Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related deaths and the most common cause of death in patients with liver cirrhosis.1-4 The management of HCC has been made difficult by the multitude of competing therapies and the absence of clear data to guide the clinician in choosing a therapy in terms of survival and cost. In general, resection is considered first-line therapy for HCC whenever it is surgically possible.5 However, this therapy is an option only if sufficient liver function can be preserved after hepatic resection; this occurs in approximately 5% of HCC cases. When resection is not possible, an ever-increasing array of therapies may be used: these include percutaneous approaches [chemoembolization, radioembolization, alcohol injection, and radiofrequency ablation (RFA)], external beam radiation, systemic chemotherapy (sorafenib) and orthotopic liver transplantation (OLT). Among these approaches, OLT has the highest 5-year survival rate and the lowest recurrence rate.
OLT for HCC has significant limitations. Foremost is the marked shortage of available donor organs; there are too few for all patients with HCC to undergo transplantation. OLT itself and posttransplant care are extremely expensive. Lastly, OLT is beneficial for only a carefully selected group of patients with HCC and cirrhosis. Such patients have to meet the so-called Milan criteria: 1 tumor ≤ 5 cm in diameter or no more than 3 tumors ≤ 3 cm in diameter.6 The US body governing the fair allocation of organs, the Organ Procurement and Transplantation Network, has instituted a policy by which patients with HCC are prioritized for liver transplantation according to the Milan criteria. Patients with an HCC < 2 cm do not qualify for this priority, and they are scored only according to the disease state of the liver, which may not be sufficiently advanced to place them high on the transplant list.
The management of patients with compensated cirrhosis who are found to have a small HCC (<2 cm) that cannot be resected is thus called into question. OLT is currently not an option for such patients because they do not receive sufficiently high priority on the transplant list. Two management options suggest themselves. First, monitor the patient and wait for the tumor to grow to 2 cm; at that point, transplant prioritization becomes available. Second, start treating the tumor immediately with transarterial chemoembolization (TACE) or RFA. Each option has strengths and weaknesses. For the monitoring strategy, the cons include the possibility that during the waiting period, the tumor will grow beyond the Milan criteria or the patient will ultimately prove not to be a liver transplant candidate. The pros include the potential for a curative treatment associated with OLT and its good 5-year survival rate. For the immediate HCC treatment strategy, the cons include a diminished overall chance of OLT because the treated HCC is less likely to advance to a point at which it meets the Milan criteria. Immediate HCC therapy also provides no cure and increases the long-term risk of new HCC formation in the cirrhotic liver. The diminished risk of tumor progression and the improved survival associated with TACE and RFA speak in favor of immediate treatment. The aim of the present study is to compare the survival and cost-effectiveness of these different treatment strategies for HCC.
The present analysis compares the cost-effectiveness of different treatment options in a patient with a newly diagnosed HCC less than 2 cm in diameter. The patient also suffers from underlying compensated liver cirrhosis (Table 1). Two different management strategies are being considered: (1) monitoring and (2) immediate treatment (Table 2). The patient management is modeled as 2 distinct Markov chains so that we can calculate the expected cumulative cost expenditures and saved life-years associated with each management strategy. The outcomes of the analysis are expressed as ratios of costs spent to life-years saved.
Table 1. Characteristics of the Patients in the Analysis
Unequivocal diagnosis of HCC < 2 cm
Not resection candidates
Potential liver transplantation candidates
Table 2. Descriptions of the Analyzed Strategies
Immediate TACE or RFA
Monitoring every 3 months, retreatment if there is progression, and referral for OLT if the HCC meets the Milan criteria
Monitoring every 3 months without treatment
If the HCC meets the Milan criteria, treatment with TACE or RFA and referral for OLT with priority
Figure 1 presents flow diagrams of the 2 management strategies. In each model, the boxes symbolize different disease states. The arrows between boxes symbolize the flow of patients between the different states. The percentages next to the arrows depict the transition probabilities associated with each arrow or patient flow. The percentages of all arrows leaving a given state add up to 100%. The models are started with a hypothetical population of 1000 patients entering each model at time zero through the state on the far left, that is, monitoring without therapy or immediate treatment. Every 3 months, the number of patients inside the models is redistributed according to the transition probabilities associated with the various arrows or patient flows. This redistribution is carried out for a total period of 10 years or 40 cycles of 3 months. For each cycle, the model accrues the number of life-years and costs encountered by a patient staying in a given state. The calculations have been carried out with an Excel spreadsheet (Microsoft, Redmond, WA).
