This chapter provides an overview of existing cost-effectiveness evidence on the use of the QAngio XA 3D/QFR and CAAS vFFR imaging software for assessing the functional significance of coronary obstructions in patients with suspected stable chest pain whose angiograms show intermediate stenosis and who may require revascularisation. The literature was systematically searched to identify and describe relevant evidence on the cost-effectiveness of the two new technologies within the indication for which these are being evaluated. This systematic review also aimed to identify the central issues associated with adapting existing decision models to address the current decision problem and to assist in the development of a new decision model drawing on the issues identified in the clinical effectiveness and cost-effectiveness review. Given that the two technologies under assessment have only recently been commercialised, it was anticipated that there would be a dearth of relevant economic evidence. Therefore, to assist the development of a new decision-analytic model, a pragmatic review of published cost-effectiveness studies evaluating ICA (alone and/or with FFR) in the management of CAD was also conducted.
Results of the review of decision models evaluating invasive coronary angiography
A total of 1740 records were identified during the initial search of economic databases, of which 1264 remained after deduplication. The first step of screening identified 25 titles as potentially relevant based on their titles and/or abstracts. After the full-text articles of these records were obtained and assessed for eligibility, 21 studies68–88 were considered to meet the selection criteria and included in the review. The studies are summarised in detail in Appendix 7, Table 56. Results of the searches and the list of excluded studies are presented in Appendix 7, Tables 54 and 55.
Given the aim of the review, a formal assessment using checklists to assess the quality of the included cost-effectiveness studies was not conducted. Instead, a narrative review of key model features, including testing and management strategies, and assumptions to support the conceptualisation and development of a de novo analytical model is presented below.
The majority of studies68–72,75,77,79,81,83,85,86,88 used a decision tree to model the diagnostic pathway and short-term outcomes, and a long-term Markov model (or multiple Markov models) to characterise disease progression. Two studies used microsimulation models74,82 that also combined a decision tree structure to model diagnostic outcomes followed by a lifetime disease progression state-transition model. Of the five studies that modelled the full-time horizon with a decision tree model, three models71,73,80 captured only short-term outcomes (1-year time horizon), whereas two others comprised longer time horizons (10 years84 and lifetime87). One study78 used a Bayesian mathematical model based on two equations to estimate costs and quality-adjusted life-years (QALYs) for each strategy under comparison over a 10-year time horizon. The equations appear to be equivalent to the calculations in a decision tree’s rollback algorithm.
Among the 21 studies, two models77 were considered to be good examples of alternative ways to evaluate diagnostic strategies in patients with suspected stable angina. These studies were selected on the basis that they encompassed many of the features identified in the other studies. The two models differed in terms of how they modelled the diagnostic pathway and subsequent long-term risks of major cardiovascular-related events and associated costs and outcomes. The first study77 was a cohort model that estimated outcomes for an average patient in clinical practice, and the second study74 was a microsimulation model that estimated outcomes for hypothetical patients at different levels of disease severity (defined in terms of number of coronary vessels affected and whether or not patients have ischaemia). A key difference of the two models was the approach taken to assess the long-term impact of the diagnostic strategies on the risk of major cardiovascular events. In one study,77 the model transition probabilities were based on risk prediction equations and patient covariates from a previously published model on angina, which allowed estimation of the occurrence of a primary cardiovascular event (with risk conditioned on factors such as age and sex) and of subsequent events conditional on having and surviving a first cardiovascular event. By contrast, the second study74 estimated the risk of primary and subsequent cardiovascular events dependent on disease severity, based on the rates of MACEs from the literature. A summary of both models is presented below.
