FormalPara Key Points for Decision Makers

• Despite a variety of available treatments, <40 % of the Medicare enrollees diagnosed with hepatocellular carcinoma undergo any treatment. There is a clear survival advantage among patients who undergo any treatment versus no treatment, even after adjusting for cancer stage and other confounders.

• Surgical treatment modalities, transplantation and resection, demonstrate a clear survival advantage over the liver-directed therapies of ablation and/or transarterial chemoembolization. In turn, liver-directed therapies demonstrate survival advantage over systemic chemotherapy and/or radiation therapies.

• Untreated patients accumulated between $US23,600 and $US38,300 in medical expenses after diagnosis with hepatocellular carcinoma, depending on stage. Average cumulative medical expenditures were nearly double for systemic chemotherapy and radiation patients. Patients treated with liver-directed therapy incurred between $US69,000 and $US97,500, resection patients approximately $US125,000, and transplant patients between $US207,000 and $US244,500.

• Among stage I, II and unstaged patients, resection patients incurred the least expenses per year of life gained over its next best alternative, liver-directed therapy. Liver-directed therapy was more cost effective than radiation and chemotherapy in the latest stage (IV) patients. Transplant was least cost effective in stage II and unstaged patients, but radiation was least cost effective in stage I. Radiation and chemotherapy were the least cost-effective modalities in stage III patients.

1 Background

Hepatocellular carcinoma (HCC) is the sixth most common cancer in the world but the third most common cause of cancer-related death, with an estimated 696,000 deaths worldwide in 2008 [1]. While nearly 85 % of cases occur in developing countries, the incidence of HCC in the USA has tripled between 1975 and 2005, from 1.6 to 4.9 per 100,000 inhabitants, and it is expected to keep rising in the next few decades [2]. The median age for HCC diagnosis was 64 years between 2001 and 2006 [3]. The age distribution of HCC has shifted downward; the greatest proportional increase in HCC incidence has been among individuals 45–60 years of age [3, 4]. Yet, incidence is still highest in older patients and continues to rise [3, 5].

HCC is a complex disease associated with many risk factors that make it difficult to evaluate and treat, even though a variety of treatments, including emerging therapies, are available [6, 7]. Due to major advances, 5-year survival rates after resection can exceed 50 % and transplantation has recorded 5-year survival rates of between 50 and 70 % [6]. Ablation is considered a safe and effective therapy for unresectable HCC, and its 5-year survival rates compare well with resection among patients who fit the optimal surgical profile [6]. Transarterial chemoembolization (TACE) (TAE when excluding chemotherapy) is indicated for intermediate-staged patients with unresectable HCC, who do not qualify for ablation therapy and who have no vascular invasion or extrahepatic spread, and has been shown to improve the survival of select patients [6, 7].

Sorafenib, a multikinase inhibitor, is currently the only US FDA-approved treatment for unresectable HCC. It has emerged as the standard of care for advanced-stage disease, having been shown to prolong median overall survival and delay time to progression in two phase III randomized controlled trials [8, 9]. The theoretically synergistic combination of sorafenib and TACE is a promising new therapeutic approach for unresectable HCC [7]. Systemic therapy, chemotherapy, hormone therapy or immunotherapy, as well as radiation treatments such as external beam radiation therapy (EBRT) and selective internal radiation therapy (SIRT), have shown no conclusive results [6, 10, 11].

In view of increasing incidence and treatment options, more evidence is needed from real-world data on the relative effectiveness and cost implications of treatments. Lang et al. [12] extrapolated from utilization patterns within a subset of HCC patients in the 1999 Surveillance, Epidemiology, and End Results (SEER)-Medicare linked database to conservatively estimate annual US healthcare costs of nearly $US406 million (year 2006 values) and an average per-patient cost of $US29,354 (year 2006 values). Arterial chemoembolization, cancer-related surgical procedures, liver transplantation and chemotherapy accounted for 8.8, 8.6, 7.1 and 4.4 % of total healthcare costs, respectively. Chemotherapy accounted for a greater proportion of healthcare costs in elderly patients (65 years or older) versus younger patients (6.4 vs. 3.0 %), while other procedures contributed similar proportions.

