Model analyses and decision problem
The analysis evaluated the cost-effectiveness of ED testing in two locations; Leeds Teaching Hospitals NHS Trust (LTHT), and Guys and St. Thomas’ NHS Foundation Trust (GSTT) in London [13, 14]. In both EDs, the electronic patient record systems were modified to include a HBV and HCV test for adults receiving a blood test as part of their care, unless they opt-out.
A decision model was developed to evaluate the cost-effectiveness and budget impact of opt-out HBV and HCV testing for those attending the EDs. The comparator was no hepatitis testing in the ED, meaning those infected and undiagnosed remained so, at least until they receive testing in another setting in the future. The model evaluated the cost-effectiveness of HBV and HCV testing as two separate decisions, each compared to no hepatitis testing in the ED. This is reasonable as there were no shared costs between the tests, and so testing for one virus may be cost-effective whilst testing for the other may not be.
The cost-effectiveness analysis was performed from a UK National Health Service (NHS) perspective. Costs are presented in British pounds (GBP) in 2020 prices, with costs inflated using the NHS inflation index where necessary [15]. Health outcomes are presented as quality-adjusted life years (QALYs). The model was run over a lifetime time horizon, with an annual cycle length. Costs and outcomes are discounted at 3.5%, as per National Institute for Health and Care Excellence (NICE) guidelines [16]. The model estimates the incremental cost-effectiveness ratio (ICER) by dividing the incremental costs by the incremental QALYs of providing hepatitis testing compared to no testing. A simple budget impact analysis was also performed from a UK NHS perspective. This analysis primarily used the decision tree of the cost-effectiveness model to evaluate the costs associated with the intervention, which were assumed to be over a one-year period.
Model structure
The cost-effectiveness model used in this analysis is an adaptation of a previous decision tree and two Markov models used to evaluate hepatitis testing in EDs [12]. The Markov model structures are shown in the Appendix.
The decision tree captures the components of ED testing, including the prevalence of infection as indicated by the diagnostic test (HBsAg+, HCV RNA+, or uninfected), the proportion of patients requiring linkage to care (i.e. new diagnoses or those previously disengaged from care), and actual linkage to care (including the proportion attending clinic and proportion receiving treatment, if indicated). The decision tree informs the starting health state in the appropriate Markov model.
For HCV, people in early disease states (up to and including compensated cirrhosis) have the opportunity to receive treatment and achieve a sustained virological response (SVR). For HBV, those diagnosed can engage in care, which assumes they receive and adhere to treatment if indicated based on their clinical status, or can disengage with care, in which case they are assumed to receive no treatment. Those with HBV or HCV follow the progressive nature of these diseases, captured through Markov health states.
Model population
The mean age of individuals entering the model with HBV was 41.1 and 48.6, and with HCV was 42.3 and 47.1, as derived from LTHT and GSTT hospitals, respectively (Table 1).
Table 1
Key intervention and clinical parameters
Base case probabilities
|
Mean value
(LTHT, Leeds)
|
Distribution
|
Mean Value
(GSTT, London)
|
Distribution
|
Source
|
HBV Parameters
|
|
|
|
|
|
Age of HBV cases
|
41.1
|
N/A
|
48.6†
|
N/A
|
LTHT[13]/GSTT[14]
|
Prevalence (HBsAg)
|
0.5%
|
Beta (α = 73, β = 15980)
|
0.9%
|
Beta (α = 235, β = 27,411)
|
LTHT/GSTT
|
Proportion of diagnoses requiring linkage to care
|
53.4%
|
Beta (α = 39, β = 34)
|
57.4%
|
