Hepatic resection versus radiofrequency ablation for very early stage hepatocellular carcinoma: A Markov model analysis

Authors


  • Potential conflict of interest: Nothing to report.

Abstract

No adequate randomized trials have been reported for a comparison between hepatic resection (HR) versus radiofrequency ablation (RFA) for the treatment of patients with very early stage hepatocellular carcinoma (HCC), defined as an asymptomatic solitary HCC <2 cm. For compensated cirrhotic patients with very early stage HCC, a Markov model was created to simulate a randomized trial between HR (group I) versus primary percutaneous RFA followed by HR for cases of initial local failure (group II) versus percutaneous RFA monotherapy (group III); each arm was allocated with a hypothetical cohort of 10,000 patients. The primary endpoint was overall survival. The estimates of the variables were extracted from published articles after a systematic review. In the parameter estimations, we assumed the best scenario for HR and the worst scenario for RFA. The mean expected survival was 7.577 years, 7.564 years, and 7.356 years for group I, group II, and group III, respectively. One-way sensitivity analysis demonstrated that group II was the preferred strategy if the perioperative mortality rate was greater than 1.0%, if the probability of local recurrence following an initial complete ablation was <1.9% or if the positive microscopic resection margin rate was >0.3%. The 95% confidence intervals for the difference in overall survival were −0.18–0.18 years between group I and II, 0.06–0.36 years between group I and III, and 0.13–0.30 years between group II and III, respectively. Conclusion: Primary percutaneous RFA followed by HR for cases of initial local failure was nearly identical to HR for the overall survival of compensated cirrhotic patients with very early stage HCC. (HEPATOLOGY 2010.)

Very early stage hepatocellular carcinoma (HCC), defined as an asymptomatic solitary small HCC <2 cm, can be an ideal indication for hepatic resection (HR) because of the low potential risk of microscopic seeding.1 According to the American Association for the Study of Liver Diseases guidelines, HR is the treatment of choice for patients with very early stage HCC.2

Although several observational studies and suboptimal randomized controlled trials have demonstrated that radiofrequency ablation (RFA) was comparable to HR with regard to the overall survival of patients with early stage HCC,3–9 few comparative studies have been reported for very early stage HCC between HR and RFA. To obtain a definitive conclusion, it would be mandatory to perform a well-designed randomized trial concerning overall survival. However, an adequate randomized controlled trial would require enrollment of an enormous sample size of several thousands of patients.

In this study, instead of performing a real randomized controlled trial, a simulated randomized trial was performed to compare the overall survival of compensated cirrhotic patients with very early stage HCC treated with HR, RFA, or the combined approach of primary RFA followed by HR for cases of initial local failure.

Abbreviations

HCC, hepatocellular carcinoma; HR, hepatic resection; RFA, radiofrequency ablation.

Patients and Methods

Study Purpose.

We tried to compare HR and percutaneous RFA for the treatment of compensated cirrhotic patients with very early stage HCC by using a Markov model wherein the primary endpoint was overall survival. In this study, the presence of asymptomatic single HCC tumor <2 cm in the absence of portal vein invasion or extrahepatic disease was defined as very early stage HCC according to the Barcelona Clinic Liver Cancer staging system.2

Computerized Simulation Methods.

We created a multistate Markov model that simulated a randomized trial for the treatment of compensated cirrhotic patients with very early stage HCC. Each hypothetical patient was randomly assigned to undergo HR (group I), primary percutaneous RFA followed by HR for cases of initial local treatment failure (group II), or percutaneous RFA monotherapy (group III), and 10,000 patients were allocated to each group. In group III, patients with initial tumor control failure did not undergo any further interventions and were transited to a state of progressive HCC. During the follow-up periods, all patients with recurrent HCC were considered candidates for RFA, regardless of the previous treatment modalities.

For this Markov model, 14 states of health were defined, seven states for the cohort of patients undergoing HR and the remaining seven states for patients treated with RFA (Fig. 1). For each state of health, the probability of transition into other states was determined according to the values extracted from the literature (Table 1). In two Markov states, patients could stay longer than one cycle, which were a tumor-free state and progressive HCC state, respectively. The cycle length of the model was set to be 1 year. Half-cycle correction was used under the assumption that each transition happened halfway during the cycle.10 The Markov cycle was assumed to be repeated for 15 cycles, as long-term mortality rates over 15 years have not been reported for cirrhotic patients in the literature.11 A commercially available software product (TreeAge Pro; TreeAge Software, Williamstown, MA) was used to generate a Markov model.12 The model used a normal distribution for continuous variables and a beta distribution for ratios. The declining exponential approximation of life expectancy was used to calculate the annual mortality rates from the median survival of pertinent published Kaplan–Meier curves.13

Figure 1.