It remains unclear how healthcare expenditures vary across HCC patients who undergo different modes of therapy. Shah et al. [5] found that, although therapy rates in the Medicare population had increased over time, <30 % of Medicare patients overall underwent non-palliative therapy, despite evidence that a greater proportion had biological characteristics favourable for treatment. The reasons why HCC therapy is underutilized are not fully understood, but it is clear that the estimated costs of HCC management are skewed by this fact. This study aims to assess the comparative and cost effectiveness of the various therapies for HCC using data on HCC patients identified in the SEER-Medicare linked database over the 2000–2007 period (Table 1).

Table 1 Characteristics of SEER-Medicare patients diagnosed with HCC within 2000–2007, by treatment modality
Full size table

2 Methods

2.1 Data Source

The National Cancer Institute SEER database collects clinical information on incident-based cancer diagnoses from cancer registries covering nearly 26 % of the US population. Data collected includes demographic characteristics, date of diagnosis, details about the cancer (e.g. histology, stage and grade) and cause of death, if applicable. Medicare enrolment and claims files from Parts A (inpatient), B (provider) and D (drugs) are linked to Medicare-enrolled SEER participants such that information can be collected regarding utilization, particularly types and timing of treatments undertaken, as well as underlying co-morbidity. The database has been found representative of the US elderly population and is described in detail elsewhere [13–15].

2.2 Cohort Definition

All subjects were initially diagnosed with primary HCC between 1 January 2000 and 31 December 2007. Patients were excluded from this study if they were not enrolled in Medicare or were enrolled in a health maintenance organization (HMO) within 12 months prior to diagnosis of HCC. The latter exclusion is to eliminate possible censoring of information due to switching between a HMO and Part A/B enrolment. Patients were also excluded if their month or year of diagnosis was unknown, if they had a history of other cancers within 5 years prior to diagnosis or if cancer was diagnosed upon death. Eligible patients were followed until week of death, or until censored due to the loss of Part A/B coverage, HMO enrolment or 31 December 2009.

2.3 Survival Analysis

Survival was calculated as the number of weeks from diagnosis until death. All-cause mortality was examined. Since the elderly population is prone to co-morbidities, we examined, alternatively, HCC-related mortality. In this case, mortality was attributed to HCC if the cause of death was indicated as ‘liver’, and all other recorded deaths were treated as censored cases.

2.4 HCC Treatment

Treatments were identified as liver transplantation, surgical resection, ablation, TACE, TAE or systemic chemotherapy using the Medicare databases and respective International Classification of Diseases, ninth revision (ICD-9) procedure or hospitalization codes. Patients were categorized as untreated if no treatment was undertaken in the follow-up period. Treated patients were categorized into five treatment modality groups according to the treatments undertaken in the follow-up period. If more than one therapy was undergone during follow-up, modality was assigned in hierarchical fashion based on the Barcelona Clinic Liver Cancer (BCLC) staging system, the most widely used system for staging and treatment [6]. Patients were assigned to the transplant modality regardless of other treatments undergone during follow-up. Similarly, all non-transplant resection patients were assigned to a treatment modality. The remaining patients were assigned to liver-directed therapy (if ablation and/or TACE were ever undergone), chemotherapy (no liver-directed therapy; chemotherapy undergone alone or prior to radiation therapy) or radiation (no liver-directed therapy; EBRT or SIRT undergone alone or prior to chemotherapy). TAE was excluded as an independent modality since no individuals received it alone. Sorafenib was not considered, since oral kinase inhibitors are covered under Medicare drug benefits, for which claims were unavailable before 2007.

The BCLC staging system links staging with treatment modalities based on variables related to tumour stage, liver function status, physical status and cancer symptoms, as well as published response rates for treatments [6]. Accordingly, we compare the effectiveness and costs of the treatment modalities among patients diagnosed at given stages of HCC. Patients were categorized as having stage I, II, III, IV or unstaged HCC at the time of diagnosis, using the American Joint Committee on Cancer (AJCC) sixth edition or the Tumor Nodes Metastasis (TNM) staging system (if prior to 2003).