Walker et al.77
Walker et al.77 developed a decision tree and Markov model structure to evaluate the cost-effectiveness of eight alternative testing sequences, including different combinations of exercise treadmill testing, SPECT, cardiovascular MR and coronary angiography, to identify patients with angina who require revascularisation (i.e. those with significant stenosis) derived from the CE-MARC (Clinical Evaluation of Magnetic Resonance Imaging in Coronary Heart Disease) trial.89 The study population included patients with angina (with and without significant stenosis) and those without angina, based on characteristics of patients in the CE-MARC trial.89 The base-case analysis considered the case of a 60-year-old man, classified as grade 2 on the Canadian Cardiovascular Society (CCS) scale, with a prior likelihood of significant stenosis requiring revascularisation of 39.5%. Patients with angina were assumed to have had no previous MI. The Markov model had a 50-year time horizon with a 3-month cycle length. The perspective of the study was NHS and Personal Social Services (PSS), and health outcomes were measured in terms of QALYs. Costs were expressed in term Great British pounds (GBP) (2010/11 price year), and costs and health outcomes were discounted at a rate of 3.5% per annum.
The aim of the diagnostic testing was to identify patients with significant coronary artery stenosis who require revascularisation [either PCI or coronary artery bypass graft (CABG)]. It was assumed that all patients suspected of having significant coronary stenosis would undergo coronary angiography as a definitive test before revascularisation. ICA was considered the reference standard test with perfect sensitivity and specificity. As ICA was performed on all patients indicated for revascularisation, the model did not consider any FP test results. The diagnostic component of the model divided the patient cohort according to their underlying disease status based on characteristics of patients in the CE-MARC trial,89 survival to interventional and diagnostic procedures, test results and subsequent clinical management conditional on test results. All patients with positive and inconclusive test results progressed to a further test in the sequence, although the type of the next test depended on whether the result was positive or inconclusive for some strategies (e.g. in strategy 8 a positive exercise treadmill test result would be followed by ICA, whereas inconclusive test results would be followed by a SPECT test). Patients whose overall testing sequence resulted in a positive result were managed with either PCI or CABG. The relative proportion of patients who underwent each type of revascularisation was sourced from UK clinical registries. Patients who tested negative at any point in the test sequence were managed with optimal medication if they had angina, or with no further medical therapy for those without angina. The decision tree captures mortality associated with both invasive tests and revascularisation, and separately applies procedure-specific mortality rates for ICA, PCI and CABG. At the end of the decision tree, patients with significant stenosis could be classified as TP, FN or dead. Patients without significant stenosis could be classified as TN with angina, TN without angina or dead. All testing strategies are assumed to take the same time and do not account for delays to revascularisation resulting from strategies that involve more tests.
The diagnostic accuracy estimates for the different tests considered in the alternative strategies were conditional on positive/inconclusive results in previous tests in the strategy, thus accounting for correlations between tests within diagnostic strategies. This is possible only with access to IPD from studies that include all the tests used across the full set of diagnostic strategies, as was the case for the CE-MARC study,89 which informed diagnostic accuracy in this model. However, the people interpreting each test were blinded to the results of previous tests in each diagnostic sequence, so the data would not have captured the influence of knowledge on previous tests on the diagnostic accuracy estimates of subsequent tests.
The long-term model is composed of three submodels. Patients with significant stenosis enter one submodel at either the TP or FN state. The key difference between TP and FN patients is that TP patients have undergone revascularisation. In the base-case analysis, the treatment effect of revascularisation is limited to a reduction from angina symptoms, with improved HRQoL for TP patients compared with FN patients, whereas the same baseline risk of cardiovascular events is applied for TP and FN patients. A proportion of FN patients are assumed be correctly diagnosed over time (conditional on their CCS grade), and transition to the TP health state. Patients can remain event free, have a primary non-fatal cardiovascular event, or die from a cardiovascular event or other causes. Patients who survive a primary non-fatal cardiovascular event transition to the non-fatal cardiovascular event state and have an increased risk of further cardiovascular events for 12 months, after which they transition to the non-fatal event post 12 months state. The risk of cardiovascular events in this state is lower than in the non-fatal event post 12 months state, but higher than the baseline risk (TP and FN states). Patients in all health states are subject to a mortality risk from non-cardiovascular death, which is sourced from UK life tables (with cardiovascular deaths removed to avoid double counting). A similar submodel to the one described above, this is used to estimate the cost and health outcomes of TN patients with angina. TN patients without angina go into a two health states (alive and dead) submodel that derived transition probabilities from sex- and age-adjusted UK life tables for all-cause mortality.
