Tyrosine kinase inhibitors after progression on atezolizumab-bevacizumab in advanced hepatocellular carcinoma: A systematic review and meta-analysis

Abstract

BACKGROUND

Atezolizumab (ATE) plus bevacizumab (BEV) is the standard first-line therapy for patients with unresectable hepatocellular carcinoma (HCC). However, the optimal second-line treatment after disease progression remains undefined. Tyrosine kinase inhibitors (TKI) established in the pre-immunotherapy era are commonly used, but their efficacy and safety following prior immunotherapy exposure are not well characterized. We conducted a systematic review and meta-analysis to evaluate the efficacy and safety of TKIs in patients with advanced HCC after progression on ATE plus BEV.

AIM

To evaluate efficacy and safety of TKIs after progression on ATE-BEV in advanced HCC.

METHODS

We conducted a systematic review and meta-analysis of studies evaluating TKIs after progression on ATE-BEV. PubMed, Scopus, Web of Science, and Cochrane Central were searched from inception to October 27, 2024. Eligible studies included retrospective or prospective cohorts, case series, and clinical trials reporting survival, radiologic response, or adverse events. Pooled means and proportions were calculated using random-effects models. Risk of bias was assessed using the Methodological Index for Non-Randomized Studies tool.

RESULTS

Nine retrospective studies including 994 patients were analyzed. Pooled mean overall survival was 10.98 months (95% confidence interval: 8.46-13.50), and progression-free survival was 3.41 months (95% confidence interval: 2.92-3.90). Subgroup overall survival was 12.99 months for lenvatinib, 10.56 months for sorafenib, and 8.79 months for cabozantinib. Objective response rate was 9% and disease control rate 66% by Response Evaluation Criteria in Solid Tumors 1.1 criteria, increasing to 28% and 69% by modified Response Evaluation Criteria in Solid Tumors criteria. Common adverse events included fatigue (52%), elevated liver enzymes (42%), hyperbilirubinemia (38%), anorexia (38%), proteinuria (30%), hand-foot syndrome (34%), and hypertension (18%). Grade 3 or higher events were uncommon, including liver enzyme elevation (9%), proteinuria (15%), hyperbilirubinemia (4%), thrombocytopenia (3%), diarrhea (2%), and hypertension (3%).

CONCLUSION

TKIs demonstrate modest efficacy and acceptable safety following progression on ATE-BEV, supporting their role in second-line management of advanced HCC, pending further prospective validation.

Key Words: Hepatocellular carcinoma; Tyrosine kinase inhibitors; Atezolizumab; Bevacizumab; Second-line therapy; Disease progression; Survival outcomes; Adverse events; Meta-analysis

Core Tip: Atezolizumab plus bevacizumab represents the standard first-line therapy for advanced hepatocellular carcinoma; however, treatment options after disease progression remain poorly defined. Tyrosine kinase inhibitors, established prior to the immunotherapy era, are frequently used in this setting, although their effectiveness after prior immunotherapy exposure is not well characterized. Therefore, a comprehensive evaluation of their role is needed. Our meta-analysis assessed the efficacy and safety of tyrosine kinase inhibitors after progression on atezolizumab plus bevacizumab in advanced hepatocellular carcinoma.

INTRODUCTION

Epidemiology and current treatment landscape

Hepatocellular carcinoma (HCC) remains a leading cause of cancer-related mortality worldwide, despite significant advances in treatment. The IMbrave150 trial demonstrated that the combination of atezolizumab (ATE), an immune checkpoint inhibitor targeting programmed death-ligand 1[1], and bevacizumab (BEV), a monoclonal antibody targeting vascular endothelial growth factor (VEGF) to inhibit angiogenesis and tumor progression[2,3], significantly outperformed sorafenib, the prior standard of care alongside lenvatinib. The combination therapy improved overall survival (OS) at 12 months (67.2% vs 54.6%, P < 0.001) and progression-free survival (PFS; 6.8 months vs 4.3 months, P < 0.001) compared with sorafenib[4]. These findings established ATE plus BEV as the standard first-line therapy for unresectable HCC, representing a paradigm shift in systemic treatment and replacing sorafenib and lenvatinib as the preferred initial options for eligible patients.

Clinical gap

Despite these advances, the optimal second-line treatment following progression on ATE-BEV remains undefined. Although several therapeutic strategies have been evaluated, current guidelines provide limited recommendations for patients who do not respond to this first-line regimen[5,6]. Consequently, identifying effective and safe second-line options is of significant clinical importance.

Study aims and clinical implications

The objective of this study is to evaluate the efficacy and safety of tyrosine kinase inhibitors (TKIs) as second-line therapy in patients with advanced HCC who have experienced progression following first-line treatment with ATE and BEV.

MATERIALS AND METHODS

A systematic review and meta-analysis were conducted in accordance with the Cochrane Collaboration Handbook for Systematic Reviews of Interventions and the Preferred Reporting Items for Systematic Reviews and Meta-Analysis Statement[7]. No protocol was prospectively registered for this systematic review.

