Evaluation of the Utility of Prognostic Models for Patients Diagnosed with Peri-hilar Cholangiocarcinoma

Abstract Background Several potential prognostic models have been developed to stratify patients with peri-hilar cholangiocarcinoma (PHC) by Overall Survival (OS). The American Joint Committee on Cancer (AJCC) staging system is a post-resectional model utilising tumour-specific pathological parameters to stratify and predict OS. The Mayo Clinical (MC) scoring system has been developed utilising primarily clinical, serological, and radiological variables to predict survival in all patients with a diagnosis of peri-hilar cholangiocarcinoma. The objective of this study was to evaluate the utility of these models in determining prognosis for all patients presenting to a tertiary treatment centre with PHC. Methods Three hundred and two patients diagnosed with PHC referred to a regional tertiary referral centre between 2008 and 2019 had their demographic and survival data retrospectively analysed from a prospectively held database linked to Hospital Episode Statistics and Somerset Cancer Registry data. One hundred and twenty seven patients were surgically explored. Eight-four patients underwent resection. One-hundred and seventy-four (57.6%) patients underwent palliative endoscopic therapy. Univariate and multivariate modelling was utilised to determine significant prognostic variables. Concordance Indices (C-Indices) were constructed for the prognostic models to determine internal validity within the cohort. Results Multivariate analysis demonstrated that: pre-interventional ECOG status (p < 0.001); serum albumin (p < 0.001); bilirubin levels (p < 0.001); CA 19-9 levels (p < 0.001) and resectional status (p < 0.001) were significant predictors of OS. Patients stratified by the MC scoring system to early-stage disease had a significantly longer OS compared to patients fulfilling late-stage criteria (p < 0.001). The predictive C-Indices for the MC model obtained significance in discriminating OS for the entire cohort (p < 0.05) and un-resected patients (p < 0.05). Neither model attained significant concordance for accurately discriminating OS in post-resectional patients. Conclusions The predictive performance of the stated prognostic models for OS have poor utility. Simple pre-interventional serological, functional and radiological variables appear to provide better prognostic indication of OS. Variables not incorporated in the AJCC registry have a significant effect upon post-resectional OS and require full incorporation in to model prognostication.