This chapter explores the cost-effectiveness of hs-cTn assays (used singly or in series, up to 4 hours from the onset of chest pain/presentation) compared with the current standard of serial Tn T and/or I testing on admission and at 10–12 hours after the onset of symptoms for the early rule-out of AMI in people with acute chest pain.
Model structure and methodology
Troponin tests considered in the model
The health-economic analysis will estimate the cost-effectiveness of different Tn testing methods for diagnosing or ruling out NSTEMI, in patients presenting at the ED with suspected NSTE-ACS, who have no major comorbidities requiring hospitalisation (e.g. as HF or arrhythmia) and in whom STEMI has been ruled out. Those diagnosed with NSTEMI will then be admitted to the hospital for AMI treatment and those diagnosed as without NSTEMI can be discharged without AMI treatment and further hospital stay. AMI treatment might include aspirin, statins and ACE inhibitors and consideration of coronary revascularisation for high-risk cases.7 Initiating AMI treatment for NSTEMI will reduce the probability of MACEs, particularly cardiac death and re-infarction.
Standard serial Tn testing, for patients with acute chest pain attributable to possible ACS, does not achieve optimal sensitivity in detecting AMI until 10–12 hours after onset of symptoms. Waiting for 10–12 hours after symptom onset is burdensome for patients and induces additional health-care costs. Therefore, various alternatives have been proposed, using more sensitive Tn tests, for the early rule-out of NSTEMI (within the 4-hour NHS ED target).89
Two hs-cTn assays (Roche Elecsys hs-cTnT and Abbott ARCHITECT hs-cTnI) are currently used in NHS laboratories in England and Wales. One additional assay (Beckman Coulter hs-cTnI) was listed in the scope for this assessment, pending CE marking. However, each of these tests can be used at different time points and with different diagnostic thresholds, resulting in multiple possible strategies for each test. Whether or not a test strategy was included in the economic model was decided based on optimal diagnostic performance, given the available evidence on accuracy for a population with STEMI ruled out, and on applicability in clinical practice (see Results of the assessment of clinical effectiveness assessment, above). The test strategies evaluated in the model are:
Standard Tn at presentation and at 10–12 hours (reference standard).
Roche Elecsys hs-cTnT at presentation: 99th centile threshold.
Roche Elecsys hs-cTnT (optimal strategy): LoB threshold at presentation followed by 99th centile threshold peak within 3 hours and/or Δ20% (compared with presentation test) at 1–3 hours (see ).
Abbott ARCHITECT hs-cTnI at presentation: 99th centile threshold.
Abbott ARCHITECT hs-cTnI (optimal strategy): LoD threshold at presentation, followed by 99th centile threshold at 3 hours (see ).
Beckman Coulter hs-cTnI at presentation: 99th centile threshold.
No testing, discharge all patients without testing or treatment (only in sensitivity analyses). A Tn test may not be indicated when clinical judgement assesses the probability that a patient is experiencing an AMI as low. Therefore, consistent with the protocol, this hypothetical strategy is included in sensitivity analyses wherein the AMI prevalence is varied.
In the base case, it was assumed that standard Tn had perfect sensitivity and specificity (reference case) for diagnosing AMI. Using this assumption, all patients testing positive on a hs-cTn test but negative on the standard Tn would be classified as FPs. This implies that their risk for adverse events would be the same as for those patients testing negative on both the hs-cTn test and the standard Tn, and that they ought to be discharged home without further immediate treatment. However, recent evidence has shown that patients with a negative standard Tn, but a positive hs-cTn, may be at higher risk for adverse events than patients who test negative on both the standard and the high-sensitive Tn (Goodacre S, Medstar Washington Hospital Center, Washington, DC, USA; Lipinski M, University of Sheffield, Sheffield, UK: 2014, personal communication). A secondary analysis was therefore performed, which attributed a higher risk of adverse events to a proportion of patients testing FP with the hs-cTn test.
Based on the available evidence, two analyses were performed:
Base-case analysis.
Secondary analysis, assuming that FPs in the hs-cTn testing strategies do not have the same risk for adverse events as TNs. Instead, these patients were assigned a higher risk for (re-)infarction and death, to reflect the idea that when the hs-cTn test gives a positive result, in some cases this must be caused by a disease process, whether or not the strict definition of AMI is met. The risk of adverse events in patients with positive hs-cTn but a negative standard Tn is higher than the patients testing negative on both the hs-cTn test and the standard Tn, but lower than risk of adverse events in patients diagnosed with NSTEMI (i.e. both positive hs-cTn and standard Tn).
