Cost-effectiveness of screening for hepatocellular carcinoma in patients with cirrhosis due to chronic hepatitis C

Authors


Dr Otto S. Lin, C3-Gas, Gastroenterology Section, Virginia Mason Medical Center, 1100 Ninth Avenue, Seattle, WA 98101, USA.
E-mail: otto.lin@vmmc.org

Summary

Background : Screening for hepatocellular carcinoma in cirrhotic patients using abdominal ultrasonography and alpha-foetoprotein levels is widely practiced.

Aim : To evaluate its cost-effectiveness using a Markov decision model.

Methods : Several screening strategies with abdominal ultrasonography or computerized tomography and serum alpha-foetoprotein at 6–12-month intervals in 40-year-old patients with chronic hepatitis C and compensated cirrhosis were simulated from a societal perspective, resulting in discounted costs per quality-adjusted life-year saved. Extensive sensitivity analysis was performed.

Results : For the least efficacious strategy, annual alpha-foetoprotein/ultrasonography, the incremental cost-effectiveness ratio (vs. no screening) was $23 043/quality-adjusted life-year. Biannual alpha-foetoprotein/annual ultrasonography, the most commonly used strategy in the United States, was more efficacious, with a cost-effectiveness ratio of $33 083/quality-adjusted life-year vs. annual alpha-foetoprotein/ultrasonography. The most efficacious strategy, biannual alpha-foetoprotein/ultrasonography, resulted in a cost-effectiveness ratio of $73 789/quality-adjusted life-year vs. biannual alpha-foetoprotein/annual ultrasonography. Biannual alpha-foetoprotein/annual computerized tomography screening resulted in a cost-effectiveness ratio of $51 750/quality-adjusted life-year vs. biannual alpha-foetoprotein/annual ultrasonography screening.

Conclusions : Screening for hepatocellular carcinoma is as cost-effective as other accepted screening protocols. Of the strategies evaluated, biannual alpha-foetoprotein/annual ultrasonography gives the most quality-adjusted life-year gain while still maintaining a cost-effectiveness ratio <$50 000/quality-adjusted life-year. Biannual alpha-foetoprotein/annual computerized tomography screening may be cost-effective.

Introduction

Almost four million Americans have chronic hepatitis C virus (HCV) infection, and in the next decade a continued increase in the incidence of hepatocellular carcinoma (HCC) is likely as this cohort matures.1 Cirrhosis because of chronic hepatitis C is a major risk factor for the development of HCC. Unfortunately, at the time of diagnosis, many patients already have advanced tumours. Routine screening of cirrhotic patients with abdominal ultrasound (US) and alpha-foetoprotein (AFP) levels has been advocated and is widely practiced.2 However, the high cost of US and follow-up studies generated by positive screening tests leads to the continued investment of health care resources for uncertain gain.

For ethical and logistical reasons, prospective controlled trials to assess the cost-effectiveness (CE) of screening for HCC are difficult to perform. The results of retrospective studies are vulnerable not only to obvious problems such as confounding, but also to subtle errors arising from lead time bias, over-diagnosis bias and length bias.3 In the absence of clinical trials, decision analysis represents an alternative approach to the evaluation of this question. The aim of this study is to assess the CE of screening for HCC in patients with chronic hepatitis C and compensated cirrhosis.

Methods

Our decision model consisted of a Markov model programmed using DATA 3.5 software (Treeage Inc, Williamstown, MA, USA). Markov models simulate the transitions between health states over long time periods in a defined population, and are used to investigate chronic medical conditions. We adhered to the recommendations of the Panel on CE in Health and Medicine.4 Results were expressed in terms of discounted costs, life-years (LYs), quality-adjusted life-years (QALYs), and incremental CE ratios for QALYs gained.

Decision model

A schematic representation of our decision model is shown in Figure 1. The two initial branches of the decision tree represent a decision node between the screening vs. the no screening strategy in a cirrhotic adult with chronic hepatitis C. All subjects begin the simulation in the compensated cirrhosis state (Child's class A). During each subsequent month, they may die, remain in the Child's class A state, progress to the decompensated cirrhosis (Child's class B or C) state, or enter other states representing various scenarios with diagnosed or undiagnosed and resectable or unresectable HCC. Because of the size and complexity of the entire tree, which consists of 28 unique subtrees, we have shown only two representative subtrees (Figures 2 and 3).

