Machine Learning Algorithm-Based Tumor-Stroma Ratio Can Stratify the Prognosis of Intrahepatic Cholangiocarcinoma.

INTRODUCTION: The study aimed to quantify tumor components through the machine learning algorithm and explore effective biomarkers to stratify the prognosis of intrahepatic cholangiocarcinoma (iCCA) after radical surgery.

METHODS: A cohort of 237 iCCA patients who underwent radical resection was recruited. The semiautomated pipeline was constructed to measure the tumor microenvironment components, including tumor, lymphocyte, and stromal cells, tumor-stroma ratio, and tumor-infiltrated lymphocytes ratio % were calculated. The overall survival (OS) and disease-free survival (DFS) were compared to evaluate their prognostic values. The predictive values for adjuvant chemotherapy were then explored.

RESULTS: The Kaplan-Meier analysis showed that high-stroma and low-tumor-infiltrated lymphocytes ratio % were associated with shorter DFS and OS, and the multivariable Cox analysis also verified the prognosis values of iCCA including DFS (hazard ratio: 1.59, 95% confidence interval: 1.10-2.30, P = 0.015) and OS (hazard ratio: 1.92, 95% confidence interval: 1.27-4.17, P < 0.001). The nomograms presented better performance than previous staging systems, including the 8th American Joint Committee on Cancer system and the Liver Cancer Study Group of Japan system. The low-stroma cohort was more likely to benefit from chemotherapy, including DFS and OS (P = 0.019 and P = 0.002).

CONCLUSIONS: Machine learning-based tumor-stroma ratio could serve as an effective prognostic biomarker for iCCA after radical surgery and potentially predict the therapeutic response of adjuvant chemotherapy.