The probabilities of fatal and non-fatal cardiovascular events in the submodels for patients with angina were estimated based on risk equations from the EUROPA (EUropean trial on Reduction Of cardiac events with Perindopril in stable coronary Artery) trial.90 This study estimated risk equations to predict (1) the risk of a first primary event, cardiovascular death, MI or cardiac arrest (see equation 1), (2) the odds of that event being fatal (see equation 2) and (3) the risk of a further primary event in the first year after a first non-fatal event (see equation 3). The equations allow for the adjustment of the rate of events dependent on the patient characteristics (age, sex, medication, comorbidities, etc.) and, importantly, accounting for the occurrence of previous MI. Walker et al.77 applied a fourth equation to model the risk of secondary cardiovascular events, which captures the excess cardiovascular risk for patients who had had a previous MI.
The model also considers cancer-related mortality due to radiation exposure during some testing procedures (ICA and SPECT) and PCI (assumed to be performed at the same time as ICA). The model quantified the average radiation exposure in each test sequence; these radiation dosages were then combined with cancer incidence and mortality estimates from the literature to calculate lifetime incidence and mortality conditional on the patient’s age when they were tested. The costs and morbidity associated with cancer were not modelled.
The HRQoL in the model was dependent on age, sex, CCS grade and whether or not the patient had undergone revascularisation. EuroQol-5 Dimensions (EQ-5D) utility weights by CCS grade from a study on angina were combined with UK-population norm EQ-5D estimates by age and sex to obtain age- and CCS-specific HRQoL estimates. The underlying assumption was that the relative impact of CCS grade on HRQoL compared with the population is the same across all age groups.
One important base-case assumption of the Walker et al.77 model is that revascularisation has no impact on the risk of cardiovascular events, and provides relief only from angina symptoms (captured by change in CCS score). HRQoL scores for patients with angina (with and without significant stenosis) are based on age- and sex-adjusted UK population scores with a relative adjustment made based on CCS grade. Data from a RCT comparing coronary angioplasty with medical management was used to link CCS scores at baseline and 6 months after intervention with the two treatments. Patients with angina and significant stenosis who receive revascularisation (TP) are attributed the HRQoL based on the average CCS grade of those following treatment with angioplasty conditional on initial CCS grade. Patients with angina and significant stenosis who are misclassified (FN) are attributed the HRQoL based on the average CCS grade of those following treatment with medical management conditional on initial CCS grade. It was assumed that angina patients without significant stenosis received the same HRQoL as FN patients, whereas the HRQoL of the other TN patients without angina was based on age- and sex-adjusted UK population scores.
Costs included in the model were those of tests and interventional procedures, treatment costs in the long-term model and health-state costs (namely fatal and non-fatal cardiovascular events, and other-cause mortality). Treatment and health-state costs were also sourced from the EUROPA trial90 (with a price year inflation adjustment). Background treatment costs were the same for all patients with angina and an additional background cost was applied for patients after a cardiovascular event. Patients without angina were assumed to have no costs in the long-term model.
The authors considered uncertainty by performing probability sensitivity analysis and scenario analysis where they varied assumptions on baseline characteristics (CCS grade, sex and age), prior likelihood of coronary heart disease requiring revascularisation, rediagnosis rate of FN patients, clinical management of TP patients, the impact of radiation exposure on cancer (risk assumed to be zero), risk of cardiovascular events following revascularisation (treatment effect from the EUROPA trial90), HRQoL decrements and the cost of diagnostic tests. The model was sensitive to prior likelihood of disease, reducing the starting age and increasing baseline CCS grade in the model, use of absolute HRQoL decrements by CCS grade, allowing for a proportion of TP patients to not receive revascularisation, reidentification rate of FN patients, and costs of tests. The prior likelihood of coronary heart disease requiring revascularisation was considered a key driver of cost-effectiveness.