Beta (α = 135, β = 100)
|
LTHT/GSTT
|
Proportion attending referral
|
69.2%
|
Beta (α = 27, β = 12)
|
71.1%
|
Beta (α = 96, β = 39)
|
LTHT/GSTT
|
Proportion accepting treatment, post-referral (if indicated)
|
86.6%
|
Beta (α = 13, β = 2)
|
86.6%
|
Beta (α = 13, β = 2)
|
Parry[9]
|
Proportion HBeAg+
|
8.3%
|
Beta (α = 5, β = 55)
|
8.3%
|
Beta (α = 5, β = 55)
|
LTHT[13]
|
Proportion inactive disease (HBeAg + seroconverted or HBeAg- inactive disease)
|
80%
|
Beta (α = 80, β = 20)
|
80%
|
Beta (α = 80, β = 20)
|
PHE
|
Proportion cirrhotic
|
12%
|
Beta (α = 3, β = 22)
|
12%
|
Beta (α = 3, β = 22)
|
Parry
|
HCV Parameters
|
|
|
|
|
|
Age of HCV cases
|
42.3
|
N/A
|
47.1†
|
N/A
|
LTHT/GSTT
|
Prevalence (HCV RNA + or Ag+)
|
1.0%
|
Beta (α = 156, β = 15897)
|
0.9%
|
Beta (α = 261, β = 27396)
|
LTHT/GSTT
|
Proportion of Ab + testing RNA+/Ag + upon reflex test
|
45.6%
|
Beta (α = 156, β = 186)
|
49.9%
|
Beta (α = 261, β = 262)
|
LTHT/GSTT
|
Proportion of diagnoses requiring linkage to care
|
94.9%
|
Beta (α = 148, β = 8)
|
86.5%‡
|
Beta (α = 217, β = 34)‡
|
LTHT/GSTT
|
Proportion attending referral
|
51.4%
|
Beta (α = 76, β = 72)
|
23.5%
|
Beta (α = 51, β = 166)
|
LTHT/GSTT
|
Proportion receiving treatment, post-referral
|
53.9%
|
Beta (α = 41, β = 35)
|
51.0%
|
Beta (α = 26, β = 25)
|
LTHT/GSTT
|
Proportion F0
|
57.5%
|
Dirichlet (42, 12, 2, 5, 12)
|
22.7%
|
Dirichlet(10,10,10,7,7)§
|
LTHT/Parry
|
Proportion F1
|
16.4%
|
Dirichlet (42, 12, 2, 5, 12)
|
22.7%
|
Dirichlet(10,10,10,7,7)§
|
LTHT/Parry
|
Proportion F2
|
2.7%
|
Dirichlet (42, 12, 2, 5, 12)
|
22.7%
|
Dirichlet(10,10,10,7,7)§
|
LTHT/Parry
|
Proportion F3
|
6.8%
|
Dirichlet (42, 12, 2, 5, 12)
|
15.9%
|
Dirichlet(10,10,10,7,7)§
|
LTHT/Parry
|
Proportion cirrhotic (F4)
|
16.4%
|
Dirichlet (42, 12, 2, 5, 12)
|
15.9%
|
Dirichlet(10,10,10,7,7)§
|
LTHT/Parry
|
Proportion current PWID
|
61.3%
|
Beta (α = 84, β = 53)
|
61.3%
|
Beta (α = 84, β = 53)
|
LTHT
|
†The mean age for individuals diagnosed in GSTT was estimated from the proportion of patients in each age band, using a midpoint to calculate the mean age. |
‡Patients who were uncontacted were assumed to require linkage to care |
§ Sample size of 44, Dirichlet(10,10,10,7,7) |
Since injecting drug use is a major risk factor for HCV, the HCV model differentiated between people who inject drugs (PWID) and other individuals in the model. Data on the proportion of current PWID was available from LTHT (61%), and in the absence of data for GSTT, this proportion was also used for GSTT hospital. A sensitivity analysis considered data from another London based testing study which reported a slightly lower proportion of current or ex-PWID (54.5%), albeit from a sample of just 11 patients [9]. In the base case model, the disease progression, background risk of mortality, reinfection rate, and the background probability of receiving testing were all different for current PWID (see Appendix Table 1 and Appendix Table 4 for parameter details).
Prevalence and linkage to care
Amongst those receiving ED testing, 0.5% in LTHT and 0.9% in GSTT were HBV HBsAg positive. The HCV RNA prevalence was 1.0% in LTHT and 0.9% in GSTT (in GSTT all HCV antigen positive confirmatory tests were HCV RNA positive). The model assumes those testing HCV antigen positive are RNA positive. The proportion of HCV antibody positive tests which tested HCV RNA or antigen positive was similar in both locations (46–50%).
The linkage to care parameters were also derived from each ED. Amongst those testing positive, only those undiagnosed or previously diagnosed but not currently engaged in care required linkage to care. This was 53–57% for HBV, and 87–95% for HCV. If a patient remained uncontacted and did not have evidence of being a known diagnosis and already being engaged in care, the model assumed that they required linkage to care.