2.5 Patient Characteristics

We adjusted for demographics, liver disease, and co-morbidity. Age, sex and race/ethnicity were recorded for all patients at diagnosis, and a variable for rural (vs. urban) county of residence was included. The extent of liver dysfunction and underlying co-morbidity was assessed using diagnoses for hospital or medical claims in the year prior to HCC diagnosis. Indicators for Hepatitis B or C, alcohol-related liver disease and moderate–severe liver disease were created. A Charlson Co-morbidity Index (CCI) was created, modified by excluding cancer- and liver-related morbidities (CCI = 0, 1 or >1).

2.6 Costs

Economic costs were assessed from Medicare’s perspective over the follow-up period (i.e. patient costs, including indirect costs, were not considered). Medicare-paid expenditures for every patient were ascertained from Parts A and B claims data. Part A benefits cover inpatient care, short- and long-stay hospitals, skilled nursing facilities, home health and hospice care. Part B benefits cover physician services, outpatient care, durable medical equipment, and home health in some cases. Total direct medical costs to Medicare per patient per month were calculated from diagnosis until the end of follow-up. Total Medicare expenditures may be somewhat underestimated for all patients. While several oral anticancer and anti-emetic drugs are/were covered under Medicare medical benefits, data for oral prescription drugs covered under Part D were not available prior to 2007 and therefore were not incorporated into the study. All expenditures were inflated to represent year 2011 values, reported in $US, using the Bureau of Labor Statistics’ annual average Consumer Price Index for medical care.

2.7 Statistical Analysis

The proportions of patients in each treatment modality were calculated, and Chi-squared tests were used to compare patient characteristics across the modalities. For each disease stage, Cox proportional hazards models, stratified by modalities, were conducted to estimate survival rates over time adjusted for demographics, underlying health and liver dysfunction. Non-stratified multivariate Cox proportional hazards analyses were performed to estimate the relative risk of mortality between modalities.

SEER-Medicare data are right-censored because the data are approximately 2–3 years old. Mean cumulative expenditures will be underestimated because expenses after censoring are unknown. The partitioned inverse probability weighted (IPW) least squares method was used to estimate mean cumulative Medicare expenditures by each modality/no-treatment cohort at each stage, adjusted for censoring and all patient characteristics identified [18]. This method entailed partitioning the study period into monthly intervals. Observed expenditures in each interval were weighted by the inverse probability of not being censored at the beginning of the interval, estimated by the Kaplan–Meier method. Some modalities had greater propensity for censoring, therefore IPWs were estimated for each modality and within each stage [19]. Separate IPW ordinary least squares regression analyses were estimated for each of the monthly partitions. The coefficients for each factor were summed across all partitions to obtain the cumulative, incremental expenditures associated with each factor [18, 19].

The cost effectiveness of a treatment modality could be assessed relative to no treatment as well as to each of the other modalities examined at each stage. Average years survived for each modality were calculated from the treatment-stratified Cox proportional hazards results. Incremental cost-effectiveness ratios (ICERs) were calculated as the difference in average cumulative Medicare expenditures divided by the difference in the average years survived after diagnosis. Confidence intervals for the mean years survived after diagnosis, cumulative expenditures and the ICERs were calculated using a bootstrap approach, in which all estimations were repeated 1,000 times using sampling with replacement, for each disease stage cohort.

3 Results

3.1 Descriptive Statistics

The final study sample comprised 11,047 SEER-Medicare patients diagnosed with HCC between 2000 and 2007. A majority of patients were male (66 %). Nearly two-thirds (63 %) of patients were Caucasian, 10 % African American, 13 % Hispanic and 15 % were other races/ethnicities; 7 % resided in rural counties. Most were aged 65–84 years (73 %), and just over one-third (37 %) had unstaged disease, 24 % stage I, 17 % stage IV, 14 % stage III and 9 % stage II. Less than half (40 %) had received any treatment. Among treated patients, 411 (9 %) received a transplant, 829 (19 %) a resection, 1,817 (41 %) liver-directed therapy, 924 (21 %) systemic chemotherapy and 483 (11 %) radiation.

There were significant differences in treatment modalities according to clinical characteristics of the patients. Treatment modalities (including no treatment) spanned all disease stages, albeit, consistent with BCLC staging, chemotherapy and radiation tended to be used to treat HCC at later disease stages while transplant and resection were more commonly used at early stages. Untreated patients were largely unstaged (46 %). Across all treatment modalities, a large proportion of patients had a high co-morbidity burden (CCI >1) prior to diagnosis, ranging from 27 % within resection patients to 58 % within transplant patients.

