Comparison of surgical resection and radiofrequency ablation for the treatment of small hepatocellular carcinoma (≤ 3 cm): an updated meta-analysis

Abstract

Objectives

This meta-analysis aims to compare the safety and efficacy of radiofrequency ablation (RFA) and surgical resection (SR) in patients with small hepatocellular carcinoma (HCC).

Methods

A systematic literature search was conducted across four databases (PubMed, Web of Science, Cochrane Library, and Embase) for studies published between January 2020 and June 2024. The included studies reported comparative outcomes of SR and RFA in patients with small HCC (≤ 3 cm), including survival, oncological, and perioperative endpoints.

Results

A total of 25 studies involving 10,322 patients were included in the analysis. Compared with RFA, SR was associated with superior survival outcomes, including three-year overall survival (OS) (HR: 0.73; 95% CI: 0.66–0.80), five-year OS (HR: 0.70; 95% CI: 0.62–0.79), one-year recurrence-free survival (RFS) (HR: 0.55; 95% CI: 0.47–0.65), three-year RFS (HR: 0.66; 95% CI: 0.59–0.73), and five-year RFS (HR: 0.76; 95% CI: 0.68–0.85). Although SR was linked to longer operative times, prolonged hospitalization, and higher complication rates, it provided significantly better local tumor control, as reflected by a markedly reduced local recurrence rate. In patients with multiple tumors or those measuring ≤ 2 cm, the three- and five-year RFS rates were comparable between the SR and RFA groups. The three-year OS was also similar between the two treatments for tumors ≤ 2 cm.

Conclusion

These findings indicate that, compared with RFA, SR offers improved survival and lower recurrence rates in patients with small HCC. However, for individuals with tumors ≤ 2 cm or multiple HCC lesions, long-term survival outcomes are comparable between the two modalities, suggesting that treatment decisions should be individualized on the basis of tumor characteristics and patient risk–benefit profiles.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13643-025-03057-x.

Introduction

Liver cancer remains the main cause of death worldwide [1–5]. Hepatocellular carcinoma (HCC) has garnered considerable attention from clinicians because of its significant incidence and remains one of the leading causes of cancer-related deaths worldwide, posing a substantial global health burden [6]. With advancements in imaging technology and the widespread implementation of cancer screening protocols, an increasing number of early-stage cancer cases, particularly small hepatocellular carcinomas (≤ 3 cm), are being identified, thus opening up opportunities for curative-intent treatment and significantly improving the prognosis for this patient population. The early detection and management of these small HCCs are crucial, as they represent a stage where curative therapies can be effectively applied [7, 8]. The Barcelona Clinic Liver Cancer (BCLC) algorithm provides various treatment options for HCC, including liver transplantation, ablation and surgical resection (SR) [9]. However, liver transplantation is constrained by donor costs, technical difficulty, and complications [10]. Therefore, the primary treatments for early HCC are SR and radiofrequency or microwave ablation.

The European Association for the Study of the Liver (EASL) and the American Association for the Study of Liver Diseases (AASLD) recommend hepatectomy for patients with early-stage HCC and adequate functional liver reserve, establishing it as the historical standard of care [9, 11]. This recommendation is rooted in the principle of complete anatomical removal of the tumor, which theoretically minimizes marginal recurrence and provides accurate pathological staging. However, thermal ablation remains a feasible alternative for many patients eligible for excision [12–14]. The core of this controversy lies in balancing the potential superior oncological outcomes of SR with the minimally invasive benefits of RA. Hepatectomy is often restricted by impaired liver function and a high complication rate [15, 16]. In contrast, radiofrequency ablation (RFA) is recognized for its simplicity and minimalism and is an acceptable choice for small HCC patients. Proponents of SRs argue that it offers more durable long-term survival and lower recurrence rates. Overall survival (OS) and recurrence-free survival (RFS) following hepatectomy are more favorable than those following RA for patients with small HCC [13, 17]. Nonetheless, a significant body of research has indicated that both RA and SR are equally effective in treating early HCC, particularly for tumors smaller than 2 cm [18–20]. Compared with hepatectomy, RA is associated with a significantly lower risk of complications [12, 21]. Additional considerations that support RA over SR include intraoperative blood loss, length of hospital stay (LOS), operative time (OT), and expenses, with notable distinctions between the two treatment modalities [12–14, 21, 22]. This conflicting evidence, derived from numerous individual studies and previous meta-analyses, creates a clinical dilemma. The heterogeneity in findings can be attributed to variations in study design, patient selection criteria, and technical expertise across different institutions. Consequently, the discussion of the most effective and safe treatments for early HCC patients is ongoing.