Patients requiring linkage to care in both EDs were contacted by various means, including phone calls, text messages, and letters to their home and registered GP. Full details are available in the original testing publications [13, 14]. Specialist outreach nurses or teams working with the homeless were also informed of those requiring linkage to care. The model captures the proportion that were engaged in care, defined as those who returned for at least one hospital appointment. For HBV, this was 69.2% in LTHT, and 71.1% in GSTT. For HCV, it was 51.4% in LTHT and 23.5% in GSTT.
For HBV patients, there was no data on the proportion that were offered treatment, or who accepted treatment. For both settings, we assumed 86.6% would accept treatment if offered, based on a separate ED testing study [9]. Treatment was only provided when clinically indicated (see treatment section below). For HCV, a proportion of those engaged would receive direct acting antiviral (DAA) treatment, and this was available from both settings. In LTHT, 53.9% of those engaged received treatment, with 51% receiving treatment in GSTT.
In GSTT, patient information was shared with local hospital homeless services, and the Find and Treat team (UCLH NHS Trust), a pan-London community inclusion health outreach team. This was successful in linking those testing positive to treatment, and is included within the above linkage to care data. A similar approach was taken in LTHT, with specialist nurse teams helping to find those with no fixed abode in GP practices and shelters, and in drug and alcohol services.
The severity of fibrosis was only available for those with HCV from LTHT. For GSTT, the fibrosis levels were derived from another ED testing study in London, although the values from LTHT were used in a sensitivity analysis [9]. The proportion of cirrhotic HBV patients was also derived from this study, since this data was not available in LTHT or GSTT.
Treatment parameters
Treatment for HBV followed NICE guidelines, with HBeAg + patients and those with HBeAg- active disease receiving pegylated interferon alpha-2a (PegIFN) for one year, followed by tenofovir disoproxil fumarate (TDF) if treatment continued [17]. Treatment sought to achieve HBeAg seroconversion or inactive disease. Full details of HBV treatment are available in the Appendix. Since many of those receiving HBV treatment will require long-term treatment, we included an annual probability of treatment disengagement (3.3% per year), to avoid overestimating the benefit associated with treatment.
All HCV patients from F0 to compensated cirrhosis health states accepting treatment were assumed to receive a pan-genotypic DAA. An estimated 93% of non-cirrhotic and 91% of cirrhotic patients achieve an SVR, based on UK outcome data [18]. We assumed that those not achieving an SVR would be re-treated once, with SVR rates of 95.3% and 81% for non-cirrhotics and cirrhotics, respectively [19].
Transition probabilities
The transition probabilities for the HBV model were derived from a UK Health Technology Assessment, and differed by HBeAg status [20]. Those on treatment were more likely to achieve inactive disease or HBeAg/HBsAg seroconversion which slowed disease progression, compared to those not receiving treatment. For early HCV states (F0 to compensated cirrhosis), transition probabilities were derived from a meta-regression of HCV progression rates, which differed for PWID and non-PWID [21]. For more advanced disease states (compensated cirrhosis progression onwards), transition probabilities were derived from a UK study [22]. For those achieving SVR, disease progression either halted (F0-F3 health states) or dramatically reduced (compensated cirrhosis state) [23, 24]. For HCV PWID, a standardised mortality ratio of 7.8 was also applied to the probability of death, which was applied for 11 years, the estimated duration of injecting [25, 26]. All transition probability values are available in the Appendix. Both models also include the risk of all-cause mortality in each cycle, based on UK life tables [27].
Background probability of testing
The background probability of testing for HBV and non-PWIDs with HCV was estimated by dividing the number of tests recorded in the PHE sentinel surveillance statistics of blood-borne virus testing in England (after adjusted for database coverage), by estimated adult population in England [28, 29]. The background probability of HBV testing was estimated to be 2.5% per year, from all HBV tests, except paediatric tests. For HBV, we adjusted the frequency in which those infected would be tested in the background rate of testing, to ensure that the proportion of positive tests was equal to the national average of 1.0% [28].
For non-PWID HCV diagnoses, the background probability of testing was estimated to be 2.1%, based on all HCV tests except paediatric tests and tests from setting which were likely to test PWID (drug dependency services, prisons, and pharmacies). For PWIDs, an estimated 26.8% received testing each year, from the PHE Unlinked Anonymous Monitoring survey of PWID [30]. We did not adjust the testing rates for HCV, as the background rate of testing was already elevated amongst PWID.
