Personalized prognosis in unresectable hepatocellular carcinoma: Development and validation of a model for transcatheter arterial chemoembolization plus lenvatinib

Abstract

BACKGROUND

Transcatheter arterial chemoembolization (TACE) combined with lenvatinib is an important modality for the treatment of unresectable hepatocellular carcinoma (HCC). To date, no prognostic analysis exists for clinical predictive models of TACE combined with lenvatinib in treating advanced unresectable HCC. A model was constructed through meta-analysis, and its validation was further enhanced by the collection of external clinical data, thereby providing guidance for clinical practice.

AIM

To identify risk factors for unresectable HCC following TACE plus lenvatinib therapy and to construct a clinical prediction model.

METHODS

We searched PubMed, Web of Science, EMBASE, and Cochrane Library databases for studies on TACE plus lenvatinib for unresectable HCC. Risk factors from the meta-analysis and sensitivity analyses were used to construct a prediction model. The validation set included clinical data from 106 eligible patients at the Affiliated Hospital of North Sichuan Medical College collected by June 1, 2023.

RESULTS

This study included 43 group studies involving 5070 patients. Tumor number, microvascular invasion, Eastern Cooperative Oncology Group performance status, Child-Pugh stage, Barcelona Clinic Liver Cancer stage, extrahepatic metastases, alpha-fetoprotein level, and hepatitis B virus status were risk factors for overall survival and progression-free survival, while triple therapy was a protective factor for both. In the validation set, the overall survival prediction model had area under the curve values of 0.616, 0.643, and 0.706 at 1 year, 2 years, and 3 years, respectively, and the progression-free survival model had area under the curve values of 0.702, 0.696, and 0.670 at the corresponding time points, demonstrating good model performance. Calibration curves, Kaplan-Meier survival analysis, and decision curves further validated the efficacy of the model.

CONCLUSION

Models based on nine variables from 43 group studies predicted the efficacy of TACE plus lenvatinib in unresectable HCC, supporting evidence-based clinical decisions and treatment strategies.

Key Words: Hepatocellular carcinoma; Transcatheter arterial chemoembolization; Lenvatinib; Prediction model; Retrospective study

Core Tip: We summarized risk factors (number of tumors, microvascular invasion, Eastern Cooperative Oncology Group performance status, Child-Pugh stage, etc.) 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, etc., confirmed their good performance, providing guidance for clinical practice.

INTRODUCTION

Hepatocellular carcinoma (HCC) is a serious threat to human health and the third leading cause of cancer-related deaths, with the sixth highest incidence worldwide[1]. Early symptoms of HCC are not obvious, and approximately 70%-80% of patients are in the middle-to-late stages upon diagnosis, beyond the optimal window for surgical resection. Consequently, HCC is associated with a high mortality rate and poor prognosis[2].

Transcatheter arterial chemoembolization (TACE) is an important treatment modality for unresectable HCC, which focuses on killing the tumor by blocking the tumor vasculature and releasing chemotherapeutic agents[3]. Some studies have shown that TACE can shrink tumors, creating an opportunity for subsequent surgery and prolonging patient survival[4]. Although TACE improves the short-term outcomes of patients, prolonged ischemia and hypoxia can lead to the overexpression of vascular endothelial growth factor, causing tumor angiogenesis[5]. Second, after TACE, the tumor rapidly establishes collateral circulation, and the heterogeneity of the tumor cells and drug resistance in later stages lead to tumor recurrence. Therefore, it is difficult to meet the needs of patients with HCC[6]. Consequently, more in-depth studies have been conducted to assess adverse events after TACE, and anti-angiogenic drugs have become a major focus of attention, with tyrosine kinase inhibitors offering new hope for patients with unresectable HCC. Treatment with lenvatinib has yielded substantial improvements in survival time for patients with unresectable HCC since it was approved as first-line treatment for patients with unresectable HCC in 2018[7]. Lenvatinib, a multi-target tyrosine kinase inhibitor, inhibits tumor growth by inhibiting the vascular endothelial growth factor receptor (VEGFR) and fibroblast growth factor receptor, thereby addressing the adverse events associated with prolonged hypoxia and ischemia after TACE treatment[7]. One study has shown that the synergistic effects of TACE and lenvatinib significantly increased the survival time of patients[8].

