Transcatheter arterial chemoembolization plus lenvatinib toward precision prognostication for unresectable hepatocellular carcinoma

Abstract

A recent study presented a timely prognostic model deriving from a meta-analysis of 43 group studies and externally validated with clinical data for predicting overall survival and progression-free survival in patients with unresectable hepatocellular carcinoma treated with transcatheter arterial chemoembolization plus lenvatinib. The model integrated nine readily available clinical variables, including number of tumors, microvascular invasion, Eastern Cooperative Oncology Group grade, Child-Pugh stage, triple therapy with programmed death 1 inhibitors, Barcelona Clinic Liver Cancer stage, extrahepatic metastasis, alpha-fetoproteinlevel, and hepatitis B virus (HBV) status, and demonstrated favorable discrimination relative to Barcelona Clinic Liver Cancer staging. This article acknowledges that the underlying evidence base is predominantly observational and derived largely from HBV-endemic regions; therefore, prospective and geographically diverse validation is required before broad clinical implementation. This article acknowledges the study’s strengths concerning large evidence synthesis, external validation, and clinical applicability while offering constructive critique on methodology, potential biases, biologic interpretation, and steps to maximize translational impact. More priorities could be proposed for refining and prospectively validating the model, integrating molecular and imaging biomarkers, addressing heterogeneity from HBV-endemic populations, and designing interventional studies to test model-guided therapy. Future work should test the model in prospective, multiethnic cohorts, examine interactions with etiologic subtypes (HBV/hepatitis C virus/non-alcoholic steatohepatitis), and integrate temporally aware analyses and biomarker/imaging data to optimize individualized treatment selection. These measures would strengthen the model’s role as a decision-support tool and accelerate evidence-based personalized care in advanced unresectable hepatocellular carcinoma.

Key Words: Hepatocellular carcinoma; Transcatheter arterial chemoembolization; Lenvatinib; Prognostic model; Nomogram; Survival prediction

Core Tip: We summarized risk factors including number of tumors, microvascular invasion, Eastern Cooperative Oncology Group performance status, and Child-Pugh stage, and protective factor (triple therapy) for unresectable hepatocellular carcinoma treated with transcatheter arterial chemoembolization plus lenvatinib, constructed and validated prognostic models. In the validation set, area under the curve values of overall survival and progression-free survival, calibration curves, confirmed their good performance, providing guidance for clinical practice.

INTRODUCTION

The recent study by Yu et al[1] has addressed the issue about individualized prognostication for patients with unresectable hepatocellular carcinoma (HCC) receiving transcatheter arterial chemoembolization (TACE) plus lenvatinib as a pressing clinical need[2-10]. In HCC, the combination of locoregional therapy with a modern multikinase inhibitor has been increasingly used nowadays[11-17]. Therefore, clinicians urgently need transparent tools with validation to balance expected benefit against toxicity, liver reserve, and competing risks[18-24]. Yu et al’s research group[1] has performed a meta-analytic approach through synthesizing 43 group studies and subsequent external validation to represent an important effort to move beyond single-cohort risk scores and the coarse granularity of Barcelona Clinic Liver Cancer (BCLC) staging.

STRENGTHS OF YU ET AL’S STUDY

Several strengths merit emphasis in Yu et al’s study[1]. First, the model they developed incorporates clinically meaningful variables which are easy-to-measure routinely across centers, including Eastern Cooperative Oncology Group (ECOG), Child-Pugh, alpha-fetoprotein (AFP), tumor burden, microvascular invasion, extrahepatic disease and hepatitis B virus (HBV) status, facilitating real-world uptake[25]. Second, based on external validation and comparison with Barcelona Clinic Liver Cancer staging strengthen, the study claims that a multivariable model adds prognostic value beyond an anatomic staging system[26-32]. Third, the authors’ attention to triple therapy in addition of programmed death-1 (PD-1) inhibitors and its classification as protective have potential to capture the contemporary evolution of HCC therapy as well as providing an immediately actionable insight for clinicians considering escalation[33-35].

