Surgical resection versus ablation for hepatocellular carcinoma ≤ 3 cm: a population-based analysis

Abstract

Background

Ablation for ≤ 3-cm hepatocellular carcinoma (HCC) has been demonstrated to be an effective treatment strategy. The present study sought to examine the outcomes of patients with ≤3 cm HCC after ablation versus resection.

Methods

Patients treated by ablation or surgical resection for ≤ 3 cm T1 HCC were identified from the National Cancer Database (2002–2011). Survival outcomes were analysed according to propensity score modelling.

Results

A total of 2804 patients underwent ablation (n = 1984) or a resection (n = 820) for solitary HCC ≤ 3 cm. Patients treated with ablation as compared with a resection had a higher frequency in alpha-fetoprotein level (AFP) elevation (46.5% versus 39.1%, P < 0.01) and the presence of cirrhosis (22.2% versus 14.5%, P < 0.01). Unadjusted overall survival (OS) at 3 and 5 years was greater after a resection (67%, 55%) versus ablation (52%, 36%, P < 0.01). After propensity score matching, the improved overall survival (OS) was sustained among the resection cohort (5 year OS: 54% versus 37%, P < 0.001). In multivariable models, a resection was independently associated with an improved OS [hazard ratio (HR): 0.62, 95% confidence interval (CI): 0.48–0.81; P < 0.01].

Conclusion

Resection of HCC ≤ 3 cm results in better long-term survival as compared with ablation. Treatment strategies for small solitary HCC should emphasize a resection first approach, with ablation being reserved for patients precluded from surgery.

Introduction

Current treatment algorithms for hepatocellular carcinoma (HCC) have been largely influenced by the Barcelona Clinic Liver Cancer (BCLC) staging system, which takes into account both patient- and tumour-related factors, when deciding the optimal therapy to pursue.¹ In the setting of early stage HCC, liver transplantation, hepatic resection (HR), or ablative therapies are considered curative approaches for this disease.² However, the relative scarcity of suitable organs has limited the dissemination of liver transplantation, which continues to result in the best long-term outcomes. Instead, most patients are offered HR or ablation as first-line therapies for early stage HCC.

Over the last decade, there continues to be an ongoing debate over the best therapeutic option, either HR versus ablation, for small HCC lesions. In the most recent treatment guidelines proposed by the American Association for the Study of Liver Diseases, HR was recommended for single HCC lesions ≤5 cm with preserved liver function, whereas ablation was proposed for multifocal HCC lesions (2–3 tumours ≤3 cm).³ However, the quality of data for which these recommendations were generated remains poor.⁴ A major limitation that has hindered the execution of well-designed randomized trials has been the difficulty in enrolling a patient cohort that would be eligible for a treatment approach and the lack of adoption of a uniform inclusion criteria.^5–7 Moreover, numerous single institution series have compared ablation to HR, but the definition of what constituted a small HCC has been variable, which probably contributed to the equivocal survival outcomes reported in the literature.^8–13 Among many of the studies, early stage HCC lesions were typically defined as a single T1 lesion up to 5 cm. However, contemporary series along with a systematic review have since proposed the 3-cm HCC tumour size as a predictive threshold for ablation efficacy.^14,15 Whether or not ablation can achieve a similar outcome as HR for tumours within the 3-cm range remains ill defined.

Aside from varying tumour sizes, retrospective studies comparing HR to ablation have also been problematic given that most patients that undergo ablation typically have altered liver function, potentially affecting the prognosis. Even with the use of multivariate analysis, comparisons between treatment groups with disparate clinicopathological characteristics may still underestimate the true treatment effect of a specific therapy.¹⁶ The objective of the present study was to compare the efficacy of ablation to HR for single HCC lesions ≤3 cm by using propensity score modelling to control for any confounders that may influence outcomes. It was hypothesized that ablation resulted in similar long-term outcomes as compared with HR for a small, single HCC.