Search strategy

We conducted a comprehensive literature search on October 27, 2024, using PubMed, Scopus, Web of Science, and the Cochrane Central Register of Controlled Trials databases. A search formula (Supplementary Table 1) was constructed using terms related to “hepatocellular carcinoma”, “atezolizumab/bevacizumab”, and “tyrosine kinase inhibitors”. Reference lists of included studies and prior reviews were manually searched to identify additional studies.

Selection criteria

We included studies (cohort studies, case control, case series, and either randomized or non-randomized trials) that evaluated TKIs as second-line therapies for HCC after progression on ATE-BEV. Outcomes of interest were OS and/or PFS, clinical response as measured by Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 or modified RECIST criteria, and adverse events. Language was restricted to articles written in English.

Data collection

Records were uploaded to https://www.rayyan.ai/, a platform that facilitates the study selection process. Duplicates were removed. Titles and abstracts were screened by two authors (Blanco Montecino R, Guevara Rodriguez N), who subsequently extracted data independently. Disagreements were resolved by a third author (Mendoza-Quispe D), who acted as a supervisor (Figure 1).

Figure 1 Preferred Reporting Items for Systematic Reviews and Meta-Analysis flow diagram of study selection. Including identification, screening, eligibility, and inclusion of studies in the meta-analysis. TKI: Tyrosine kinase inhibitor.

Statistical analysis

The following data was retrieved: (1) Study and sample characteristics: First author, year of publication, study design, sample size, and key demographic variables (age and sex). When reported, additional clinical characteristics were collected, including Eastern Cooperative Oncology Group (ECOG) performance status, viral etiology, Barcelona Clinic Liver Cancer (BCLC) stage, Child-Pugh class, presence of metastases, and vascular invasion status; and (2) Outcomes: For the primary outcome, mean and standard deviation values were extracted when reported. When studies provided medians and interquartile ranges instead of means, these were converted to mean and standard deviation estimates using the Data Estimation and Conversion for Meta-Analysis tool[8]. Pooled OS and PFS estimates were generated using a random-effects model. All statistical analyses were conducted using Stata version 18.

Clinical response was evaluated according to RECIST 1.1 or modified RECIST criteria, whereas adverse events were classified based on the National Cancer Institute’s Common Terminology Criteria for Adverse Events. Adverse events were graded as follows: Grade 1, mild (asymptomatic or mild symptoms, no intervention required); grade 2, moderate (requiring minimal, local, or noninvasive intervention); grade 3, severe or medically significant but not immediately life-threatening (potentially disabling or limiting activities of daily living); grade 4, life-threatening (requiring urgent or emergent intervention); and grade 5, death related to the adverse event.

For the secondary outcomes, clinical response and adverse events, a meta-analysis of proportions was performed to estimate pooled event rates and assess consistency across studies, thereby quantifying the overall prevalence of each outcome within the analyzed population. Heterogeneity was assessed using the I² statistics.

Quality assessment

We evaluated risk of bias using the Methodological Index for Non-randomized Studies (MINORS) tool, a validated instrument designed to assess risk of bias in observational research. MINORS includes 12 items - 8 applicable to non-comparative studies and four additional items for comparative designs - each scored from 0 to 2 (0 = not reported, 1 = reported but inadequate, 2 = reported and adequate), yielding a maximum score of 16 for non-comparative studies. Higher scores indicate better methodological quality (Supplementary Table 2). Two reviewers (Kilani Y and Kamal SAF) independently evaluated each study. After discussing the reasons for the discrepancy, disagreements were resolved through consensus with the help of a third author (Figure 2).

Figure 2 Methodological quality assessment of included studies using the Methodological Index for Non-Randomized Studies tool.

RESULTS

Study selection

Among 1652 records, 33 studies were selected for full-text review. Of these, 24 studies were excluded due to incorrect study design (n = 14), inappropriate patient population (n = 7), or publication in a non-English language (n = 3). Ultimately, nine studies with a total of 994 patients were included (Preferred Reporting Items for Systematic Reviews and Meta-Analysis flow diagram shown in Figure 1).

Description of the included studies

The characteristics of the included studies are summarized in Table 1. All nine studies were retrospective observational cohorts conducted in routine clinical practice and included patients with advanced HCC who received TKIs following progression on ATE-BEV.

Table 1 Baseline characteristics of patients treated with tyrosine kinase inhibitors following atezolizumab-bevacizumab in advanced hepatocellular carcinoma, n (%).