Resection patients had the fewest cases of alcohol-related liver disease and moderate–severe liver disease, while transplant patients had the largest amount of both and the highest prevalence of hepatitis C (63 %). On the other hand, the prevalence of hepatitis B, hepatitis C or liver diseases was similarly low in untreated patients and in those receiving chemotherapy.

Fewer than 12 patients aged 75 years or older, who constituted just under half of the sample, received transplant and most (68 %) received no treatment at all. Liver-directed therapy (13 %) was the treatment administered most frequently to this age group. Proportionately more patients from rural areas (68 %) received no treatment.

Unobserved heterogeneity between patients eligible and ineligible for transplantation, or with resectable and unresectable disease may confound comparisons in the subsequent analyses. To avoid confounding as much as possible, 25 patients who were diagnosed at stage III and received a transplant were excluded from the stage III analysis, and 40 resection patients were excluded from the stage IV analysis. All modalities were compared among earlier staged (I and II) and unstaged patients.

3.2 Comparative Effectiveness of Treatment Modalities

Figure 1 presents adjusted survival rates across disease stages for each of the modalities. Figure 1 in the electronic supplementary material presents the results when using HCC-related mortality. Table 2 provides the relative risks of all-cause and HCC-related mortality for each treatment modality compared with no treatment. At all disease stages, all-cause and HCC-related mortality is significantly lower for each treatment (P < 0.001). For stage I, II and unstaged patients, the most effective mode of therapy with regard to all-cause or HCC-related mortality was transplant, followed by resection, liver-directed therapy, radiation and chemotherapy. Chi-squared tests indicated that the mortality risks associated with the radiation and systemic chemotherapy modalities were not significantly different in these stages. Among stage II patients only, the risk of HCC mortality associated with liver-directed therapy was not significantly different from the risk associated with radiation (P = 0.175) and chemotherapy (P = 0.140).

Fig. 1
figure 1

Adjusted survival curves, stratified by disease stage and the following treatment modalities: transplantation, surgical resection, liver-directed therapy (ablation and/or TACE), systemic chemotherapy, radiation therapy (external radiation therapy and/or selective internal radiation therapy) and no treatment. All-cause survival is measured in weeks from HCC diagnosis and estimated using stratified Cox proportional hazards analysis. HCC hepatocellular carcinoma, TACE transarterial chemoembolization

Full size image
Table 2 Adjusted risk of all-cause and HCC-related mortality associated with treatment modalities for stages I, II, III, IV and unstaged HCC: multivariate Cox proportional hazards analyses
Full size table

The same pattern in relative effectiveness of the treatment modalities persisted among patients diagnosed at stages III and IV. The risk of mortality was lowest for resection patients (stage III), followed by liver-directed therapy, then radiation and chemotherapy patients. Again, the relative risks of all-cause mortality between radiation and chemotherapy patients were not significantly different. However, when considering HCC-related mortality, the survival benefit associated with chemotherapy in stage IV patients was significantly greater than radiation (hazard ratio [HR] 0.35, 95 % confidence interval [CI] 0.29–0.44 vs. HR 0.50, 95 % CI 0.41–0.61; P = 0.011) and not significantly different from liver-directed therapy (HR 0.31, 95 % CI 0.24–0.40; P = 0.408). Differences in estimated survival benefits between any two-way comparisons between modalities that have not been noted were statistically significant (i.e. P < 0.05 for any comparison).