Patients and methods

After approval by the Medical College of Wisconsin Institutional Review Board and completion of a data use agreement with the American College of Surgeons, a retrospective cohort study was conducted using the National Cancer Data Base (NCDB). Created as a joint project of the American Cancer Society and the Commission on Cancer of the American College of Surgeons, the NCDB has developed into a comprehensive oncology dataset that captures 70% of all malignancies in the United States.

Patients with HCC were identified from the NCDB from 2002 to 2011 using the following parameters: site code C22.0 (liver) and histology codes 8170–8175 (HCC). To ensure that only early stage, solitary HCC without vascular invasion were included in the study, patients with T1N0M0 (AJCC 6th/7th edn) disease were first captured. Patients were then excluded if the tumour size exceeded 3 cm. The final study cohort consisted of solitary, ≤ 3 cm HCC. The group was then further stratified by type of treatment, ablation versus resection, according to treatment codes 11–17 (ablation) and 20–90 (resection). Patients with metastatic disease, lymph node positivity, or went on to transplantation were excluded from the analysis. Data on clinicopathologic characteristics, including patient age, gender, ethnicity, Charlson's Comorbidity Score (CCS), alpha-fetoprotein level (AFP) and the presence of cirrhosis were obtained.

Statistical analyses were performed with Stata/MP 10.0 for Windows (StataCorp, College Station, TX, USA). Continuous variables were described as medians and interquartile ranges. Categorical variables were described as totals and frequencies. Comparisons between treatment arms were conducted using the Mann–Whitney U-test or chi-squared test as appropriate. Logistic models were used to evaluate the association of covariates with ablation. Alpha was set at 0.05, and all tests were two-tailed. Overall survival (OS) was defined as the number of months from diagnosis to the time of death. Kaplan–Meier methods and Cox's proportional hazards modelling were used to evaluate survival.

To control for the potential treatment effect by covariates, propensity score (PS) modelling using the greedy matching algorithm (1 : 1 match) was applied.¹⁷ A multivariate logistic model was constructed, which generated a PS score that reflected the conditional probability of undergoing ablation.

Results

A total of 2855 patients met inclusion criteria; 1984 patients (69.5%) underwent ablation versus 871 patients (30.5%) were resected. In the ablation cohort, the commonest treatments utilized were by a heat-radio-frequency approach (n = 1674, 84.4%), or percutaneous ethanol injection (PEI) (n = 72, 3.6%). Demographic and clinical characteristics for both the unmatched and propensity-matched cohort by treatment groups are described in Table1.

Table 1.

Demographic and tumour characteristics of patients with hepatocellular carcinoma (HCC) before and after propensity score matching, 2002–2011

Variable, n (%)	Total (2855)	Unmatched cohort		P-value	Propensity matched cohort		P-value
		Ablation (n = 1984)	Resection (n = 871)		Ablation (n = 871)	Resection (n = 871)
Median Age (IQR)	61 (55–70)	61 (55–70.5)	61 (55–69)	0.267	61 (55–70)	61 (55–69)	0.717
Gender, n (%)
Male	2007 (70.1)	1388 (70)	619 (71.1)	0.551	608 (69.8)	619 (71.1)	0.564
Female	848 (29.9)	596 (30)	252 (28.9)		263 (30.2)	252 (28.9)
Race, n (%)
White	2006 (70.2)	1463 (73.7)	543 (62.3)	<0.001	560 (64.3)	543 (62.3)	0.682
Black	396 (13.9)	246 (12.4)	150 (17.2)		145 (16.7)	150 (17.2)
Other	453 (51.9)	275 (13.9)	178 (20.5)		166 (19)	178 (20.5)
Charlson's Comorbidity Score, n (%)
0	1293 (45.3)	892 (45)	401 (46)	0.045	404 (46.4)	401 (46)	0.976
1	845 (29.6)	568 (28.6)	277 (31.8)		270 (31)	277 (31.8)
≥2	658 (23)	485 (24.4)	173 (19.9)		175 (20.1)	173 (19.9)
Unknown	59 (2.1)	39 (2)	20 (2.3)		22 (2.5)	20 (2.3)
Alpha-fetoprotein level, n (%)
Normal	760 (26.6)	513 (25.9)	247 (28.4)	0.001	233 (26.7)	247 (28.4)	0.414
Elevated	1262 (44.2)	922 (46.5)	340 (39)		328 (37.7)	340 (39)
Unknown	833 (29.2)	549 (27.7)	284 (32.6)		310 (35.6)	284 (32.6)
Cirrhosis, n (%)
No	188 (6.6)	85 (4.3)	103 (11.8)	<0.001	84 (9.7)	103 (11.8)	0.335
Yes	571 (20)	441 (22.2)	130 (14.9)		131 (15)	130 (14.9)
Unknown	2096 (73.4)	1458 (73.5)	638 (73.3)		656 (75.3)	638 (73.3)
Median Tumor Size, cm (IQR)	2.2 (1.7–2.6)	2.2 (1.8–2.6)	2.2 (1.7–2.7)	0.824	2.2 (1.8–2.6)	2.2 (1.7–2.7)	0.940