In the monitoring strategy, all patients with an HCC less than 2 cm in diameter are initially closely monitored by repeated computed tomography scans and laboratory work. Cancer growth is assessed according to the so-called Milan criteria for OLT. If the HCC has reached an overall diameter greater than 2 cm but less than 5 cm, the patient transitions from monitoring without therapy to the state of being within the Milan criteria, and he becomes considered a candidate for OLT. If the HCC grows rapidly and reaches an overall diameter greater than 5 cm during monitoring, the patient transitions to the state of being outside the Milan criteria. Patients in this state are no longer considered transplant candidates for OLT, and all eventually die from their liver disease. Patients stay within the Milan criteria unless they undergo transplantation or die from their liver disease; in addition, these patients receive either TACE or RFA according to the standard of care whether they are ultimately OLT candidates or not. Patients after OLT stay in the state of OLT or die for various reasons.
In the immediate treatment strategy, all patients undergo HCC therapy from the onset. In comparison with the monitoring strategy, this management strategy has the advantage of a smaller number of immediate deaths because treatment is started earlier and with smaller cancers. The disadvantage of immediate HCC therapy is the smaller number of patients who, with their HCC exactly meeting the Milan criteria, obtain the elusive opportunity to undergo OLT.
In either strategy, the patient may be decompensated as part of the natural history of cirrhosis. At this point, the patient enters the liver decompensation state and may undergo OLT because of the liver disease itself (rather than HCC). The rate of decompensation is slightly higher with the immediate treatment strategy because the treatments themselves (TACE and RFA) have a small rate of procedure-related complications resulting in decompensation. Patients who become decompensated as a result of treatment may still get a transplant.
Table 3 lists the transition probabilities used in the present models. The transitions are taken from the literature. Because the transition probabilities are frequently available for various times periods, to be used in the Markov chain model, they need to be converted into rates per cycle with the actuarial method. All actuarial conversions have been easily executed on an Excel spreadsheet with its built-in functions. Two examples for such conversions are given next.
Table 3. Transition Probabilities Used in the Models
Rate per Cycle
Progression from HCC < 2 cm to being outside the Milan criteria
First, it has been estimated that 7% of all monitored patients advance to a cancer state outside the Milan criteria within 6 months:
The following equation can be solved to calculate the monthly rate:
Placing this rate back into the first equation, we discover the following:
As expected, the cycle rate is roughly half that of the initial 6-month rate.
Second, with immediate treatment of HCC by TACE, the 5-year survival rate is 39%; in other words, 1–39% of patients transition into the state of death:
The following equation can be solved to calculate the monthly rate:
Placing this rate back into the first equation, we find the following:
When patients present with compensated cirrhosis and small HCC (and they are not resection candidates), they can be considered potential OLT candidates. Not all patients in this pool will actually be listed for OLT, and not all of those listed will undergo transplantation. For the purposes of this model, we calculate the probability that any potential OLT candidate will actually undergo transplantation. This probability is equal to the chance that an individual who is evaluated for transplantation will actually be put on the transplant list (listing rate) multiplied by the transplant rate for patients with HCC. Although transplant rates for listed patients are known, listing rates are not available, and they vary from center to center. Our center has a listing rate of approximately 50% (2005-2008). Personal communications with transplant centers on the East coast (1), in the upper Midwest (1), in the lower Midwest (1), and in the West (3) have revealed listing rates ranging from less than 50% to nearly 80% (most in the 50% range). We have thus chosen 50% as a listing rate for the purposes of this model. Patients are declined for transplantation because of financial problems, previously unrecognized medical problems, or psychosocial problems; any of these can preclude listing for transplantation. Data from the Scientific Registry of Transplant Patients indicate that patients listed for OLT in the United States with the indication of HCC undergo transplantation at a rate of 71% in 2 years.7 Thus, the chance for a potential OLT candidate to actually obtain a liver transplant for HCC equals 35.5% (50% × 71%), which corresponds to the probability used in the present model. This rate has been applied to all newcomers who enter the state of being within the Milan criteria. The transplant rate for patients listed for transplantation with an indication other than HCC (decompensated liver disease) is 49%.7 Thus, the chance for a potential OLT candidate to actually obtain a liver transplant for decompensated liver disease equals 25% (50% × 49%).