3.3 Cost Effectiveness of Treatment Modalities

Table 3 reports estimates for mean years survived after diagnosis, mean cumulative Medicare expenditures and ICERs when considering all-cause mortality. The outcomes varied somewhat across the groups of patients diagnosed at different disease stages. Within each stage, average cumulative expenditures tended to be higher for treatment modalities associated with greater survival benefit. Cumulative expenditures and length of survival for patients who did not receive treatment ranged from $US23,791 (95 % CI 22,400–25,373) over approximately 5 months (0.38 years, 95 % CI 0.34–0.42) in stage IV to $US38,265 (95 % CI 32,755–44,621) over 11 months (1.06 years, 95 % CI 0.96–1.18) in stage II. Outcomes for systemic chemotherapy and radiation patients tended to be similar within each stage. For chemotherapy patients, average expenditures and survival ranged from $US48,148 (43,167–53,359) over nearly 9 months (0.71 years, 0.62–0.79) in stage IV to $US68,824 (56,670–82,488) over 16 months (1.29 years, 1.07–1.55) in stage I, while radiation patients accumulated between $US49,638 (42,361–56,910) over 7 months (0.62 years, 0.53–0.70) in stage IV to $US78,333 (55,157–105,968) over 17 months (1.39 years, 1.05–1.86) in stage II. Liver-directed therapy patients incurred between $US69,084 (59,403–78,976) over 1.32 years (1.09–1.58) of survival by stage IV to $US95,566 (88,525–102,878) over 2.67 years (2.49–2.86) in stage I.

Table 3 Adjusted mean years survived and mean cumulative medical expenditures ($US, year 2011 values) after HCC diagnosis with ICERs: estimates using all-cause mortality
Full size table

Medicare expenditures incurred by resection patients were similar across stages I, II, III and unstaged patients, while mean years survived after diagnosis tended to decrease with disease progression. Average cumulative expenditures ranged between $US123,807 (112,259–136,291) and $US126,738 (106,752–152,704), and mean years survived ranged between 2.81 years (2.23–3.46) in stage III to 4.51 years (4.01–4.98) in stage I. Transplant patients lived longer than resection patients, on average, among stage I (5.98 years, 5.16–6.73) and II patients (5.85 years, 95 % CI 3.94–7.30 vs. 3.08 years, 95 % CI 2.62–3.62), but not among unstaged patients (3.82 years, 95 % CI 3.05–4.59 vs. 3.93 years, 95 % CI 3.29–4.61). Cumulative expenditures among transplant patients were much higher than for all other modalities among stage I, II and unstaged patients; ranging between $US207,473 (180,720–240,539) in stage I to $US244,442 (186,512–317,738) if unstaged.

There was substantial overlap of the confidence intervals on the ICERs within each disease stage. Nonetheless, among stage I, III and unstaged patients, resection tended to incur the lowest cost to Medicare per additional year of survival over its next best alternative, liver-directed therapy. For stage II patients, the survival benefit of resection over liver-directed therapy (ICER $US29,736 [5,436–73,728]) was about the same as liver-directed therapy compared with radiation (ICER $US25,657 [−15,904 to 87,634]). Transplant was the least cost-effective modality in stage II (ICER $US38,337 [20,723–123,184]) and unstaged patients (ICER $US95,351 [54,304–188,436]). The benefit of transplant over resection in stage I patients cost $US55,066 (28,470–145,118) per year gained, but radiation was the least cost-effective therapy in stage I, costing $US74,404 (37,765–230,357) per year over no treatment. For stage IV patients, the cost per year of additional life associated with liver-directed therapy over systemic chemotherapy (ICER $US33,940 [18,421–55,025]) or radiation (ICER $US27,445 [11,316–44,851]) was significantly lower than the cost per year gained by chemotherapy/radiation over no treatment. A similar result was found in stage I (ICER $US25,345 [13,225–40,000]).

Table 1 in the electronic supplementary material presents survival, expenditures and ICERs estimated using HCC-related mortality only, which treats all other mortality as censored cases. The estimated magnitudes of mean years survived and cumulative expenditures as well as their confidence intervals are consistently larger than in the all-cause mortality scenario. This is expected given that the Medicare population has a greater co-morbidity burden than the general population, and therefore a sizable portion of patients are likely to succumb to a variety of illness and complications beyond just those linked primarily to HCC. Nonetheless, despite inconclusive results for the transplant modality, ICERs for resection were lower than for liver-directed therapy in stages I and II, while ICERs for resection and liver-directed therapy were similar in stage III and unstaged patients. The ICER for liver-directed therapy was still significantly lower than ICERs for radiation and chemotherapy in stage IV.