Although a combination of TACE and lenvatinib has shown promising results, several factors may influence its therapeutic efficacy. A high number of tumors, the presence of microvascular invasion, and extrahepatic metastases are usually indicative of the highly malignant biological behavior of the tumor. These pathological features are significantly associated with increased tumor aggressiveness, leading to shortened overall patient survival and poor clinical outcomes[9-12]. The Eastern Cooperative Oncology Group (ECOG) performance status and Child-Pugh stage reflect patients’ physical and hepatic reserves, which directly affect their tolerance to combination therapy[13]. Alpha-fetoprotein (AFP) is an important tumor marker, and high levels of AFP are associated with poor prognosis[14]. Hepatitis B virus (HBV) infection is a major cause of unresectable HCC and can influence treatment outcomes by affecting the immune status of the body and tumor microenvironment[15]. The existing Barcelona Clinic Liver Cancer (BCLC) staging system has limitations in evaluating the performance of combination therapy, focusing excessively on anatomical staging elements (e.g., number of tumors and microvascular invasion) and failing to adequately consider the dynamic role of host immune status (e.g., Child-Pugh stage) and virological characteristics (e.g., HBV DNA load)[16]. Microvascular invasion is not only associated with the development of satellite foci after TACE[17], but also impairs the antiangiogenic efficacy of lenvatinib through the VEGFR pathway, particularly in patients with AFP levels ≥ 400 ng/mL, thereby reducing disease control by 38%[18]. Therefore, not all patients benefit from TACE combined with lenvatinib.

However, existing prognostic prediction models for the combination of TACE and lenvatinib have many limitations, such as low coverage of risk factors, small sample sizes, and lack of external validation. In this study, we conducted a meta-analysis to integrate data from 43 group studies and analyzed key prognostic factors with the aim of constructing and validating prognostic risk prediction models for patients with unresectable HCC treated with TACE and lenvatinib, thereby providing an evidence-based foundation for the development of individualized treatment plans and optimization of patient survival management strategies.

MATERIALS AND METHODS

Meta-analysis

Search strategy: This study was conducted in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses guidelines[19]. As of June 1, 2025, we searched PubMed, Web of Science, EMBASE, and Cochrane Library databases for studies consistent with the topic of this article. The main search terms were “transcatheter arterial chemoembolization”, “TACE”, and “lenvatinib”. The search was performed without language restrictions. The detailed search process is presented in Supplementary Table 1.

Inclusion criteria: (1) Imaging or pathological diagnosis of HCC; (2) Treatment with a combination of TACE and lenvatinib therapy with or without programmed death-1 (PD-1) inhibitors; (3) Availability of data for outcomes such as overall survival (OS) or progression-free survival (PFS); and (4) Evaluation of at least one prognostically relevant risk factor or protective factor.

Exclusion criteria: (1) Inclusion of other malignancies in addition to HCC; (2) Inclusion of patients who had undergone prior treatment; (3) Case reports, letters, and reviews; and (4) Incomplete information that could have affected the subsequent analysis.

Data extraction: The data were first extracted separately by two independent authors and then organized in the form of a table. The final data for analyses were obtained by comparing the final data recorded by two independent authors; in case of disagreement, data extraction was performed by a third author. The extracted baseline data included the name of the first author, year of publication, country, study method, mean age, sex, sample size, and associated risk factors.

Quality assessment

The Newcastle-Ottawa Scale was used for quality assessment. This scale evaluates three dimensions: Selection of study participants, comparability between groups, and measurement of outcome indicators, with eight items evaluated using a point system. The highest score for comparability was 2, and the highest score for the remaining items was 1. The total score was 9, with higher scores indicating better study quality[17]. Two researchers independently performed this quality assessment; studies with scores between 5 and 9 were considered high quality, whereas those with scores below 5 were classified as low quality.

Model development and validation

Construction of the prediction model: First, we screened for statistically significant influencing factors using meta-analysis. The robustness of the influencing factors was then assessed using sensitivity analysis, and the influencing factors that were statistically significant but showed poor robustness were excluded. Influencing factors that were statistically significant and showed stable results in the sensitivity analysis were used for modeling. When constructing predictive models for ordered categorical variables (e.g., the number of tumors), different studies may use different critical values (cutoff values) for grouping, resulting in data heterogeneity. To address this issue, we adopted the following rule: When a variable had two cut-off value options, the most frequently used cut-off value was selected as the criterion. The median was selected as the critical value for the model when a variable had three or more critical value options. The hazard ratio (HR) values of all influencing factors were extracted, and the β values of each influencing factor were calculated by obtaining the combined HR values and 95% confidence interval (CI) according to the formula β = ln(HR), which was followed by multiplying the β values by 10 and rounding to one decimal place to derive the scores for each influencing factor according to the method reported by Jiang et al[20]. A risk scorecard as a prognostic prediction model of TACE combined with lenvatinib was established for patients with unresectable HCC. The total score was the sum of the scores for each influencing factor. Patient prognosis was predicted based on the total score. To construct the model, risk factors that were statistically significant and showed stable results in the sensitivity analysis but whose calculated final score was 0 were excluded.