IQR, Interquartile Range.

In multivariate models, patients with an elevated AFP [odds ratio (OR) 1.4, 95% confidence interval (CI): 1.14–1.72, P < 0.001] or underlying cirrhosis (OR 4.03, 95% CI: 2.83–5.73, P < 0.001) were independently associated with ablation (Table2). Black (OR 0.58, 95%CI: 0.46–0.73, P < 0.001) and other ethnicities (OR 0.56, 95%CI: 0.45–0.70, P < 0.001) were less likely to undergo ablation. In light of the marked clinicopathological differences between the treatment cohorts, PS modelling was applied. Variables adjusted in the model included patient age, gender, race, CCS, AFP level and the presence of cirrhosis. The final propensity matched cohort resulted in treatment groups with similar baseline clinicopathological variables. As reflected in Table1, PS methods identified 1742 patients [Ablation: 871 patients (50%), Resection: 871 patients (50%)] with similar baseline characteristics.

Table 2.

Multivariate Logistic Regression of factors associated with receipt of ablation

Variable	OR (95% CI)	P-value
Age at diagnosis	1.01 (0.99–1.01)	0.141
Gender
Male	Ref	–
Female	1.00 (0.83–1.20)	0.992
Race
White	Ref	–
Black	0.58 (0.46–0.73)	<0.001
Other	0.56 (0.45–0.70)	<0.001
Alpha-fetoprotein level
Normal	Ref	–
Elevated	1.40 (1.14–1.72)	<0.001
Unknown	0.94 (0.76–1.17)	0.595
Charlson Comorbidity Score
0	Ref	–
1	0.89 (0.73–1.07)	0.223
≥2	1.17 (0.94–1.45)	0.161
Unknown	1.07 (0.60–1.90)	0.812
Cirrhosis
No	Ref	–
Yes	4.03 (2.83–5.73)	<0.001
Unknown	2.78 (2.04–3.78)	<0.001

Overall survival

Unadjusted OS rates at 1, 3 and 5 years were 84%, 52% and 36%, respectively, for patients that underwent ablation, versus 84%, 67% and 55% of resected patients (P < 0.001). After PS matching, OS remained statistically significant between the two treatment groups (OS at 1, 3, and 5 years: Ablation 84%, 52%, 36%, Resection 84%, 66%, 54%; P < 0.001, Fig.1). After multivariate adjusted Cox's proportional hazards modelling, a resection demonstrated a persistent survival benefit as compared with ablation [hazard ratio (HR): 0.62, 95% CI 0.48–0.81, P < 0.001, Table3]. Poor prognostic factors included older age (HR 1.03, 95% CI 1.02–1.04, P < 0.001) and underlying cirrhosis (HR 1.95, 95% CI 1.01–3.74, P = 0.046). After excluding patients that underwent PEI and other forms of ablative therapies, subgroup analysis of only those treated by a heat-radio-frequency approach (n = 732, 84%) demonstrated a similar survival disadvantage when compared with the resection cohort (OS at 1, 3 and 5 years: 84%, 49% and 36%, P < 0.001).