To overcome the limitations of individual studies—such as limited sample size and insufficient statistical power to detect differences in key subgroups—a comprehensive and up-to-date synthesis of the available evidence is imperative. Therefore, we performed a systematic review and meta-analysis to assess the safety and effectiveness of SRs and RAs in the medical management of patients with small HCC lesions to provide a more comprehensive practical reference. Our study synthesizes data from a large cohort of patients, incorporates the most recent literature up to June 2024, and specifically investigates crucial subgroups to provide contemporary, high-level evidence that can guide personalized clinical decision-making.

Methods

The PROSPERO database has a prospective registration of this review (CRD42024575603). The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement and AMSTAR (Assessing the Methodological Quality of Systematic Reviews) guidelines were followed in the reporting of this review [23, 24].

Search strategy

We systematically searched four databases (PubMed, Web of Science, Cochrane Library, and Embase) from January 1, 2020, to June 30, 2024. This timeframe was selected for two principal reasons: First, this study is intended as a focused update of the evidence, aiming to consolidate research data published since prior key syntheses [25–27]. Second, technological and management strategies for HCC have evolved rapidly in recent years. Limiting the search to contemporary studies helps ensure that the included evidence reflects current clinical practice, thereby enhancing the timeliness and clinical relevance of our findings. The search string merged medical subject headings (MeSH) with text words: (Carcinoma, Hepatocellular OR HCC OR Liver Cell Carcinoma) AND (Hepatectomy OR surgical resection OR Partial hepatectomy) AND (radiofrequency ablation OR Ablation OR Radiofrequency Ablation). Moreover, we manually retrieved references from studies that were relevant to the topic to expand the search. Each featured study was evaluated separately by two reviewers (L.D. and R.Z.), who resolved any disagreements by consensus.

Inclusion/exclusion criteria

Studies that complied with the inclusion criteria were as follows: (1) patients diagnosed with small HCC (diameter ≤ 3 cm and ≤ 3 nodules) with well-preserved liver function; (2) comparisons between the SR and RA; (3) prospective controlled trials and retrospective studies should be included in addition to randomized controlled trials (RCTs); and (4) papers that include at least one outcome parameter, such as OS, RFS, OT, LOS, overall complication rate, and local recurrence rate. We excluded patients with distant or vascular invasion, recurrent HCC, or HCC treated in combination with other modalities. Furthermore, reviews, letters, case reports, and studies with incomplete data were not included. The publication period is limited from January 1, 2020, to June 30, 2024.

Data extraction

The data extracted for the meta-analysis included the leading author, research type, publication date, research country, research interval, number of samples, and research outcomes such as OS and RFS. We used the 95% confidence interval (CI) hazard ratio (HR) described by Tierney et al. as the effect size for OS and RFS [28]. When continuous variables are presented in different formats in the main research, we calculate the mean and standard deviation [29].

Outcomes

The time interval between the onset of treatment and HCC recurrence or tumor mortality is referred to as the RFS. The time interval from the first therapy (ablation and resection) to death from any factor is referred to as the OS. Detecting an enhancing lesion next to the ablation or surgery area that was absent from the first postprocedure scan was the definition of local recurrence. Local recurrence for SR was defined as recurrence in the same liver region following local removal, whereas for RA, recurrence was regarded as any lesion inside 2 cm of the ablation location.

Quality assessment

The Newcastle‒Ottawa quality assessment scale was utilized to evaluate the quality of studies that were not randomized. [30]. Three parameters were used to evaluate the studies: selection, comparability, and exposure/outcome. High-quality studies were considered to be indicated by a score of at least 7. For RCTs, the Cochrane Collaboration Risk of Bias Tool was applied. We assessed every trial and assigned it either a high, low, or unclear risk of bias on the basis of random sequence generation, allocation concealment, blinding of participants and personnel to the study protocol, blinding of outcome assessment, incomplete result data, or selective reporting [31]. Discussions between the two commentators resolved all disagreements.

Statistical analysis

Stata 14.0 software was used to process the meta-analysis statistically in this study. For OS and RFS, if the study did not directly report the HR value and 95% confidence interval, we used the provided data for calculation. The weighted mean difference (WMD) and odds ratio (OR) were calculated for continuous and dichotomous variables with 95% confidence intervals (CIs), respectively. Given the anticipated clinical and methodological diversity among the included studies, all meta-analyses were performed via a random effects model, which provides a more conservative and generalizable estimate under such conditions. Heterogeneity among studies was quantified via the I² statistic alongside the χ² test, with I² > 50% considered indicative of substantial heterogeneity, although this metric was interpreted descriptively and did not determine the choice of model. Statistical significance was defined as p < 0.05. We conducted sensitivity analyses to further measure the robustness of the outcomes.

Subgroup analysis

Subgroup analyses were carried out for one-, three-, and five-year OS and 1-, 3-, and five-year RFS on the basis of tumor diameter (≤ 3 cm and ≤ 2 cm), number of patients (> 100 and ≤ 100), number of tumors (solitary and multiple), and study type (RCT, propensity score matching (PSM) and retrospective).