Validation of the prediction model: A total of 106 patients who underwent TACE combined with lenvatinib for unresectable HCC at the Affiliated Hospital of North Sichuan Medical College between February 2019 and June 2025 were retrospectively enrolled as the validation set. The validation set-specific process is illustrated in Figure 1. The inclusion criteria were as follows: (1) Diagnosis of HCC by imaging or pathology; (2) No previous surgery or other treatment; (3) No other malignant tumor except HCC; (4) Previous lenvatinib use for at least two months; and (5) At least one TACE treatment during the follow-up period. The exclusion criteria were as follows: (1) Incomplete clinical data; or (2) Loss to follow-up. Factors affecting the prognosis of TACE combined with lenvatinib for unresectable HCC were collected, including the number of tumors, microvascular invasion, patients’ physical condition (ECOG performance status), hepatic reserve function (Child-Pugh stage), clinical stage of HCC (BCLC stage), presence or absence of extrahepatic metastases, AFP level, HBV-positivity status, and patients’ choice of treatment. This study was approved by the Medical Ethics Committee of the Affiliated Hospital of the North Sichuan Medical College (Approval No. 2025ER258-1).

Figure 1 Validation cohort patient screening flowchart. HCC: Hepatocellular carcinoma; TACE: Transcatheter arterial chemoembolization.

Statistical analysis

Heterogeneity was analyzed using the Q-test combined with the I² test, with I² < 50% or P > 0.05 serving as the criterion for low heterogeneity. During data analysis, the results were uniformly interpreted using a random-effects model, regardless of the level of heterogeneity. Publication bias was detected using the Egger’s and Begg’s tests. To assess the robustness of the results, sensitivity analyses were performed by implementing a method of excluding individual studies. Survival outcomes are reported in the form of HRs and 95%CIs.

To validate the model, the receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA) were used to assess prediction performance. Additionally, we divided the patients into low-risk, moderate-risk, and high-risk groups based on their total scores according to the interquartile spacing method. Kaplan-Meier survival curves were generated for each group to further assess the model performance.

All included studies were analyzed using the R software (version 4.5.0; University of Akron, New Zealand), Stata/MP (version 17.0; STATA Corp, TX, United States), and GraphPad Prism (9.5.1; GraphPad Software, CA, United States). P < 0.05 was considered statistically significant.

RESULTS

Meta-analysis

Study selection and study characteristics: A total of 1810 papers related to the topic of this study were obtained from the databases, of which 1109 papers were excluded as duplicates. We then screened pathology reports, letters, and reviews, thereby excluding 344 studies. After reading titles and abstracts, 296 studies were excluded. Finally, full-text intensive reading was performed to exclude 18 papers related to the outcome indicators and those that did not meet the inclusion criteria. Finally, 43 studies were included. See Figure 2 for the details of the search process.

Figure 2 Literature search and screening process for prognostic risk factors associated with receiving transcatheter arterial chemoembolization in combination with lenvatinib for patients with unresectable hepatocellular carcinoma. AFP: Alpha-fetoprotein; ECOG: Eastern Cooperative Oncology Group; HBV: Hepatitis B virus; BCLC: Barcelona Clinic Liver Cancer.

A total of 43 studies with a total sample size of 5070 were included, of which four were prospective studies and 39 were retrospective studies. Among these, 31 studies reported the mean or median age of patients, which ranged from 45 to 75 years. All the studies reported both male and female sample sizes, and the male sample sizes were larger than the female sample sizes. The baseline data for all included studies are detailed in Supplementary Table 2.

Quality assessment: The quality of the included studies was assessed using Newcastle-Ottawa Scale. Twelve studies included propensity score matching analyses controlling for confounders and scored nine points. Thirteen studies scored 7 due to insufficient follow-up. The average score for the remaining studies was 8. All studies scored greater than 5, which is the threshold for high quality (Supplementary Table 3).

