N.2.1.1. Comparators

The NGC liver disease pathway model (LDPM) was developed for this guideline and for the NICE non-alcoholic fatty liver disease guideline. The model is composed of 3 modules, covering steatosis, fibrosis and cirrhosis, and follows the progression of people with liver disease through the course of their lifetime. For this economic analysis the cirrhosis module was used, and was adapted for separate populations with alcohol-related liver disease (ALD), non-alcoholic fatty liver disease (NAFLD), hepatitis B (HBV) and hepatitis C (HCV).

For this analysis, 23 single tests, identified in the relevant clinical review, and 4 combinations of tests, were compared for 1 or more of the 4 populations of interest. These are summarised below. Several tests were not considered for modelling due to the absence of sensitivity and specificity data in the relevant papers (only area under the curve figures reported). For each aetiology population the diagnostic tests are also compared against the reference standard, liver biopsy.

Two further strategies were also considered which did not include any tests:

• no test, monitor all patients in the relevant population assuming they have cirrhosis
• no test, monitor no-one, assuming none have cirrhosis until later clinical presentation.

In the hepatitis C cohort, for modelling purposes, there was an additional no-testing strategy for which in addition to no monitoring there would also be no treatment given for hepatitis C.

N.2.1.1.1. Combinations of more than 1 test

In planning the model structure, the inclusion of combinations of tests was considered. Four algorithms were identified in papers included for the hepatitis C population (at the bottom of Table 52 above) and these were included alongside the single tests. The GDG also considered using 2 of the single tests (excluding liver biopsy) consecutively. The GDG considered that combinations should include 1 blood test and 1 imaging test as these would be likely to give independent results. The most promising combination would be one using a blood test with high sensitivity (to maximise true positives and minimise false negatives) followed by an imaging test with high specificity (to rule out true negatives). However, when viewing the diagnostic accuracy values found in the clinical review (see Section N.2.3.2 below) no such combination could be found. Consequently there was no reason to believe any combination of 2 tests would give more accurate results than the best single tests, but with an increased cost for using 2 tests instead of 1. Therefore no such combinations were modelled.

Table 52

Tests included in the model by disease aetiology.

N.2.1.2. Population

The model considers people aged 50 years at the start of the model with one of the 4 major underlying causes of cirrhosis (hepatitis B, hepatitis C, alcohol-related liver disease, non-alcoholic fatty liver disease) who are therefore at risk of developing cirrhosis. Patients with different aetiologies are treated as separate patient cohorts in the model. Hepatitis B patients are further separated in 2 cohorts (positive or negative hepatitis B e Antigen, HBeAg). Hepatitis C patients are further separated by disease genotype (Genotypes 1–4).

N.2.1.3. Time horizon, perspective, discount rates used

The analysis follows the standard assumptions of the NICE reference case including discounting at 3.5% for costs and health effects, and the perspective of the UK NHS and personal social services. A sensitivity analysis will also be conducted using a discount rate of 1.5% for costs and health benefits. A lifetime horizon has been chosen to fully capture the adverse outcomes derived from incorrect diagnosis.

N.2.2.1. Model structure

Figure 199

Graphical depiction of the decision tree.

N.2.2.2. High-level model structure

Initially, a decision tree determines the proportion of people with cirrhosis who receive a correct diagnosis (true positive – TP) and an incorrect diagnosis (false negative – FN); and the proportion of people without cirrhosis who receive a correct diagnosis (true negative – TN) and an incorrect diagnosis (false positive – FP) depending on the diagnostic accuracy of every test. People diagnosed as not having cirrhosis were assumed to have advanced fibrosis (F3 on the METAVIR scale).

It is assumed that 27% of people with cirrhosis will already have medium or large varices at the time when they are first diagnosed with cirrhosis. People will receive endoscopic surveillance for oesophageal varices immediately following a positive diagnosis of cirrhosis. It is assumed that this is 100% successful at identifying medium or large varices.

Consequently, patients enter the Markov model through 6 health states:

• advanced fibrosis with a true negative diagnosis – (F3-TN)
• advanced fibrosis with a false positive diagnosis of cirrhosis– (F3-FP)
• compensated cirrhosis with a true positive diagnosis – (comp-TP)
• compensated cirrhosis with a false negative diagnosis of advanced fibrosis only – (comp-FN)
• compensated cirrhosis with oesophageal varices with a true positive diagnosis, hence immediately receiving prophylactic measures to prevent variceal bleeding – (VarTP-Pr)
• compensated cirrhosis with oesophageal varices with a false negative diagnosis of advanced fibrosis only, and hence not assessed or receiving treatment for varices – (Var-FN)

It is assumed that everyone with cirrhosis at the start of the model has compensated cirrhosis, as decompensated cirrhosis would have previously been identified by a clinician's observations without the need for the diagnostic tests examined here. Under GDG guidance, retesting for those with a negative diagnosis was set at 2 years for all populations. The cost-effectiveness of decreasing the retesting frequency to 1 year was examined in sensitivity analyses for each population.

