Surgery and Survival in Patients with Intrahepatic Cholangiocarcinoma: A Bayesian Inference

Background: Conventional approaches often struggle to draw inferences from limited data. We aimed to determine whether Bayesian perspectives improve computation and increase clarity when inferring from small series.
Methods: We conducted a retrospective longitudinal study. The participants were who underwent curative liver resection for ICC. Bayesian proportional hazards models were used to analyse postsurgical survival. Three scenarios were used to model overall survival; all estimates were then compared with the largest meta-analysis.
Results: The final cohort included 14 patients with resectable ICC with median follow-up duration of 14.68 months (IQR 10.77 - 19.84). While the model with non-informative priors revealed unstable estimates, the models with informative and less informative priors demonstrated consistent and reliable effects. The estimates from the latter models aligned with those reported in the meta-analysis. The posterior probabilities of null value exceedance (i.e., HR > 1.0) for age, sex, lymph node metastasis, and positive surgical margin were 98%, 93%, 100%, and 48%, respectively, albeit (in frequentist terms) non-significant for all but lymph node metastasis.
Conclusion: The Bayesian framework offers a promising strategy that overcomes the limitations of the frequentist approach, contributing to the advancement of inferences from limited data.