Meta-analysis results: In this study, 17 influencing factors were summarized, and the results summarized in a forest plot showed that the influencing factors associated with OS outcome were the number of tumors, tumor size, microvascular invasion, ECOG grade, Child-Pugh stage, triple therapy, BCLC stage, extrahepatic metastases, AFP level, HBV positivity, age, and aspartate aminotransferase (AST) level. The final combined HRs and heterogeneity were as follows: Number of tumors (HRs = 1.22, heterogeneity I² = 32.9%; P = 0.069); tumor size (HRs = 1.07, heterogeneity I² = 53.7%; P = 0.107); microvascular invasion (HRs = 1.61, heterogeneity I² = 25.8%; P = 0.069); ECOG grade (HRs = 1.30, heterogeneity I² = 0.0%; P = 0.768); Child-Pugh stage (HRs = 1.48, heterogeneity I² = 27.7%; P = 0.083); triple therapy (HRs = 0.41, heterogeneity I² = 24.3%; P = 0.113); BCLC stage (HRs = 1.77, heterogeneity I² = 6.2%; P = 0.379); extrahepatic metastases (HRs = 1.40, heterogeneity I² = 1.4%; P = 0.443); AFP level (HRs = 1.55, heterogeneity I² = 0.0%; P = 0.501); HBV positivity (HRs = 1.36, heterogeneity I² = 0.0%; P = 0.974); age (HRs = 0.99, heterogeneity I² = 0.0%; P = 0.906); and AST level (HRs = 1.01, heterogeneity I² = 0.0%; P = 0.451). The influencing factors associated with PFS outcomes included the number of tumors, tumor size, microvascular invasion, ECOG grade, Child-Pugh stage, triple therapy, BCLC stage, extrahepatic metastases, AFP level, and HBV positivity. The final combined HRs = and heterogeneity were as follows: Number of tumors (HRs = 1.23, heterogeneity I² = 33.9%; P = 0.070); tumor size (HRs = 1.06, heterogeneity I² = 42.5%; P = 0.033); microvascular invasion (HRs = 1.38, heterogeneity I² = 48.0%; P = 0.007); ECOG grade (HRs = 1.27, heterogeneity I² = 0.0%; P = 0.465); Child-Pugh stage (HRs = 1.45, heterogeneity I² = 40.5%; P = 0.015); triple therapy (HRs = 0.45, heterogeneity I² = 55.7%; P = 0.000); BCLC stage (HRs = 1.36, heterogeneity I² = 31.0%; P = 0.109); extrahepatic metastases (HRs = 1.26, heterogeneity I² = 33.2%; P = 0.063); AFP level (HRs = 1.28, heterogeneity I² = 15.6%; P = 0.235); and HBV positivity (HRs = 1.18, heterogeneity I² = 0.0%; P = 0.996). The results of the meta-analysis are detailed in Supplementary Table 4, Supplementary Figures 1-12.

The results of the sensitivity analysis suggested that tumor size was unstable for both OS and PFS; therefore, tumor size was not used as a relevant influencing factor for OS and PFS during modeling (Supplementary Figure 2). The results of the publication bias analysis showed that for the outcome metric OS, the ECOG grade (Egger’s test P = 0.037; Begg’s test P = 0.020), triple therapy (Egger’s test P = 0.001; Begg’s test P = 0.016), and extrahepatic metastases (Egger’s test P = 0.007; Begg’s test P = 0.011) showed publication biases. For the outcome indicator PFS, the ECOG grade (Egger’s test P = 0.029; Begg’s test P = 0.011), extrahepatic metastases (Egger’s test P = 0.029; Begg’s test P = 0.011), and HBV status (Egger’s test P = 0.011; Begg’s test P = 0.004) showed publication biases. However, after correction for publication bias, the pooled estimates for ECOG grade, triple therapy, extrahepatic metastases, and HBV status remained consistent with the original findings. This suggests that the influence of publication bias was minimal and unlikely to compromise the robustness of our results (Supplementary Table 5).

Development of the prediction models for OS and PFS

On the basis of the results of the meta-analysis and the sensitivity analysis, 11 influencing factors were used to construct the OS model: Number of tumors > 1 (HR = 1.22, 95%CI: 1.06-1.40; β-coefficient = 0.199), microvascular invasion (HR = 1.61, 95%CI: 1.41-1.85; β-coefficient = 0.476), ECOG grade > 0 (HR = 1.30, 95%CI: 1.12-1.52; β-coefficient = 0.262), Child-Pugh stage > A (HR = 1.48, 95%CI: 1.26-1.74; β-coefficient = 0.392), triple therapy (HR = 0.41, 95%CI: 0.36-0.46; β-coefficient = -0.892), BCLC stage > B (HR = 1.77, 95%CI: 1.51-2.08; β-coefficient = 0.571), extrahepatic metastases (HR = 1.40, 95%CI: 1.23-1.60; β-coefficient = 0.336), BCLC stage > B (HR = 0.41, 95%CI: 0.36-0.46; β-coefficient = -0.892), AFP level ≥ 400 ng/mL (HR = 1.55, 95%CI: 1.40-1.72; β-coefficient = 0.438), HBV positivity (HR = 1.36, 95%CI: 1.17-1.58; β-coefficient = 0.307), age ≥ 60 years (HR = 0.99, 95%CI: 0.98-1.00; β-coefficient = -0.010), and AST level ≥ 35 IU/L (HR = 1.01, 95%CI: 1.00-1.01; β-coefficient = 0.010). The score for each influencing factor was derived by multiplying the β-value by 10 and rounding to one decimal place according to the method reported by Jiang et al[20], and the final scores for AST level ≥ 35 IU/L as well as for age ≥ 60 years were 0. Therefore, these two factors were not considered in the model construction. Finally, the remaining nine influencing factors were included to construct the OS model. The specific model constructions and corresponding scores are listed in Table 1.

Table 1 The β-coefficient and score for overall survival prediction model.