Nevertheless, several critical considerations need to be proposed for next steps to fully realize the model’s potential and aid clinicians and trialists in adopting it responsibly. Heterogeneity of source data and generalizability should be considered firstly. The present meta-analysis draws primarily from studies conducted in China and HBVendemic populations that HBV status emerged as a prognostic variable, but its effect size and interaction with other variables, such as AFP and immune contexture might differ in hepatitis C virus (HCV)-predominant or non-alcoholic steatohepatitis (NASH)-predominant cohorts typical of Western registries[36-38]. Therefore, it’s essential to conduct prospective validation in geographically and etiologically diverse cohorts before broad application by including studies likely used heterogeneous TACE protocols with conventional vs drugeluting beads, variable lenvatinib dosing, and differing criteria for adding PD-1 inhibitors[39,40]. These treatment heterogeneities could confound prognostic associations, which need stratified analyses or metaregression on treatment specifics for quantifying these effects. Otherwise, the authors acknowledge publication bias for several predictors including ECOG, triple therapy, extrahepatic metastases, and HBV. Although sensitivity analyses reportedly maintained pooled estimates after correction, retrospective meta-analyses remain vulnerable to selective reporting and unmeasured confounding[41-43]. To help readers assess robustness, Yu et al[1] should consider to add a transparent list of included studies and forest plots for each predictor accompanying with study-level covariates like study design, sample size, and baseline liver function. Time-dependent biases from immortal time and leadtime can affect overall survival and progression-free survival estimates in observational datasets[44-47], especially when defining exposures such as triple therapy, which need to be optimized by clearer handling of temporality, for example, landmark analyses or timedependent covariates in future model derivation. Overall, a summary table with critical appraisal and future directions of Yu et al’s study[1] for HCC prognostication has been applied with the proposed roadmap for the field (Table 1).

Table 1 Appraisal of the prognostic model and priorities for future validation.

Domain	Strengths of current model	Limitations and future priorities
Evidence synthesis	Large-scale integration of 43 group studies encompassing 5070 patients	Underlying evidence base remains predominantly observational
Model variables	Successfully integrates nine readily available clinical parameters (e.g., tumor number, microvascular invasion, Child-Pugh stage, BCLC stage)	Lacks integration of dynamic molecular data and advanced imaging biomarkers
Population diversity	Validated externally with real-world clinical data	Skewed toward HBV-endemic regions; requires testing across geographically diverse cohorts and other etiologies (HCV, NASH)
Clinical utility	Demonstrates favorable discrimination relative to standard BCLC staging	Requires well-designed interventional studies to definitively test model-guided therapy outcomes

BCLC: Barcelona Clinic Liver Cancer Group; HBV: Hepatitis B virus; HCV: Hepatitis C virus; NASH: Non-alcoholic steatohepatitis.

Generalizability and etiologic heterogeneity

A primary limitation of the current meta-analytic derivation is that most source studies originate from HBV endemic regions (predominantly China). Because HBV associated HCC differs biologically from HCV- or NASH associated HCC, the prognostic effect size of HBV status and its interactions with markers such as AFP and immune contexture may not translate to Western or NASH predominant cohorts[48-50]. We therefore advocate prospective external validation across geographically and etiologically diverse cohorts (HBV, HCV, NASH), and preplanned subgroup analyses by etiology. Only after such prospective validations should widespread clinical adoption be considered. In particular, testing the model’s discrimination and calibration in cohorts with differing background liver disease, antiviral treatment prevalence, and PD-1 inhibitor use is essential.

Transparent reporting of included studies and heterogeneity assessment

To help readers judge robustness, we recommend that authors provide a transparent list of included studies (study ID, country, years, design, sample size, baseline liver function distribution, TACE protocol - conventional vs drugeluting bead - lenvatinib dosing, and criteria for adding PD-1 inhibitors)[51,52]. Accompanying forest plots for each pooled predictor and study-level covariates would permit visual inspection of heterogeneity and potential smallstudy effects. Metaregression stratified by key treatment variables (TACE type, lenvatinib dose, concurrent PD-1 use) should be reported where feasible to quantify treatmentlevel confounding.

Handling of time-dependent exposures

Observational datasets are susceptible to immortal time and lead-time biases, particularly when exposures such as triple therapy are defined post-baseline. Future derivations should explicitly handle temporality using methods such as landmark analyses, time-dependent Cox models (treating triple therapy as a time-varying covariate), or emulation of target trials to avoid biased survival estimates[44,53-55]. Authors should report the timing/distribution of therapy initiation relative to index date and sensitivity analyses using alternative exposure definitions.

Model building and overfitting safeguards

The nine-variable model balances accessibility and complexity, but additional modelling rigor would strengthen confidence. We recommend: (1) Reporting the exact variable selection strategy (stepwise vs penalized regression); (2) Use of penalized methods (Least Absolute Shrinkage and Selection Operator, elastic net) and prespecified shrinkage or bootstrap internal validation to reduce overfitting; (3) Exploration of nonlinear effects for continuous predictors (e.g., restricted cubic splines for AFP); and (4) Formal testing of clinically plausible interactions (for example HBV × AFP, HBV × PD-1 use) as these may uncover effect modification relevant to therapeutic decisions.

