Hepatocellular carcinoma (HCC) is among the most common cancers worldwide, and is associated with high morbidity and mortality rates (1). The primary treatment options for HCC include surgical liver resection (LR) and radiofrequency ablation (RFA) (2). LR involves removing the tumor and surrounding liver tissue; however, compared to RFA, LR may be associated with higher perioperative risks, including morbidity and mortality (3). RFA is a minimally invasive technique that destroys cancer cells using high-frequency alternating currents. It is often used as an alternative to surgical LR, especially in patients with small tumors or contraindications to surgery (4).

LR and RFA are considered to be effective treatments for early stage HCC (5). Recent studies have compared effectiveness and outcomes of LR versus RFA in the treatment of HCC, although with varying results. Some studies have reported that LR results in better survival rates, whereas others have described comparable outcomes between the 2 approaches (6).

Despite various studies comparing the effectiveness of LR and RFA, the findings have not consistently favored one treatment over the other. Consequently, systematic reviews and meta-analyses are needed to provide more robust evidence-based recommendations for the optimal management of HCC (7).

However, there are some limitations to previous meta-analyses, including differences in patient selection criteria, surgical techniques, and outcome measures, which may have affected the results (8). As such, this systematic review and meta-analysis aimed to provide a comprehensive evaluation of the available evidence regarding the effectiveness of LR versus RFA in the treatment of HCC and to address existing limitations in the literature.

The literature search retrieved 1790 studies. After duplicates were removed and titles and abstracts were screened, 1432 studies remained, of which 61 full-text articles were read. In total, 45 studies (14,849 patients; 7567 RFA and 7282 LR procedures) were included in the analysis ( Figure 1 ).

A summary of the 45 included studies, of which OS was reported in 39, is presented in Table 1 . The systematic review included 10 RCTs (16–25) and 35 cohort studies (26–60), with 14 cohort studies using propensity score matching (PSM). Two RCTs demonstrated a high risk of bias, 7 studies indicated some concerns regarding the risk of bias, and 1 study had a low risk of bias. Eight non-randomized studies had a serious risk of bias, and 27 studies had a moderate risk of bias ( Supplementary Figure S1 ).

LR significantly prolonged patient survival compared with RFA. The 1-, 3-, and 5-year OS rates were used to compare survival outcomes between RFA and LR. The 1-year OS for RFA and LR was similar (LR versus [vs.] RFA, OR 1.10 [95% CI 0.96 – 1.27]; P = 0.18, I² = 8%) ( Figure 2 ), while LR was associated with better 3-year OS (LR vs. RFA, OR 1.34 [95% CI 1.22 – 1.47]; P<0.00001, I² = 51%) ( Figure 3 ), and 5-year OS (LR vs. RFA, OR 1.66 [95% CI 1.49 – 1.84]; P<0.00001, I² = 42%) ( Figure 4 ) compared with RFA. The recurrence rate for LR was consistently much lower than that of RFA (OR 0.61 [95% CI 0.54 – 0.70]; P<0.00001, I² = 54%) ( Figure 5 ).

RFA demonstrated a significant advantage over LR in terms of intraoperative outcomes. Operative duration was significantly shorter in the RFA vs. LR groups (LR vs. RFA, MD 117.80 [95% CI 113.30 – 122.30]; P<0.00001, I² = 97%) ( Figure 6 ). EBL was significantly lower in the RFA group than that in the LR group (LR vs. RFA, MD 99.67 [95% CI 93.56 – 105.77]; P<0.00001, I² = 95%) ( Figure 7 ).

The short-term outcomes of RFA were better than those of LR. The RFA group experienced fewer postoperative complications than the LR group (LR vs. RFA, OR 3.35 [95% CI 2.52 – 4.45]; P<0.00001, I² = 42%) ( Figure 8 ). The postoperative length of hospital stay was consistently shorter in the RFA group (LR vs. RFA, MD 5.36 [95% CI 4.95 – 5.77]; P<0.00001, I² = 93%) ( Figure 9 ). However, mortality rates were similar between the LR and RFA groups (LR vs. RFA, OR 1.29 [95% CI 0.38 – 4.34]; P = 0.68, I² = 0%) ( Figure 10 ).