Overall, the model attempts to represent the natural history of the disease, from compensated cirrhosis without varices to the development of varices (which may lead to bleeding), HCC and other decompensation events, and finally to a post-liver transplant state or to death.

N.2.2.2.2. Surveillance for oesophageal varices

Variceal bleeding is one the most common complications of cirrhosis and is considered a decompensating event. Endoscopic surveillance for the development and the size estimation of oesophageal varices is believed to have a substantial patient benefit as those identified with medium and large varices receive a band ligation procedure that offers prophylactic benefits against variceal bleeding.

In the model base case, all patients diagnosed with cirrhosis will be monitored every 3 years for varices. This was set after agreement with the GDG that this reflects common current practice in the NHS. A 2-yearly and an annual surveillance strategy will also be tested for their cost-effectiveness compared to a 3-yearly strategy to contribute to the relevant clinical review question.

People who developed medium or large varices whilst in either compensated or decompensated cirrhosis states were represented in separate health states (depicted as Var and dcVar respectively in the model structure, Figure 200). As presented in Figure 201 below, people with cirrhosis are separated between those who have developed varices since their most recent endoscopy and so have not been yet been identified as having varices or offered a prophylactic band ligation (VarTP-Un, dcVar-Un) and those who have received an endoscopy since they have developed varices and so are assumed to have been correctly identified as having varices and consequently protected against bleeding by prophylactic band ligation (VarTP-Pr, dcVar-Pr). Similarly bleeding has been separated from the other decompensating events (ascites, hepatic encephalopathy, jaundice) and is represented as a separate state, which individuals are in for a single Markov cycle, after which if still alive they are transferred to a decompensated state, but with their varices now protected (dcVar-Pr). Prophylactic band ligation was taken to reduce the risk of bleeding by 50%, as found by a literature review provided by the GDG (Berzigotti 2013). The prevalence of varices (of any size) in people diagnosed with cirrhosis (40%) and annual rates of varices development in people with compensated or decompensated cirrhosis but without varices of 6% and 10% respectively were also sourced from this study. Those figures were adjusted accordingly to represent the proportion of people with cirrhosis with medium or large varices, which was set to 67% of the overall cohort of people with cirrhosis with varices of any size (assumption by Stevenson 2012).

Figure 200

Graphical depiction of the Markov model.

Figure 201

Surveillance for varices structure.

The cost of a diagnostic endoscopy is accordingly added to any Markov cycle during which surveillance for varices is conducted, depending on the frequency chosen. Under GDG guidance it was also assumed that if the endoscopy identified medium to large varices, a band ligation was offered immediately at the same visit. In the described scenario the cost of endoscopy was not applied to avoid double counting as band ligation is conducted endoscopically.

No relevant economic evaluations were identified in our systematic literature review.

N.2.2.3. Population cohorts

N.2.2.3.4. Hepatitis C (HCV)

A new generation of polymerase inhibitor drugs for hepatitis C has been recently assessed by NICE in technology appraisals and are entering NHS practice. In order for the present economic model to reflect the most up-to-date NICE recommendations, 2 recently published drug combinations (part of TA330 and TA ID742) covering the 4 most prevalent UK HCV genotypes are included in the modelling of the cirrhosis patient pathway. Ombitasvir-paritaprevir-ritonavir is also an option for genotypes 1 and 4. We chose ledipasvir-sofosbuvir as that is at least as effective and with a similar price. Note that the economic results would not be altered by this choice of drug as both effectiveness and cost are very similar. People with HCV without cirrhosis are assumed to receive the appropriate pegylated interferon and ribavirin regimes since polymerase inhibitor drugs are not currently recommended for these patients.

With the introduction of the new antiviral treatments and their inclusion to the present model, the GDG has made a similar assumption as the one described for the HBV model cohort, that for patients to be suspected for cirrhosis they must be new referrals and therefore not appropriately treated before for the underlying cirrhosis cause (since antiviral treatments would dramatically decrease the progression rate to cirrhosis). Therefore all of the patients in the model cohort will be treated with an antiviral agent.

The treatment effectiveness of the antiviral drugs is represented in the model by their sustained viral response (SVR). This figure is the rate of patients who have responded to treatment and therefore were ‘cured’ of the virus. The SVR was consequently applied to the probability of a patient progressing to the next state in the Markov model as patients that are free from the virus are assumed not to progress to more severe liver disease states. They were also assumed to only receive HCC surveillance and not varices surveillance; this was based on GDG guidance that there is still high uncertainty over the risk of HCC in ‘cured’ patients treated with the new drug combinations. SVRs per genotype where sourced from the evidence reports of TA330 and TA ID742.

Table 53

Sustained viral response per genotype.