Simplified outline for the Markov state transition model used in this study. Each circle represents a state of health. After the local tumor control failed initially, the patients were assumed to be transited to a state of progressive HCC (group III), or to receive subsequent hepatic resection (group II), which was not depicted on the diagram due to the complexity of the flow chart. The repeatability of RFA for local recurrence or needle tract seeding was assumed to be the same as that for remote intrahepatic recurrence of HCC. The repeatability of recurrent HCC was assumed to be same for each patients treated with HR or RFA. Straight arrows represent the changes that may occur during each cycle or a very short time interval. In contrast, circular arrows mean that the patients may remain in the same Markov state for more than one cycle.

Table 1. Estimated Values of the Variables Used for the Markov Model Extracted from the Literature
VariableHRRFA
Annual mortality rate of general population (65 years old)0.05116
Annual mortality rate of cirrhotic patients0.02217
Median survival time (in years) for progressive HCC1.73 (range, 1.16–1.73)28, 29
Probability of procedure-related mortality0.008 (0–0.05)19–250.001 (0–0.005)26, 27
Probability of initial tumor control failure0 (0.02–0.10)30–320.0183 (99% CI 0–0.041)
Probability of needle tract seeding during RFA0.002 (0–0.125)41–43
Probability of local recurrence following initial complete ablation0.0093 (99% CI 0–0.025)
Probability of distant intrahepatic recurrent HCC within 5 years0.734–36
Probability of performing RFA for recurrent HCC0.6 (0.68–0.80)39, 400.6 (0.48–0.8)4, 37, 38

Sensitivity analyses were performed by changing the variable values used in the model to identify those values that had the greatest effect on survival. One-way sensitivity analysis was performed to evaluate the effects of changing each single variable value, while the values of other parameters remained constant. Subsequently, two-way sensitivity analysis was performed to evaluate the effect of simultaneous changes of two variable values, whereas the values of other parameters remained constant. Finally, a second-order Monte Carlo probabilistic sensitivity analysis was performed to evaluate the uncertainties associated with the parameter estimations altogether.12, 14

Selection of Trials for Parameters Estimation.

We selected all articles published as abstracts or full papers in English from 1978 to July 2009 in peer review journals that assessed a survival benefit or tumor response derived from HR or percutaneous RFA as a primary treatment of early or very early stage HCCs. All of the estimates of the variables used in this model were extracted by a systematic review of published articles (Table 1). Whenever possible, the estimates were extracted from randomized trials and, if not possible, from quasirandomized trials, prospective cohorts, retrospective cohort studies, and case series in that order.12

Studies were identified by searching MEDLINE on PubMed, the Cochrane Library database and CANCERLIT (National Cancer Institute) using “hepatocellular carcinoma,” “liver cancer,” or “primary liver carcinoma” as common text words combined with “resection,” “hepatectomy,” or “radiofrequency ablation.” This search was supplemented by manual research and review of reference lists. We were not masked to authors, institutions, journals, or interventions while we selected trials or extracted the data.15

Summary of Data from the Literature and Assumptions.

As HR is the present standard therapy for very early stage HCC, we assumed the best scenario for HR and the worst scenario for RFA in the parameter estimations. The age of patients in the cohort of this study was assumed to range from 45 years to 75 years, while the mean age of patients was assumed to be 65 years.3, 12 The estimated annual mortality rates were calculated as the sum of the annual mortality of the general population and the liver-related annual mortality of cirrhotic patients, respectively. The non–liver-related annual mortality rate for the hypothetical cohort was assumed to be equal to that of a 10-year younger generation in the general population (see Supporting Information for details).16 The best reported 10-year survival rate for cirrhotic patients was 80%.17 Under the assumption that half of cirrhotic patients died of cancer,18 the tumor-free liver-related mortality rate of compensated cirrhotic patients was estimated as 1.1%, while extrapolated for the entire period of follow-up in this Markov model. We estimated the procedure-related mortality of each procedure19–27 and the annual mortality of progressive HCC28, 29 under the assumption of a beta distribution (see Supporting Information for details).

