Enhancing Recurrence-Free Survival Prediction in Hepatocellular Carcinoma: A Time-Updated Model Incorporating Tumor Burden and AFP Dynamics.

Abstract

Background: Existing models to predict recurrence-free survival (RFS) after hepatectomy for hepatocellular carcinoma (HCC) rely on static preoperative factors such as alpha-fetoprotein (AFP) and tumor burden score (TBS). These models overlook dynamic postoperative AFP changes, which may reflect evolving recurrence risk. We sought to develop a dynamic, real-time model integrating time-updated AFP values with TBS for improved recurrence prediction.

Patients and methods: Patients undergoing curative-intent hepatectomy for HCC (2000-2023) were identified from an international, multi-institutional database with RFS as the primary outcome. AFP trajectory was monitored from preoperative to 6- and 12-month postoperative values, using time-varying Cox regression with AFP as a time-dependent covariate. The predictive accuracy of this time-updated model was compared with a static preoperative Cox model excluding postoperative AFP.

Results: Among 1911 patients, AFP trajectories differed between recurrent and nonrecurrent cases. While preoperative AFP values were similar, recurrent cases exhibited higher AFP at 6 and 12 months. Multivariable analysis identified TBS (hazard ratio (HR):1.043 [95% confidence interval (CI): 1.002-1.086]; p = 0.039) and postoperative log AFP dynamics (HR:1.216 [CI 1.132-1.305]; p < 0.001) as predictors. Contour plots depicted TBS's influence decreasing over time, while postoperative AFP became more predictive. The time-varying Cox model was created to update RFS predictions continuously on the basis of the latest AFP values. The preoperative Cox model, developed with age, AFP, TBS, and albumin-bilirubin score, had a baseline C-index of 0.61 [0.59-0.63]. At 6 months, the time-varying model's C-index was 0.70 [0.67-0.73] versus 0.59 [0.56-0.61] for the static model; at 12 months, it was 0.70 [0.66-0.73] versus 0.56 [0.53-0.59]. The model was made available online ( https://nm49jf-miho-akabane.shinyapps.io/AFPHCC/ ).

Conclusions: Incorporating postoperative AFP dynamics into RFS prediction after HCC resection enhanced prediction accuracy over time, as TBS's influence decreased. This adaptive, time-varying model provides refined RFS predictions throughout follow-up.