In addition, under GDG guidance it was assumed that, for patients falsely identified as having cirrhosis, the drug effectiveness is identical as for the correctly diagnosed with cirrhosis patients. For patients falsely diagnosed as negative, the drug effectiveness of the fibrosis-HCV treatment options was adjusted to 50% in order to depict their lower efficacy in patients with cirrhosis.

N.2.2.4. Uncertainty

The model was built probabilistically to take account of the uncertainty around input parameter point estimates. A probability distribution was defined for each model input parameter. When the model was run, a value for each input was randomly selected simultaneously from its respective probability distribution; mean costs and mean QALYs were calculated using these values. The model was run repeatedly – 5,000 times for the base case and 5,000 times for each sensitivity analysis – and results were summarised.

The way in which distributions are defined reflects the nature of the data, so for example utilities were given a beta distribution, which is bounded by 0 and 1, reflecting that a quality of life weighting will not be outside this range. All of the variables that were probabilistic in the model and their distributional parameters are detailed in Table 54 and in the relevant input summary tables in Section N.2.3. Probability distributions in the analysis were parameterised using error estimates from data sources.

Table 54

Description of the type and properties of distributions used in the probabilistic sensitivity analysis.

In addition, various deterministic sensitivity analyses were undertaken to test the robustness of model assumptions. In these, one or more inputs were changed and the analysis rerun to evaluate the impact on results and whether conclusions on which intervention should be recommended would change.

N.2.2.4.1. Deterministic sensitivity analysis

Apart from assigning distributions to most of the model parameters, deterministic sensitivity analysis was also performed for a variety of variables.

Table 55

Summary of parameters tested in DSA.

N.2.3.1. Summary table of model inputs

Model inputs were based on clinical evidence identified in the systematic review undertaken for the guideline, supplemented by additional data sources as required. Model inputs were validated by clinical members of the GDG. A summary of the model inputs used in the base case (primary) analysis is provided in Table 56 and Table 57 below. Health state costs are presented separately in the relevant cost section. More details about sources, calculations and rationales for selection can be found in the sections following this summary table.

Table 56

Summary of base case model inputs.

Table 57

Overview of parameters and parameter distributions used in the model.

N.2.3.2. Diagnostic accuracy

The characteristics of liver biopsy, when serving as a reference standard, were carefully specified in the diagnostic review protocol. Therefore, after agreement with the GDG, only studies reporting a liver biopsy with at least 6 portal tracts and a length of 15 mm or more were considered in the review of the literature. When there were not enough studies (fewer than 3) around the diagnostic accuracy of a specific test for pooled sensitivity and specificity estimates, the corresponding 2×2 diagnostic table was selected from a single study that was believed to represent the best quality evidence. For the ALD cohort and transient elastography at a 11–<13 threshold, to represent the uncertainty around its diagnostic accuracy and because the log-normal distribution could not fit onto a test with a 100% sensitivity, its 2×2 table was adjusted by adding 0.1 patients in each of the four diagnostic outcomes. This brought down its sensitivity from 100 to 99. A similar approach was followed for the NAFLD cohort and transient elastography at a 15 threshold. Selection criteria for the chosen sources are presented in Table 58 below.

Table 58

Source selection when <3 studies identified.

To account for uncertainty around diagnostic accuracies and correlation between sensitivity and specificity a joint distribution was used when making diagnostic accuracies probabilistic. First of all the diagnostic odds ratio (DOR) was calculated for the diagnostic test:

The standard error of the log DOR was calculated using the absolute values for the number of TP, TN, FP and FN:

Using these equations a normal distribution was fitted around the log of the DOR.

Once the DOR is calculated the sensitivity can become a function of the DOR and the specificity:

Finally a beta distribution was fitted around the specificity, therefore when probabilistic sensitivity analysis is conducted the specificity will change in accordance to the overall diagnostic uncertainty and its relationship with the sensitivity.

When reviewers identified more than 2 studies for a specific test, pooled diagnostic accuracy figures were estimated with the use of Bayesian methods. To account for uncertainty around these figures random samples were drawn from the original joint posterior distribution (WinBUGS iterations) for the purposes of probabilistic sensitivity analysis.

Diagnostic accuracy data for the HBV cohort were sourced from the NICE Hepatitis B guideline (CG165).

N.2.3.3. Baseline transition probabilities

Relevant transition rates were sought in the literature and were confirmed by the GDG as appropriate for use in the current model. All transition rates were transformed to 6-monthly transition probabilities.

N.2.3.3.1. Hepatitis B and hepatitis C

Table 59

HBV – 6-monthly transition probabilities.

Table 60

HCV – 6-monthly transition probabilities.