Model structure
This assessment uses the HTA report by Goodacre et al.7 as a starting point for cost-effectiveness modelling. The Goodacre report compared the cost-effectiveness of several diagnostic strategies for ACS. The assessment group received the health-economic model (in SIMUL8 2011, Simul8 Corporation, Boston, MA, USA) that this HTA was based on, and this model was used as a starting point to develop a de novo model (in Microsoft Excel 2003: Microsoft Corporation, Redmond, WA, USA) adapted to better fit the scope of the current assessment. In the health-economic model the mean expected costs and QALYs were calculated for each alternative strategy. These long-term consequences were estimated based on the accuracy of the different testing strategies followed by AMI treatment or discharge from the hospital without AMI treatment for patients presenting at the ED with suspected NSTE-ACS, including patients with NSTEMI and patients without NSTEMI, who are further subdivided into ‘No ACS, no UA’ and ‘Unstable angina’. For this purpose a decision tree and a Markov model were developed. The decision tree was used to model the 30-day outcomes after presentation, based on test results and the accompanying treatment decision. These outcomes consisted of ‘No ACS, no UA’, ‘Unstable angina’, ‘Non-fatal AMI (untreated)’, ‘Non-fatal AMI (treated)’ and ‘Death’. The decision tree is shown in .
The long-term consequences in terms of costs and QALYs were estimated using a Markov cohort model () with a lifetime time horizon (60 years). The cycle time was 1 year, except for the first cycle, which was adjusted to 335.25 days (365.25–30) to ensure that the decision tree period (30 days) and the first cycle combined summed to 1 year. The following health states were included:
Markov model structure. a, During the first year post AMI, a distinction is made between treated and untreated AMI.
‘No acute coronary syndrome and no unstable angina (no ACS, no UA)’
‘Unstable angina’
‘Post AMI (treated and untreated)’
‘Post AMI with re-infarction’
‘Death’.
Model parameters
Estimates for the model input parameters were retrieved from the literature and by consulting experts for unpublished data. Accuracy estimates were derived from the systematic review component of this assessment (see Results of the assessment of clinical effectiveness assessment, above).
Transition probabilities
An overview of transition probabilities is provided in .
Decision tree
The proportions of patients testing positive or negative (and thus commencing AMI treatment or being discharged from the hospital) were based on the estimated accuracy of the testing strategies considered () and the estimated prevalence of NSTEMI in the UK [17.0% with standard error (SE) 2.8%; see ].39,40,46,64 This prevalence was higher than that derived from the RATPAC trial78 and used in the Goodacre model,7 because the RATPAC study population was a low-risk population.79,87 The proportion of TPs, FPs, FNs and TNs were calculated as follows:
TP = NSTEMI prevalence × sensitivity
FP = (1 – NSTEMI prevalence) × (1 – specificity)
FN = NSTEMI prevalence × (1 – sensitivity)
TN = (1 – NSTEMI prevalence) × specificity.
Subsequently, the proportions of patients who receive AMI treatment (TP + FP), and who are discharged without AMI treatment (TN + FN) were calculated. These results are listed in .
After treatment, TP patients in the decision tree were allocated to ‘Non-fatal AMI (treated)’ and FP patients were further subdivided between ‘No ACS, no UA’ and ‘Unstable angina’ (based on the proportion of UA among non-NSTEMI patients; see ). After being discharged, TN patients were also subdivided between ‘No ACS, no UA’ and ‘Unstable angina’, whereas FN patients were allocated to ‘Non-fatal AMI (untreated)’. The proportions of FNs, reported in , can be considered as the proportions of AMIs that would have been missed when assuming that standard Tn testing has perfect accuracy. Finally, to calculate the total number of deaths in the decision tree, the probability of 30-day mortality was assigned based on abovementioned subdivision (see ). It was assumed that UA is always correctly diagnosed, hence the mortality probability for treated UA was used.