Risk factors	Pooled HR	95%CI	β-coefficient	Score
Number of tumors
Single	Reference1			0
Multiple	1.22	1.06-1.40	0.199	2
Microvascular invasion
No	Reference1			0
Yes	1.61	1.41-1.85	0.476	5
ECOG performance status
≤ 0	Reference1			0
> 0	1.30	1.12-1.52	0.262	2.5
Child-Pugh stage
≤ A	Reference1			0
> A	1.48	1.26-1.74	0.392	4
Triple therapy
TACE + LEN	Reference1			0
TACE + LEN + PD-1	0.41	0.36-0.46	-0.892	-9
BCLC stage
≤ B	Reference1			0
> B	1.77	1.51-2.08	0.571	6
Extrahepatic metastases
No	Reference1			0
Yes	1.40	1.23-1.60	0.336	3.5
AFP
< 400 ng/mL	Reference1			0
≥ 400 ng/mL	1.55	1.40-1.72	0.438	4.5
HBV
Negative	Reference1			0
Positive	1.36	1.17-1.58	0.307	3

¹Reference group.

ECOG: Eastern Cooperative Oncology Group; TACE: Transcatheter arterial chemoembolization; LEN: Lenvatinib; PD-1: Programmed death-1; BCLC: Barcelona Clinic Liver Cancer; AFP: Alpha-fetoprotein; HBV: Hepatitis B virus; HR: Hazard ratio; CI: Confidence interval.

On the basis of the results of the meta-analysis and sensitivity analysis, a total of nine influencing factors were used to construct the PFS model: Number of tumors > 1 (HR = 1.23, 95%CI: 1.07-1.40; β-coefficient = 0.207), microvascular invasion (HR = 1.38, 95%CI: 1.18-1.62; β-coefficient = 0.322), ECOG grade > 0 (HR = 1.27, 95%CI: 1.14-1.42; β-coefficient = 0.239), Child-Pugh stage > A (HR = 1.45, 95%CI: 1.23-1.70; β-coefficient = 0.372), triple therapy (HR = 0.45, 95%CI: 0.39-0.53; β-coefficient = -0.799), BCLC stage > B (HR = 1.36, 95%CI: 1.15-1.60; β-coefficient = 0.307), extrahepatic metastases (HR = 1.26, 95%, CI: 1.07-1.50; β-coefficient = 0.231), BCLC stage > B (HR = 0.45, 95%CI: 0.39-0.53; β-coefficient = -0.799), AFP level ≥ 400 ng/mL (HR = 1.28, 95%CI: 1.16-1.41; β-coefficient = 0.247), and HBV positivity (HR = 1.18, 95%CI: 1.02-1.36; β-coefficient = 0.166). The specific model constructions and corresponding scores are listed in Table 2.

Table 2 The β-coefficient and score for progression-free survival prediction model.

Risk factors	Pooled HR	95%CI	β-coefficient	Score
Number of tumors
Single	Reference1			0
Multiple	1.23	1.07-1.40	0.207	2
Microvascular invasion
No	Reference1			0
Yes	1.38	1.18-1.62	0.322	3
ECOG performance status
≤ 0	Reference1			0
> 0	1.27	1.14-1.42	0.239	2.5
Child-Pugh stage
≤ A	Reference1			0
> A	1.45	1.23-1.70	0.372	4
Triple therapy
TACE + LEN	Reference1			0
TACE + LEN + PD-1	0.45	0.39-0.53	-0.799	-8
BCLC stage
≤ B	Reference1			0
> B	1.36	1.15-1.60	0.307	3
Extrahepatic metastases
No	Reference1			0
Yes	1.26	1.07-1.50	0.231	2.5
AFP
< 400 ng/mL	Reference1			0
≥ 400 ng/mL	1.28	1.16-1.41	0.247	2.5
HBV
Negative	Reference1			0
Positive	1.18	1.02-1.36	0.166	1.5

¹Reference group.

ECOG: Eastern Cooperative Oncology Group; TACE: Transcatheter arterial chemoembolization; LEN: Lenvatinib; PD-1: Programmed death-1; BCLC: Barcelona Clinic Liver Cancer; AFP: Alpha-fetoprotein; HBV: Hepatitis B virus; HR: Hazard ratio; CI: Confidence interval.

Baseline characteristics of the validation group

The validation set consisted of 106 patients with HCC from the Affiliated Hospital of North Sichuan Medical College who met the inclusion criteria (mean age: 55.9 ± 12.5 years). Among them, 92 were male and 14 were female. This cohort included 49 patients with AFP level ≥ 400 ng/mL, 89 HBV-positive patients, 81 patients with > 1 tumors, 61 patients showing microvascular invasion, 90 patients with ECOG grade > 0, 39 patients with Child-Pugh stage > A, 34 patients who underwent TACE + lenvatinib + PD-1 inhibitor triple therapy, 72 patients with BCLC stage > B, and 34 patients with extrahepatic metastases. The median follow-up period was 13 months. Specific baseline data for the validation cohort are presented in Supplementary Table 6.