For patients characterized by microscopic tumor infiltration of the resection margin (R1), it was assumed that no further interventions were possible because of progressive HCC.12 In the literature, the R1 rates were reported to range between 2% and 10% for patients with early stage HCC,30–32 but no data has been available for very early stage HCC. We assumed the R1 rate as 0% for very early stage HCC, reflecting that microscopic tumor seeding is not very frequent for this stage of HCC.1, 33

There was only one article identified that evaluated local tumor response of patients with solitary small HCCs <2 cm treated with primary percutaneous RFA.3 As there was a chance probability of favorable outcomes for RFA due to a sampling error, we assumed the highest value within the 99% confidence interval for the initial tumor control failure and the local recurrence rates derived from the data in this article, which were calculated as 4.1% and 2.5%, respectively.

The incidence of intrahepatic recurrence distant from the original tumor has been known to be at least 70% during the 5-year follow-up periods, and the annual incidence of recurrence was estimated from a declining exponential approximation.34–36 Although the rates to treat recurrent HCC by RFA have been reported to be somewhat variable,4, 37–40 the variation seems to originate from a random effect or a selection bias.10 We assumed the same rate for both patients treated with HR or RFA.

Needle tract seeding is also a well known complication of RFA.41–43 However, over half of tumor seeding cases have been successfully treated with local procedures.41, 42 To simplify the Markov model, the repeatability of RFA was assumed as 60% for a local recurrence or needle tract seeding, which was the same as that for remote intrahepatic recurrence (Table 1).

The validity of our model was tested by estimating the mortality data from the literature (see the Supporting Information for details).3, 44–46

Results

Expected Values for Overall Survival.

With the preset values listed in Table 1, the expected values of overall survival were 7.577 years, 7.564 years, and 7.356 years in group I, group II, and group III, respectively. The expected 5-year overall survival rates were calculated as 62.5%, 62.3%, and 60.3% for group I, group II, and group III, respectively (Fig. 2).

Figure 2.

Overall survival curves of compensated cirrhotic patients with very early stage HCC, stratified according to the treatment options. The patients were treated with HR (group I), primary percutaneous RFA followed by HR for cases of initial local treatment failure (group II), or percutaneous RFA monotherapy (group III). Note that the survival curves of group I and II were nearly identical and were slightly better than that of group III.

One-Way and Two-Way Sensitivity Analysis.

One-way sensitivity analysis for age demonstrated that group I was the preferred strategy for all ages of patients from 30 to 80 years when other variables values remained constant (Supporting Fig. 1). However, group II was the preferred strategy if the perioperative mortality rate was >1.0%, if the probability of local recurrence following initial complete ablation was <1.9%, if RFA could be performed for a recurrent HCC at least 70.2% of the time, if the median survival of patients with progressive HCC was >2.73 years, or if the R1 rate was >0.28% (Table 2). Other variables did not alter the preferred treatment option from HR. The analysis also demonstrated that group II was always superior to group III for overall survival and that group III was the preferred strategy over group I if the perioperative mortality rate was >3.8% or if the R1 rate was >4.2%.

Table 2. One-Way Sensitivity Analysis: List of Variables and Respective Threshold Values Influencing the Treatment Strategy Between HR Versus Primary RFA Followed by HR for Cases of Initial Local Treatment Failure
VariableThreshold
Median survival (in years) of progressive HCC2.73
Probability of perioperative mortality1.0%
Probability of local recurrence following RFA1.9%
Probability of positive microscopic resection margin0.28%
Probability of performing RFA for recurrent HCC70.2%

Two-way sensitivity analysis demonstrated that the overall survival of patients with a perioperative mortality rate of 1% for group I was the same as that of patients with a local recurrence rate of 2.5% for group II, when other variables values remained constant at preset values (Fig. 3). The analysis also showed that a 3% increase in the local recurrence rate following RFA was equivalent to a 1% increase in the perioperative mortality rate following HR concerning the effect on overall survival. This finding was due to the fact that many of the local recurrent tumors could be successfully treated with repeated RFA.

Figure 3.

Two-way sensitivity analysis of the probability of perioperative mortality and the local recurrence rate for the overall survival of compensated cirrhotic patients with very early stage HCC. The selection of treatment options was between group I and group II. Group III was not selected in any cases, because group II was always superior to group III. Note that the analysis demonstrated that a 3% increase of the local recurrence rate following RFA was comparable to a 1% increase of perioperative mortality with regard to overall survival.

