-ICC as surrogate index compared with radiological response.

Based on pretreatment specimens, 189 patients receiving chemoimmunotherapy were included. Univariate Cox analysis demonstrated that the PS-ICC status from a pathomics-driven ML model was associated with overall survival (OS) in patients with unresectable ICC undergoing chemoimmunotherapy (training cohort: hazard ratio (HR) = 0.09, 95% CI, 0.05-0.14, p <0.001; validation cohort: HR = 0.20, 95% CI 0.08-0.47, p <0.001). PS-ICC characterized tumor microenvironment heterogeneity, with lower scores correlating with reduced M0 macrophage infiltration (p = 0.043) and programmed death-ligand 1 expression and programmed death-1 checkpoint pathways (p = 0.002). In addition, PS-ICC status could serve as a surrogate marker for OS compared with the Response Evaluation Criteria in Solid Tumors Version1.1 at 3 months post treatment (training cohort: AUC at 1 year: 0.868 vs. 0.701; 2 year: 0.787 vs. 0.694). The PS-ICC-integrated nomogram outperformed conventional prognostic model in accuracy (Concordance index: 0.80 vs. 0.71).

The PS-ICC demonstrates potential as a surrogate endpoint for survival prediction in patients with ICC undergoing chemoimmunotherapy, with biological plausibility evidenced by its tumor microenvironment associations. Prospective trials are warranted to confirm clinical utility.

This study developed and validated a machine learning-based pathomics signature that accurately predicts overall survival in patients with intrahepatic cholangiocarcinoma (ICC) receiving chemoimmunotherapy. The pathomics signature for ICC provides a biologically grounded, pretreatment biomarker to stratify patients for chemoimmunotherapy, potentially reducing overtreatment and guiding personalized strategies. By demonstrating strong correlation with overall survival, this signature could serve as a surrogate endpoint in clinical trials, thereby accelerating drug development. Furthermore, its link to immune pathways could inform novel therapeutic targets in ICC. However, prospective validation is needed for clinical adoption.