Markov model
The age-dependent AMI incidence in the UK92 was used to model the occurrence of AMI for patients in the health states ‘No ACS’ and ‘Unstable angina’. It was assumed that all AMIs in the Markov trace are diagnosed correctly and thus receive treatment. For patients in the ‘Post-MI’ health state, the probability of re-infarction after treated AMI was retrieved from a UK record linkage study (n = 387,452), which assessed long-term survival and recurrence after AMI.93 For the current assessment the probabilities for females and males were weighted according to the estimated proportion of females and males in the population (males = 58.1%7). The re-infarction probability for the ‘Post-MI with re-infarction’ health state is equal to the re-infarction probability for the ‘Post-MI’ health state. The re-infarction RR for people with untreated AMI compared with treated AMI was calculated from a recent study by Mills et al.83 based on patients with a Tn concentration of 5–19 ng/l. This RR was assumed only for the first year after presentation at ED, after which no increased risk was assumed (i.e. RR = 1.0 for untreated vs. treated AMI after year 1).
Age-dependent mortality from the general population was used for patients in the ‘No ACS, no UA’ health state.91 For the ‘Post-MI’ and ‘Post-MI with re-infarction’ health states, mortality was extracted from the record linkage study.93 Again, the study by Mills et al.83 was used to calculate the mortality RR for untreated AMI compared with treated AMI for the first year, after which an RR of 1.0 was used. Finally, a multivariate adjusted mortality hazard ratio (HR) for UA compared with NSTEMI was retrieved from a study by Allen et al.94 to calculate mortality after UA.
All input parameters for the Markov model are reported in .
Health-state utilities
Age-dependent utility scores, from the UK general population, were calculated for patients in the ‘No ACS, no UA’ health state based on a linear regression model.85 These age-dependent utility scores from the general population were combined with age-dependent disutilities for AMI82 to calculate utilities for the ‘Post-MI’ health states (with or without re-infarction). Utility scores for the ‘Unstable angina’ health state were calculated based on Post-MI utility scores and a utility increment of 0.010 ().85
Resource use and costs
Test-specific resource use consisted of the number of tests performed and the duration of hospital stay (hours) before discharge/AMI treatment ().
Resource use (test specific)
Health-state costs () were mainly retrieved from previous economic evaluations conducted in the UK.85,97 Health-state costs for the ‘Unstable angina’, ‘Post-MI’ and ‘Post-MI with re-infarction’ consisted of costs for three 15-minute general practitioner consultations and medication costs.85 For the first year in the ‘Unstable angina’ health state, costs for clopidogrel (for 60%) and hospitalisation (for 50%) were added to this. The first year costs for both ‘Post-MI’ health states were based on resource data from the Nottingham Heart Attack Register.97
Health-state costs, event costs and unit prices
Additionally, costs of fatal events, retrieved from a UK economic evaluation,85 were accumulated for all fatal AMIs. For this purpose, it was assumed that all 30-day deaths after ‘true’ NSTEMI were due to a fatal AMI event. In addition, AMI treatment costs were calculated based on the national tariff for non-elective AMI without complications [Healthcare Resource Group (HRG) code: EB10Z].98 To calculate the hospital stay costs for patients, based on the number of hours before the test results become available, non-elective NHS reference costs for the general medical ward were used (HRG code: EB01Z).98 For this purpose, it was assumed that doctors were available on demand, and the time to discharge was delayed because of time between arrival at the ED and start of first sampling (1 hour) and the time between sampling and the results being available (2 hours). In the case of multiple testing, the 1-hour delay between arrival at the ED and start of sampling was applied to only the first test; however, this also affected the timing of the second test if applicable. The 2-hour delay before test results become available applies to all tests performed. Incorporating these time delays effectively implies that only tests at presentation and tests performed 1 hour after presentation could inform decisions within the NHS 4-hour ED target. All other multiple testing strategies, as well as standard Tn testing at 10–12 hours, would require a transfer from the ED to the general ward (patients are transferred to the general ward 4 hours after presentation at the ED). Finally, the test costs include panel (including reagent, machine and maintenance), calibration and quality control costs. Depending on the annual number of panels, the test costs varied between £16.18 and £21.33, for annual rates of testing of 1500 and 3000, respectively.80 Based on clinical expert input, the average test costs were estimated to be £20 (2011 price level).7,80
Overview of main model assumptions
The main assumptions in the health-economic analyses were:
Serial Tn testing (comparator) has perfect accuracy (sensitivity = 1.0 and specificity = 1.0).
For the Roche Elecsys hs-cTnT and Abbott ARCHITECT hs-cTnI optimal strategies it was assumed that the sensitivity and specificity for the subpopulation not discharged after the presentation test is equal to the sensitivity and specificity for the initial group (presenting at the ED).