As presented in the above tables, the majority of the transition probabilities originated from the Wright 2006 UK HTA and an economic evaluation on HBV drugs conducted by Dakin et al 2010. For use in the current model those figures where sourced from the Crossan 2015 HTA. The figures on the prevalence of varices in people with cirrhosis were sourced from a review conducted by Berzigotti 2013 and were adjusted by assuming that two thirds of patients develop medium to large varices; this adjustment was applied to all subgroups evaluated in the model. Bleeding rates were obtained from a prospective study of 321 patients with cirrhosis and varices and no history of bleeding conducted by the North Italian Endoscopic Club (NIEC 1988). The HCC incidence rate was assumed to be constant across all patients with cirrhosis (compensated or decompensated), an approach also followed by the Crossan 2015 HTA. Bleeding mortality was sourced from Stevenson 2012 and it was based on clinical judgement. The decompensation rates were adjusted for people with and without varices with a ±25% adjustment to the baseline rate that was based on GDG expert opinion. This adjustment was considered appropriate by the GDG and was applied to all the subgroups considered in the model.

N.2.3.3.2. NAFLD

Table 61

NAFLD – 6-monthly transition probabilities.

As presented in the above table, for the progression of NAFLD patients to cirrhosis a transition probability was obtained from the Singh 2015 meta-analysis of studies with a paired biopsy study design. The decompensation rate was sourced from Hui 2003, a study observing the long-term outcomes of cirrhosis in non-alcoholic steatohepatitis (NASH) patients. The figures on the prevalence of varices in people with cirrhosis were sourced from a review conducted by Berzigotti 2013 and were adjusted by assuming that two thirds of patients develop medium to large varices. Bleeding rates were obtained from a prospective study of 321 patients with cirrhosis and varices and no history of bleeding conducted by the North Italian Endoscopic Club (NIEC 1988). Bleeding mortality was sourced from Stevenson 2012 and it was based on clinical judgement. The incidence of HCC was obtained from Ascha 2010, a study evaluating the incidence and risk factors of HCC in 195 NASH patients. It was assumed that this rate applied to both compensated and decompensated patients. Due to the lack evidence in the remaining transition probabilities, those from the hepatitis cohorts were used after agreement with the GDG.

N.2.3.3.3. ALD

Table 62

ALD – 6-monthly transition probabilities.

As presented in the table above, progression to cirrhosis was obtained from Pares 1986, a study on the histological course of alcoholic hepatitis. The decompensation rate was sourced from an epidemiologic analysis of patients from the UK General practice research database conducted by Fleming 2010. The figures on the prevalence of varices in people with cirrhosis were sourced from a review conducted by Berzigotti 2013 and were adjusted by assuming that two thirds of patients develop medium to large varices. Bleeding rates were obtained from Stevenson 2012 where separate rates were reported for drinkers and abstainers (for abstainers this was based on clinical judgement). Those were adjusted for decompensated cirrhosis patients according to the proportional increase reported in NIEC 1988 that was used in the HBV, HCV and NAFLD model cohorts. The HCC incidence rate was assumed to be constant across all people with cirrhosis (compensated or decompensated), an approach also followed by the Crossan 2015 HTA. Bleeding mortality was sourced from Stevenson 2012 and it was based on clinical judgement. Due to the lack of evidence in the remaining transition probabilities, the mean between the HBV and HCV cohorts were used after agreement with the GDG.

N.2.3.4. Life expectancy and mortality rates

Life tables for England and Wales, published by the Office of National Statistics (ONS) based on 2011–2013 mortality data were used to establish population mortality rates for men and women for ages 45 to 100 years.⁵⁶¹ ONS 2013 mortality statistics for England and Wales by cause of death⁵⁶⁰ were used to calculate the proportion of deaths for each 5-year age group which were due to liver related or non-liver related causes. These proportions were applied to the mortality rates to give the risk of death due to non-liver related causes for each annual age group for both men and women.

N.2.3.5. Utilities

N.2.3.6. Resource use and cost

N.2.3.6.1. Diagnostic test costs

The majority of the unit costs were sourced from the 2 relevant published HTAs.^177,207 The cost of ARFI VTq was built on top of the ultrasound NHS tariff (NHS reference costs 2013–14) assuming an extra kit has to be acquired in order to perform an ARFI examination. The cost of the kit was sourced from the relevant NICE M-Tec assessment.⁵³⁸ A machine lifespan of 5 years with 500 ultrasound or ARFI scans per year was assumed after GDG guidance. Point shear wave elastography cost was assumed to be similar to ARFI due to technology similarities and a lack of available evidence around it.

Table 63

Cirrhosis test unit costs.

N.2.3.6.2. Surveillance for complications costs

Table 64

Unit costs of surveillance.

N.2.3.6.3. Drugs

Unit costs were sourced from BNF 69. The dosages were either taken from the relevant NICE technology appraisals or were based on GDG guidance.

Table 65

Unit costs of drugs.

