Page 81 - Read Online

Page 12 of 22 Ma et al. Hepatoma Res. 2026;12:43

Integrating the
multidimensional To integrate the
molecular copy number,
Whole-genome mutation, and
The number of characteristics of
mtDNA sequencing, plasma-free fragment omics
HDR prediction mutations, Capture-based mitochondrial DNA characteristics of
model and HPR 2025 mtDNA copy mtDNA provides a highly plasma-free OS and PFS
prediction model [55] sequencing, Cox mitochondrial DNA
number, HPP sensitive, real-time,
score proportional and noninvasive for real-time
hazards model monitoring of TACE
prognostic
monitoring tool for efficacy and
Frontier liquid biopsy prognosis
and
future The key variables
models Excellent are selected
Intratumoral classification through recursive
feature elimination,
artery, Corona Recursive feature feature processing and the
enhancement, elimination, ability and strong interpretability of
DCP, fat in mass, ensemble anti-overfitting
CatBoost model [56] 2025 model prediction is ASP
INR, Neu, tumor learning performance can realized by using
number, ALT, BMI, (gradient effectively integrate SHAP, which
necrosis or severe boosting), SHAP multi-source ultimately serves
ischemia, AFP clinical and image the risk
data
stratification of
patients
HCC: Hepatocellular carcinoma; TACE: transarterial chemoembolization; AST: aspartate aminotransferase; AFP: alpha-fetoprotein; HBV: hepatitis B
virus; BCLC: Barcelona Clinic Liver Cancer; mtDNA: mitochondrial DNA; HPP score: HCC Prognosis Prediction Model; DCP: des-γ-carboxy
prothrombin; INR: international normalized ratio; Neu: neutrophil count; ALT: alanine aminotransferase; BMI: body mass index; SHAP: SHapley
Additive exPlanations; OS: overall survival; PFS: progression-free survival; ASP: advanced-stage progression; mRECIST: modified Response
Evaluation Criteria in Solid Tumors.

model specifically developed for ideal TACE candidates (BCLC A unsuitable for curative therapies and
BCLC B), it demonstrated that tumor burden alone effectively stratifies outcomes. However, its limitations
include derivation from a predominantly Chinese HBV cohort with limited generalizability, and the absence
of post-treatment response or alpha-fetoprotein (AFP), limiting dynamic risk assessment.

To address these limitations, they proposed the 6-and-12 model 2.0 , which incorporates baseline AFP as a
[47]
continuous variable: size + number + 1.5 × log (AFP). Using AFP-dependent cut-offs (e.g., 6/12 for AFP
10
400-2,000 ng/mL), patients were stratified into three risk strata. In a multi-ethnic validation, the 2.0 model
showed improved discrimination and calibration compared with the original version and other existing
models, with mOS of 45.0, 30.0, and 15.8 months in the training cohort. Limitations include retrospective
design, lack of dynamic post-treatment variables, and the need for prospective validation in the current era of
targeted therapy plus immunotherapy.

Pre-TACE-Predict model
Pre-TACE-Predict is a multinational, multicentre prognostic model developed from 4,621 patients with
[48]
HCC treated with TACE across 19 centers in 11 countries. The model predicts overall survival using seven
baseline variables: tumor number, tumor size (log ), AFP (log ), albumin, bilirubin (log ), vascular invasion,
10
10
10
and etiology. Based on the linear predictor, patients were stratified into four risk categories (using the 16th,
50th, and 84th percentiles as cut-offs), with mOS as follows: risk category 1 (lowest risk) approximately 41
months (range 35-47 months across cohorts), risk category 2 approximately 26-34 months, risk category 3
approximately 17-18 months, and risk category 4 (highest risk) approximately 8-9 months. A free online
calculator (TACE-Predict) is available to generate individualized survival probabilities. As a well-validated
pre-treatment tool, Pre-TACE-Predict helps identify high-risk patients unlikely to benefit from TACE,
facilitating early alternative treatment decisions.