When the study by Kim (34) was excluded, the OR and 95% CI changed significantly from 1.66 (1.49 – 1.84) to 1.83 (1.64 – 2.04), indicating that the study by Kim (34) was the main source of bias ( Figure 11 ). To assess the robustness of primary outcomes, we performed comprehensive sensitivity analyses. Exclusion of studies with high risk of bias and non-propensity-score-matched cohorts consistently demonstrated superior outcomes for liver resection over radiofrequency ablation (OR 1.68, 95% CI 1.20 – 1.94; P<0.01, I²=57.7%, Supplementary Figure S2 ). Similarly, stratification by study design revealed concordant results: analysis restricted to randomized trials maintained significant advantage for resection (OR 1.60, 95% CI 1.08 – 2.37; P<0.0001, I²=40.2%), while observational studies alone yielded comparable effect sizes (OR 1.70, 95% CI 1.44 – 2.00; P<0.0001, I²=44.4%, Supplementary Figure S3 ). These methodologically distinct approaches collectively demonstrate the stability of our core findings across analytical frameworks. Trim-and-fill analysis indicated potential publication bias for the outcome of 5 year overall survival (OS), with imputation of 2 hypothetical studies reducing the HR magnitude from (LR vs. RFA, OR 1.66 [95% CI 1.49 – 1.84]; P<0.00001, I² = 42%) to (LR vs. RFA, OR 1.63 [95% CI 1.41 – 1.90]; P<0.0001, I² = 44%). While this suggests our pooled effect may overestimate LR’s benefit, the adjusted HR remained statistically significant and clinically relevant. Nevertheless, the possibility of unpublished null findings warrants caution in interpreting the magnitude of survival advantage ( Figure 12 ). Moreover, publication bias resulted in asymmetry of the funnel plot.

This systematic review and meta-analysis compared the efficacy and safety of LR and RFA for the treatment of HCC ( Table 2 ). Our analysis included 45 studies comprising 14,849 patients, of whom 7567 underwent RFA and 7282 underwent LR. Results of analysis revealed that LR significantly prolonged OS of patients with HCC compared with RFA. The recurrence rate after LR was significantly lower than RFA. The intraoperative outcomes favored RFA, with a significantly shorter operative duration, reduced EBL, fewer postoperative complications, and shorter postoperative length of hospital stay.

Our meta-analysis revealed that LR was associated with a better OS rate than RFA (61). This finding is consistent with those of several previous investigations. One possible explanation is that surgical LR offers complete tumor removal with sufficient margins to reduce the risk for recurrence (62, 63). However, RFA relies on thermal energy to destroy tumors, which may not be completely effective in eliminating HCC (64). Our results are important for clinical decision-making because they provide support for recommending LR for patients with HCC who are physically able to tolerate invasive surgical procedures.

However, RFA had a superior effect on intra- and postoperative outcomes compared with LR. Our study and several RCTs suggest that RFA minimizes operative duration and reduces intraoperative EBL (65). This finding may have important implications, especially in reducing operative risk in patients with poor liver function, performing repeated treatments, or managing more challenging lesions, such as large tumors or those located near vital structures (66). In addition, our findings demonstrated that RFA resulted in shorter hospital stays and fewer postoperative complications. These are important benefits for improving patient outcomes and reducing healthcare costs (67).

One of the strengths of our study is its large sample size, which provides robust data for the comparison between LR and RFA in the treatment of HCC. We also included high-quality studies that minimized the impact of bias and increased the reliability of the results (16). Furthermore, although the positive results from the sensitivity and publication bias analyses suggested that there may have been some degree of bias, the fact that the conclusion of the meta-analysis remained favorable for long-term survival after bias adjustment indicates that the conclusion of this meta-analysis is robust.

However, this study also had several limitations. First, the heterogeneity of the included studies may have affected the consistency of findings. Second, although we performed a subgroup analysis to reduce heterogeneity, results may have been affected due to the various surgical techniques and devices used. Although our findings demonstrate LR’s survival advantage in broad HCC populations, further research is needed to clarify its benefit in specific clinical scenarios—particularly among elderly patients, those with marginal liver reserve (Child-Pugh B), or complex tumor locations where RFA’s minimally invasive profile may offset oncologic trade-offs. Future individual patient data meta-analyses or propensity-matched cohort studies targeting these subgroups are warranted.

In conclusion, based on pooled evidence from randomized and high-quality observational studies, liver resection demonstrates superior survival outcomes compared to RFA, particularly for patients with preserved liver function and resectable tumors. However, given the inherent selection bias in non-randomized comparisons and heterogeneity in patient populations, treatment decisions should be individualized, considering comorbidities, tumor location, and local expertise. LR represents a preferred curative-intent option where clinically feasible, rather than a universal ‘first-line’ approach.