N.2.3.6.4. Health states

Health state costs were constructed with GDG guidance so they represent a reference patient pathway. These include staff, test, procedure and drug costs where relevant. When pegylated interferon was used as a drug treatment a more intensive management is assumed according to current clinical protocols. Staff costs were sourced from the NHS reference cost 2013/14 schedules and PSSRU 2014. A multi-speciality staff mix was also agreed with the GDG so that it better represents current care arrangements. Test costs where sourced from a relevant HTA (Donnan 2009). Complication costs related to cirrhosis were sourced from a HTA on HCV patients (Wright 2006) and were assumed to be relevant to all aetiologies. Under GDG guidance, complication costs of patients with ALD were assumed to be 50% higher than those in the other cohorts. Liver transplant costs for hepatitis B or C patients were sourced from Brown 2006 and Wright 2006. An average of those figures was used for the NAFLD and ALD aetiologies.

Table 66

Unit costs of staff.

Table 67

6-monthly health state costs (based on GDG guidance).

N.3.1.1. People with NAFLD

Table 69

Number of events & time spent in health states.

Table 70

Life years and results.

Figure 202

Cost-effectiveness plot: NAFLD.

Across the 6 strategies compared, the non-invasive tests ranked on top with transient elastography at a <15.0 threshold ranking first having an NMB of £165,224. All 3 non-invasive strategies delivered similar QALY figures and slightly differed in the overall mean costs. The confidence intervals in the rankings only excluded liver biopsy and the ‘no-test’ strategies from ranking first, highlighting the uncertainty in the cost-effectiveness of the 3 non-invasive tests. In the probabilistic analysis transient elastography at <15.0 ranked first in 51% of the simulations followed by ARFI and TE at 10.0 < 13.0 (36% and 12% respectively). ICERs comparing all strategies against ‘no test – no monitoring’ ranged from £13,868 to £14,577 for the non-invasive strategies and at £98,051 for liver biopsy.

N.3.1.2. People with ALD

Table 71

Number of events & time spent in health states.

Table 72

Life years and results.

Figure 203

Cost-effectiveness plot: ALD.

In the ALD cohort, testing for cirrhosis was not cost-effective at a £20,000 threshold with the 2 ‘no-test’ strategies ranking higher. The ‘no-monitor’ strategy had the highest NMB value of £68,697 and the ‘monitor-all’ strategy followed with £68,321. The diagnostic test that ranked first was transient elastography at 11.0–<13.0 with an NMB of £67,644. All diagnostic test strategies delivered considerably higher QALY values compared to no testing (up to 0.5 more QALYs) but at increased mean costs. All strategies apart from liver biopsy had wide confidence intervals of their ranks ranging from first or second to fifth and depicting the high uncertainty in the results. ICERs comparing all strategies against ‘no test – no monitoring’ ranged from £22,438 to £22,977 for the non-invasive strategies and at £25,405 for liver biopsy.

N.3.1.3. People with HBV: HBV− antigen

Table 73

Number of events & time spent in health states.

Table 74

Life years and results.

Figure 204

Cost-effectiveness plot: HBV− antigen.

In the HBeAg-negative cohort, APRI at 1.0 ranked first with an NMB of £103,281. Transient elastography at 11.0 kPa and FibroTest at 0.74 followed with NMBs of £103,237 and £103,210 respectively. Transient elastography delivered similar QALYs to APRI at 1.0 but for an incremental cost of £62 per patient. FibroTest was less costly then transient elastography and APRI at 1.0 but less effective too. Liver biopsy ranked lowest across all strategies particularly due to its high overall mean costs. In the confidence intervals of the ranks, transient elastography, FibroTest, APRI at 1.0 and ‘no test – no monitoring’ could all rank first with APRI at 1.0 ranking first in 45% of the simulations followed by FibroTest (23%). ICERs comparing all strategies against ‘no test – no monitoring’ ranged from £14,406 to £16,439 for the non-invasive strategies and at £67,013 for liver biopsy.

N.3.1.4. People with HBV: HBV+ antigen

Table 75

Number of events & time spent in health states.

Table 76

Life years and results.

Figure 205

Cost-effectiveness plot: HBV+ antigen.

In the HBeAg-positive cohort, FibroTest at 0.74 ranked first. Transient elastography at 11.0 kPa and APRI at 1.0 followed as second and third options. NMB for FibroTest was £158,366 and for transient elastography at 11.0 kPa and APRI at 1.0 was £158,362 and £158,358 respectively. Liver biopsy ranked lowest across all strategies particularly due to its high mean costs. The top 4 options (transient elastography, FibroTest, APRI at 1.0 and ‘no test – no monitoring’) had NMBs sufficiently close that it is impossible to be sure which of these should be preferred in terms of cost-effectiveness. Each could rank first within the confidence intervals. In the probabilistic analysis, ‘no test – no monitoring’ ranked first in 36% of the simulations with transient elastography, FibroTest, APRI at 1.0 all ranking first in about 20% of the simulations, each highlighting the high uncertainty in the results. ICERs comparing all strategies against ‘no test – no monitoring’ ranged from £19,039 to £25,418 for the non-invasive strategies and at £423,351 for liver biopsy.