Figure 1.

A schematic representing the general Markov decision analysis model tree structure, showing the 15 Markov states, each of which leads to a unique subtree (not shown in this diagram) that determines transition between Markov states during each cycle. HCC, hepatocellular carcinoma; M, Markov node.

Figure 2.

A simplified schematic of a typical sub-tree in the Markov model. This sub-tree arises from the ‘Undiagnosed Resectable HCC (Child's class A)’ state. Child's classes are noted in parentheses. HCC, hepatocellular carcinoma.

Figure 3.

A simplified schematic of a typical sub-tree in the Markov model. This sub-tree arises from the ‘Resected HCC (Child's class A)’ state. Child's classes are noted in parentheses. HCC, hepatocellular carcinoma.

In the no screening strategy, diagnostic tests are performed only when the patient develops symptoms suggestive of a tumour or when laboratory or radiologic tests register an incidental finding. In the screening strategy, screening with outpatient high-resolution abdominal US and/or serum AFP levels is performed at regular intervals. The screening test is considered positive if either AFP or US is positive. Confirmatory diagnostic tests will be performed in patients with positive screening results, symptoms suggestive of HCC or incidental laboratory or radiologic findings. These tests include triphasic abdominal computerized tomography (CT) or magnetic resonance imaging, as well as US-guided fine needle biopsy in selected patients according to criteria from the European Association for the Study of the Liver 2001 Consensus Conference.5 Child's class is defined using standard modified Child-Pugh criteria. The ‘Milan criteria’ were used to determine eligibility for liver transplantation.6 Patients diagnosed with HCC were treated with tumour resection for resectable lesions (as determined by size, location and number of lesions, comorbidity and portal hypertension) or palliative treatment – transarterial chemoembolization (TACE), percutaneous ethanol injection (PEI) or thermoabalation – for unresectable cases.

In the primary analysis, liver transplantation was not modelled because currently many medically eligible patients never receive liver transplants because of donor organ shortages, i.e. the decision as to whether or not to perform liver transplant on eligible patients is often determined not by economic or medical considerations but by organ availability. However, in the secondary analysis we included liver transplantation as a possible therapy for all eligible patients with or without HCC.

Incremental comparisons were performed by rank ordering the alternatives by increasing efficacy after eliminating those that were more costly and less efficacious than an alternative (i.e. dominated). The incremental CE ratio was defined as the additional cost per additional gain in QALY for any particular screening strategy vs. the next less efficacious (in terms of QALYs gained) screening strategy. The least efficacious screening strategy was compared against the no screening strategy.

Study population

Our base case subject was a 40-year-old patient with no risk factors for HCC except HCV-related cirrhosis. We selected this age so that screening can start 15–20 years before the mean age of HCC occurrence.

Screening strategies

Based on tumour doubling time,7 6 months has been postulated as the most reasonable screening interval. Three plausible strategies were modelled: AFP and US every 6 months (biannual AFP/US screening), AFP and US every 12 months (annual AFP/US) and AFP every 6 months with US every 12 months (biannual AFP/annual US). The last strategy is the most commonly used screening method, according to national surveys.2 Only direct medical costs were considered; indirect costs from lost productivity were not modelled. Discounting was implemented at 3% per year for QALYs and costs in the standard manner. The background age-adjusted mortality was derived from United States tables of vital statistics.8 The Markov cycle length was 1 month.

Model assumptions

Based on available data in the literature, the following assumptions were made: (i) the progression from Child's class A cirrhosis to Child's class B or C cirrhosis is independent of the presence of HCC; (ii) only Child's class A patients can undergo resection of HCC and therefore only these patients are screened; (iii) postresection patients will only be eligible for palliative treatment regardless of size or location of the tumour; (iv) the probability of HCC recurrence, cirrhosis progression, HCC growth, and mortality remain constant over time, because no reliable data on the temporal pattern of these values are available; (v) treatment compliance is assumed to be 100% in the primary analysis, although suboptimal compliance is evaluated in the secondary analysis; and (vi) screening is stopped at the age of 70.