4 Discussion

This study followed 11,047 Medicare patients with HCC for up to 10 years and examined the comparative impact of largely widespread and available treatments on likelihood of survival. The results demonstrated a clear survival advantage to patients who had undergone liver transplantation or resection, even after adjusting for cancer stage, liver function and other patient characteristics. Moreover, there is a clear survival advantage among patients who had undergone any treatment versus no treatment.

Cumulative Medicare expenditures after HCC diagnosis were greater for more effective treatments. After adjusting for other confounders, on average, and depending on disease stage at diagnosis, untreated patients accumulated between $US23,600 and $US38,300 in medical expenses after diagnosis with HCC. Average cumulative medical expenditures were nearly double for systemic chemotherapy and radiation patients. Patients treated with liver-directed therapy incurred between $US69,000 and $US97,500, resection patients approximately $US125,000, and transplant patients between $US207,000 and $US244,500.

Cumulative expenditures can be attributed to longer survival or greater intensity of medical resource use associated with a treatment modality. ICERs were used to compare the modes of treatments, including no treatment, on the basis of cumulative medical expenditures per additional year of life. Considering all-cause mortality, resection patients tended to incur the lowest expenses per year of life gained over its next best alternative, liver-directed therapy, followed by liver-directed therapy over its next best alternative(s) radiation or systemic chemotherapy. Transplant tended to be the least cost effective of the modalities for stage II and unstaged patients, while radiation was least cost effective in stage I. While the relative risks of mortality and cumulative expenditures observed when examining all-cause mortality tended to persist when restricting the analysis to HCC-related mortality, fewer uncensored cases in each stage led to largely inconclusive results, overall.

The analysis is based on a prospective non-concurrent cohort study of Medicare beneficiaries who resided within SEER registry areas and were diagnosed with HCC between 2000 and 2007; followed-up through end of 2009. SEER registries are population based and are one of the most comprehensive and authoritative data sources on cancer. The general demographics of the study population reflect the general epidemiology of HCC in the Medicare population. Linked Medicare claims data allowed us to observe long-term survival in tandem with real-world healthcare expenditures across different modes of therapy or no therapy. Claims data also allowed us to identify information on liver function and co-morbidity which could be confounders for both survival and healthcare utilization. Thus, our results can be reasonably applied to the HCC Medicare population and used to inform practice and HCC management within a significant portion of the overall HCC population in the USA.

We divided the cohort into mutually exclusive therapy groups using a hierarchy based on HCC guidelines [6]. By design, this study was not an intent-to-treat analysis. Patients were not precluded from subsequent therapy use to treat tumour recurrence or palliative treatment. Nor were they precluded from receipt of bridge therapy prior to transplantation or treatment to downgrade the tumour prior to surgical resection.

It is important to moderate our message with the caveat that sorafenib (FDA approved in 2005) was not included as a mode of therapy in the analysis. Although several oral anticancer and anti-emetic drugs are covered under Part B, sorafenib is covered under Medicare Part D (drug) benefits, for which data were not available prior to 2007. Under HCC guidelines, sorafenib is the current standard of care for advanced HCC. To the extent that sorafenib is an innovative first-line alternative only for patients with advanced HCC, and does not supplant the common modes of HCC therapy included in this analysis, the addition of sorafenib to the set of therapeutic options does not diminish the relevance of our findings to the current management of HCC in the Medicare population.Footnote 1 Cumulative Medicare expenditures were underestimated to some extent, also due to the absence of information on prescription medications used during follow-up. Future studies on comparative and cost effectiveness of liver cancer treatments will benefit as data on Part D benefits are updated and released.

Age and liver disease can influence treatment decisions: very few patients over the age of 75 years had received a transplant because many patients older than 70 years are not considered suitable candidates for liver transplantation. Transplant and resection are both favoured at earlier cancer stages, but the majority of those receiving a transplant had hepatitis C, moderate–severe liver disease and/or high co-morbidity burden. Compared with transplant patients, resection patients were relatively healthier and more frequently females. Given this, the survival benefit of transplant exceeded that of all other treatments. As cancer stage advances, systemic chemotherapy and radiation become the treatments of choice, regardless of patient health status or age. Consistent with prior literature, we found that <60 % of those diagnosed did not receive any treatment [5]. Over one-third of untreated patients were diagnosed with stage I or II HCC, and untreated patients were prevalent at all ages and levels of co-morbidity and liver dysfunction. This was also consistent with the prior findings [5].