Publication bias

Egger's test and Begg's test were used to examine publication bias. In the event of publication bias, we employed the trim-and-fill method to calculate the number of missing studies and recalculate the outcomes [32].

Results

Search results

Our search method resulted in the identification of 4573 articles, 1192 of which were duplicates. The remaining papers underwent additional screening following rigorous inclusion and exclusion criteria. According to the titles and abstracts provided, 3232 articles were removed, and 149 full-text articles remained. Overall, 25 papers containing 10,322 patients were included in the present study after unrelated articles were eliminated, with between 5,062 and 5,260 patients in the SR and RA groups, respectively (Fig. 1) [13, 17, 33–55]. The included studies included one RCT, nine PSM studies, and fifteen retrospective studies. Eight of the trials involved multiple HCCs, and seven involved a single HCC. We assembled the pertinent information in Table 1. Additionally, Supplementary Table S1 displays the quality assessment of the included RCTs.

Fig. 1.

Flow diagram of studies identified, included, and excluded

Table 1.

Baseline characteristics of the included studies and methodological assessment

Study	Region	Intervention year	Study type	Patients (SR/RA)	Age	Tumor diameter (cm)	number of tumors	CTP		Outcome	NOS
								SR	RA
Dong et al 2020 [54]	Korea	2008.01- 2009.12	Retrospective	145/178	53.3 ± 10.0/ 53.3 ± 10.0	≤ 3	solitary	Child‒Pugh A:131 Child‒Pugh B:14	Child‒Pugh A:156 Child‒Pugh B:22	complication	8
Hsiao et al. 2020 [55]	Taiwan	2007.01- 2015.12	Retrospective	156/231	58.8 ± 11.7/ 62.2 ± 12.3	< 2	solitary	Child‒Pugh A:156	Child‒Pugh A:231	OS、RFS	7
Lin et al. 2020 [13]	Taiwan	2010–2016	Retrospective	36/39	NR	≤ 2	solitary	Child‒Pugh A:36	Child‒Pugh A:36	OS、RFS、LR	8
Magistri et al. 2020 [50]	Italy	2014.01- 2019.10	Retrospective	24/36	63.67 ± 8.95/ 58.00 ± 17.50	< 3	solitary	NR	NR	OS、LOS、complication	6
Delvecchio et al. 2021 [43]	Italy	2009.01- 2019.01	PSM	26/26	75.02 ± 3.13/ 75.26 ± 3.11	≤ 3	solitary	Child‒Pugh A:21 Child‒Pugh B:5	Child‒Pugh A:26	OS、RFS、OT、 LOS、complication	8
Lee et al. 2021 [17]	Korea	2014.01- 2016.12	Retrospective	251/315	57.5 ± 9.3/ 60.8 ± 9.6	≤ 3	solitary	NR	NR	OS、RFS、LOS、 complication	7
Li et al. 2021 [34]	Taiwan	2007- 2018	PSM	58/58	59.33 ± 11.11/ 58.33 ± 10.88	≤ 2	solitary	Child‒Pugh A:56	Child‒Pugh A:57	OS、RFS	7
Ogiso et al. 2021 [21]	Japan	2005- 2016	Retrospective	85/136	67.67 ± 9.27/ 71.60 ± 8.17	≤ 3	multiple	Child‒Pugh A:73 Child‒Pugh B:12	Child‒Pugh A:110 Child‒Pugh B:26	OS、RFS、LOS、 LR、complication	8
Suh et al. 2021 [49]	South Korea	2008- 2014	Retrospective	722/731	NR	≤ 3	solitary	Child‒Pugh A:722	Child‒Pugh A:731	OS	6
Wu et al. 2021 [48]	Taiwan	2006.01- 2016.12	Retrospective	83/73	NR	≤ 2	solitary	NR	NR	OS、RFS、LR、	8
Conticchio et al. 2022 [39]	Italy	2009.01- 2019.01	PSM	58/58	76.92 ± 4.06/ 76.78 ± 4.09	≤ 3	solitary	Child‒Pugh A:50 Child‒Pugh B:8	Child‒Pugh A:52 Child‒Pugh B:6	OS、RFS、OT、 LR、complication	8
Ivanics et al. 2022 [44]	Canada	2000.02- 2018.11	Retrospective	25/83	61.41 ± 9.05/ 60.99 ± 7.52	≤ 3	solitary	Child‒Pugh A:25	Child‒Pugh A:79 Child‒Pugh B:4	RFS	7
Seong et al. 2022 [56]	Korea	2014.01- 2016.12	PSM	60/29	55.8 ± 9.0/ 60.0 ± 9.8	≤ 3	solitary	NR	NR	OS、RFS、LOS、 complication	6
Lee et al. 2022 [53]	Korea	2005.01- 2015.12	Retrospective	232/159	NR	≤ 3	solitary	Child‒Pugh A:232	Child‒Pugh A:159	OS、RFS、LOS、 complication	7
Takayama et al. 2022 [52]	Takayama	2009–2015	RCT	150/151	68.67 ± 8.24/ 68.04 ± 8.37	≤ 3	multiple	Child‒Pugh A:139 Child‒Pugh B:10	Child‒Pugh A:149 Child‒Pugh B:2	RFS、OT、LOS、 LR、complication	-
Zhang et al. 2022 [33]	China	2009–2018	PSM	67/67	57.51 ± 8.37/ 57.78 ± 10.97	≤ 3	solitary	Child‒Pugh A:67	Child‒Pugh A:67	OS、RFS、OT、 LOS、complication	8
Kang et al. 2023 [35]	Korea	2009–2018	Retrospective	36/40	57.8 ± 11.70/ 61.6 ± 13.72	≤ 3	solitary	Child‒Pugh A:36	Child‒Pugh A:40	OS、RFS、LOS、 LR、complication	8
Kim et al. 2023 [57]	Korea	2011–2016	PSM	115/115	70 ± 4.45/ 69.40 ± 3.72	< 3	multiple	Child‒Pugh A:110 Child‒Pugh B:5	Child‒Pugh A:112 Child‒Pugh B:3	OS、RFS	7
Maher et al. 2024 [36]	Australia	2016–2020	Retrospective	25/51	60.6 ± 9.2/ 62.6 ± 8.8	≤ 3	solitary	NR	NR	OS、RFS、LOS、 complication、LR	8
Min et al. 2023 [58]	Korea	2014–2016	PSM	78/78	59 ± 9/ 60 ± 9	≤ 3	solitary	Child‒Pugh A:75 Child‒Pugh B:3	Child‒Pugh A:76 Child‒Pugh B:2	OS、RFS、LOS、 LR、complication	7
Alessandro et al. 2024 [59]	Italy	2008.11- 2020.12	Retrospective	296/240	NR	≤ 3	multiple	Child‒Pugh A:268 Child‒Pugh B:28	Child‒Pugh A:218 Child‒Pugh B:22	OS	7
Zhang et al. 2020 [47]	China	2016.7- 2019.7	Retrospective	85/90	63.5 ± 7.6/ 62.8 ± 8.5	≤ 2	multiple	NR	NR	OS、RFS、LOS、 complication	8
Rasic et al. (1) 2023 [46]	USA	2004–2015	PSM	876/876	NR	≤ 2	multiple	NR	NR	OS	8
Rasic et al. (2) 2023 [46]	USA	2004–2015	PSM	1235/1235	NR	≤ 3	multiple	NR	NR	OS	8
Liu et al. 2022 [45]	Taiwan	2011–2018	PSM	103/103	63 ± 11.86/ 61.75 ± 11.82	≤ 3	multiple	Child‒Pugh A:102 Child‒Pugh B:1	Child‒Pugh A:102 Child‒Pugh B:1	OS、RFS、LR、 complication	8
Wei et al. 2024 [40]	China	2013.8- 2018.1	Retrospective	72/68	45.7 ± 11.9/ 43.4 ± 12.8	≤ 3	multiple	Child‒Pugh A:58 Child‒Pugh B:14	Child‒Pugh A:55 Child‒Pugh B:13	OS、RFS、OT、 LOS、complication	8