Probability, utility and cost values

Probability, utility and cost values used in our decision analysis are described in Table 1. The incidence of new HCC in cirrhotic patients has been modelled to increase with the duration of cirrhosis based on data from a large prospective study.9 Mortality in cirrhotic and HCC patients was derived from prospective and retrospective follow-up studies.9–18 The mortality of HCC patients treated with surgical resection or TACE/PEI was based on the results of controlled or uncontrolled studies and meta-analyses.19–27 As the main cause of excess mortality after successful surgical resection is recurrence of HCC, postresection mortality in our model (which does not reflect deaths due to tumour recurrence) is only slightly higher than that of cirrhotic patients without HCC. Tumour recurrence, and the mortality associated with it, is modelled separately (see Figure 3). The probability of cirrhosis progression and growth of HCC were derived from natural history studies.7, 10, 11, 17, 19 The sensitivity and specificity of combination US and AFP was based on cross-sectional studies specifically measuring the accuracy of using these two modalities in combination,28–31 while that of AFP was determined from a study using receiver-operating-characteristic curve analysis (a cutoff value of 150 ng/mL was used).32 As clinicians often use a much lower cut-off of 20 ng/mL, sensitivity analysis was performed using sensitivity values of 21–80% and specificity of 60–98% based on meta-analyses.33, 34

Table 1.  Input variables – probabilities, costs and health state utilities
DefinitionBase case valueRangeStrength of evidence1References
  1. 1 Strength of evidence ranking (adapted from Petitti) 83: A, prospective controlled trial, meta-analysis; B, prospective uncontrolled or observational studies, cross-sectional studies; C, retrospective controlled or uncontrolled studies; D, very small studies, whether prospective or retrospective (<20 subjects); E, case series/expert opinion.

  2. 2 Excess mortality is defined as observed mortality minus age-adjusted baseline mortality (as determined by US vital statistics reports). Postresection mortality does not include mortality after recurrence of HCC (recurrence is modelled independently in the decision tree).

  3. 3 Annual incidence of new HCC increases based on duration of cirrhosis: Year 1 – 0.02; Year 2 – 0.02; Year 3 – 0.02; Year 4 – 0.03; Year 5 – 0.05; Year 6 – 0.05; Year 7 – 0.07; Year 8 and thereafter – 0.1. This is based on data from a large prospective trial.22

  4. 4 Confirmatory tests include triphasic abdominal computerized tomography in all patients, as well as abdominal magnetic resonance imaging and/or ultrasound-guided biopsy (in selected patients).

  5. 5 Costs are in 2003 US dollars.

  6. 6 Standard preoperative tests include electrolyte panel, hepatic function panel, complete cell count, coagulation parameters and electrocardiogram.

  7. 7 These were derived using time-tradeoff and standard gamble techniques.39

  8. AFP, alpha-foetoprotein; US, ultrasound; TACE, transarterial chemoembolization; PEI, percutaneous ethanol injection; HCC, hepatocellular carcinoma; CT, computerized tomography.