The rate of complications arising from TACE range from 2% to 5%. Death as a result of TACE occurs at a rate of 0.14% to 0.5%,8-10 whereas liver failure occurs at a rate of 0.26% to 0.7% (included in these calculations is hepatic insufficiency, liver failure, or the development of liver failure–related complications such as ascites). The rate of complications arising from RFA ranges from 9% to 11%.11-13 The majority of these complications (eg, pleural effusion, renal failure, abscesses, bilomas, and biliary fistulas) do not result in death or hepatic decompensation. The rate of dying as a result of RFA ranges from 0.3% to 0.86% (18 of 3670,11 1 of 582,12 and 3 of 60813). Survival rates after TACE and RFA take into account death as a complication of the procedure. However, liver failure as a result of TACE or RFA can result in transplantation for liver decompensation. The model accounts for this possibility by increasing the rate of liver decompensation (patients would be eligible for OLT) with the immediate treatment strategy. Patients who become decompensated as a result of TACE or RFA with the monitoring strategy are already within the Milan criteria and thus are eligible for transplantation.
The six monthly interval recommended for HCC surveillance is based on average HCC doubling time. A small percentage of liver cancers have considerably faster doubling times (see Table 6), which supports the supposition in the model that a small number of HCCs will grow out of Milan criteria even when monitored closely.
Table 4 enumerates costs taken from the paired Surveillance, Epidemiology, and End Results–Medicare database,14 which has detailed Medicare costs for the treatment of HCC. The costs include facility and physician fees as well as drugs and materials used during procedures such as computed tomography scans, TACE, and RFA. The most important Medicare costs for the purposes of this model include the costs of liver transplantation ($278,765), TACE ($24,304), and RFA ($18,386). For the cost analysis using the Markov chain, the itemized cost is multiplied by the frequency of its annual use and divided by the number of cycles per year to obtain the cost per cycle associated with various states. Patients undergoing TACE or RFA receive on average 3 or 1.5 treatment sessions per life, respectively. The average number of treatments per year is calculated via the division of the number of treatments per patient by the average length of survival in a given state (eg, receiving HCC therapy or being within the Milan criteria).
Table 4. Costs Associated with Managing HCC
Frequency per Year (or Patient)
Cost per Item
NOTE: Costs are based on the paired Surveillance, Epidemiology, and End Results–Medicare database.
TACE (per patient)
RFA (per patient)
None of the submodels includes the cost of chemotherapy in the fraction of patients with terminal HCC. We do not consider the occurrence of complications of interventional procedures or unexpected events, such as infection or gastrointestinal bleeding. In estimating the costs, we also assume an evidence-based practice that rationally and diligently uses health care resources. In medical routines, additional costs may accrue from the redundant use of diagnostic tests or the overzealous use of ineffective therapy.
Economic Analysis and Sensitivity Analysis
The time horizon of the present analysis is 10 years. The costs and life-years gained during this time period are accumulated for each strategy separately. Future costs and future life-years are discounted by an annual rate of 3%. Two strategies for calculating the incremental cost-effectiveness ratio (ICER) are compared:
where subscripts 1 and 2 refer to the first and second strategies, respectively. A strategy is considered not cost-effective if its implementation is associated with more than $50,000 to $100,000 spent per life-year gained. A strategy is also considered not cost-effective if it becomes relatively or absolutely dominated by other management strategies.15
In a sensitivity analysis, all costs and probability values built into the model have been varied over a broad range to test the overall robustness of the outcomes of our analysis.
Figure 2 depicts the flow of patients with the 2 management strategies. The number of patients in the initial state of monitoring drops precipitously within 2 years. The majority of patients transition into 1 of 2 states: being within the Milan criteria or undergoing OLT. The time-dependent rise in the fraction of deaths is similar with the 2 management strategies. The 2 fractions of patients (those within the Milan criteria and those undergoing OLT) are smaller after immediate treatment versus the monitoring strategy.
The relationship between cost and life-years gained is shown in Fig. 3. Two hypothetical strategies are included (an immediate evaluation for OLT under the assumption that no HCC size requirement for priority exists and immediate OLT for all-comers) to demonstrate the relative or absolute dominance of immediate treatment over the monitoring strategy. Table 5 presents the costs spent and life-years gained with the different strategies. The percentage of patients undergoing OLT increases from 21.4% with HCC therapy to 31.3% among patients subjected to the monitoring strategy. Costs and life-years are analyzed separately with TACE or RFA used to treat HCC. With TACE, the monitoring strategy saves more lives than the immediate treatment strategy (4.324 versus 4.269 years of life on average) but at the prohibitively high cost of $739,602 per life-year gained. Thus, when TACE is used, the immediate treatment strategy relatively dominates the monitoring strategy. When RFA is the treatment modality, the monitoring strategy is more expensive and saves fewer lives than the immediate treatment strategy; hence, the monitoring strategy is absolutely dominated by the immediate treatment strategy.6
Table 5. Costs and Life-Years Gained with Different HCC Management Strategies
Table 6. Doubling Times for Small HCC
Initial Tumor Size (mm)
Median Doubling Time (Days)
For 10 of 30 patients, the doubling time was less than 3 months.