N.3.1.5. People with HCV: genotype 1

Table 77

Number of events & time spent in health states.

Table 78

Life years and results.

Figure 206

Cost-effectiveness plot: HCV genotype 1.

In the HCV genotype 1 cohort liver biopsy ranked first with a NMB value of £216,472. Transient elastography at 13.0–<15.0 and the Castera algorithm followed with £215,580 and £215,251 respectively. Liver biopsy dominated all the other strategies apart from ‘no test – no monitoring or treatment’ by having the highest QALY value and the second lowest mean costs. Transient elastography at 13.0–< 15.0 delivered slightly lower QALYs for an incremental cost of £656. From all strategies it was only liver biopsy and transient elastography at 13.0–<15.0 that could rank first according to the ranking confidence intervals with liver biopsy ranking first in 90% of the simulations. ICERs comparing liver biopsy and TE at 13.0-<15.0 to ‘no testing – no monitoring or treatment’ were £2,720 and £2,897 respectively.

N.3.1.6. People with HCV: genotype 3

Table 79

Number of events & time spent in health states.

Table 80

Life years and results.

Figure 207

Cost-effectiveness plot: HCV genotype 3.

In the HCV genotype 3 cohort, liver biopsy ranked first with an NMB value of £225,611. Transient elastography at 13.0–<15.0 and the Castera algorithm followed with almost identical NMBs at £223,277 and £223,199 respectively. Liver biopsy dominated the Castera algorithm being more effective and less costly. Transient elastography at 13.0–<15.0 delivered marginally more QALYs but for a considerable incremental cost of £2,575. From all strategies it was only liver biopsy and transient elastography at 13.0–<15.0 that could rank first according to the ranking confidence intervals with liver biopsy ranking first in 97% of the simulations. The ‘no testing – no monitoring or treatment’ strategy was dominated by liver biopsy and TE at 13.0-<15.0 in direct comparisons.

N.3.1.7. People with HCV: all genotypes

Table 81

HCV diagnostic tests – top 3 ranked in every genotype.

N.3.2.1. Frequency of HCC surveillance

Table 82

ICERs comparing 6-monthly surveillance against annual surveillance.

The cirrhosis test used in each case was that recommended by the GDG following its consideration of the results of Section N.3.1. Where more than 1 test was recommended, the most cost-effective of those tests was used.

Across all aetiologies 6-monthly surveillance for HCC was overall more costly and more effective compared to the annual strategy. At a £20,000 threshold, 6-monthly surveillance was cost-effective only in the HCV genotype 1 cohort (ICER at £18,657). The ICERs in the remaining cohorts ranged between £23,220 and £28,352.

N.3.2.2. Frequency of oesophageal varices surveillance

Table 83

ICERs comparing annual and 2-yearly surveillance against 3-yearly surveillance.

The cirrhosis test used in each case was that recommended by the GDG following its consideration of the results of Section N.3.1. Where more than 1 test was recommended, the most cost-effective of those tests was used.

Surveillance for the presence of oesophageal varices every 2 years was not cost-effective for all cohorts but the two HCV. The ICER comparing 2-yearly with 3-yearly surveillance was below £20,000 for HCV genotype 1, dominating for genotype 3, but not cost-effective for NAFLD, ALD or HBV. Annual surveillance was not cost-effective for any aetiology at a £20,000 threshold.

N.3.3.1. NAFLD

Table 84

NAFLD model – Cost-effectiveness rank under different scenarios.

Across all scenarios transient elastography at >15.0 ranked first apart from when its unit cost was increased 20% and when its diagnostic accuracy was set at the low CI value. ARFI ranked first in both the aforementioned scenarios showing the amount of uncertainty between the 2 tests. Liver biopsy and the 2 ‘no test’ strategies remained last in all scenarios without a change in their rank.

N.3.3.2. ALD

Table 85

ALD model – Cost-effectiveness rank under different scenarios.

The ‘no test – no monitor’ strategy remained first in all scenarios. Transient elastography at 11.0–<13.0 remained the diagnostic test ranking first in 9 out of the 10 tested scenarios. In the remaining scenarios transient elastography at >15.0 ranked higher when the diagnostic accuracy of transient elastography at 11.0–<13.0 was set at its low CI.

N.3.3.3. HBeAg-negative

Table 86

HBV- model – Cost-effectiveness rank under different scenarios.

APRI at a 1.0 threshold remained first in 6 out of 10 scenarios and came second in the 5 remaining ones. FibroTest ranked first in 3 scenarios and second or third in the remaining ones. Transient elastography ranked first in 2 scenarios (20% lower fibroscan unit costs or diagnostic accuracy of transient elastography at its high CI) and ranked from second to fourth in the remaining scenarios. No substantial ranking changes are observed in the other test strategies.

N.3.3.4. HCV genotype 3

Table 87

HCV genotype 3 model – Cost-effectiveness rank under different scenarios.