Probabilities
Excess Annual Mortality2
 Child's class A cirrhosis0.020.01–0.05B,C9–11
 Child's class B/C cirrhosis0.210.09–0.36C10, 12
 Unresectable HCC (Child's class A/B/C)0.960.92–0.99B,C13–16, 45
 Resectable untreated HCC (Child's class A)0.050.04–0.23A,B15, 17, 18
 Resected HCC (Child's class A)0.040.02–0.15A,B,C19, 20, 43, 70
 Resected HCC (Child's class B/C)0.220.14–0.35B46
 HCC treated with TACE/PEI (Child's class A)0.110.05–0.5B,D21, 22, 24, 26, 27
 HCC treated with TACE/PEI (Child's class B/C)0.300.25–0.65A21, 22, 24, 26, 27
 Postliver transplant for HCC (first year)0.10.05–0.21A6, 20, 71
 Postliver transplant for HCC (subsequent years)0.040.02–0.15A6, 20, 71
Procedure-related mortality
 Perioperative HCC resection mortality0.040.02–0.08A20, 72, 73
 Laparotomy mortality0.010.005–0.04C74
 TACE/PEI procedure-related complication rate0.030.01–0.06B27, 75
 Perioperative mortality of liver transplant0.060.03–0.12B20
Annual progression
 Progression from Child's class A to Child's class B/C cirrhosis0.060.03–0.12B,C9– 11
 Progression from resectable to unresectable HCC (Untreated)0.30.1–0.6C7, 17, 19
 Progression from resectable to unresectable HCC (TACE treated)0.20.05–0.4C75, 76
Annual incidence
 Incidence of new HCC0.02–0.130.01–0.05 to 0.04–0.2A,B,C9, 11, 41, 43, 50, 52, 77
 Incidence of HCC recurrence after resection0.210.15–0.37B47, 78
Test accuracy
 Sensitivity of US and AFP0.850.55–0.95B28–31, 79–81
 Specificity of US and AFP0.80.7–0.9B28–31, 79–81
 Sensitivity of AFP alone0.540.21–0.8B32, 33
 Specificity of AFP alone0.930.6–0.98B32, 33
 Sensitivity of confirmatory tests40.950.9–1.0B82
 Specificity of confirmatory tests40.990.9–1.0B82
Costs, $5
Laboratory tests
 AFP screening5226–104 Micro-costing
 Preoperative tests69849–196 Micro-costing
Radiologic tests
 US screening18191–362 Micro-costing
 Abdominal CT (triphasic)312156–624 Micro-costing
 Hepatic angiography/portography487244–974 Micro-costing
Interventional procedures
 US-guided biopsy285143–670 Micro-costing
 Palliative treatment – TACE (once)612306–1224 Micro-costing
 Palliative treatment – PEI (once)255128–510 Micro-costing
Surgical procedures
 Partial hepatectomy26 60913 305–39 915 Micro-costing
 Laparotomy98034902–19 606 Micro-costing
 Liver transplantation86 07843 039–129 117 Micro-costing
Outpatient care
 Cirrhosis-related outpatient care (per month)14573–290 Micro-costing
 Post-transplant outpatient care (per month)246123–492 Micro-costing
Terminal care
 Death from surgery31 00015 500–62 000 35, 36
 Cirrhosis/HCC-related terminal care34 00017 000–68 000 35, 36
Health State Utilities, QALY7
 Child's class A (compensated cirrhosis)0.780.6–0.95 39
 Child's class B/C (decompensated cirrhosis)0.650.5–0.85 37– 39
 Presence of HCC (all Child's classes)0.50.2–0.7 37– 39
 Toll for TACE/PEI or surgery−1 day N/A

The true costs of laboratory and radiologic tests, interventional radiologic and surgical procedures, and outpatient follow-up were derived using micro-costing techniques based on a hospital cost accounting system (Transition I, Transition Systems Inc., Boston, MA, USA). Cancer-related end-of-life terminal care costs were estimated from the pertinent literature.35, 36 When calculating costs, we included the costs of non-reusable materials consumed (such as AFP test kits), costs of non-tangible hospital resources (e.g. operating room time), salaries of ancillary personnel such as nurses and technicians, and physician fees. We assumed that commonly available institutional infrastructure was already present (e.g. laboratories, surgical facilities, terminal care facilities, computerized tomography machines, ultrasound machines).

Utilities represent an individual's preference for a given health state and are scaled from 0 to 1. QALYs are calculated by multiplying the time spent in a given health state with the utility for that health state. The health-state utilities for different stages of liver disease and HCC were based on quality of life studies using standard gamble and time-tradeoff techniques.37–39

Screening and treatment efficacy

Although studies have not definitively proven the superiority of surgical resection over other treatment modalities such as TACE/PEI, the advantages of treatment over no treatment in both symptomatic or asymptomatic patients is well established. In our model, the main benefits of screening are the result of survival differences between asymptomatic HCC patients (discovered by screening) who undergo resection or TACE/PEI vs. those who had no treatment (which would be the case in unscreened asymptomatic patients). In comparison, the survival benefits of undergoing surgical resection vs. TACE/PEI in HCC patients are assumed to be relatively minor.

Secondary analysis

As 25% of gastroenterologists reported performing HCC screening with CT,2 we also modelled screening using AFP and triphasic CT, instead of US. Finally, we repeated the analysis assuming that medically eligible patients would be placed on liver transplant lists regardless of insurance or donor organ availability issues. Median waiting times reported by the United Network for Organ Sharing (UNOS) in the United States were used (data accessed from http://www.unos.orgas of August 2003). While waiting for transplant, patients with HCC were treated with TACE every 6 months.

Sensitivity analysis

One-way sensitivity analysis, using plausible ranges based on the lowest and highest numbers reported in different studies, was done on all probability and utility values. For costs, sensitivity analysis was performed using 50 to 150–200% of the base case value, if plausible. In addition, extensive two- and three-way sensitivity analyses were performed.