Three of 11 patients were outside the Milan criteria in less than 188 days.
For 8 of 22 patients, the doubling time was less than 3 months.
Two of 28 patients were outside the Milan criteria within 3 to 6 months.
One of 9 patients was outside the Milan criteria in less than 6 months.
In several sensitivity analyses, we have varied the assumptions built into the models over a broad range. The relationships between the 2 strategies remain virtually insensitive to changes in the discount rate or the time horizon of the analysis. Because monitoring and immediate treatment both rely similarly on the use of TACE or RFA for the treatment of patients who do not undergo OLT, changing the number of procedures per patient, cost per procedure, and survival associated with TACE or RFA does not affect the size of the relative difference between these 2 strategies. Even with average rates of TACE (3) or RFA (1.5) per patient, the treatment cost per patient is greater than $100,000 per life-year gained. In a sensitivity analysis in which the average number of procedures per patient is increased, costs with both management strategies rise for each treatment. As the number of procedures rises, the OLT rate with the immediate treatment strategy declines because more patients become decompensated, and the transplant rate for patients with decompensated cirrhosis is lower than that for patients with HCC. For example, when the average number of TACE procedures per patient is 3, the OLT rate is 21.35%; when the number of TACE procedures increases to 10 per patient, the OLT rate drops to 20.97% with the immediate treatment strategy. This reflects the increased number of patients with procedure-related liver decompensation. A similar phenomenon occurs when RFA is the treatment modality.
The rate of OLT is the only probability value affecting the relationship (Fig. 4). As the OLT rate increases, the number of life-years gained increases with the immediate treatment and monitoring strategies. This OLT-dependent rise is more pronounced with the monitoring strategy versus immediate HCC therapy. The threshold or crossover between the 2 lines occurs at an OLT rate of 35.8% with TACE and at an OLT rate of 59.6% with RFA versus a baseline value of 35.5%. Above these 2 threshold values, the monitoring strategy may save more lives but still costs more than HCC therapy. As the OLT rate comes close to the threshold value, the ICER of monitoring versus immediate treatment trends toward infinity. More importantly, even if the rate of OLT approaches the unrealistically high value of 100%, the ICER remains considerably higher than $100,000 per life-year saved.
Our study suggests that the immediate treatment of small HCC (<2 cm) with TACE or RFA in patients with compensated cirrhosis results in similar or longer survival and decreased cost in comparison with a strategy of expectant monitoring of these patients with the intention of liver transplantation at a later tumor stage. The results of the analysis remain robust and unaffected by variations of most of the assumptions built into the model. These results are relevant in populations in which most cases of HCC occur in cirrhotic livers and OLT is readily available. In such populations, HCC is the leading cause of death among patients with cirrhosis.4, 16 Because the prevalence of cirrhosis and HCC secondary to hepatitis C virus infection is still rising,17 this issue is likely to intensify in the coming years, especially in light of the relatively stagnant rates of OLT.7
When we compare RFA and TACE as treatment modalities, it is clear from multiple studies that RFA provides superior survival benefit versus TACE and should in all cases be preferred to TACE when it can be used. We do not compare RFA to TACE in the model because of the clear and accepted superiority of RFA. Many patients, however, are not able to undergo RFA for technical reasons (eg, the tumor is near the liver capsule, it is too high in the dome of the liver, it is too close to a large blood vessel or bile duct, or there are multiple tumors), and the next best treatment for these patients is TACE.