The life expectancy, quality of life and costs for FP patients is, in the base-case analysis, equal to the life expectancy, quality of life and costs of TN patients. This assumption was amended in the secondary and sensitivity analyses.
In contrast with AMIs occurring during the decision tree period, all AMIs (either first or re-infarction) occurring in the Markov trace are diagnosed correctly and thus treated.
UA is always correctly diagnosed and thus treated.
The re-infarction probability for the ‘Post-MI with re-infarction’ health state is equal to the re-infarction probability for the ‘Post-MI’ health state.
The increased ‘Post-MI’ re-infarction and mortality probabilities for untreated AMI were assumed to last 1 year: afterwards a RR of 1.0 was applied (for untreated vs. treated AMI).
There is no additional benefit of starting treatment early, so treatment effect for high-sensitive strategies is equal to treatment effect for standard Tn strategy.
All 30-day deaths (after presentation at the ED) are due to fatal AMI events and will receive the associated costs.
Model analyses
Expected costs, LYs and QALYs were estimated for all Tn testing methods. Discount rates of 3.5% and a half-cycle correction were applied for both costs and effects. Incremental cost and QALYs for each strategy compared with standard Tn, and compared with the next best alternative, were calculated. The ICER was then calculated by dividing the incremental costs by the incremental QALYs. PSAs (10,000 simulations) were performed, and cost-effectiveness acceptability curves (CEACs) and cost-effectiveness acceptability frontiers (CEAFs) were constructed. Although CEACs can be used to illustrate decision uncertainty, the option with the highest probability of being cost-effective may not necessarily be the most cost-effective option according to the expected values. Moreover, CEAFs can be used to illustrate the decision uncertainty surrounding the most cost-effective option.100
Secondary analysis
For the base case, it was assumed that patients who tested negative on standard Tn and positive on hs-cTn tests would experience life expectancy and quality of life equal to TN patients. This assumption is, however, debatable, as unpublished data (Goodacre S, Lipinski M, personal communication) show that patients with a negative standard Tn test and positive hs-cTn test have an increased risk of (re-)infarction and mortality compared with those who test negative on both standard Tn and hs-cTn tests. Although this risk was not as high as in patients with both positive standard Tn and positive hs-cTn tests, it could still be considered prognostically important. Therefore, in this secondary analysis the risk of re-infarction and mortality was adjusted for patients who tested FP (see ). It was assumed that for this proportion of patients, the relative treatment benefit would be equal to that for TP patients. As the prevalence of this ‘higher risk subgroup’ is likely to be the same for all comparators, it was assumed that this proportion was equal to the lowest proportion of FP patients for all hs-cTn tests (see ). This ‘higher risk subgroup’ was assumed to be treated for all hs-cTn tests (as they tested positive with these tests) and untreated for the standard Tn test (as they tested negative with this test), thus affecting the probability of adverse outcomes and treatment costs. In addition, the post-MI utility and health-state costs were used for this ‘higher-risk subgroup’.
Sensitivity analysis
For both the base case and the secondary analysis, the following one-way sensitivity analyses were performed to assess the impact of model assumptions and input parameters on the estimated outcomes:
Model assumptions:
The assumption that the increased post AMI re-infarction and mortality probabilities for untreated AMI lasts for only 1 year was replaced by the assumption that these probabilities would remain elevated for a lifetime.
The assumption that a doctor will be available on demand and thus that a decision could be made immediately (as in the base case) was replaced with an assumed delay (1, 2 or 3 hours) before a doctor is available and a decision could be made.
As for the previous sensitivity analysis, except that the delay (1, 2 or 3 hours) applies only once patients are transferred to the general ward 4 hours after presentation (no delay in the ED).
A total delay of 1.5 hours is assumed (includes delay from the time at which sampling could be performed to the time at which results became available and delay between arrival at hospital and Tn assessment commencing) rather than assuming a total delay of 3 hours (base case).
AMI treatment costs are applied for patients who tested FP rather than using no treatment costs, as assumed in the base-case analysis.
In addition to the health-state costs of UA during the first year, the AMI treatment costs are also applied for patients with UA (during the first year), rather than assuming no additional treatment costs.