Results

Primary analysis

Table 2 shows the additional gains in QALY, additional costs, and incremental CE ratios for the three principal screening strategies considered, each compared with the next less efficacious strategy, or, in the case of the least efficacious strategy, to the no screening strategy. The least efficacious strategy was annual AFP/US screening, which resulted in an expected life-expectancy of 8.965 LYs (or 6.569 QALYs) for the screened group, with a lifetime cost of $53 145. This translated to a gain of 0.3 QALYs, or 110 quality-adjusted life-days, and an incremental CE ratio of $23 043/QALY or $20 885/LY. For the strategy most commonly used in the United States (biannual AFP/annual US screening),2 the incremental QALY gain from screening was 0.048 QALYs, resulting in an incremental CE ratio of $33 083/QALY vs. annual AFP/US screening. The most efficacious strategy (biannual AFP/US screening) entailed a higher incremental CE ratio of $73 789/QALY vs. biannual AFP/annual US screening.

Table 2.  Cost-effectiveness of strategies for hepatocellular carcinoma screening in patients with compensated cirrhosis
StrategyLifetime cost ($)Additional cost ($)Expected QALYQALY gainedExpected LYLY gainedIncremental CE ratio ($/QALY)Incremental CE ratio ($/LY)
  1. 1 Incremental CE ratios are calculated by dividing incremental cost ($) over incremental outcome (QALYs). Each incremental value is determined by subtracting the value of the strategy of the next less effective strategy (as measured by QALYs gained) from that of the strategy under consideration. The incremental CE ratio of the least efficacious screening strategy (US and AFP every 12 months) is calculated against the no screening strategy.

  2. 2 CT screening: screening strategies with triphasic abdominal CT and AFP and compared against each other as well as against strategies using US.

  3. US12AFP6, US at 12-month intervals and AFP levels at 6-month intervals; US6AFP6, US and AFP levels at 6-month intervals; US12AFP12, US and AFP levels at 12-month intervals; QALY, quality-adjusted life-year; US, abdominal ultrasonography; AFP, serum alpha-foetoprotein level.

Base case
 No screen46 23206.26908.6340
 US12AFP1253 14569136.5690.3008.9650.33123 04320 885
 US12AFP654 73315886.6170.0489.0210.05633 08328 357
 US6AFP657 16824356.6500.0339.0800.03673 78967 639
After inclusion of CT screening strategies2
 No screen46 23206.26908.6340
 US12AFP1253 14569136.5690.3008.9650.33123 04320 885
 CT12AFP1253 6555106.5830.0148.9830.01836 42928 333
 US12AFP654 73310786.6170.0349.0210.03831 70628 368
 CT12AFP655 1474146.6250.0089.0310.01051 75041 400
 US6AFP657 16820216.6500.0259.0800.02680 84077 731
 CT6AFP658 23210646.6100.0119.0930.01396 72781 846

Secondary analysis

Screening using triphasic abdominal CT and AFP resulted in better survival compared with the corresponding strategy using US and AFP, with incremental CE ratios ranging from approximately $23 000 to $96 000 per QALY (see Table 2). We also modelled treatment compliance rates of 70–99%; this did not change the rank ordering of the three different strategies by efficacy, nor did it affect the incremental CE ratios in a significant manner (data not shown). As expected, compliance rates below 70% led to progressive increases in the CE ratios.

Finally, we repeated the analysis assuming all medically eligible patients (with or without HCC) were placed on a liver transplant waiting list. In this situation, screening using annual AFP/US, biannual AFP/annual US screening and biannual AFP/US screening resulted in incremental CE ratios of $44 883/QALY, $38 684/QALY and $95 913/QALY respectively, compared with the next most efficacious screening strategy.

Sensitivity analysis

Extensive one-way sensitivity analysis on all probability and cost variables did not change the rank ordering of the different strategies. Incremental CE ratios ranged between $15 000 and $42 000/QALY for annual AFP/US screening, $14 000 and $89 000/QALY for biannual AFP/annual US screening vs. annual AFP/US screening (Figure 4), and $16 000 and $201 000/QALY for biannual AFP/US screening vs. biannual AFP/annual US screening. It should be noted that certain potentially critical variables whose values are subject to uncertainty because of the lack of reliable studies, such as the recurrence rate after resection and mortality of HCC patients treated with palliative chemoembolization, did not affect CE ratios markedly on one-way analysis. The most influential variable was the postresection mortality of HCC patients.