SR surgical resection, RA radiofrequency ablation, CTP Child-Turcotte-Pugh, OS overall survival, RFS recurrence-free survival, NOS Newcastle–Ottawa Quality Assessment Scale, OT operative time, LOS length of hospital stay, LR local recurrence, PSM propensity score matching, NR not reported

Survival outcomes

One-year OS

Sixteen studies were used to extract data on one-year OS, with high heterogeneity among studies (I² = 69.9%, P < 0.001) [17, 21, 33–35, 37, 39, 43, 48, 50, 51, 53]. The result of the final meta-analysis was one-year OS for RA patients vs. one-year OS for SRs; hazard ratio (HR) = 0.86; 95% CI = 0.63–1.18; P = 0.256 (Fig. 2A).

Fig. 2.

Forest plot and meta-analysis of OS. A One-year OS. B Three-year OS. C Five-year OS

Three-year OS

Among the twenty-one studies on three-year OS, obvious heterogeneity was observed (I² = 52.4%, P = 0.002) [17, 21, 33–37, 40–43, 45, 47–51, 53, 55]. Our final results demonstrated a significant survival benefit for the SR group: three-year OS for the SR group vs. the RA group; HR = 0.73; 95% CI = 0.66 to 0.80; P < 0.001 (Fig. 2B).

Five-year OS

Fourteen studies reported the five-year OS [13, 33–35, 37, 40, 42, 43, 45, 48, 49, 53, 55]. The heterogeneity test revealed low and nonsignificant heterogeneity (I² = 32.5%, P = 0.115), and the meta-analysis demonstrated a superior five-year OS for the SR group vs. the RA group (HR = 0.70; 95% CI = 0.62–0.79; P < 0.001) (Fig. 2C).