Because cirrhosis itself is a risk factor for the development of HCC, it intuitively makes sense to remove by OLT the premalignant cirrhotic condition along with the cancer itself. However, not all patients with cirrhosis and HCC have the same propensity to develop another HCC. A recent study18 has revealed that gene expression profiles of cirrhotic livers might be able to predict the development of another de novo HCC. Therefore, a group of patients with cirrhosis and a low risk of developing a second HCC may fare just as well with effective treatment of the first HCC without OLT. Indeed, the survival data of the present model clearly indicate that immediate treatment with either TACE or RFA is superior to waiting for a potential liver transplant. It is unclear if treatment (with TACE or RFA) prior to transplantation results in less recurrence after transplantation. For the purposes of this model (which we feel describes clinical reality), we assume that patients (even if they are listed) may ultimately not get a transplant (the rate of transplantation for listed patients is approximately 75%). Given this fact, we have designed the model so that patients will undergo treatment as soon as possible (either immediately with the immediate treatment strategy or when a patient's HCC meets the Milan criteria with the monitoring strategy). In this way, patients who do not ultimately get a liver transplant (for whatever reason) do not miss the opportunity for HCC treatment at the earliest time possible when outcomes are better.
The results of this model do not apply to patients who are already on the transplant list. Most patients who are discovered to have HCC while they are on the transplant list are patients with decompensated cirrhosis who will benefit from OLT even if the HCC can successfully be treated with TACE or RFA. Patients on transplant lists are obviously much more likely to undergo OLT than patients who are potential OLT candidates. Our model also does not consider the recurrence of HCC in the absence of death. The recurrence-free survival of patients with OLT is much longer than that of patients treated with TACE or RFA. Because survival is the most meaningful endpoint for measuring the effectiveness of HCC treatment, studies of HCC management should concentrate on this outcome parameter.19 A Markov model designed to analyze the best treatment for HCC was previously described,20 but the model did not include OLT as a treatment option. Another recent Markov model was used in an attempt to determine the general strategy for managing a 1- to 2-cm nodule noted from the imaging of patients with cirrhosis.21 In contrast to the latter study, our analysis is based on the assumption that the diagnosis of HCC has already been clearly established. Such an assumption has become increasingly relevant because our general ability to establish the diagnosis of HCC has markedly improved with advancing imaging techniques.
A potential limitation of any given decision analysis is that its model does not perfectly reflect clinical reality in all its complexity. The clinical management of HCC is quite heterogeneous, varies substantially among different hepatology centers, and depends on local expertise. For example, at some centers, yttrium 90 embolization is favored as the primary treatment for small HCCs, whereas other centers favor TACE. Marked heterogeneity exists in the application of TACE itself, with several types of embolic materials or chemotherapies used. Few centers employ external beam radiation for HCC therapy. In clinical practice, patients may receive more than 1 kind of treatment for their HCC over the course of their illness. Data supporting these practices are scarce, although some centers have reported good outcomes with multimodal therapy.22 The model presented here assumes that, in addition to OLT, only 1 other form of therapy will be used per patient. The primary reason for this assumption is the relative wealth of data for survival with 1 type of therapy and the lack of reliable data for a large number of potential combinations.
Obviously, the results of the model depend on the reliability of its input data. The data on 5-year survival rates for very early HCC treated with TACE, RFA, and OLT are quite reliable and have been reproduced in various populations. Data on HCC progression rates, unfortunately, are less well validated. The values input into the present model are based on some of the best data available and are sufficient to produce meaningful results from the model. The OLT rate is varied in the models as noted in the sensitivity analyses, and this is important because listing rates likely vary between transplant centers. When the OLT listing rate for HCC is set at 100%, the monitoring strategy becomes more life-saving (6.09 versus 5.39 years with immediate TACE and 6.59 versus 6.45 years with immediate RFA) but at a high cost ($178,913 per life-year saved versus immediate TACE and $1,071,550 per life-year saved versus immediate RFA). Thus, with RFA as the HCC treatment and with an OLT listing rate of 50%, the immediate treatment strategy absolutely dominates (survival and cost) the monitoring strategy; at an OLT listing rate of 100%, the immediate treatment strategy relatively dominates (cost per life-year saved) the monitoring strategy.
The model suggests that immediate treatment with TACE would result in 10% fewer liver transplants in comparison with the monitoring strategy, whereas treatment with RFA would result in 12% fewer transplants. The application of the immediate treatment strategy would potentially free up a significant number of donor livers that could be transplanted into other recipients in need of transplantation.
In conclusion, the present survival and cost-effectiveness analysis strongly suggests that immediate treatment with TACE or RFA for small HCC is a superior strategy versus waiting for additional tumor growth to gain a higher priority score for liver transplantation. As the methods of HCC diagnosis and the non-OLT treatment of HCC improve, the superiority of immediate treatment of HCC will increase. Clinical studies for the confirmation of these findings are warranted, but they may be difficult to perform because of the rapidly evolving treatment modalities and expertise in HCC management.
The authors thank Dr. Jonathan Schwartz for his guidance at the inception of this project.