In 13 out of the 15 scenarios liver biopsy remained the first ranked strategy. It came twentieth where HCV treatment was not provided and second where the drug treatment costs were reduced by 50%. The Castera algorithm remained second in 12 out of the 15 scenarios and third, fourth and nineteenth in the remaining 3 ones. Transient elastography at 11.0–<13.0 ranked third in 12 out of the 15 scenarios and first, second and fifth in the remaining 3 ones. Rankings in the remaining test strategies did not differ substantially across the scenarios tested apart from the ‘no HCV treatment’ scenario which seemed to favour the ‘no testing – no monitoring’ strategy.

N.3.3.5. Frequency of cirrhosis testing

Table 88

ICERs comparing 2-yearly testing against annual testing.

Annual testing was not found to be cost-effective at a £20,000 per QALY threshold. ICERs comparing annual and 2-yearly testing ranged from £25,975 to £1,060,920.

N.3.3.6. HCC surveillance frequencies

Table 89

ICERs comparing 6-monthly against annual surveillance.

Lowering the HCC surveillance costs had a moderately small effect on the ICERs with only the HCV cohorts being lower than the 20,000 threshold. Increasing the effectiveness of 6-monthly surveillance had a slightly larger effect still making 6-monthly surveillance cost-effective only in the HCV cohorts.

N.3.3.7. Oesophageal varices surveillance frequencies

Table 90

ICERs compared to 3-year surveillance.

Variation of the surveillance costs and the RR on the bleeding probability had little effect on the overall cost-effectiveness of more frequent surveillance for oesophageal varices. Increasing the frequency to 2 years was only cost-effective for the ALD cohort in 2 out of the 3 tested scenarios.

N.4.2.1. NAFLD

Transient elastography at a threshold of 15.0 kPa ranked first mainly due to having the highest diagnostic accuracy among the non-invasive tests. ARFI followed second being slightly less accurate but also having lower test unit costs. Transient elastography at 10.0 - <13.0 kPa ranked third having similar specificity to the other 2 tests but lower sensitivity. All 3 non-invasive tests had similarly wide confidence intervals (1 to 4).

In the deterministic sensitivity analysis, rankings were sensitive to increases in the transient elastography and ARFI unit costs and in the decrease of the diagnostic accuracy of transient elastography at 15.0 kPa. Therefore, no safe conclusion can be made over the most cost-effective option among the 3 comparators.

N.4.2.2. ALD

Testing people with alcohol-related liver disease for cirrhosis was not cost-effective compared to ‘no test – no monitor’ and ‘no test – monitor all’ at a cost-effectiveness threshold of £20,000 per QALY gained. However, it was cost-effective at a threshold of £30,000 per QALY gained: the ICERs for the 3 non-invasive liver tests were £22,438–£22,977). All 3 non-invasive tests had similarly wide confidence intervals (from first or second to fifth place).

In none of the deterministic sensitivity analysis scenarios did a test strategy rank higher than third. Ranking among the 3 non-invasive liver tests slightly varied across the different scenarios with transient elastography at 11.0–<13.0 remaining third in ranking for 9 out of the 10 tested scenarios.

N.4.2.3. HBV

For the HBeAg negative group, APRI at 1.0 ranked first, most probably due to its low test unit costs and its moderate diagnostic accuracy (second best after transient elastography). Transient elastography and FibroTest ranked second and third. APRI at 2.0 ranked last among the non-invasive liver tests mainly due to its considerably lower sensitivity. All non-invasive liver tests had similarly wide 95% confidence intervals.

In the HBeAg positive group, FibroTest ranked first with transient elastography and APRI at 1.0 ranking second and third. All non-invasive liver tests had similarly wide 95% confidence intervals. In the probabilistic analysis, the 3 tests also shared similar probabilities ranking first (20–23%).

Deterministic sensitivity analysis was only conducted for the HBeAg negative group. Rankings between the deterministic and the probabilistic analyses varied particularly for the FibroTest and transient elastography tests, highlighting how incorporating the uncertainty of the input parameters in the model affects the cost-effectiveness results. APRI at 1.0 ranked first or second in all scenarios. FibroTest and transient elastography followed with alternating first to fourth positions. The cost-effectiveness of APRI at 1.0 was sensitive to the decrease of HBV prevalence, the presence of varices at the point of cirrhosis diagnosis and changes to the cost and accuracy of transient elastography.

N.4.2.4. HCV

For all 4 genotypes, liver biopsy ranked first with substantially higher NMB values compared to the second options. This is mainly attributable to the fact that liver biopsy was assumed to have perfect sensitivity and specificity, and that cirrhosis misdiagnosis is associated with the incorrect administration of the highly costly polymerase inhibitor drugs. This led to the economic model particularly favouring the test with the highest diagnostic accuracy irrespective of its unit cost. In genotypes 1 and 3 where detailed results are presented, liver biopsy ranked first in 90% and 97% of the simulations respectively. Transient elastography at 13.0–<15.0 and the Castera algorithm ranked second and third in genotypes 1–4 and the ‘transient elastography or ARFI’ strategy ranked third in genotype 4.