Tornado diagrams showed that the overall survival outcomes were most sensitive to the probability of remote intrahepatic recurrence or the repeatability of RFA for recurrent HCCs, which were not related to the initial treatment options (Supporting Fig. 2). In contrast, survival outcomes were less sensitive to variables related to initial treatment options such as perioperative mortality or local recurrence.

Second-Order Monte Carlo Simulation.

The probability distributions of overall survival for the cohort in this study demonstrated that the expected overall survival for group I and group II were nearly identical, but were longer than that for group III (Fig. 4). The 95% confidence intervals were 7.18–7.96, 7.15–7.94, and 6.96–7.73 years for group I, group II, and group III, respectively. The 95% confidence intervals for the difference in overall survival were −0.18–0.18 years between group I and II, 0.06–0.36 years between group I and III, and 0.13–0.30 years between group II and III. The difference between group I and II was insignificant (P = 0.309), but the difference between group I and III and between group II and III was statistically significant (P = 0.003 and P = 0.000, respectively) (Supporting Fig. 3).

Figure 4.

The probability distribution of the overall survival for the cohort of compensated cirrhotic patients treated by (A) HR (group I), (B) percutaneous RFA followed by HR for cases of initial tumor control failure (group II), or (C) percutaneous RFA monotherapy (group III), as simulated by a second-order Monte Carlo simulation. Note that the probability distribution was nearly identical between group I and group II.

This apparent discrepancy occurred because the overall survival outcomes were more sensitive to variables not related to the initial treatment options. More importantly, this finding indicates that the preference for the treatment strategy between groups I/II and group III would not be affected by the uncertainties in the parameter estimations.

Discussion

Usually, RFA is inferior to HR in terms of local recurrence, which is known to be a significant adverse prognostic factor for survival.3 However, if initial local treatment failure following RFA can be further treated by HR, as in group II, survival outcomes of patients treated with primary RFA may be improved.3 The results of this study matched such an expectation, while group I and group II were nearly identical concerning overall survival of patients with very early stage HCC. Considering that RFA is much less invasive as compared with HR, this study highly suggests that RFA may deserve to be considered as a primary treatment for very early stage HCC.

In this study, the best scenario for HR and the worst scenario for RFA were assumed. First, we assumed no cases of microscopic tumor infiltration of the resection margin for very early stage HCC. Second, the initial tumor control failure rate and the local recurrence rate following RFA were assumed as the highest values within the 99% confidence intervals. Third, the annual mortality rate of cirrhotic patients due to liver-related disease was assumed to be constant, at a low rate of 1.1% during the entire period of follow-up. HR is relatively advantageous as compared with RFA concerning overall survival when the annual mortality rate is low.12 However, even with this scenario, the overall survival outcomes of group I and group II were nearly identical.

Limitations of this study are as follows. First, the data for the Markov model were not extracted from studies for very early stage HCC because of lack of information. Considerable uncertainties could exist concerning the parameter estimations associated with the simulated model. However, the second-order Monte Carlo simulation showed that the preference among the groups would not be changed by the uncertainties in the parameter estimations. Second, there was only a single study for RFA regarding treatment of solitary small HCCs <2 cm, and the favorable clinical outcomes for RFA might have been a result of a sampling error. However, we assumed the highest values within the 99% confidence interval as the preset values of local tumor control to minimize the sampling error. Third, both HR and RFA may not be feasible in a number of patients and that this may influence the treatment selection. Especially, RFA may not be applicable to all of the single small HCCs because of difficulties to access tumors or an expectation of severe adverse effects.47 Masses within 5 mm from the liver hila or common bile duct are not usually indicated for RFA.47, 48 Fourth, a subcapsular location or presence of a large vessel contiguous with a tumor are known to be significant adverse prognostic factors for local recurrence.49, 50 For tumors with a high risk of recurrence, RFA may not be considered as a primary treatment modality. Finally, the pathological information obtained at resection, such as satellite formation and/or microvascular invasion, may allow enlisting for rescue transplantation because of risk of recurrence and this is not feasible with RFA.2, 51 For the treatment of potential candidates for transplantation, resection can be more advantageous when compared with RFA.

In conclusion, primary percutaneous RFA followed by HR for cases of initial local treatment failure was nearly identical to HR regarding overall survival of compensated cirrhotic patients with very early stage HCC, even with the best scenario for HR and the worst scenario for RFA. We hope that this study will be helpful for further investigations concerning survival outcomes of early stage HCC.

Ancillary