Model input parameters (varied to lower and upper boundary of the 95% CI unless stated otherwise):
test costs [test costs was varied over a wider range (£5–40) than the 95% CI]
AMI treatment costs (±25%)
post-MI first-year health-state costs
utility increment for UA compared with AMI
post-MI disutility compared with no ACS
mortality (30 day) treated AMI (decision tree)
mortality (30 day) untreated AMI (decision tree)
annual re-infarction (after initial AMI)
RR re-infarction (untreated vs. treated AMI)
annual post-MI mortality
annual post-MI mortality after re-infarction
HR mortality (UA vs. NSTEMI)
RR mortality (untreated vs. treated AMI).
Subgroup analysis
For both the base case and the secondary analysis, a number of subgroup analyses were performed. The main subgroup analyses were based on age- and sex-dependent re-infarction probabilities, mortality probabilities (for all health states), AMI incidence and quality of life, and could be applied to all test strategies. Accuracy was thus assumed to be subgroup independent (equal to the base case values). The following subgroups were identified:
Sex.
Age (45, 55, 65, 75 and 85 years).
People with a history of previous NSTEMI. For this purpose, a proportion of 0% UA was assumed and the probabilities for the initial ‘Post-MI’ health state were used for the ‘No ACS, no UA’ health state and the probabilities for ‘Post-MI with re-infarction’ were used for the ‘Post-MI’ and ‘Post-MI with re-infarction’ health states. This subgroup analysis was performed for only the base case, as for the secondary analysis this would lead to lower mortality probabilities for FP patients than TN patients (which seems implausible).
Subgroups with varying AMI prevalence (1%, 5%, 10%, 20%, 30%). In these analyses the no-testing strategy was included as a comparator, as a Tn test may not be indicated when clinical judgement assesses that the probability that a patient is experiencing an AMI is low. For the no-testing strategy it is assumed that patients will be discharged immediately.
It should be noted that the main subgroup analyses (described above) differ from the subgroups described in the systematic review component of this assessment (see Chapter 3, Presentation samples), for which specific accuracy and prevalence data were available. Additional subgroup analyses were performed based on these subgroup-specific accuracy data. However, these analyses could be performed for only the Roche Elecsys hs-cTnT assay at presentation sample, using the 99th centile diagnostic threshold, compared with standard Tn testing; no subgroup-specific accuracy data were available for the other two hs-cTn assays. The following subgroups were considered:
age ≤ 70 years and age > 70 years
patients with pre-existing CAD and patients without pre-existing CAD
symptom onset at < 3 hours before presentation and symptom onset at ≥ 3 hours before presentation.
The subgroups with high pre-test probability and low-to-moderate pre-test probability were not considered, as the prevalence data for these subgroups were unknown.
Results of cost-effectiveness analyses
This section describes the results using probabilistic analyses for the base-case analysis and the secondary analysis. In addition, the sensitivity analyses (deterministic) and subgroup analyses are described (these deterministic analyses are also presented in tabulated form in Appendices 5–9.
Base-case analysis
The base-case analysis includes six test strategies. and show the probabilistic results of this analysis. Standard Tn testing was both most effective (15.101 LYs, 11.730 QALYs) and most expensive (£2697). The Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was least effective (15.076 LYs, 11.712 QALYs) and least expensive (£2253). Compared with standard Tn testing, hs-cTn testing resulted in ICERs ranging between £90,725 and £24,019 savings per QALY lost.
Probabilistic results for base-case analysis: LYs
Probabilistic results for base-case analysis: costs and QALYs
Comparisons based on the next best alternative showed that for willingness-to-pay values of < £6600 per QALY, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, would be cost-effective. For thresholds between £6600 and £30,631 per QALY, the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective; above £30,631 per QALY the Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective. Standard Tn becomes cost-effective at a threshold of £90,725 or higher (see ).
At willingness-to-pay thresholds of £20,000 and £30,000 per QALY, the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, had probabilities of being cost-effective of 47% and 35%, respectively. Although the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective at a willingness-to-pay threshold of £30,000 per QALY, the Abbott ARCHITECT hs-cTnI optimal strategy had the highest probability of being cost-effective (35%) at this threshold ( and ).
Cost-effectiveness acceptability curve and incremental cost-effectiveness plane (incremental costs and QALYs compared with standard Tn) for base-case analysis.
Cost-effectiveness acceptability frontier for base-case analysis.
