Dynamic Prediction of Survival After Curative Resection of Intrahepatic Cholangiocarcinoma: A Landmarking-Based Analysis

Background The current study aimed to develop a dynamic prognostic model for patients undergoing curative-intent resection for intrahepatic cholangiocarcinoma (ICC) using landmark analysis. Methods Patients who underwent curative-intent surgery for ICC from 1999 to 2017 were selected from a multi-institutional international database. A landmark analysis to undertake dynamic overall survival (OS) prediction was performed. A multivariate Cox proportional hazard model was applied to measure the interaction of selected variables with time. The performance of the model was internally cross-validated via bootstrap resampling procedure. Discrimination was evaluated using the Harrell's Concordance Index. Accuracy was evaluated with calibration plots. Results Variables retained in the multivariable Cox regression OS model included age, tumor size, margin status, morphologic type, histologic grade, T and N category, and tumor recurrence. The effect of several variables on OS changed over time. Results were provided as a survival plot and the predicted probability of OS at the desired time in the future. For example, a 65-year-old patient with an intraductal, T1, grade 3 or 4 ICC measuring 3 cm who underwent an R0 resection had a calculated estimated 3-year OS of 76%. The OS estimate increased if the patient had already survived 1 year (79%). The discrimination ability of the final model was very good (C-index: 0.80). Conclusion The long-term outcome for patients undergoing curative-intent surgery for ICC should be adjusted based on follow-up time and intervening events. The model in this study showed excellent discriminative ability and performed well in the validation process.

Automated summary

The study aimed to develop a dynamic prognostic model for patients undergoing curative-intent resection for intrahepatic cholangiocarcinoma (ICC) using landmark analysis. The model incorporated selected variables and their interaction with time to predict overall survival (OS) over time. The model demonstrated good discrimination ability and accuracy in internal validation.