One-year RFS

A total of nineteen studies reported one-year RFS [13, 17, 21, 33–39, 41, 43, 44, 47, 48, 52, 55, 60]. Low heterogeneity between studies was detected by the heterogeneity test (I² = 44.6%, P = 0.019). The one-year RFS for the SR group was superior to that for the RA group (HR = 0.55; 95% CI = 0.47–0.65; P < 0.001) (Fig. 3A).

Fig. 3.

Forest plot and meta-analysis of RFS. A One-year RFS. (B) Three-year RFS. (C) Five-year RFS

Three-year RFS

Nineteen studies provided data on three-year RFS, with significant heterogeneity observed among them (I² = 57.3%, P = 0.001) [13, 17, 21, 33–41, 43, 45, 47, 48, 52, 55]. The analysis of all eligible studies revealed a significant survival benefit for HCC patients receiving SR versus alternative treatments; HR = 0.66; 95% CI = 0.59–0.73; P < 0.001 (Fig. 3B).

Five-year RFS

Five-year RFS was reported in sixteen studies comprising 2,749 patients; the heterogeneity test indicated a high degree of heterogeneity (I² = 65.4%, P < 0.001) [13, 21, 33–35, 37, 38, 40, 41, 43, 45–48, 52, 55]. A significant improvement in five-year RFS was observed for the SR group versus the RA group (OR = 0.76; 95% CI = 0.68 to 0.85; P = 0.001) (Fig. 3C).

Perioperative outcomes

Overall complications

Among the sixteen studies on overall complications [17, 21, 38, 39, 43, 45, 47, 50, 51, 53]. Assessment of heterogeneity indicated negligible heterogeneity that was not statistically significant (I² = 1.4%, P = 0.436). Our research revealed that RA was significantly associated with a greater risk of overall complications (OR = 2.60; 95% CI = 2.02 to 3.35; P < 0.001) (Fig. 4A).

Fig. 4.

Forest plot and meta-analysis. A Overall complications. B Length of hospital stay. C Operative time. D Local recurrence rate

Length of hospital stay

Thirteen studies documented the LOS of patients with HCC [17, 21, 33, 35, 36, 38, 40, 41, 43, 47, 50, 52, 53]. High heterogeneity within studies was identified through the heterogeneity test (I² = 97.5%, P < 0.001). The analysis revealed a significant increase in the length of stay for the RA team compared with the SR team; WMD = 4.83 days; 95% CI = 3.25–6.40; P < 0.001 (Fig. 4B).

Operative time

The data on OT were gathered from five studies [33, 39, 40, 43, 52]. The included studies exhibited high heterogeneity (I² = 91.1%, P < 0.001); nevertheless, the meta-analysis consistently revealed a significantly shorter operative time in the RA group than in the SR group; WMD = −180.50 min; 95% CI = −212.41– −148.60; P < 0.001 (Fig. 4C).

Local recurrence rate

Nine studies with 1777 patients were examined [13, 21, 35, 36, 39, 41, 45, 48, 52]. The included studies exhibited high heterogeneity (I² = 44.7%, P = 0.070). The results revealed that the SR was associated with a significantly lower local recurrence rate than the RA was; OR = 0.24; 95% CI = 0.15 to 0.40; P < 0.001 (Fig. 4D).

Publication bias and sensitivity analysis

Sensitivity analysis was conducted for one-, three-, and five-year OS and 1-, 3-, and five-year RFS. The stability of our findings was demonstrated by the absence of a corresponding difference following the methodical removal of every study (Fig. 5). A funnel plot was used to assess publication bias among the studies. The findings revealed that some publication bias persisted even when the distribution of research was roughly conical (Fig. 6). Begg's test failed to detect any significant publication bias (one-, three-, or five-year OS: P = 0.163, 0.159, 0.228; 1-, 3-, or five-year RFS: P = 0.208, 0.162, 0.444) or Egger's test (one-, three-, or five-year OS: P = 0.304, 0.290, 0.355; 1-, 3-, or five-year RFS: P = 0.107, 0.237, 0.199).

Fig. 5.

Sensitivity analysis. A One-year OS. B Three-year OS. C Five-year OS. D One-year RFS. E Three-year RFS. F Five-year RFS

Fig. 6.

Funnel plot. A One-year OS. B Three-year OS. C Five-year OS. D One-year RFS. E Three-year RFS. F Five-year RFS

Subgroup analyses

Subgroup analysis revealed no significant differences in three-year OS, three-year RFS, or five-year RFS between RAs and SRs for patients with HCC tumors measuring ≤ 2 cm. For solitary tumors, the SR offered an equal benefit in terms of one-year OS, as indicated by subgroup analysis on the basis of tumor count. Studies with sample sizes of ≤ 100 participants revealed no differences in three-year OS or five-year OS between the two treatment groups according to subgroup analysis on the basis of sample size. The findings from the subgroup analysis by study type were in line with the overall meta-analysis results (Table 2, Supplementary figures S1–S6).