Validation of the prediction models for OS and PFS in the validation group

In the validation cohort, the areas under the curve (AUC) at 1 year, 2 years, and 3 years using the predictive model for OS were 0.616, 0.643, and 0.706, respectively (Figure 3A). In the PFS prediction model, the AUCs at 1 year, 2 years, and 3 years were 0.702, 0.696, and 0.670, respectively (Figure 3B). The results show that both models have good predictive abilities. We generated the ROC curve of the traditional HCC prognostic model based on BCLC staging in order to elucidate the performance of the new model. AUCs for OS were 0.486, 0.497, and 0.585, respectively (Supplementary Figure 13A), and those for PFS were 0.594, 0.677, and 0.646, respectively, at 1 year, 2 years, and 3 years (Supplementary Figure 13B). For the OS-related 1 year, 2 years, and 3 years, the corresponding AUC differences were 0.130, 0.146, and 0.121, respectively. For PFS at 1 year, 2 years, and 3 years, the corresponding AUC differences were 0.108, 0.019, and 0.024, respectively. The revised model’s clinical incremental value was demonstrated by a final AUC difference > 0. In addition, the calibration curve results showed a good agreement between the observed and predicted values for both models (Figure 4). The influencing factors were categorized into low-risk, moderate-risk, and high-risk groups using the interquartile spacing method. The Kaplan-Meier survival curves showed significant differences in OS and PFS among patients in different risk strata, further validating the models (Figure 5). The DCA curves showed that both models provided a net benefit to patients (Figure 6).

Figure 3 Receiver operating characteristic curves for prediction models in the validation cohort. A: Overall survival; B: Progression-free survival. ROC: Receiver operating characteristic; AUC: Areas under the curve.

Figure 4 Prediction models for in the validation cohort, with 1-year, 2-year, and 3-year calibration curves. A: Overall survival; B: Progression-free survival.

Figure 5 Kaplan-Meier survival curves for prediction models for the three risk groups in the validation group. A: Overall survival; B: Progression-free survival.

Figure 6 Prediction models for in the validation cohort, with 1-year, 2-year, and 3-year decision curves. A: Overall survival; B: Progression-free survival.

DISCUSSION

In this study, we aimed to develop and validate a multivariate-based prognostic prediction model for the combination of TACE and lenvatinib in patients with unresectable HCC. First, we conducted a systematic review and meta-analysis of 43 studies that met the inclusion criteria and summarized the risk factors associated with both OS and PFS: Number of tumors, microvascular invasion, ECOG grade, Child-Pugh stage, BCLC stage, extrahepatic metastases, AFP level, HBV status, and triple therapy. OS and PFS were modeled by combining the HR values and calculating the corresponding scores. External validation was performed by collecting data from 106 unresectable HCC patients who met the inclusion criteria. Our results showed that both models had good predictive abilities, and the calibration and Kaplan-Meier survival curves further validated these models. The DCA curves indicated that both models provided a net benefit to the patients.

Our findings showed that the presence of more than one tumor was a risk factor for OS and PFS. Xia et al[21] retrospectively investigated the efficacy of TACE in combination with lenvatinib in unresectable HCC by collecting data from 211 patients who met the inclusion criteria for Cox regression analysis and consistently showed that the presence of more than one tumor was an independent risk factor for OS and PFS. However, our results showed that tumor size, which also represents tumor load, was not a risk factor. This may be related to the following factors. First, a high number of tumors often suggests that the tumor originates from multiple centers and may be combined with intrahepatic metastases, which are highly invasive. One study showed that multiple tumors are more likely to invade the portal and hepatic veins, further increasing the risk of vascular invasion and distant metastasis[22]. In addition, a study by Xie et al[23] showed that extrahepatic metastases were closely associated with the number of tumors but not significantly associated with tumor size. In the BCLC staging system for HCC, the number of tumors plays a major role in the middle and late stages (BCLC stages B/C), whereas tumor size dominates only in the early stages (BCLC stage 0/A)[1]. Therefore, the number of tumors is a more direct reflection of the biological behavior of tumors during combination therapy. Second, TACE is a local treatment that embolizes the blood-supplying arteries of tumor vessels, and lenvatinib acts as an anti-angiogenic agent by inhibiting VEGFR[14,22]. Complete embolization of all tumors using TACE may be difficult in patients with multiple tumors. However, for large tumors, although TACE may yield residual lesions owing to the generation of collateral circulation or incomplete embolization, the antiangiogenic effect of lenvatinib compensates for this deficiency[7]. Therefore, when TACE is combined with lenvatinib for the treatment of advanced unresectable HCC, the number of tumors becomes more important, and the effect of tumor size may be masked by the anti-angiogenic effects of lenvatinib.