Ref.	Country	Study type	TKI	n	Age, years (mean)	Female	ECOG 0	ECOG 1	ECOG 2	Viral/nonviral3	BCLC B	BCLC C	Child-Pugh A	Child-Pugh B	Extrahepatic metastases	Vascular invasion
Falette-Puisieux et al[21], 2023	France	Retrospective	Regorafenib	29	63	6 (20.7)	1 (3.4)	26 (89.7)	2 (6.9)	12/17 (41.4/58.6)	2 (6.9)	27 (93.1)	27 (93.1)	2 (6.9)	25 (86.2)	8 (27.6)
			Sorafenib	41	62	9 (21.9)	5 (12.2)	30 (73.2)	6 (14.6)	12/29 (29.3/70.7)	2 (4.9)	39 (95.1)	34 (82.9)	7 (17.1)	34 (82.9)	20 (48.8)
			Cabozantinib	4	46	1 (25.0)	0 (0.0)	3 (75.0)	1 (25.0)	2/2 (50.0/50.0)	0 (0.0)	4 (100.0)	4 (100.0)	0 (0.0)	3 (75.0)	1 (25.0)
			Lenvatinib	8	65	1 (12.5)	0 (0.0)	5 (62.5)	3 (37.5)	1/7 (12.5/87.5)	2 (25.0)	6 (75.0)	3 (37.5)	5 (62.5)	4 (50.0)	4 (50.0)
Muto et al[22], 2023	Japan	Retrospective	Lenvatinib	20	70	3 (15.0)	12 (60.0)	8 (40.0)	0 (0.0)	6/14 (30.0/70.0)	6 (30.0)	14 (70.0)	16 (80.0)	4 (20.0)	9 (45.0)	9 (45.0)
Chon et al[23], 2023	South Korea	Retrospective	Lenvatinib	36	61	4 (11.1)	23 (63.9)	13 (36.1)	13 (36.1)	27/9 (75.0/25.0)	4 (11.1)	32 (88.9)	33 (91.7)	3 (8.3)	27 (75.0)	14 (38.9)
			Sorafenib	36	63	5 (13.8)	21 (58.3)	15 (41.7)	15 (41.7)	26/10 (72.2/27.8)	4 (11.1)	32 (88.9)	32 (88.9)	4 (11.1)	29 (80.6)	16 (44.4)
Hiraoka et al[24], 2023	Japan	Retrospective	Lenvatinib	101	72	24 (23.8)	76 (75.2)	19 (18.8)	4 (3.9)	46/55 (45.5/54.5)	35 (34.7)	61 (60.4)	82 (81.2)	18 (17.8)	45 (45.6)	30 (29.7)
Kimura et al[25], 2024	Japan	Retrospective	Lenvatinib	13	73	8 (61.5)	10 (76.9)	3 (23.1)	0 (0.0)	6/7 (46.2/53.8)	NR	NR	NR	NR	NR	NR
Persano et al[9], 2024	Italy	Retrospective	Lenvatinib	86	1	19 (22.1)	85 (98.8)	85 (98.8)	1 (1.2)	43/43 (50.0/50.0)	33 (38.4)	53 (61.6)	85 (98.8)	1 (1.2)	38 (44.2)	NR
			Sorafenib	51	2	7 (13.7)	50 (98.0)	50 (98.0)	1 (2.0)	26/25 (50.99/49.01)	17 (33.3)	34 (66.7)	48 (94.1)	3 (5.9)	19 (37.2)	NR
Kuzuya et al[26], 2023	Japan	Retrospective	Cabozantinib	19	67	4 (21.1)	10 (52.63)	6 (31.57)	3 (15.78)	10/9 (52.6/47.4)	4 (21.1)	15 (78.9)	12 (63.2)	7 (36.8)	11 (57.9)	NR
Yano et al[27], 2023	Japan	Retrospective	Lenvatinib	24	75	4 (16.7)	19 (79.2)	4 (16.66)	1 (4.16)	16/8 (66.7/33.3)	16 (66.7)	0 (0.0)	NR	NR	11 (45.8)	8 (33.3)
Lee et al[28], 2025	South Korea	Retrospective	Sorafenib	324	61	60 (18.5)	124 (38.3)	192 (59.3)	7 (2.2)	244/59 (80.6/19.5)	28 (8.6)	247 (76.2)	235 (72.5)	80 (24.7)	106 (32.7)	90 (27.8)
			Lenvatinib	154	61	26 (16.9)	77 (50.0)	71 (46.1)	6 (3.9)	117/34 (76.0/22.1)	22 (14.3)	125 (81.2)	135 (87.7)	13 (8.4)	11 (7.1)	42 (27.3)
			Regorafenib	36	55	8 (22.2)	15 (41.7)	21 (58.3)	0 (0)	33/1 (91.6/2.7)	1 (2.8)	12 (33.3)	34 (94.5)	2 (5.6)	11 (30.6)	5 (13.9)
			Cabozantinib	12	54	1 (8.3)	2 (16.7)	10 (83.3)	0 (0)	9/1 (75.0/8.3)	1 (8.3)	9 (75.0)	11 (91.6)	1 (8.3)	8 (66.7)	2 (16.7)

¹Authors reported age distribution as follows: 41.9% of patients were < 70 years old and 58.1% were ≥ 70 years old.

²Authors reported age distribution as follows: 56.9% of patients were < 70 years old and 39.2% were ≥ 70 years old.

³Viral etiology was defined as infection with hepatitis B virus and/or hepatitis C virus. Non-viral etiologies included alcohol-associated liver disease, metabolic dysfunction-associated steatotic liver disease, and other chronic liver diseases (including autoimmune hepatitis, primary biliary cholangitis, primary sclerosing cholangitis, Wilson disease, Hereditary hemochromatosis, cryptogenic cirrhosis).

TKI: Tyrosine kinase inhibitor; NR: Not reported; ECOG: Eastern Cooperative Oncology Group; BCLC: Barcelona Clinic Liver Cancer stage.