Table 2.

Subgroup analysis of survival outcomes

Outcome	Variable	No. of studies	Model	HR (95% CI)	p value	I2 (%)
One-year OS	Total	16	Random	0.86 (0.63, 1.18)	0.256	70.1
	Tumor diameter
	≤ 3 cm	13	Random	0.88 (0.63, 1.21)	0.296	69.0
	≤ 2 cm	3	Random	0.72 (0.23, 2.26)	0.749	82.4
	Number of tumors
	Solitary	11	Random	0.41 (0.27, 0.64)	0.014	50.1
	Multiple	5	Random	1.84 (1.18, 2.88)	0.145	50.9
	Patients
	> 100	12	Random	0.91 (0.65, 1.26)	0.492	74.2
	≤ 100	4	Random	0.56 (0.21, 1.50)	0.266	55.1
	Study type
	PSM	6	Random	0.48 (0.22, 1.05)	0.266	56.8
	Retrospective	10	Random	0.96 (0.69, 1.36)	0.577	75.0
Three-year OS	Total	22	Random	0.73 (0.66, 0.80)	< 0.001	52.4
	Tumor diameter
	≤ 2 cm	6	Random	0.75 (0.65, 0.86)	< 0.001	79.0
	≤ 3 cm	16	Random	0.71 (0.63, 0.80)	0.002	25.2
	Number of tumors
	Solitary	14	Random	0.58 (0.47, 0.71)	< 0.001	48.9
	Multiple	8	Random	0.77 (0.70, 0.85)	< 0.001	43.6
	Patients
	> 100	17	Random	0.73 (0.67, 0.80)	< 0.001	52.1
	≤ 100	5	Random	0.54 (0.31, 0.93)	0.054	58.0
	Study type
	PSM	9	Random	0.74 (0.67, 0.83)	< 0.001	9.8
	Retrospective	13	Random	0.69 (0.58, 0.81)	0.034	65.4
Five-year OS	Total	14	Random	0.70 (0.62, 0.79)	< 0.001	32.5
	Tumor diameter
	≤ 2 cm	4	Random	0.48 (0.34, 0.66)	< 0.001	31.0
	≤ 3 cm	10	Random	0.75 (0.66, 0.86)	< 0.001	0.0
	Number of tumors		Random
	Solitary	10	Random	0.66 (0.56, 0.78)	0.001	48.6
	Multiple	4	Random	0.75 (0.62, 0.90)	0.002	0.0
	Patients
	> 100	11	Random	0.71 (0.63, 0.81)	< 0.001	18.1
	≤ 100	3	Random	0.47 (0.23, 0.95)	0.110	65.2
	Study type
	PSM	6	Random	0.69 (0.52, 0.91)	0.009	0.0
	Retrospective	8	Random	0.70 (0.61, 0.81)	0.006	58.9
One-year RFS	Total	19	Random	0.55 (0.47, 0.65)	< 0.001	44.6
	Tumor diameter
	≤ 2 cm	5	Random	0.43 (0.28, 0.64)	0.022	59.8
	≤ 3 cm	14	Random	0.58 (0.49, 0.69)	< 0.001	37.2
	Number of tumors
	Solitary	14	Random	0.49 (0.40, 0.60)	< 0.001	40.6
	Multiple	5	Random	0.69 (0.52, 0.91)	0.008	41.3
	Patients
	> 100	14	Random	0.59 (0.50, 0.70)	< 0.001	25.1
	≤ 100	5	Random	0.29 (0.17, 0.49)	0.001	55.5
	Study type
	PSM	8	Random	0.47 (0.36, 0.61)	< 0.001	0.0
	Retrospective	10	Random	0.56 (0.45, 0.71)	0.001	62.9
Three-year RFS	Total	19	Random	0.66 (0.59, 0.73)	< 0.001	57.3
	Tumor diameter
	≤ 2 cm	5	Random	0.64 (0.50, 0.82)	0.177	78.8
	≤ 3 cm	14	Random	0.66 (0.58, 0.75)	< 0.001	44.0
	Number of tumors
	Solitary	13	Random	0.60 (0.52, 0.69)	< 0.001	41.8
	Multiple	6	Random	0.75 (0.63, 0.89)	0.178	72.1
	Patients
	> 100	14	Random	0.69 (0.61, 0.77)	< 0.001	54.8
	≤ 100	5	Random	0.39 (0.27, 0.57)	< 0.001	24.7
	Study type
	PSM	8	Random	0.56 (0.46, 0.69)	< 0.001	0.0
	Retrospective	10	Random	0.66 (0.58, 0.76)	0.001	70.7
Five-year RFS	Total	16	Random	0.76 (0.68, 0.85)	0.001	65.4
	Tumor diameter
	≤ 2 cm	5	Random	0.71 (0.58, 0.88)	0.171	80.9
	≤ 3 cm	11	Random	0.78 (0.68, 0.89)	0.003	54.5
	Number of tumors
	Solitary	11	Random	0.67 (0.58, 0.77)	< 0.001	46.6
	Multiple	5	Random	0.91 (0.77, 1.08)	0.618	76.8
	Patients
	> 100	12	Random	0.79 (0.71, 0.89)	0.013	59.8
	≤ 100	4	Random	0.46 (0.31, 0.66)	0.019	64.6
	Study type
	PSM	7	Random	0.65 (0.53, 0.79)	< 0.001	0.0
	Retrospective	8	Random	0.76 (0.65, 0.88)	0.036	76.8