Figure 4.

Tornado diagram representing the incremental cost-effectiveness ratios of one-way sensitivity analysis on the biannual AFP/annual US screening strategy vs. the annual AFP/US screening strategy. The vertical line represents the incremental cost-effectiveness ratio under base case conditions. AFP, alpha-foetoprotein; US, ultrasound; TACE, transarterial chemoembolization; PEI, percutaneous ethanol injection; HCC, hepatocellular carcinoma; CT, computerized tomography.

Two-way sensitivity analysis, focusing on clinically related pairs of variables, resulted in incremental CE ratios between $17 000 and $47 000/QALY for annual AFP/US screening vs. no screening, $22 000 and $91 000/QALY for biannual AFP/annual US screening vs. annual AFP/US screening, and $52 000 and $218 000/QALY for biannual AFP/US screening vs. biannual AFP/annual US screening. Table 3 describes the results of two-way sensitivity analysis on four selected pairs of variables: (i) postresection mortality of HCC patients and mortality of Child's A patients with untreated HCC; (ii) mortality of HCC patients treated with TACE/PEI and mortality of Child's A patients with untreated HCC; (iii) HCC recurrence rate and mortality of Child's A patients with untreated HCC; and (iv) postresection mortality of HCC patients and mortality of HCC patients treated with TACE/PEI. Of note, the highest CE ratio of $91 001/QALY for biannual AFP/annual US screening vs. annual AFP/US screening resulted from assuming annual mortality of 0.15 after HCC resection and 0.04 for untreated HCC; however, these values are highly unlikely to occur in combination (although considered separately the ranges are realistic) because the mortality after resection of HCC is unlikely to be worse than that of untreated HCC.

Table 3.  Two-way sensitivity analyses on selected variable groups
  CE Ratio ($/QALY)1CE Ratio ($/QALY)1CE Ratio ($/QALY)1
  1. 1 All CE ratios are incremental.

  2. 2 Palliative treatment includes transarterial chemoembolization or percutaneous ethanol injection.

  3. US12AFP6, US at 12-month intervals and AFP levels at 6-month intervals; US6AFP6, US and AFP levels at 6-month intervals; US12AFP12, US and AFP levels at 12-month intervals; QALY, quality-adjusted life-year; HCC, hepatocellular carcinoma.

Postresection mortality (Child A)Untreated HCC mortality (Child A)US12AFP12 vs. No ScreenUS12AFP6 vs. US12AFP12US6AFP6 vs. US12AFP6
0.020.0421 41833 01173 568
0.020.2318 91223 71652 555
0.150.0446 81491 001218 025
0.150.2333 51845 507106 655
Postpalliative treatment mortality (Child A)2Untreated HCC mortality (Child A)   
0.050.0423 24438 02085 319
0.50.2319 22125 21956 164
0.050.0425 11236 62385 377
0.50.2321 96627 50260 015
HCC recurrenceUntreated HCC mortality (Child A)   
0.150.0420 19831 46976 633
0.150.2317 77122 74353 483
0.370.0433 98456 346111 220
0.370.2327 83434 80578 145
Postresection mortality (Child A)Post palliative treatment mortality (Child A)2   
0.020.0519 79429 31667 608
0.020.521 81330 49873 995
0.150.0541 32078 996200 110
0.150.542 69165 088150 300

Finally, multiple three-way analyses were performed on combinations of clinically related variables, including, amongst others: (i) postresection mortality of HCC patients, mortality of Child's A patients with untreated HCC and mortality of HCC patients treated with TACE/PEI; (ii) incidence of HCC, progression rate of cirrhosis and growth rate of HCC; and (iii) cost of US screening, cost of AFP levels and cost of surgical resection. The incremental CE ratios ranged between approximately $16 000 and $47 000/QALY for annual AFP/US screening vs. no screening, $20 000 and $112 000/QALY for biannual AFP/annual US screening vs. annual AFP/US screening, and $36 000 and $265 000/QALY for biannual AFP/US screening vs. biannual AFP/annual US screening.