HCC is an aggressive and complex disease and remains difficult to evaluate and treat. Guidelines recommend HCC therapies based on patient prognosis, including stage. A substantial fraction of patients were unstaged (36.5 %) because there was not enough information to indicate a stage. The SEER-Medicare database is limited regarding information on more nuanced clinical data that may be used by physicians to determine severity or progression of the disease and select treatment. Furthermore, patient, physician or hospital preferences for therapy were not known and may influence treatment options. While the survival model cannot control for all unobserved heterogeneity, and treatment selection bias can never be fully excluded, we did use propensity scores to balance the untreated and individual modality samples in each strata based on stage. The propensity score models yielded little explanatory power of selection into treatment, and the results after balancing the co-variates showed no difference in comparative effectiveness. Thus, we relied on multivariate regression to adjust for any confounding related to observable factors. Furthermore, we stratified the analysis by disease stage (and unstaged patients) to mitigate, as much as possible, effects of underlying heterogeneity related to unobserved factors that may vary systematically over stages (unstaged patients). Thus, all results should be interpreted within the context of disease stage or unstaged status.

This study is the first to use SEER-Medicare linked data to estimate the cumulative medical expenditures associated with specific treatment modalities available at disease stages. Thus, it is difficult to compare our estimates of resource use with past studies using SEER-Medicare. Lang et al. [12] and White et al. [20] estimated annual per patient costs to be between $US27,200 and $US36,300 (adjusted to 2011 values) depending on initial stage. Given average cumulative expenditures, years survived and the number of patients in each modality, our estimates would translate to around $US36,100–64,600 per patient, depending on initial stage. However, our averages were computed during a more recent period (Lang et al. used 1999 data, while White et al. used 1991–2007 data). Further, survival in our sample may be somewhat longer than in other studies, given that their populations were restricted to patients aged 66 years or older and our sample is drawn from the entire SEER-Medicare population; 16.4 % of whom were <65 years of age. For instance, Nathan et al. [21] used 1988–2005 data and reported a median survival of 45 months among resection patients diagnosed at early-stage HCC. Figure 1 reports median survival in stage I resection patients of 46 months (200 weeks) and diminishing in later stages. Davila et al. reported median survival in palliative care patients in a 2000–2005 cohort to be 14, 5 and 2 months in patients who received TACE, chemotherapy and no treatment, respectively, while median survival in stage III and IV patients in our sample, was 11–14, 6–8 and 3 months, respectively.

The factors driving cumulative Medicare expenditures for patients <65 years of age may be distinctly different from older age cohorts. Patients younger than 65 years are eligible for Medicare coverage only in the case of disability or end-stage renal disease. Given the growing incidence of HCC at younger ages [3, 4], and the substantial fraction of Medicare patients with HCC in this age cohort (16.9 % in our sample), we believe that including this sub-group of patients in our analysis would add relevant data and will better inform variations in survival and corresponding Medicare expenditures across treatments for HCC. It was assumed that multivariate adjustments for age and co-morbidity were likely to control for much of the variation in expenditures attributed to disability or renal disease.

5 Conclusion

Through its use of the SEER-Medicare linked database, this population-based study adds to the evidence of clinical trials and other comparative effectiveness studies of the various treatments available to HCC patients, and can be applied to the largest segment of the HCC population in the USA. Since the data we use are continuously updated, we were able to contemporaneously observe the long-term survival of both treated and untreated HCC patients in all stages of the disease, and over a decade. We controlled for several potential confounders for survival, such as cancer stage and patient co-morbidity, including hepatitis C. This study provides additional information that may be valuable to clinical practices, healthcare organizations and governments concerned about the management and treatment of HCC.

Despite a variety of available treatments and a clear survival advantage over no treatment, <40 % of the Medicare enrolees diagnosed with HCC underwent therapy. More effective treatments incurred greater Medicare expenditures overall. Among early staged and unstaged patients, surgical resection was associated with the least Medicare expenses per year of life gained over its next best alternative, liver-directed therapy. In late stage patients, liver-directed therapy was more cost effective than radiation and chemotherapy.