Treatment distribution

Across cohorts, lenvatinib was administered to 442 of 994 patients (44.5%), while sorafenib was used in 452 patients (45.5%), together accounting for 90.0% of post-ATE-BEV treatment exposure. Regorafenib was prescribed in 65 patients (6.5%), and cabozantinib in 35 patients (3.5%). Several studies reported outcomes for more than one TKI, resulting in heterogeneity in treatment allocation across study populations.

Baseline demographics and liver function

Mean age ranged from 46 years to 75 years, and the study populations were predominantly male. Liver function was generally preserved: Among evaluable patients, Child-Pugh class A was present in 791 patients of 957 patients (82.7%), while Child-Pugh class B accounted for a smaller proportion [150 patients of 957 patients (15.7%)]. Baseline performance status was overall favorable, with the majority of patients classified as ECOG 0-1, with a minority of patients with ECOG 2 reported in several cohorts.

Etiology of liver disease

Where reported, the underlying liver disease etiology varied substantially across cohorts. The proportion of viral etiologies - defined as hepatitis B virus and/or hepatitis C virus infection - ranged from approximately 12.5% to over 80%, with higher viral prevalence observed predominantly in East Asian cohorts. In contrast, nonviral etiologies, including alcohol-associated liver disease, metabolic dysfunction-associated steatotic liver disease (MASLD), and other chronic liver diseases, accounted for a substantial proportion of cases and ranged from approximately 20% to 87.5%, particularly higher in Western studies.

Tumor stage and disease burden

Tumor stage was predominantly advanced across studies. BCLC stage C disease accounted for more than 60% of patients in the majority of cohorts, whereas BCLC stage B represented a smaller proportion. Indicators of advanced tumor burden were frequently observed: Extrahepatic metastases were reported in up to 86% of patients, and macrovascular invasion was present in up to 50%, although reporting varied across studies.

Geographic distribution and heterogeneity

The included studies originated from both Eastern (Japan and South Korea) and Western cohorts, introducing variability in patient characteristics, underlying liver disease etiology, and treatment selection. Together with differences in baseline liver function, performance status, TKI choice, and reporting completeness, this resulted in substantial clinical heterogeneity across studies, which was accounted for in the pooled analyses.

Quality assessment

Methodological quality was evaluated using the Methodological Index for Non-Randomized Studies. Among the included studies, eight fulfilled 12 out of 16 possible points, demonstrating adequate inclusion of consecutive patients, appropriate endpoint selection, unbiased assessment of outcomes, suitable follow-up duration, and minimal loss to follow-up (< 5%). One study (Persano et al[9]) scored 11 out of 16, primarily due to a loss-to-follow-up rate exceeding 5%.

Because all studies were retrospective in design, none satisfied the MINORS items related to prospective data collection or prospective calculation of study size, which are inherently non-applicable to retrospective cohorts. Overall, methodological quality was moderate, with consistency across key domains but predictable limitations related to retrospective study design.

Primary outcomes: OS and PFS

The pooled mean OS among patients treated with TKIs after progression on ATE-BEV was 10.98 months [95% confidence interval (CI): 8.46-13.50], based on a random-effects model. Subgroup analyses showed pooled mean OS of 8.79 months for cabozantinib (95%CI: 5.74-11.85; I² = 7.2%), 12.99 months for lenvatinib (95%CI: 7.76-18.23; I² = 99.2%), and 10.56 months for sorafenib (95%CI: 6.56-14.57; I² = 99.53%; Figure 3A).

Figure 3 Forest plots of primary outcomes following tyrosine kinase inhibitor therapy after progression on atezolizumab-bevacizumab. A: Overall survival; B: Progression-free survival. REML: Restricted maximum likelihood; CI: Confidence interval.

The pooled mean PFS following TKI therapy after ATE-BEV was 3.41 months (95%CI: 2.92-3.90). Subgroup analyses demonstrated variability across agents. For cabozantinib, pooled mean PFS was 3.41 months (95%CI: 1.26-5.56; I² = 66.7%; P = 0.06). The pooled estimate for lenvatinib was 3.95 months (95%CI: 3.63-4.27; I² = 77.2%; P < 0.001), based on seven cohorts. Mean PFS for sorafenib was 2.80 months (95%CI: 1.76-3.83; I² = 99.58%), and for regorafenib, 3.27 months (95%CI: 2.22-4.33; I² = 81.8%; Figure 3B).

Secondary outcomes

Clinical response: We evaluated clinical response using both RECIST 1.1 and modified RECIST criteria, with pooled estimates derived from a meta-analysis of proportions.

RECIST 1.1: Under RECIST 1.1 criteria, the pooled objective response rate (ORR), defined as complete or partial response, was 9% (95%CI: 5%-12%), with high heterogeneity (I² = 57.2%). Stable disease was observed in 49% of patients (95%CI: 37%-62%), with substantial between-study heterogeneity (I² = 91.1%). Progressive disease occurred in 25% of patients (95%CI: 14%-35%), while unevaluable response assessments were reported in 11% (range, 5%-18%). The pooled disease control rate (DCR)-defined as complete or partial response or stable disease-was 66% (95%CI: 52%-79%; Figure 4).