OS overall survival, RFS recurrence-free survival, HR hazard ratio

Discussion

Main interpretation

The optimal treatment for early-stage HCC remains unclear. Several RCTs from China, Japan, and Hong Kong have compared the benefits of RA and the SR for potentially resectable HCC [19, 52, 61–65]. The results of these studies have been mixed; some suggest that SR is superior, whereas others find similar outcomes for both treatments. Previous meta-analyses have provided valuable insights, but they are limited in number and scope and potentially lack comprehensive conclusions [66–69]. Xu et al. performed a meta-analysis to compare the long-term survival results of patients with small HCC receiving either SR or RA [69]. Their findings indicated a decrease in five-year OS with RA compared with that with SR, although trial sequence analysis suggests that more trials are needed to validate this finding. Additionally, the use of HR is recommended for analyzing time‒event results [28]. RCTs and propensity score-matched (PSM) trials were included in Zhang et al.'s recent network meta-analysis, which revealed that SRs performed better than RAs did in terms of OS and disease-free survival [70]. These results align with our findings. However, our study includes high-quality clinical studies published after 2020, reflecting advances in medical technology, making our updated meta-analysis particularly relevant. Furthermore, our analysis incorporates sensitivity and subgroup, providing robust evidence for our conclusions. Importantly, we examined specific subgroups that previous meta-analyses had not explored, including tumor number and tumor size, providing a more detailed understanding of treatment outcomes.

Survival outcomes are the primary focus of our study. On the basis of our findings, the SR outperforms the RA in terms of one-year, three-year, and five-year RFS in addition to three-year and five-year OS. The local recurrence rate was considerably lower in the SR group than in the RA group, but there was no significant difference in the one-year OS between the two therapies. RA can provide an OS similar to that of the SR in patients with small HCC, as shown by some studies [17, 71]. For example, Lee et al. reported that the three-year RFS rate was significantly higher in the SR group than in the RA group and that the local recurrence rate was noticeably lower in the SR group. Nevertheless, there was no discernible difference in the three-year OS between the two groups. The lower recurrence rate in the SR group resulted in a significantly lower average number of interventional treatments required for reoccurring HCC following initial treatment failure, which could account for the group’s similar OS results [17]. The risk of tumor spread after treatment has been linked to RA, which can negatively impact patient survival [72]. The higher local recurrence rate associated with RA is primarily due to incomplete ablation [73]. Patients with HCC are more likely to experience recurrence if they receive RA as a local therapy because it can overlook minor lesions and has poor targeted accuracy. The complex network of blood vessels in the liver poses additional challenges during RA treatment, as targeting these surrounding large blood vessels can lower the treatment temperature, significantly impacting clinical efficacy and increasing the likelihood of residual tumors and recurrence [74, 75]. Furthermore, there is a risk of needle-track seeding after RA [76]. In contrast, hepatectomy may eliminate multiple tumors and tumor thrombi from the same anatomical location, which is essential for limiting tumor recurrence [77, 78]. This comprehensive approach contributes to the superior long-term survival results observed with the SR.

Tumor size is an independent indicator of OS after RA. Similarly, our subgroup analysis revealed no significant differences in three-year OS, three-year RFS, or five-year RFS between the SR and RA patients with liver tumors ≤ 2 cm. This may be because smaller tumors allow the electrode to achieve the desired ablation margin, including the microsatellite nodules surrounding the tumor. Interestingly, we also found no significant difference in three-year or five-year RFS between the two treatments for multiple tumors. This could be attributed to the limited number of clinical trials involving multiple HCCs included in our analysis. More prospective multicenter clinical studies are needed to validate these findings.

Several variables, such as age and cancer stage, affect how HCC is treated. Notably, surgery consistently proves superior to RA for small HCCs smaller than 3 cm in elderly patients [53, 79]. For early HCC in older patients, a more aggressive surgical approach is recommended because of its favorable outcomes and potential for extended life expectancy. However, while the three-year RFS rates for RA and surgery are similar, surgery is superior to RA after 5 years. Therefore, RA may be considered a viable option for older patients with a life expectancy of five or fewer years.