Figure 1.

Overall survival comparing surgical resection with ablation for hepatocellular carcinoma (HCC) ≤ 3 cm after propensity score matching

Table 3.

Cox's proportional hazards regression analysis of variables associated with overall survival in the propensity matched cohort

Variable	Univariate		Multivariate
	HR (95% CI)	P-value	HR (95% CI)	P-value
Age at diagnosis	1.03 (1.02–1.04)	<0.001	1.03 (1.02–1.04)	<0.001
Gender
Male	Ref	0.941
Female	1.01 (0.78–1.31)
Ethnicity
White	Ref	–	Ref	–
Black	0.98 (0.70–1.36)	0.895	1.13 (0.80–1.59)	0.478
Other	0.55 (0.38–0.78)	<0.001	0.53 (0.37–0.76)	0.001
Alpha-fetoprotein level
Normal	Ref	–	Ref	–
Elevated	0.97 (0.65–1.44)	0.884	1.08 (0.73–1.61)	0.695
Unknown	0.97 (0.68–1.38)	0.855	0.87 (0.60–1.26)	0.469
Charlson Comorbidity Score
0	Ref	–	Ref	–
1	1.09 (0.80–1.47)	0.580	1.04 (0.76–1.42)	0.796
≥2	1.04 (0.72–1.49)	0.838	0.94 (0.64–1.37)	0.740
Unknown	0.98 (0.64–1.50)	0.919	1.30 (0.82–2.03)	0.256
Cirrhosis
No	Ref	–	Ref	–
Yes	1.94 (1.03–3.68)	0.041	1.95 (1.01–3.74)	0.046
Unknown	1.39 (0.86–2.25)	0.177	1.51 (0.92–2.50)	0.103
Treatment
Ablation	Ref	–	Ref	–
Resection	0.61 (0.47–0.79)	<0.001	0.62 (0.48–0.81)	<0.001

HR, hazards ratio.

Discussion

Evolving therapies for HCC have contributed to the increasing complexity surrounding the management of the disease. As highlighted by the BCLC staging system, both patient- and tumour-related factors have a major influence on the type of treatment pursued.^3,18 Aside from liver transplantation as an additional treatment option, in the setting of early stage HCC, a hepatic resection and ablation remains the cornerstone of therapy. However, there continues to be an ongoing debate as to which therapy is most effective for single, early stage HCC tumours. In this large, population-based cohort, patients with unifocal ≤3 cm HCC tumours experienced a significant survival benefit if treated by resection as compared with those that underwent ablation. Although the data reveals that ablation is more commonly performed in patients with poor performance status or the presence of cirrhosis, propensity score modelling enabled us to control for confounders, while demonstrating a sustained treatment effect by resection that was superior to ablation.

Since its introduction, attempts at establishing a tumour size threshold that could accurately predict outcomes after ablation has resulted in a spectrum of size cutoffs. Sizes up to 10 cm have been reported, but the reasoning for these breakpoints was variable.¹⁹ In recent years, ablation series have proposed a 3-cm tumour size as a predictive breakpoint, which was based on data specific to the ablation technology, and was further supported by clinical outcomes.^14,20 Studies utilizing microwave ablation have demonstrated a mean maximal ablation diameter of approximately 5.5 cm when using a multiple antennae approach. For tumours up to 3 cm, this would translate into roughly a 1 cm rim of normal liver parenchyma incorporated into the ablation zone. Furthermore, the study by Groeschl et al. utilized the Surveillance, Epidemiology, and End Results (SEER) Database to capture all patients with singular HCC tumours that underwent ablation (all types). After stratifying the patients by 1 cm tumour size intervals, the authors were able to demonstrate that disease-specific survival did not vary significantly until tumours were >3 cm. Therefore, when attempting to compare systematically treatment efficacy between ablation versus resection, it would be logical that an upper limit of 3 cm for HCC tumours be used among studies. However, as recently demonstrated in two meta-analyses, which compared radiofrequency ablation with resection for early stage HCC, the definition of what constituted a small HCC remained heterogenous.^21,22 Moreover, after pooling the results of 24 studies, Wang et al. found that overall and disease-free survival was longer for the resection cohort as compared with patients that underwent ablation when tumours were greater than 3 cm (5 year OS: OR 0.43, 95%CI 0.25–0.73, P = 0.002).