Figure 4 Forest plots of clinical response outcomes according to Response Evaluation Criteria in Solid Tumors version 1.1. A: Complete response; B: Partial response; C: Stable disease; D: Progressive disease; E: Objective response rate; F: Disease control rate; G: Not evaluable. REML: Restricted maximum likelihood; CI: Confidence interval.

Modified RECIST: Under modified RECIST criteria, which more accurately reflect viable tumor burden in HCC, the pooled ORR was 28% (95%CI: 15%-41%), with substantial heterogeneity (I² = 88.2%). Complete response was observed in 2% of patients (95%CI: 0%-4%), while partial response occurred in 28% (95%CI: 8%-48%). Stable disease was reported in 31% of patients (95%CI: 17%-46%; I² = 82.46%), and progressive disease in 25% (95%CI: 6%-43%; I² = 93.6%). The pooled estimate for patients who could not be evaluated was 12% (95%CI: 2%-22%). The overall DCR using modified RECIST was 69% (95%CI: 52%-85%), with substantial heterogeneity (I² = 93.16%; Figure 5).

Figure 5 Forest plots of clinical response outcomes according to modified Response Evaluation Criteria in Solid Tumors (modified Response Evaluation Criteria in Solid Tumors). A: Complete response; B: Partial response; C: Stable disease; D: Progressive disease; E: Objective response rate; F: Disease control rate; G: Not evaluable. REML: Restricted maximum likelihood; CI: Confidence interval.

Treatment-related adverse events

We assessed 21 treatment-related adverse events reported among patients receiving TKIs following progression on ATE-BEV for HCC. For each adverse event, pooled proportions were synthesized using a random-effects model and stratified by severity according to Common Terminology Criteria for Adverse Events criteria (any grade and grade ≥ 3). This framework enabled estimation of absolute event rates and characterization of the overall toxicity profile across the included studies (Table 2 and Supplementary Figures 1-21).

Table 2 Adverse events, %.

Adverse event	Any grade	95%CI (any grade)	I2 (any grade)	≥ Grade 3 %	95%CI (≥ grade 3)	I2 (≥ 3)
AST/ALT elevation	42	-25 to 100	96.4	9	-10 to 29	67.5
Bilirubin elevation	38	12-65	0.00	4	-6 to 13	0.00
Thrombocytopenia	25	-15 to 65	97.6	3	0-6	0.00
Diarrhea	21	11-31	74.7	2	0-4	0.00
Anorexia	38	27-49	44.9	8	4-11	0.01
Proteinuria	30	18-42	55.3	15	10-20	0.00
Fatigue	52	39-65	76.42	6	3-9	0.01
Hypertension	18	7-29	85.32	3	1-6	0.05
Nausea	8	-7 to 22	72.65	4	-6 to 13	0.00
Anemia	54	27-81	0.00	4	-6 to 13	0.00
Hand-foot syndrome	34	21-47	80.6	3	1-5	0.01
Hypothyroidism	25	14-36	26.5	3	-1 to 6	0.03
Neutropenia	31	6-56	0.00	4	-6 to 13	0.00
Vomiting	5	1-10	0.01	2	-1 to 5	0.00
Gastrointestinal bleeding	1	-1 to 3	0.00	1	-1 to 3	0.00
Fever	12	2-23	0.01	2	-2 to 7	0.01
Non-gastrointestinal bleeding	12	2-23	0.01	3	-2 to 9	0.03
Liver failure	6	0-13	52.93	1	-1 to 3	0.00
Acute kidney injury	5	-6 to 15	64.90	1	-1 to 3	0.03
Abdominal pain	1	-1 to 3	0.01	1	-1 to 3	0.01
Hoarseness	6	-2 to 13	0.00	2	-2 to 7	0.01

CI: Confidence interval.

Hepatic and hematologic toxicity

The pooled incidence of any-grade aspartate aminotransferase (AST)/alanine aminotransferase (ALT) elevation was 42% (95%CI: -25% to 100%) with substantial heterogeneity (I² = 96.4%). Grade ≥ 3 elevations occurred in 9% (95%CI: -10% to 29%; I² = 67.5%). Any-grade bilirubin elevation was reported in one study, with a pooled incidence of 38% (95%CI: 12%-65%; I² = 0%), and grade ≥ 3 events in 4% (95%CI: -6% to 13%; I² = 0%).

Liver failure of any grade was reported in 6% (95%CI: 0% to 13%; I² = 52.9%). Only one study reported grade ≥ 3 liver failure, with a pooled incidence of 1% (95%CI: -1% to 3%; I² = 0%).

Thrombocytopenia was observed in 25% (95%CI: -15% to 65%; I² = 97.6%) of patients, while grade ≥ 3 events were infrequent (3%; 95%CI: 0%-6%; I² = 0%). Neutropenia was only reported in one study, with a pooled incidence of 31% (95%CI: 6%-56%; I² = 0%) for any grade and 4% (95%CI: -6% to 13%; I² = 0%) for grade ≥ 3. Anemia was also common (54%; 95%CI: 27%-81%; I² = 0%), though grade ≥ 3 anemia was rare (4%; 95%CI: -6% to 13%; I² = 0%).