Previous studies have demonstrated that most patients who undergo RA have a short hospital stay of 2–3 days, with some patients leaving the same day [80]. This aligns with our findings, where the RA team had significantly fewer LOSs than the SR team did. Additionally, postoperative complications are less likely to occur with SR, and the primary advantage of RA over SR is its minimally invasive nature. RA causes minimal damage to the surrounding healthy liver tissue, thereby maximizing liver preservation [81]. This results in a significantly reduced complication rate and shorter hospital stays. The mild nature of RA-induced damage, minimal bleeding, and the typical absence of a need for general anesthesia contribute to patients being able to eat and move sooner, further reducing their hospital time. This explains the decreased complication rate observed in the RA group.

Limitations

This meta-analysis was carried out following the strict PRISMA statement standards, yet some limitations persist. First, there was heterogeneity in certain outcomes, including one-year OS, five-year RFS, LOS, and OT. This heterogeneity may stem from variations in surgical approaches (laparoscopic or robot-assisted hepatectomy, laparoscopic or percutaneous RA) and differences in surgeon proficiency across regions. Consequently, interpreting our conclusions requires caution. Second, the majority of the studies that were included were retrospective, with only one being an RCT. Therefore, further large-scale prospective clinical trials are essential to confirm our findings. Third, tumor heterogeneity is an unavoidable factor. Although all studies involved HCCs ≤ 3 cm in diameter, there were variations in the number and size of tumors among patients in the included trials. While we performed subgroup analyses on the basis of tumor size and number, the scope of these analyses was limited by the available data. Nevertheless, our meta-analysis provides robust evidence by including clinical studies published within the last four years, thereby minimizing the impact of generational differences in surgical and ablation techniques on the results. The comprehensive analysis of outcome measures further strengthens the confidence in our findings.

Toward a comprehensive clinical decision

The choice between SR and RFA for small HCC transcends a simple comparison of survival metrics and necessitates a comprehensive, patient-centric evaluation. Our findings provide essential data to inform this nuanced decision-making process. For fit patients with a single tumor > 2 cm, where superior long-term tumor control is the paramount goal, the SR demonstrates clear oncological benefits despite its greater initial invasiveness. Conversely, for patients with tumors ≤ 2 cm or multiple nodules, where long-term survival outcomes are comparable, the significantly lower invasiveness, shorter recovery time, and reduced complication profile of RFA have become compelling factors [82, 83]. This decision must be further individualized by considering critical patient-specific variables not fully captured in our analysis, such as tumor location relative to vital structures (which influences the technical feasibility and safety of both modalities), underlying liver function beyond the Child‒Pugh score, patient age, comorbidities, and personal preferences regarding recovery time and risk tolerance. Therefore, an optimal treatment strategy should be collaboratively determined through a multidisciplinary team discussion, integrating our robust comparative efficacy and safety data with the patient's unique clinical scenario and values [84].

Conclusion

This meta-analysis revealed that for small HCCs (≤ 3 cm), the SR offers better OS and RFS and lower local recurrence rates than does the RA. However, for tumors ≤ 2 cm or multiple HCCs, the SR and RA have similar long-term survival outcomes. The advantages of the SR need to be balanced against its potential complications, extended OT, and longer LOS.

Abbreviations

HCC

Hepatocellular carcinoma

SR

Surgical resection

RA

Radiofrequency ablation

OS

Overall survival

RFS

Recurrence-free survival

BCLC

Barcelona Clinic Liver Cancer

EASL

European Association for the Study of the Liver

AASLD

American Association for the Study of Liver Diseases

LOS

Length of hospital stay

OT

Operative time

Mesh

Medical Subject Headings

HR

Hazard ratio

WMD

Weighted mean difference

OR

Odds ratio

Cis

Confidence intervals

PSM

Propensity score matching

Authors’ contributions

Long-Ao Dai contributed to protocol/project development, data collection and management, data analysis and manuscript writing/editing. Min Sun contributed to data analysis and manuscript writing/editing. Tian Li contributed to manuscript writing/editing. Dong Wei contributed to manuscript editing. Ren-Chao Zou contributed to protocol/project development/management and manuscript editing. All authors have read and approved the final manuscript.

Funding

Kunming Medical Joint Project-surface Project (202201AY070001-107), Yunnan Special Research Projects-Youth Program (202201AU070198), Doctor of Scientific Research Project of the Second Affiliated Hospital of Kunming Medical University (2023BS08,2024BS07), Scientific Research Fund Project of Yunnan Provincial Department of Education (2022J0260), International Cooperation Research Project of the Second Affiliated Hospital of Kunming Medical University (2022dwhz05), General Project of Yunnan Province Health Science and Technology Program (2025GDYB001), and Clinical Research Project within the Second Affiliated Hospital of Kunming Medical University (ynIIT2023009), Key Project of Yunnan Provincial Health Science and Technology Plan (2024YNLCYXZX0526, 2024YNLCYXZX0373).

Data availability

The data are available from the corresponding author upon reasonable request.

Declarations

Competing interests

None declared.