Costs
The costs associated with the intervention, including test costs, treatment related costs, and costs of healthcare appoints are provided in Table 2.
Table 2
Intervention and treatment costs
Costs (per year, except where noted)
|
Cost
|
Cost year
|
Distribution
|
Source
|
LTHT, Leeds
|
|
|
|
|
HBsAg test
|
£2.26
|
2019/20
|
N/A
|
LTHT[13]
|
HBsAg confirmation test
|
£13.36
|
2019/20
|
N/A
|
LTHT
|
HCV antibody test (initial)
|
£4.19
|
2019/20
|
N/A
|
LTHT
|
HCV antibody confirmation test
|
£9.58
|
2019/20
|
N/A
|
LTHT
|
HCV RNA test
|
£17.72
|
2019/20
|
N/A
|
LTHT
|
Nurse cost to contact positive case
|
£55.41
|
2019/20
|
N/A
|
Curtis[15]
|
GSTT, London
|
|
|
|
|
HBsAg test & confirmation
|
£5.79
|
2019/20
|
N/A
|
GSTT[14]
|
HCV antibody test + HCV antigen confirmation test
|
£6.67
|
2019/20
|
N/A
|
GSTT
|
HCV RNA test
|
£73.87†
|
2015/16
|
N/A
|
Bradshaw[10]
|
Nurse cost to contact positive case
|
£40.10
|
2019/20
|
N/A
|
Curtis[15]
|
Find and Treat – Engagement
|
£75.18†
|
2017/18
|
Uniform(£60.14, £90.21)
|
Ward [32]
|
Find and Treat – Peer support for hospital visit
|
£126.32†
|
2017/18
|
Uniform(£101.06, £151.59)
|
Ward [32]
|
Both locations
|
|
|
|
|
PegIFN (Annual)
|
£3672
|
2019/20
|
N/A
|
BNF[34]
|
TDF (Annual)
|
£366
|
2019/20
|
N/A
|
BNF[34]
|
TDF + Emtricitabine (Annual)
|
£1299
|
2019/20
|
N/A
|
BNF[34]
|
DAA treatment
|
£10,000
|
2019/20
|
N/A
|
Hurley[35]
|
DAA re-treatment
|
£15,000
|
2019/20
|
N/A
|
Hurley[35]
|
Cost of background test appointment
|
£33.19
|
2019/20
|
Uniform(£16.60, £49.79)
|
Curtis[15]
|
Pre-treatment evaluation (initial)
|
£207.86
|
2019/20
|
Uniform(£168.80, £253.20)
|
NHS reference costs 2019/20[33]
|
Pre-treatment evaluation (follow-up)
|
£164.75
|
2019/20
|
Uniform(£156, £234)
|
NHS reference costs 2019/20
|
DAA treatment monitoring
|
£823.75
|
2019/20
|
Uniform(£780, £1170)
|
NHS reference costs 2019/20
|
†Costs presented have been inflated to 2019/20, using NHS cost inflation index (2015/16 = 1.08, 2017/18 = 1.046) |
Test costs
The test costs were derived from each ED separately. In LHTH, costs were estimated using the base test cost, plus the laboratory add-on costs and bio-medical scientist time. In GSTT, individual test costs were provided. HBV testing consisted of a HBsAg test, plus a confirmatory test for those who test positive. For LTHT, these costs were incurred separately (£2.26 and £13.36 respectively), whilst in GSTT, a single cost was applied for each test (£5.79), whether a confirmatory test was required or not. The HCV testing approach differed in each setting. In LTHT, a HCV antibody test was performed (£4.19), with a confirmatory antibody test performed for any initial positive test (£9.58). Following two positive antibody tests, patients received an RNA test (£17.72). In GSTT, HCV testing consisted of an HCV antibody test, followed by a confirmatory HCV antigen test for those antibody positive, with a total cost of £6.67 (whether confirmatory testing occurred or not). For those testing antigen positive, an RNA test was performed, although the source of this cost was another ED testing study from London (£73.87) [10]. The costs of additional tests for viral markers or genotype were assumed to be included in the cost of the hospital visit, for those linked to care.