Gastrointestinal adverse events

Diarrhea was reported in 21% of patients (95%CI: 11%-31%; I² = 74.8%), with variation by agent: 30% for cabozantinib, 20% for lenvatinib, 14% for regorafenib, and 24% for sorafenib. Grade ≥ 3 diarrhea was uncommon (2%; 95%CI: 0%-4%; I² = 0%). Anorexia occurred in 38% (95%CI: 27% to 49%; I² = 44.8%), including 53% for cabozantinib and 35% for lenvatinib. Grade ≥ 3 anorexia occurred in 8% (95%CI: 4%-11%; I² = 0.01%).

Nausea occurred in 8% of patients when reported (95%CI: -7% to 22%; I² = 0%), while vomiting occurred in 5% (95%CI: 1%-10%; I² = 0.01%). Grade ≥ 3 nausea and vomiting were both rare (4% and 2%, respectively). Abdominal pain was reported in 1% (95%CI: -1% to 3%) for both any grade and ≥ grade 3 severity (I² = 0.01%).

Renal and endocrine toxicity

Proteinuria occurred in 30% (95%CI: 18%-42%), with 15% (95%CI: 10%-20%) experiencing grade ≥ 3 events. Acute kidney injury of any grade was observed in 5% (95%CI: -6% to 15%; I² = 64.9%), and in 1% (95%CI: -1% to 3%; I² = 0.03%) for grade ≥ 3 events.

Hypothyroidism occurred in 25% of patients (95%CI: 14%-36%; I² = 26.5%), with comparable rates across TKIs. Severe hypothyroidism (grade ≥ 3) occurred in 3% (95%CI: -1% to 6%; I² = 0.03%).

Cardiovascular and constitutional events

Hypertension occurred in 18% (95%CI: 7%-29%; I² = 85.3%) for any grade and in 3% (95%CI: 1%-6%; I² = 0.05%) for grade ≥ 3. Fatigue was among the most frequent toxicities, affecting 52% (95%CI: 39%-65%; I² = 76.42%) of patients; grade ≥ 3 fatigue occurred in 6% (95%CI: 3%-9%; I² = 0.01%). Fever was reported in 12% (95%CI: 2%-23%; I² = 0.01%), with grade ≥ 3 events in 2% (95%CI: -2% to 7%; I² = 0.01%).

Bleeding and mucocutaneous events

Non-GI bleeding occurred in 12% (95%CI: 2%-23%; I² = 0.01%) for any grade and in 3% (95%CI: -2% to 9%; I² = 0.03%) for grade ≥ 3.

Hand-foot syndrome was reported in 34% (95%CI: 21%-47%; I² = 80.6%) for any grade and in 3% (95%CI: 1%-5%; I² = 0.01%) for grade ≥ 3. Hoarseness occurred in 6% (95%CI: -2% to 13%; I² = 0%) for any grade, and in 2% (95%CI: -2% to 7%; I² = 0.01%) for grade ≥ 3.

DISCUSSION

HCC continues to pose a significant global health challenge due to its high incidence, rising prevalence, and considerable mortality. It is the sixth most commonly diagnosed cancer and the third leading cause of cancer-related death worldwide, with an overall five-year survival rate of approximately 18%[10,11]. HCC predominantly develops in individuals with chronic liver disease and cirrhosis, which confer an annual risk of 2%-4%, varying by etiology, geographic region, sex, age, and severity of liver injury. Chronic hepatitis B virus and hepatitis C virus infections are the primary contributors, accounting for nearly 80% of global HCC cases[12]. Concurrently, the incidence of HCC related to nonalcoholic fatty liver disease is increasing rapidly, particularly in Western populations. In the United States, incidence is projected to rise by 122% between 2016 and 2030, from 5510 to 12240 cases[11,13].

The extent of underlying liver dysfunction is a critical determinant of long-term outcomes in HCC. A matched study of patients undergoing curative hepatectomy demonstrated that cirrhosis is associated with significantly higher cumulative recurrence and worse disease-specific survival compared to normal liver parenchyma[14]. This finding highlights the profound influence of the underlying liver milieu on patient outcomes, beyond tumor biology alone. Additionally, large cohort studies indicate that compensated and decompensated cirrhosis present markedly different risk profiles; decompensation accelerates HCC onset and reduces survival[15]. Therefore, the interplay among tumor burden, liver reserve [as measured by Child-Pugh class and albumin-bilirubin (ALBI) grade], and systemic therapy eligibility is essential when interpreting treatment outcomes and making sequencing decisions.

Systemic therapy for advanced HCC has undergone significant evolution. For over a decade, the multikinase inhibitor sorafenib served as the standard first-line treatment, though it provided only modest improvements (median OS of approximately 10-12 months). The IMbrave150 trial shifted this paradigm by demonstrating that ATE plus BEV improved OS and PFS compared to sorafenib, establishing immunotherapy combined with anti-angiogenic agents as the preferred first-line approach[4]. However, optimal sequencing of therapies following ATE-BEV remains largely undefined. Second-line agents such as regorafenib (RESORCE), cabozantinib (CELESTIAL), and ramucirumab (REACH-2) were evaluated in the post-sorafenib setting, but not specifically after first-line immune checkpoint inhibitor/VEGF combination therapy[16-18]. Consequently, a significant evidence gap exists for clinicians selecting systemic therapy for patients who progress on ATE-BEV.