Deterministic sensitivity analysis was only conducted for the genotype 3 group. Liver biopsy remained first in all but 2 scenarios. These were the ‘no HCV treatment’ and the ‘drug treatment cost - 50% lower’ scenarios, also highlighting how crucial the drug treatment element is for the HCV diagnostic model.

N.4.2.5. Frequency of HCC surveillance

At a cost-effectiveness threshold of £20,000 per QALY gained, 6-monthly surveillance was cost-effective compared to annual surveillance only for the HCV genotype 1 group. Although this group had the fewest liver-related deaths, risk of HCC progression was particularly high in this group compared to other model cohorts, making more frequent surveillance cost-effective at the specified threshold. However, at a cost-effectiveness threshold of £30,000 per QALY gained, 6-monthly surveillance was cost-effective compared to annual surveillance for all groups: ICERs £18,657–28,352. Variation in the ICERs was mainly due to differences in cirrhosis prevalence, risk of progression to HCC, and competing risks of other complications in each aetiology.

In the deterministic sensitivity analysis, changes in the surveillance costs or the 6-monthly surveillance effectiveness reduced the ICERs by up to £2,000 per QALY gained. Such reductions made 6-monthly surveillance cost-effective at a £20,000 per QALY gained threshold only for the 2 HCV cohorts.

N.4.2.6. Frequency of oesophageal varices surveillance

Annual surveillance was not cost-effective compared to 3-yearly surveillance for any of the model cohorts with the ICERs either exceeding £45,000 per QALY gained or showing it being dominated by the 3-year frequency option (more costly and less effective). Two-yearly surveillance was cost-effective compared to 3-yearly surveillance at a cost-effectiveness threshold of £20,000 per QALY in the 2 HCV cohorts. In the deterministic sensitivity analysis, changes in the surveillance costs or the RR applied on the bleeding probability had considerable effect on the ICERs of the higher frequencies. However with the base case ICERs of the deterministic analysis being far beyond the £20,000 threshold, any reductions in the ICERs made 2-yearly surveillance cost-effective only for the ALD cohort.

N.4.3.1. Cirrhosis diagnostic tests

Three relevant studies identified in our literature review attempted to assess the cost-effectiveness of diagnostic tests for cirrhosis, with contrasting findings.

Canavan 2013¹¹⁹ found TE to have an ICER of £6,557 when compared with no testing in chronic HCV patients. Liver biopsy was dominated by both these strategies. This is in contrast to the result of this model, however, the Canavan evaluation did not include the recently launched HCV treatments which particularly enhance the cost-effectiveness of highly accurate tests (such as liver biopsy) irrespective of their cost due to the very high treatment cost.

Steadman 2013⁷⁰⁹ concluded that liver biopsy was more costly and more effective compared to TE with a cost per additional correct diagnosis between £1,136 and 3,841 in the HBV, HCV and NAFLD groups. However, no safe conclusions or comparisons can be made based on these figures since important factors such as the follow-up costs and the health-related quality of life following correct or incorrect diagnoses were not included in this economic evaluation.

Stevenson 2012⁷¹¹ compared 6 relevant diagnostic strategies and concluded that only liver biopsy was cost-effective at a cost-effectiveness threshold of £20,000 per QALY gained for people with ALD. This is in contrast with the results of this model which indicated that neither non-invasive liver tests nor liver biopsy were considered cost-effective at the £20,000 threshold. The 2 models followed similar perspectives so result differences are mainly attributable to dissimilarities in the model structure and the input parameters (such as the strict liver biopsy quality criterion for the study selection followed by the present analysis).

N.4.3.2. Frequency of HCC surveillance

Two relevant studies identified in our literature review attempted to assess the cost-effectiveness of HCC surveillance in different frequencies.

Cucchetti 2012¹⁷⁸ compared annual versus 6-monthly surveillance and concluded that 6-monthly is not cost-effective for either groups with compensated or decompensated cirrhosis at a cost-effectiveness threshold of £20,000 per QALY gained (ICERs of £21,230 and £40,540 respectively). These figures are similar to the ones in the present analysis, which produced ICERs ranging between £20,000 and £30,000 across the different groups.

Thompson Coon 2008^732,733 compared 7 relevant strategies (including annual and 6-monthly frequencies) and concluded that only the ‘no surveillance’ strategy was cost-effective at a cost-effectiveness threshold of £20,000 per QALY gained. ICERs for the non-dominated strategies varied from £25,490 to £83,333. When 6-monthly strategies were directly compared with the annual ones, ICERs were beyond £27,500. The latter results are also in line with those in the present analysis.

N.4.3.3. Frequency of varices surveillance

No relevant studies were identified in the literature.