Contacting patients and healthcare costs
The model assumed that all patients testing positive were contacted, whether they required linkage to care or not. The costs of contacting patients from the ED was estimated from the salary costs of part-time nurses in both locations (band 6, 0.5 full-time equivalent for LTHT, band 7, 0.4 full-time equivalent for GSTT). Salaries were derived from UK personal social services, and divided by the number of positive tests to contact in each setting, giving an average cost of £55.41 per positive case in LTHT, and £40.10 in GSTT [15]. The costs were assumed equal for contacting HBV and HCV cases, although for HCV additional costs for outreach activities are applied. A sensitivity analysis considered a higher cost to contact each positive case, with a full-time band 6 nurse assumed to be contacting cases, with an average cost of £110.83 in LTHT, and £81.66 in GSTT, per positive case.
In addition, for HCV we included the cost of an inclusion health team, who were responsible for outreach services to link patients to care, based on collaboration between GSTT and UCL Find and Treat, with costs derived from a previous economic evaluation of their service [31, 32]. The average costs of engaging RNA positive patients was £75.18 per person, and peer support for engagement for each hospital visit was £126.32 per person. This was applied to 31.3% of those requiring linkage to care (31/99), and 37.3% of those requiring a hospital visit (19/51), based on data from GSTT. To be conservative, the same costs and proportions were applied to LTHT, since nurses had close contact with specialist GPs for the homeless, and close links to drug and alcohol centres, but no data for proportions were recorded as these links were informal and existed previously.
For HBV and HCV patients returning and engaged in care, the model assumes an initial healthcare visit (£207.86) derived from NHS reference costs [33]. For HBV, those who are fully engaged (and would receive treatment if indicated) incurred a second visit cost (£164.75), whether they did receive treatment or not, based on their health status [33]. For HCV, those returning for DAA treatment also incurred a second visit cost (£164.75).
Treatment costs
HBV treatment costs were derived from the British National Formulary (BNF) [34]. For HCV, the NHS has negotiated a confidential price reduced on the costs of DAA treatment, expected to be approximately £5,000 per successful treatment [35]. To remain conservative, we assumed a cost of £10,000 per DAA treatment, and £15,000 for re-treatment, with the cost incurred only upon SVR, as per NHS policy [36]. We explored lower DAA costs in sensitivity analyses. DAA treatment monitoring costs were assumed to consist of five visits for HCV patients (£823.75) [33]. HBV treatment monitoring was assumed to be included in the specific health state costs.
Health state costs
The health state costs for HBV and HCV were derived from two UK HTA reports and other literature sources (Appendix Table 5) [20, 22, 37]. The health state costs were only incurred for those patients diagnosed. Health state costs were lower for those achieving seroconversion or inactive disease (HBV) or achieving SVR (HCV).
Utility
Utility values were assigned to each health state to estimate the health-related quality of life. For early HBV health states (pre-cirrhotic), utilities were derived from a study of over 400 HBV patients [38]. For early HCV health states, utilities were derived from a large meta-analysis of studies in HCV patients, with utility estimated using the EQ-5D-3L [39]. The same source was used for later health states (compensated cirrhosis onwards), and the utility values for each health state were assumed the same for HBV and HCV patients [39].
Cost-effectiveness analyses
One-way deterministic sensitivity analyses were performed on parameters of interest, by individually changing parameter values and observing the impact upon the ICER. A probabilistic sensitivity analysis (PSA) was performed to capture the parameter uncertainty in the model. Distributions were assigned to appropriate model parameters, with each sampled simultaneously across 10,000 Monte Carlo simulations. The parameters included, and distributions used, are provided in Table 1 and Table 2, and Appendix Tables 1–6.
A probabilistic threshold analysis was also performed, to consider the minimum prevalence at which HBV and HCV testing would remain 90% cost-effective. This involved running the same PSA as described above, with prevalence values increasing by 0.05% increments. This has been described as a two-level Monte Carlo approach [40]. This was performed for both settings individually, and when combining results for the two settings.
Budget impact analysis
A budget impact analysis was performed to estimate the costs associated with HBV and HCV testing and linkage to care. The analysis assumes testing is performed for one year (at the same testing rate as the two studies) [13, 14]. For both HBV and HCV, the budget impact estimated the costs of the intervention, including all tests, contacting positive patients, and healthcare visits for those engaging with care (Table 2). The budget impact analysis only focused on the ED-based costs of the intervention, including testing and linking patients into care, and therefore did not consider treatment costs.