Discussion

Several features of HCC make it a potentially viable target for screening: first, HCC occurs in a well-defined risk population. The primary risk factor is cirrhosis, particularly that related to viral hepatitis, alcoholic liver disease or haemochromatosis. Secondly, HCC has a protracted subclinical phase. Natural history studies have shown that once cirrhosis is present, up to 20% of patients may develop HCC during the next 10 years.40 During the subclinical phase, there are often no distinctive symptoms that distinguish patients with HCC from those with only cirrhosis.41, 42 By the time diagnosis is made, over 85% of tumours are unresectable because of size, multifocality, hepatic decompensation or invasion of the portal vein or surrounding structures.13, 41–44 Thirdly, HCC has an improved prognosis if subjected to early vs. late treatment. The prognosis of large, unresectable HCC is dismal;13, 23, 45 however, for resectable HCCs, the outlook appears to be considerably brighter.44, 46–48

Because of the above reasons, HCC screening has been repeatedly advocated. Such screening is routinely performed in other nations on patients with chronic hepatitis B or cirrhosis. In the United States, national surveys show that 84% of gastroenterologists regularly screen their cirrhotic patients, especially those with viral hepatitis, alcoholic liver disease and haemochromatosis.2

As HCC screening is considered the standard of care, it is difficult to perform controlled prospective trials to assess its efficacy or CE. To date, there has been only one randomized controlled trial in Chinese patients with hepatitis B virus (HBV); this study showed increased rates of HCC detection, increased resectability and improved 2-year survival.49 Additional data on screening efficacy come from various studies, mostly uncontrolled.9, 28, 41–44, 50–54 Five studies have reported that screened patients enjoy longer survival compared with unscreened patients.55–59 Two of these were retrospective case–control studies that evaluated Chinese patients with HBV-related HCC, while a third study was a prospective 16-year follow-up of native Alaskans with HBV infection in an AFP screening programme, whose survival was compared with those of historical controls (also from Alaska) who were not screened. The last two studies were non-randomized controlled studies of Italian patients with cirrhosis.58, 59 Studies have shown that the proportion of resectable tumours is as high as 57–81% with screening,41–43, 51 but only one study has been controlled.41 Although a few studies have reported no increase in the proportion of resectable HCCs,28, 50 the preponderance of evidence suggests that screening increases the proportion of diagnosed HCCs that are resectable and improves survival.

There have been few CE evaluations of HCC screening performed as part of an actual clinical study. A retrospective cohort study found that survival was improved in the screened group, but at a cost of $112 993/LY saved.59 In addition, a simplified calculation using 1989 dollars arrived at an estimated cost of $216 000 per patient ‘cured’ of HCC.60

There have been three formal decision analyses on this topic. The first one was a Swiss study that utilized a Markov model to evaluate the CE of screening for HCC in patients with Child's class A cirrhosis of all causes.61 The investigators found that the CE ratio for screening with US and AFP at 6-month intervals ranged from $48 000 to $284 000/LY gained. This study differed from ours in several important ways: (i) liver transplantation was not modelled; (ii) life-expectancy was not adjusted for quality; (iii) it is unclear whether charges or costs were used, or whether the estimates were based on American or European health care costs; (iv) palliative therapy for HCC, such as TACE or PEI, was not modelled; and (v) Confirmatory diagnostic tests were assumed to be 100% accurate and without morbidity. More recently, an American study also utilized a Markov model to evaluate the CE of screening for HCC in transplant-eligible patients with cirrhosis.62 In this study, the CE ratio (compared against no screening) was $26 689/QALY for screening with alternating US and AFP (US or AFP were performed every 12 months but were staggered 6 months apart) and $25 232/QALY for CT and AFP. This model was structurally more similar to ours than the Swiss model; however, it featured unusual screening regimens, unlike the strategies in our study which are commonly used in general practice.2 Furthermore, this model included Child's class B patients and assumed that all patients would be listed for liver transplant. Our study consists of two separate analyses based on whether or not liver transplantation was a realistic option for cohort members. The third model is focused exclusively on UNOS status 3 cirrhotic patients who were already awaiting liver transplant.63 Four screening strategies were compared – AFP alone (the referent strategy), US alone (incremental CE ratio of $60 300/LY), AFP and US ($74 000/LY), and CT alone ($101 100/LY), all done at 6-month intervals. There was no comparison with a no screening strategy because the authors felt that all patients on transplant waiting lists should undergo some form of HCC screening.