Secondary analysis
The secondary analysis includes the same six test strategies. This analysis assumed that in a proportion of patients with a FP hs-cTn test (i.e. positive hs-cTn test and a negative standard Tn test), there is prognostic significance [i.e. it is associated with an increased risk of adverse events (mortality and re-infarction)].
Standard Tn testing was least effective (14.785 LYs, 11.464 QALYs) and most expensive (£3058). The Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was the least effective hs-cTn test strategy (14.833 LYs, 11.501 QALYs) and, overall, the least expensive strategy (£2781). The Abbott ARCHITECT hs-cTnI optimal strategy was most effective (14.855 LYs, 11.518 QALYs). Standard Tn testing was dominated by all hs-cTn testing strategies.
Comparisons based on the next best alternative showed that for willingness-to-pay values of < £13,623 per QALY, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective. For thresholds between £13,623 and £14,562 per QALY, the Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, was cost-effective; above £14,562 per QALY the Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective ( and ).
Probabilistic results for secondary analysis: LYs
Probabilistic results for secondary analysis: costs and QALYs
At willingness-to-pay thresholds of £20,000 and £30,000 per QALY, the Abbott ARCHITECT hs-cTnI optimal strategy had the highest probability of being cost-effective (53% and 67%, respectively; and ).
Cost-effectiveness acceptability curve and incremental cost-effectiveness plane (incremental costs and QALYs compared with standard Tn) for secondary analysis.
Cost-effectiveness acceptability frontier for secondary analysis.
Sensitivity analysis
The deterministic analysis for the base-case analysis is presented in Appendix 5. When it was assumed that the post-MI re-infarction and mortality probabilities would remain elevated for untreated AMI for a life-time period, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds of < £1642 per QALY, at which point the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, became cost-effective up to a threshold of £7602 per QALY. The Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for thresholds between £7602 and £26,532 per QALY. Standard Tn testing was cost-effective for thresholds of > £26,532 per QALY. Consistent with the base-case analysis, all ‘no doctor on demand’ sensitivity analyses (1, 2 or 3 hours) showed that the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds between approximately £8000 and £40,000 per QALY. Similarly, where the total delay decreased to 1.5 hours (and assuming availability of a doctor on demand), the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds between £7778 and £29,653 per QALY, at which point the ARCHITECT hs-cTnI optimal strategy became cost-effective. Adding AMI treatment costs for the patients with a FP test substantially impacted upon the results: standard Tn testing was cost-effective for all threshold values of > £16,050 per QALY. Adding AMI treatment costs to the UA health state for the first year had a negligible impact on the incremental outcomes.
The following input parameters had a noticeable impact on the estimated cost-effectiveness: 30-day mortality for treated and untreated AMI (decision tree) and the mortality RR for treated AMI compared with untreated AMI (Markov trace). Varying the remaining parameters did not have a substantial impact on the results (i.e. the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds between approximately £10,000 and £35,000 per QALY).
The deterministic analysis for the secondary analysis is presented in Appendix 6. When assuming that the post-AMI re-infarction and mortality probabilities would remain elevated for untreated AMI for a life-time period, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds of < £1853 per QALY, at which point the Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, became cost-effective up to a threshold of £2017 per QALY. The Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds between £2017 and £5889 per QALY. The Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for thresholds of > £5889 per QALY. For all ‘no doctor-on-demand’ sensitivity analyses, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds of < £18,000 per QALY for 1, 2 and 3 hours’ delay. The Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds between £18,000 and £19,000, £20,000 and £22,000 per QALY in case of 1, 2 and 3 hours’ delay, respectively. The Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for higher thresholds. Similarly to the deterministic base case, for which the total delay decreased to 1.5 hours (assuming availability of a doctor on demand), the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds of below £14,956, at which point the ARCHITECT hs-cTnI optimal strategy became cost-effective. Adding AMI treatment costs for all patients with a FP test gave similar results to the deterministic analysis: the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for all threshold values of < £15,508 per QALY, at which point the Abbott hs-cTnI optimal strategy became the preferred option. Adding AMI treatment costs to the ‘Unstable angina’ health state for the first year had a negligible impact on the incremental outcomes.
The following input parameters had a noticeable impact on the estimated cost-effectiveness of the secondary analysis: increased test cost (of £40 per test), 30-day mortality for treated and untreated AMI (decision tree), and the re-infarction and mortality RR for treated AMI compared with untreated AMI (Markov trace). Varying the remaining parameters did not have a substantial impact on the results.
