DEVELOPMENT AND INTERNAL VAL?DAT?ON OF A PREOPERAT?VE CL?N?CAL PROGNOST?C MODEL FOR OVERALL SURV?VAL AND RECURRENCE AFTER CURAT?VE RESECT?ON OF HEPATOCELLULAR CARC?NOMA

Ipek Balikci Cicek*, Zeynep Kucukakcali

ABSTRACT

Background: Established staging systems for hepatocellular carcinoma (HCC), including the Barcelona Clinic Liver Cancer (BCLC), Cancer of the Liver Italian Program (CLIP), and Tumor-Node-Metastasis (TNM) classifications, were primarily developed to inform treatment allocation and offer comparatively limited discriminative ability for individualized outcome prediction after curative resection. We aimed to develop and internally validate a prognostic model based exclusively on routinely available preoperative variables for overall survival (OS) and time-to-recurrence (TTR) following hepatic resection for HCC. Methods: We retrospectively analyzed an openly available cohort of 535 patients who underwent curative resection for histologically confirmed HCC. Multivariable Cox proportional hazards regression was used to construct separate models for OS and TTR from a candidate set of preoperative clinical and laboratory variables. Model performance was benchmarked against the BCLC, CLIP, and TNM staging systems using Harrell's concordance index (C-index), internally validated via 500-repetition bootstrap resampling with optimism correction, and evaluated for calibration and clinical utility using calibration plots and decision curve analysis (DCA). Because death without recurrence constitutes a competing risk for TTR, we additionally compared naive Kaplan-Meier estimates with Aalen-Johansen cumulative incidence functions. Results: Over a median follow-up of 42 months, 219 patients (40.9%) died and 245 (45.8%) developed recurrence. The multivariable OS model achieved a C-index of 0.658, and the TTR model achieved a C-index of 0.574; both exceeded the BCLC (0.556/0.539), CLIP (0.511/0.529), and TNM (0.502/0.493) systems. Bootstrap-corrected C-indices were 0.636 (OS) and 0.559 (TTR), indicating limited overfitting. Calibration was satisfactory across risk strata, and DCA demonstrated positive net benefit across a clinically plausible range of risk thresholds (5%-50%). Aalen-Johansen analysis indicated that naive Kaplan-Meier estimation overstated 5-year recurrence risk by approximately 5 percentage points relative to the competing-risk-adjusted estimate. Multiple tumor number was independently associated with lower, rather than higher, mortality risk in the multivariable model (hazard ratio 0.66); sensitivity analyses did not support this being a statistical artifact. Conclusions: A prognostic model built exclusively from routinely available preoperative variables demonstrated better discrimination than established staging systems, particularly for overall survival, together with satisfactory calibration and positive clinical net benefit after HCC resection. Discrimination for TTR remained modest despite outperforming the evaluated staging systems. External validation in independent cohorts is required before clinical implementation.

Keywords: hepatocellular carcinoma; hepatectomy; prognostic model; nomogram; competing risks; recurrence; decision curve analysis.