Our meta-analysis of nine retrospective studies involving 994 patients treated with TKIs after progression on ATE-BEV provides valuable insights into this evolving therapeutic landscape. The pooled mean OS of 10.98 months and mean PFS of 3.41 months serve as important benchmarks for this clinical scenario. Notably, these outcomes are comparable to those reported in post-sorafenib second-line trials, where OS was approximately 10-11 months. These findings indicate that prior exposure to ATE-BEV does not abrogate the efficacy of TKIs, reinforcing the feasibility of TKI-based salvage therapy.

In subgroup analyses, Lenvatinib had the highest mean OS (12.99 months), followed by sorafenib (10.56 months) and cabozantinib (8.79 months). A similar trend was observed for PFS, with lenvatinib demonstrating the longest mean PFS (3.95 months), followed by cabozantinib, regorafenib, and sorafenib. Although direct head-to-head comparisons are not available, these findings raise the hypothesis that lenvatinib may offer superior survival in the post-immuno-oncology/VEGF setting. Lenvatinib also achieved a strong modified RECIST ORR and DCR in the pooled data (ORR = 34%, DCR = 78%), comparable to its first-line performance in the REFLECT trial, which reported an ORR of approximately 24% by modified RECIST[19]. These findings support the potential repositioning of lenvatinib as a second-line option after ATE-BEV, particularly for patients with preserved liver function and an ECOG performance status of 0-1.

Radiologic response evaluation highlights the significance of assessment criteria. Using RECIST 1.1, the ORR for TKIs as a group was 9% with a DCR of 66%, whereas modified RECIST identified an ORR of 28% and a DCR of 69%. This discrepancy reflects modified RECIST’s increased sensitivity in detecting viable tumor necrosis and anti-angiogenic effects, rather than solely measuring tumor shrinkage, and supports its application in HCC TKI trials. Clinically, disease stabilization, rather than substantial tumor reduction, may provide meaningful survival benefits, especially in patients with cirrhosis, where maintaining liver reserve is crucial.

The toxicity profile observed in the pooled analysis aligns with the established safety characteristics of TKIs. Common any-grade adverse events included fatigue (52%), anorexia (38%), proteinuria (30%), hand-foot syndrome (34%), and hypertension (18%). Hepatic laboratory abnormalities were also frequent, with any-grade AST/ALT elevation in 42% and bilirubin elevation in 38%. Most of these abnormalities were low grade; clinically significant hepatic events were infrequent, with grade ≥ 3 AST/ALT elevation in 9% and grade ≥ 3 bilirubin elevation in 4%. Except for proteinuria (15%), high-grade non-hepatic toxicities were uncommon, including thrombocytopenia (3%), diarrhea (2%), hand-foot syndrome (3%), hypertension (3%), and neutropenia (4%). Serious hepatic decompensation events were rare, with liver failure grade ≥ 3 reported in only 1% of patients. The low incidence of severe hepatic toxicity (grade ≥ 3 AST/ALT in 9%) suggests that, in appropriately selected patients (Child-Pugh A), TKIs remain both effective and tolerable following prior ATE-BEV exposure. These safety findings are consistent with earlier TKI trials and real-world data[17,19], supporting the feasibility of TKI salvage therapy in patients with preserved liver function and good performance status. However, caution is warranted when extrapolating these results to real-world HCC populations, many of whom have cirrhotic decompensation and comorbidities.

Integration of these results with existing evidence yields several important insights. First, the consistency of OS and PFS with historical second-line benchmarks indicates that ATE-BEV failure does not necessarily predict a substantially worse prognosis for salvage therapy, which is clinically reassuring. Second, the superior numerical outcomes observed with lenvatinib may be attributable to its potent multikinase inhibition, favorable response profile, and potential selection bias; however, these findings require prospective validation. Third, the variability in radiologic response highlights the necessity for standardized assessment criteria in HCC trials. Fourth, the safety data might support the acceptability of TKI sequencing in experienced clinical settings, with manageable toxicities even after prior immuno-oncology therapy.

Clinical guidelines from the National Comprehensive Cancer Network (2024) and American Association for the Study of Liver Diseases/American Gastroenterological Association acknowledge TKIs as options following ATE-BEV therapy but do not prioritize specific agents or define sequencing after immuno-oncology/VEGF therapy, reflecting a significant data gap. This meta-analysis provides preliminary real-world evidence of TKI as a valid strategy for patients with preserved hepatic reserve, pending prospective validation.

Despite these findings, important gaps and challenges remain. The majority of included studies were conducted in Asia and Europe, limiting racial and ethnic diversity and reducing generalizability. United States population-based analyses have identified significant disparities in outcomes: Black patients experience worse survival compared to non-Hispanic White patients [pooled hazard ratio (HR) = 1.08, 95%CI: 1.05-1.12], while Hispanic (pooled HR = 0.92, 95%CI: 0.87-0.97) and Asian patients (pooled HR = 0.81, 95%CI: 0.73-0.88) demonstrate improved survival[20]. These disparities may partially reflect differences in stage at diagnosis, as Black patients have lower odds of early-stage HCC detection compared to White patients [odds ratio (OR) = 0.66, 95%CI: 0.54-0.78], whereas no significant differences are observed among Asian (OR = 1.01, 95%CI: 0.97-1.05) or Hispanic patients (OR = 0.87, 95%CI: 0.74-1.01)[20].