Our findings show that HBV infection is an independent risk factor that significantly affects the prognosis of patients with advanced HCC treated with TACE in combination with lenvatinib. Our results are consistent with those of previous studies[24,25]. This relationship is primarily mediated by the following factors: First, HBV-associated HCC can inhibit the regenerative ability of hepatocytes by activating the caspase apoptosis pathway[26]. The liver requires compensatory regeneration after TACE; thus, impaired regenerative ability impedes recovery and affects patient prognosis. Second, HBV infection activates the nuclear factor kappa-light-chain-enhancer of activated B cells pathway to promote the release of inflammatory factors, creating an inflammatory environment that accelerates the progression of HCC[27]. A relevant study has shown that HBV infection can stimulate the expression of collagen triple helix repeat containing 1 by activating the nuclear factor kappa-light-chain-enhancer of activated B cells pathway, and high expression of collagen triple helix repeat containing 1 regulates the phosphatidylinositol 3-kinase/protein kinase B/mammalian target of rapamycin pathway, which induces the expression of vascular endothelial growth factor and enhances tumor angiogenesis, thus affecting the efficacy after TACE[28]. Third, the G1896A mutation associated with HBV infection can activate the extracellular signal-related kinase/mitogen-activated protein kinase signaling pathway and enhance the resistance of HCC cells to targeted drugs, such as lenvatinib and sorafenib[29], thereby affecting the efficacy and prognosis of patients with unresectable HCC. Another issue worth considering is the effect of the ongoing HBV antiviral therapy on the efficacy of TACE combined with lenvatinib in patients with HBV-positive advanced HCC. Additional prospective studies that dynamically monitor patient indicators of antiviral therapy efficacy (e.g., HBV DNA levels) are warranted to explore the relationship between viral load and efficacy in patients with HCC.

ECOG performance status is an important indicator of the physical condition of patients and can reflect the tolerance of patients with HCC to a combination of TACE and lenvatinib treatment. Patients with high scores were less tolerant to treatment and had relatively more post-treatment complications, which affected their prognosis[30]. The results of the previous studies are consistent with ours[31]. In patients with HCC, Child-Pugh classification is an important indicator of hepatic reserve, which can severely affect drug metabolism and the patient’s physical status when it is insufficient[32]. It is metabolized by hepatic cytochrome P450 enzymes (e.g., cytochrome P450 3A4). A high Child-Pugh stage indicates reduced liver metabolism and decreased activity of hepatic metabolizing enzymes. The resultant accumulation of lenvatinib in the body reduces the binding of the drug to the target site and the synergistic effects of TACE and lenvatinib, thereby affecting patient[33]. Extrahepatic metastasis, which usually occurs in the lungs, bones, and lymphatic sites, is an important marker of advanced-stage HCC. Our results showed that extrahepatic metastasis was an independent risk factor in patients with advanced HCC treated with TACE in combination with lenvatinib. This can be attributed to several factors. First, TACE can only locally embolize the tumor through intrahepatic blood vessels and has no effect on extrahepatic metastases. A study comparing the efficacy of TACE and placebo in treating patients with extrahepatic metastases showed no benefit in the TACE treatment group[34]. Second, the anti-angiogenic effect of lenvatinib is achieved through the anti-VEGFR pathway, whereas in extrahepatic metastases, tumors can grow through other pathways (e.g., the platelet-derived growth factor receptor pathway), resulting in poor systemic therapeutic efficacy of lenvatinib[35,36]. AFP is a serological tumor marker for HCC, and AFP levels ≥ 400 ng/mL usually indicate a more aggressive tumor. AFP may influence patient prognosis by inhibiting antitumor immunity through the suppression of natural killer cell activity and antagonizing the local immune activation effect of TACE[37]. Our findings suggest that microvascular invasion is an independent risk factor of advanced HCC treated with TACE and lenvatinib. According to both the National Comprehensive Cancer Network and BCLC guidelines, TACE is not recommended for patients with HCC and microvascular invasion because when the hepatic blood vessels themselves are embolized, treatment with TACE exacerbates the embolism of the hepatic blood vessels, resulting in a drastic decrease in liver function due to severe ischemia and poor prognosis. In contrast, we found relevant studies showing that TACE is safe for patients with HCC and portal vein cancer embolism[38]. A prospective study by Ding et al[39] demonstrated that the combination of TACE and lenvatinib was effective in treating patients with HCC and microvascular invasion. Indeed, this discrepancy is mainly due to the fact that the treatment process is influenced by a number of factors, including the degree of spongiosis and the locations of the portal vein thrombus and the tumor.