Other CE studies in related areas include a four-stage ‘micro-simulation’ model from Taiwan, published in abstract form. The detailed structure of the model was not described, but it appears to be a Monte Carlo simulation. In this study, the CE ratio of community screening using AFP and US in a population endemic for hepatitis B was found to be $190 000 per life-year gained.64 Another study evaluated screening for HCC with US and AFP at 10-month intervals in all HBV carriers regardless of cirrhosis.65 This study looked only at a surrogate end-point, cost per resectable tumour detected, and used charges at a Singaporean hospital.

Our decision analysis attempted to address the limitations of previous studies. In our study, true costs were used instead of charges and life-years were adjusted for quality. The effect of liver transplantation was analysed in the secondary analysis, and we modelled the effects of palliative therapy with TACE or PEI. Studies have shown that TACE may improve the 12-month survival from 18–54% to up to 82%,21, 22, 24, 26 and we feel that inclusion of palliative therapy in our model is essential because such therapy in unresectable patients is the standard of care in many parts of the world.

In the US health care system, charges are often poorly correlated with actual costs. For our study, we collaborated closely with our hospital accounting department in order to derive true costs using micro-costing techniques for all diagnostic and therapeutic procedures. This has a significant impact on the results. For example, the actual cost of a triphasic abdominal CT is only $312, which is substantially lower than the charge of $616. It should be noted that our costs are based on those at an academic medical center; community hospital costs may be somewhat lower. Unlike procedure costs, which were itemized in detail, terminal care costs in our model were based on literature reports of average cancer-related terminal care costs,35, 36 because it was impossible to itemize terminal care costs due to their immense variability. Our sensitivity analysis shows that terminal care costs do not significantly affect the marginal CE ratio, because most patients in the model, screened or unscreened, incur terminal-care costs prior to dying.

Several assumptions in our model require discussion. In practice, it is unlikely that rates of mortality or progression of cirrhosis remain constant over time. However, no reliable studies have described the temporal pattern of changes in the probability of development of cirrhosis, progression of cirrhosis and HCC-related mortality. Hence we have used constant probabilities based on published rates. The survival results of our model are validated by their agreement with the survival curves of long-term natural history studies.10 We also assumed that all medically eligible patients with resectable HCC will undergo tumour resection, and that all remaining patients will undergo palliative treatment. In practice, many medically eligible patients never undergo treatment for various reasons.66 We are reassured by the fact that sensitivity analysis did not significantly affect our results as long as treatment compliance rates remained above 70%.

We assumed that only Child's class A patients are eligible for tumour resection and thus Child's class B and C patients are not screened. In actuality, carefully selected Child's class B patients may survive resection, although the perioperative mortality is high in the presence of decompensated cirrhosis.67 Therefore, we have modelled screening in Child's class B patients only in the secondary analysis where transplantation is a viable therapy.

Decision analysis studies indicate that liver transplantation as first-line treatment leads to improved survival and is potentially cost-effective in selected patients when compared with other therapies.68, 69 Unfortunately, the majority of medically eligible patients never receive a transplant because the number of potential recipients continues to vastly outnumber the number of donor organs. As the applicability of liver transplantation is limited not only by financial considerations but also by donor organ supply, it was not included in the primary analysis. From the results of the secondary analysis, however, it appears that liver transplantation increases the CE ratio for screening, but not dramatically.

As described in the methods section, our study models a cohort of patients with known compensated cirrhosis. The most appropriate strategy for identifying HCV-infected patients who are cirrhotic is a complicated and controversial issue that is beyond the scope of this study. In addition, we have not modelled treatment with interferon and ribavirin.

In summary, this study suggests that in a well-defined cirrhotic population, screening for HCC is at least as cost-effective as other screening practices that are considered standard of care. The best screening protocol has yet to be formally defined, but based on the results of our model, US at 12-month intervals and AFP at 6-month intervals is a reasonable strategy, offering the greatest gain in life-expectancy while still maintaining an incremental CE ratio <$50 000/QALY. More frequent screening with US provides some additional benefit, but is more expensive. Screening in patients awaiting liver transplant is also cost-effective. Screening with CT instead of US is more efficacious and appears to be cost-effective, and deserves further study.

Acknowledgements

This study was supported by an American College of Gastroenterology Clinical Research Award (Dr Otto Lin).

Ancillary