Subgroup analysis
Additional analyses were performed for subgroups based on age, sex, people with a history of previous NSTEMI, and AMI prevalence. These deterministic subgroup analyses (for the base case) analysis are presented in Appendix 7. Consistent with the base-case analyses, analyses based on age and sex subgroups indicated that, up to an age of 75 years, the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds between approximately £10,000 and £35,000 per QALY. The Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for higher thresholds up to £115,000–170,000, at which point standard Tn testing became cost-effective. For females aged > 85 years, the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds between £15,793 and £74,597 per QALY; the Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for thresholds between £74,597 and £259,592 per QALY, and standard Tn testing was cost-effective for thresholds of £259,592 per QALY and higher. For males aged > 85 years, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds of < £28,711 per QALY; the Beckman Coulter hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds between £28,711 and £143,225 per QALY and the Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for thresholds between £143,225 and £503,476 per QALY, at which point standard Tn testing became cost-effective. The results for the subgroup with a history of previous NSTEMI were almost identical to the base-case analysis.
For subgroup analyses considering AMI prevalence, no testing was included as additional comparator. For an AMI prevalence of 1%, the no-testing strategy was cost-effective up to thresholds of £27,409 per QALY, at which point the Beckman Coulter hs-cTnI (99th centile) test became cost-effective up to a threshold of £447,934 per QALY. For an AMI prevalence of 5–20%, the no-testing strategy was cost-effective up to thresholds of £8759–11,703 per QALY, at which point the Beckman Coulter hs-cTnI (99th centile) test became cost-effective up to thresholds of £32,042–97,709 per QALY. For an AMI prevalence of 30%, the no-testing strategy was cost-effective up to a threshold of £8431 per QALY, at which point the Beckman Coulter hs-cTnI (99th centile) test became cost-effective up to a threshold of £24,745 per QALY. The Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for thresholds between £24,745 and £70,942 per QALY.
In addition, cost-effectiveness estimates for the subgroups, described in Chapter 3 (see Presentation samples), based on subgroup-specific accuracy and prevalence, are reported in Appendix 9 (only comparing the Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, and standard Tn testing). The results of these analyses indicated that differences in accuracy and AMI prevalence between subgroups had a substantial impact on the cost-effectiveness of the Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, compared with standard Tn testing (ICER range: £22,111–355,571; deterministic base case: £41,233).
The deterministic subgroup analyses for the secondary analysis are presented in Appendix 8. For females aged 45 years and males aged 45 or 55 years, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds of < £16,023–17,836 per QALY. The Abbott ARCHITECT hs-cTnI optimal strategy became cost-effective for higher thresholds. For females aged 55 or 65 years and males aged 65 or 75 years, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds of < £13,064–16,994 per QALY. From this threshold up to £18,999–25,149 per QALY, the Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, was most cost-effective. The Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for higher thresholds. For females aged 75 or 85 years, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective up to thresholds of £12,392£21,140 per QALY, at which point the Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, became cost-effective up to thresholds of £16,407–26,911 per QALY. The Abbott ARCHITECT hs-cTnI optimal strategy became cost-effective for thresholds of > £24,020–45,709 per QALY. For males aged 85 years, the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, was cost-effective for thresholds of < £66,418 per QALY. The Abbott ARCHITECT hs-cTnI optimal strategy became cost-effective for higher thresholds.
For subgroup analyses considering AMI prevalence, no testing was included as additional comparator. For an AMI prevalence of 1%, the no-testing strategy was cost-effective up to a threshold of £4563 per QALY, at which point the Abbott ARCHITECT hs-cTnI assay at presentation, using the 99th centile diagnostic threshold, became cost-effective up to a threshold of £109,991 per QALY, where the Abbott ARCHITECT hs-cTnI optimal strategy became cost-effective. Similarly, for AMI prevalences of 5% and 10% the thresholds were £5209 and £35,574, and £5820 and £22,684, respectively. For a AMI prevalences of 20% and 30%, the Abbott ARCHITECT hs-cTnI optimal strategy was cost-effective for thresholds of > £16,319 and £15,410, respectively.
In contrast with the base-case analysis (described above), the subgroup-specific accuracy and prevalence (only comparing the Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, and standard Tn testing) did not have an important impact on the cost-effectiveness (see Appendix 9). The Roche Elecsys hs-cTnT assay at presentation, using the 99th centile diagnostic threshold, was dominant for all subgroups.