The evolving epidemiology of HCC also requires careful attention. MASLD and nonalcoholic steatohepatitis are becoming increasingly prominent causes of HCC, particularly in Western populations. Notably, MASLD-related HCC often develops in the absence of cirrhosis and is more frequently observed in older individuals with significant cardiometabolic comorbidities[13]. These trends underscore the necessity for more inclusive and representative study populations in future research.

Background liver function remains a central consideration. Cirrhosis severity, ALBI grade, portal hypertension, and decompensation stage each independently influence survival, regardless of tumor burden. For example, earlier resection studies showed that patients with cirrhosis experienced significantly higher recurrence and worse survival than those without cirrhosis[14]. Retrospective data further indicate that decompensated cirrhosis is associated with a shorter time to HCC development and poorer survival compared to compensated cirrhosis[15]. In this meta-analysis, most patients were classified as Child-Pugh A, but heterogeneity in liver reserve and prior therapy exposure limited the ability to conduct more detailed subgroup analyses. Future studies should stratify outcomes by ALBI or Child-Pugh class, degree of portal hypertension, and cirrhosis etiology to enable more individualized therapy selection.

Fundamentally, the therapeutic sequence following immuno-oncology/VEGF therapy remains an area in need of innovation. There is an urgent need for biomarker-driven selection strategies (such as angiogenic gene signatures and circulating tumor DNA dynamics), novel therapeutic combinations (e.g., TKI plus checkpoint inhibitor rechallenge), and clinical trials in populations with compromised liver function (Child-Pugh B). Given the numerical superiority observed with lenvatinib, a prospective randomized trial evaluating TKI sequencing after ATE-BEV would be both scientifically and clinically justified.

From a methodological perspective, this meta-analysis is strengthened by synthesizing the largest available real-world cohort following ATE-BEV therapy. However, limitations include reliance on retrospective data, non-randomized study designs, heterogeneity in patient populations (including tumor burden, prior therapies, and liver reserve), variability in TKI dosing and regimens, and inconsistent reporting of radiologic and adverse events. Although median-to-mean conversion techniques were employed as needed, the inherent variability of real-world observational studies necessitates cautious interpretation. Publication bias is also possible, given the recent emergence of the post-ATE-BEV salvage setting.

Strengths and limitations: This meta-analysis offers several strengths that enhance its clinical relevance. It represents one of the largest syntheses of real-world data evaluating the efficacy and safety of TKIs after progression on ATE plus BEV, thereby addressing a significant gap in current treatment sequencing for advanced HCC. By aggregating data from diverse populations and reporting a broad range of clinical outcomes - including OS, PFS, objective response, DCRs, and adverse event profiles - the study provides a comprehensive assessment of TKI utility in this treatment context. Stratified analyses by TKI agent, along with the inclusion of both RECIST and modified RECIST criteria, further enhance clinical interpretability and provide insight into radiologic response expectations.

Nonetheless, several limitations should be acknowledged. All included studies were retrospective and observational, which limits control over unmeasured confounders and precludes definitive causal inference. Considerable heterogeneity existed across studies in terms of patient selection, TKI regimens, prior lines of therapy, radiologic assessment methods, and liver function staging, potentially introducing bias. Small and unequal sample sizes in specific subgroups, particularly for cabozantinib, further limit the robustness of between-agent reports.

Additionally, key clinical factors such as cirrhosis severity, ALBI grade, and underlying liver disease etiology were inconsistently reported, limiting their incorporation into subgroup analyses. Publication bias remains possible due to the recent introduction of post-ATE-BEV treatment and selective reporting. Patient selection was largely restricted to individuals with preserved hepatic function, predominantly Child-Pugh class A, introducing selection bias toward patients more likely to tolerate and respond to systemic therapy and limiting generalizability to those with more advanced liver dysfunction.

Despite these limitations, this study provides preliminary real-world observations on second-line TKI use following ATE-BEV and underscores the need for prospective trials to better define treatment strategies in this setting.

CONCLUSION

This meta-analysis provides preliminary evidence of TKIs as a viable and effective salvage option for patients with advanced HCC who have progressed on ATE-BEV. Among the agents evaluated, lenvatinib exhibits encouraging real-world outcomes in selected patients, with favorable OS and DCRs, although definitive comparative data are limited. Disease stabilization, as reflected in consistently high DCRs, may represent a critical therapeutic objective in this context, particularly given the complexity of managing patients with underlying cirrhosis.

As the first-line therapeutic landscape evolves and the treated population becomes increasingly heterogeneous, future research should prioritize prospective, biomarker-driven studies, greater inclusion of underrepresented patient groups, and improved implementation of evidence-based sequencing strategies. Until such data are available, these findings provide real-world context supporting the feasibility of second-line TKI therapy in appropriately selected patients in the post-ATE-BEV era.