In the present study, a predefined breakpoint of 3 cm was utilized to minimize the potential for an incomplete ablation associated with larger tumours. Furthermore, only patients with T1N0M0 (solitary) HCC were identified in an attempt to limit the influence of tumour biology. When comparing the two cohorts, unadjusted OS was significantly different, favouring a resection over ablation (5 year OS: 55 versus 36%, P < 0.001). However, similar to other observational series, ablation was more frequently performed in patients with a poorer performance status, or a worse liver function, highlighting the treatment bias that occurs with therapy selection.^23,24 Thus, propensity score modelling was applied, creating treatment cohorts with similar clinicopathological variables. Despite adjusting for confounding variables, there remained a persistent survival advantage among patients receiving HR as compared with ablation, suggesting HR to be the most effective treatment modality among the two.

A major advantage of large population-based data, such as the NCDB, is its ability to capture a large cohort, which in turn improves a study's generalizability.¹⁶ However, when attempting to compare the efficacy of two therapies, several individual nuances specific to each treatment, which may have an impact on outcomes, are often unable to be accounted for in these registries. If a tumour was located in the periphery versus centrally located, its proximity to major bile ducts, a patient's body habitus, or the provider's experience has the potential to influence the outcome of a treatment. Unlike HR, whereby tumour removal can be confirmed by inspection of the gross specimen, completeness of ablation is dependent on the familiarity of ultrasound interpretation. Studies have demonstrated a significant learning curve associated with ablation therapy.²⁵ Hildebrand et al. found that ablation rates and survival outcomes were correlated with the number of ablations performed by the operator. Furthermore, the NCDB is unable to account for additional operations/procedures a patient may undergo after the index therapy. Recurrence data are currently unavailable. As such, the present study is unable to comment on the role of additional therapies and their impact on survival.

An additional area lacking in the NCDB is the assessment of underlying liver disease, which continues to play a critical role not only in prognosis, but also in treatment selection. At present, the NCDB identifies cirrhosis as either being present or absent. The severity of cirrhosis as described by the Child–Pugh Classification System is not currently captured. Moreover, site-specific variables introduced by NCDB, such as cirrhosis, are often subject to incomplete reporting as demonstrated by > 70% of the study cohort having an unknown status. Whether the inability to account for these variables are enough to explain the significant difference in survival rates between the two treatments remains ill defined.

Lastly, all types of ablation (i.e. RFA, MWA, ethanol and cryoablation) were captured in this study, which may have influenced the outcomes. However, the vast majority of procedures were coded as ‘Heat-Radio-Frequency’ ablation (84%), which studies have demonstrated to be the most effective among the various ablation techniques.²⁰ Moreover, in the subgroup analysis, of only patients that underwent Heat-Radio-Frequency ablation, there was no statistical change in the survival difference between ablation compared with a resection. Unfortunately, owing to coding limitations, the present study is unable to address the relative efficacy pertaining to a specific type of thermal ablation (MWA versus RFA).

In conclusion, the present study provides further evidence supporting HR over ablation for patients with small, unifocal, HCC tumours. Numerous factors, both patient and tumour specific, continue to have a major influence on treatment allocation. Therefore, it would be short-sighted to suggest that ablation has no role for this subset of HCC tumours. Instead, in the setting of a good surgical candidate, treatment strategies should continue to emphasize a HR first approach for single, <3 cm HCC tumours, with ablation being reserved for patients less suited for the operating room.

Conflict of interest

None declared.

Funding sources

None.