Beyond discrimination concerning calibration and net benefit

Reporting area under the curves is necessary but insufficient. Calibration plots (observed vs predicted risk across deciles), calibration-in-the-large and calibration slope should be provided to assess absolute risk accuracy. Decision-curve analysis should be used to quantify net clinical benefit across plausible threshold probabilities and to identify actionable cutoffs (e.g., predicted 1-year survival thresholds that might prompt escalation to triple therapy or palliation)[56,57]. These analyses help translate model outputs to decision rules.

Implementation pathway and prospective testing

For clinical uptake, convert the model into userfriendly tools (nomogram, or web calculator) and deploy pilot implementation studies that measure whether modelinformed decisions change management and improve patientcentered outcomes and cost-effectiveness. Ultimately, prospective, modelstratified or biomarkerenriched randomized trials (for example adaptive or biomarkerstratified randomization to test upfront triple therapy vs TACE + lenvatinib) are needed to test whether modelguided treatment assignment improves outcomes.

Open data and reproducibility

To support validation and secondary analyses, we recommend that meta-analytic datasets, study-level extraction tables, and code for pooled analyses be shared in a public repository or as supplementary material, subject to ethical and copyright considerations.

The nine-variable model from the meta-analysis balances clinical accessibility with complexity, but several potential improvements warrant exploration, including interactions, nonlinear effects, and performance overestimation. The biologic interplay between HBV, AFP, and immune checkpoint efficacy is plausible that HBV-driven immune exhaustion might modify PD-1 response, so the testing interaction terms could reveal clinically relevant effect modifiers. The continuous variables like AFP level might have nonlinear associations, so transformation or flexible modeling could improve calibration[58]. The stepwise vs penalized regression and internal bootstrap validation could better guard against overfitting through specifying details of variable selection methodology. Additionally, reporting of area under the curves is informative for discrimination, but calibration about how predicted risk compares to observed outcomes across risk strata and decision-curve analysis are pivotal to determine net benefit and clinically meaningful thresholds for action, for example, what predicted absolute survival probabilities would prompt escalation to triple therapy or suggest palliative management only? Further integration into clinical workflows requires userfriendly tools, for instance, nomogram, web calculator, electronic hospital record integration with proof that model-informed decisions change management and outcomes[59]. As for biological underpinnings and translational extensions, the model’s predictors suggest underlying clues that microvascular invasion and tumor multiplicity reflect invasive phenotype. High AFP and HBV status indicate distinct tumor biology and immune microenvironment, while Child-Pugh and ECOG reflect host tolerance. These implications could be strengthened by discussing mechanistic hypotheses and proposing exploratory biomarkers for augment prognostication, including tumor gene signatures, circulating tumor DNA, and immune profiling[60]. Moreover, some imaging biomarkers like radiomics, perfusion metrics, and response after initial TACE offer dynamic and noninvasive data to refine early prediction and identify responders to combination therapy[61,62]. The prospective studies combining clinical, molecular and imaging data should be prioritized since the authors have correctly noted literature on radiomics score integration.

Yu et al’s study[1] has showed implications about model-based stratification for trials and future clinical practice. The study has found that triple therapy is protective invites hypothesisdriven trials rather than ad hoc addition of PD-1 inhibitors, inducing questions about whether the model could identify subgroups most likely to benefit from upfront triple therapy, for example, high HBV and AFP but preserved Child-Pugh. These finding could be tested by adaptive trial designs or biomarkerstratified randomization. In settings with resource intensity and toxicity risk of lenvatinib plus TACE, a validated prognostic score could rationalize patient selection, but it still requires demonstration for triaging based on the model in improving patient-centered outcomes and cost-effectiveness. Some recommendations for future research and reporting need to be proposed, with prospective and multicenter external validation in non-HBV predominant regions, preplanned subgroup and interaction analyses should be considered, especially for etiologic background like HBV/HCV/NASH, presence of portal vein tumor thrombus, and type of TACE. The decision-support tools with calibration and net-benefit reporting and pilot implementation studies can be developed to assess clinical impact in order to incorporate timevarying exposures and standardized outcome definitions for bias reduction.

CONCLUSION

Yu et al[1] have provided an important step toward individualized prognostication for unresectable HCC treated with TACE plus lenvatinib, whose model has been grounded in a large body of observational evidence and externally validated with clear potential to influence clinical decisions and trial design. Concerning the expectation moving from a promising predictive model to a practice-changing instrument, some approaches are required including prospective validation in diverse populations, careful attention to treatment heterogeneity and temporality, exploration of biologic and imaging biomarkers, and demonstration of clinical utility through implementation studies. For the next steps, the model would be considered as a foundation for hypothesis-driven and biomarker-enriched trials to advance precision therapy in advanced HCC.