The BCLC staging system is commonly used for the diagnosis of HCC. In this system, patients with HCC are accurately localized to provide effective therapeutic options for clinical management[16,40]. The results of this study showed that BCLC stage > B was an independent risk factor for the combination of TACE with lenvatinib in the treatment of advanced HCC. In patients with BCLC stage > B, TACE treatment showed difficulty in completely removing the tumor owing to the complex rows of the tumor. Moreover, even when TACE was combined with lenvatinib, the combined therapeutic effect was limited despite the anti-angiogenic effect of lenvatinib. Therefore, the addition of PD-1 inhibitors to the combination of TACE and lenvatinib has become a focus of interest. Our results showed that triple therapy with TACE combined with lenvatinib and a PD-1 inhibitor was protective in patients with advanced HCC. A previous meta-analysis compared the efficacy of TACE combined with lenvatinib and a PD-1 inhibitor in patients with advanced HCC, and its findings showed that triple therapy significantly prolonged the OS and PFS of patients with HCC compared to non-triple therapy[41]. In addition, a related study compared the efficacy of TACE combined with lenvatinib with or without PD-1 administration in patients with advanced HCC, and the results showed that the addition of PD-1 inhibitors significantly prolonged the OS and PFS of patients compared to those without PD-1 inhibitors[42]. All the above findings further validate the findings of this study.

The prognostic risk prediction model proposed in this study was constructed on the basis of a meta-analysis of 43 group studies that yielded a large sample size with reliable results. In addition, we integrated the risk factors for OS and PFS, thereby comprehensively covering the relevant factors influencing the outcomes of TACE combined with lenvatinib for advanced HCC and accurately predicting patient prognosis. Previous studies have focused more on prognostic risk factor prediction models for a single treatment[43-45], and no prediction model for the combination of TACE and lenvatinib has been proposed to date. Therefore, this study focused on the combination of TACE and lenvatinib, a common and effective combination therapy, to ensure that the results were closer to the clinical reality. Furthermore, we generated a separate ROC curve for the BCLC staging system to compare the incremental value of our model in predicting the survival period of patients with advanced unresectable HCC treated with TACE combined with lenvatinib. The final AUC difference was greater than zero, indicating that the new model that integrated multiple variables had a significant advantage over the traditional BCLC staging model. This model serves as an important risk assessment tool, allowing clinicians to provide patients with individualized prognostic assessments and precise treatment strategies.

However, this study has some limitations. First, the meta-analysis included 40 retrospective and three prospective studies. Although 12 of the 40 retrospective studies conducted propensity score matching analyses to reduce selection bias, they could not eliminate the effect of selection bias on the results. A larger number of prospective studies are needed to verify the predictive efficacy of the model. Second, as all the studies were from Asia, the generalizability and representativeness of the results are limited, and more future studies including non-Asian populations may be required for further validation. Third, differences in the methodologies and experimental designs of the included studies led to unavoidable heterogeneity, which could be reduced by further sensitivity and subgroup analyses. Fourth, some of the findings indicated publication bias, and we consider that this was mainly due to an insufficient sample size, which needs to be expanded for further validation in the future. Fifth, all patient information gathered from the validation cohort was obtained from a single facility. The universality of the findings was unavoidable despite rigorous screening of the patients in compliance with the inclusion and exclusion criteria. Because all patients in this study came from the Affiliated Hospital of North Sichuan Medical College, the baseline characteristics of the patients were concentrated in this area, making the model more appropriate for the local or Asian population and having weaker applicability to populations in other regions. Therefore, multiple hospitals in and outside Asia should be included for multi-center verification, unify the inclusion and exclusion criteria, and further verify the stability of the model through subgroup analysis. The AUC for OS in the validation groups were 0.616, 0.643, and 0.706. Although the survival risk of patients can be distinguished to a certain extent, the shortcomings of the AUC index of the model itself should be addressed squarely, such as sensitivity to data category imbalance and the inability to reflect the linear relationship of the risk gradient. To solve these problems, it is necessary to further integrate dynamic biomarkers (such as HBV DNA load and circulating tumor DNA) and use machine learning models (such as random forest and extreme gradient boosting) to integrate multimodal data.

CONCLUSION

This study successfully developed and validated a prognostic risk prediction model for the treatment of unresectable HCC with a combination of TACE and lenvatinib, and clarified the effects of the number of tumors, microvascular invasion, ECOG grade, Child-Pugh stage, triple therapy, BCLC stage, extrahepatic metastasis, AFP level, and HBV status on OS and PFS. This model can serve as a powerful tool for individualized clinical treatment and prognostic assessments. Future studies should expand the sample size, conduct prospective multicenter validations, explore the applicability of the model in different populations, and investigate the interactions and potential mechanisms between risk factors to continuously optimize prediction models and provide more accurate treatment and management strategies for patients with advanced HCC.