Genders et al.74
The model developed by Genders et al.74 was a microsimulation model comprising a decision tree and a lifetime state-transition model to assess the cost-effectiveness of invasive and non-invasive testing strategies for patients with stable chest pain. The base-case population consisted of 60-year-old patients with a 30% pretest probability of obstructive CAD (defined as ≥ 50% stenosis on at least one vessel) who had never undergone revascularisation procedures and had no prior history of CAD. The study presents cost-effectiveness results for the separate jurisdictions. We refer here to inputs and results specific to the analyses under the UK NHS perspective, as they are more relevant to our study. Costs were calculated in GBP (2011 price year), and health outcomes were calculated as QALYs. Both costs and QALYs were discounted at an annual rate of 3.5%.
The diagnostic strategies in the model are evaluated under two different diagnostic workups. In the invasive workup, patients with obstructive CAD on CCTA and patients with inducible ischaemia on cardiac stress imaging were referred for ICA prior to a decision regarding medical management. In the conservative workup, only patients identified as having higher CAD severity by CCTA or cardiac stress imaging would be referred to ICA and patients with milder forms of the disease managed with OMT. Patients with normal arteries or mild CAD (< 50% stenosis) received no further testing under either diagnostic workup.
The decision tree starts by classifying patients according to eight categories of disease severity based on percentage stenosis, number of vessels affected, location of lesion (left main trunk or not) and severity of inducible stenosis (where present). Patient distribution across disease severity categories was sourced from hospital records for patients who had undergone CCTA and ICA. Diagnostic accuracy estimates derived from published meta-analyses were then applied to split patients according to the test results for each diagnostic strategy. For the purpose of applying these estimates, patients who were considered correctly classified with a negative result had normal coronary arteries or mild CAD (< 50% stenosis). Patients correctly identified with a positive result had moderate CAD, severe CAD or three-vessel disease/left main coronary stenosis. The model did not consider inconclusive test results. The authors assumed independence of diagnostic accuracy estimates for CCTA and cardiac stress imaging, and further assumed that FP results were possible only for mild CAD and mild inducible ischaemia (under the conservative diagnostic workup). Patients with FP results are assumed to receive unnecessary optimal medication for the full time horizon, incurring a treatment cost and utility decrement in the long-term model. As in Walker et al.,77 adverse events from testing and revascularisation procedures were considered. However, in this model adverse events are not limited to procedural mortality, but also include non-fatal MI with ICA. This adverse event had a cost attributed to it, but did not translate into an increased risk of further events in the long-term model.
The decision tree splits the patient population according to disease severity, test results and survival to testing (ICA and FFR) and revascularisation procedures. It also allows quantifying the average exposure to radiation with the different tests and PCI.
In the ICA strategy, all patients were tested with ICA. Those who tested negative received risk factor management and those who tested positive would be tested with FFR to decide treatment. ICA is assumed to be a perfect test, and FFR appears to allow prefect distinction between disease severity categories, although this is not explicitly stated in the paper. OMT was then given to patients with mild ischaemia and moderate to severe CAD, PCI was given to patients with severe CAD and severe ischaemia, and CABG was given to patients with three-vessel disease or left main coronary stenosis. Revascularised patients would also receive OMT, and all individuals in the model received risk factor management.
Subsequent to the decision tree, patients entered a state-transition model comprising three health states: alive, post MI and dead. Patients enter the model through the alive state, where they could remain until death or suffering a non-fatal MI. Patients who suffered a non-fatal MI would transition to the post-MI state, where they could remain or transition to the dead state. The transition probabilities were derived from published trial data that reported risk of MACE (cardiovascular death, non-fatal MI and repeated revascularisation) in patients treated with CABG, PCI and OMT. The rates of MACE were dependent on disease severity and whether patients were treated with optimal medication or revascularisation. All FN patients were assumed to be correctly identified and treated by the end of the first year, with the exception of those with moderate CAD without ischaemia, of whom only 25% were rediagnosed. Patients who experienced a primary cardiovascular event would have a higher risk of subsequent cardiovascular events, which was modelled by applying a HR of 1.44 to their baseline risk. The model also considered mortality from non-cardiovascular causes. This was estimated based on age- and sex-specific general mortality data from which deaths attributed to cardiovascular causes had been removed to avoid double counting. The mortality, morbidity and costs due to cancer incidence were not modelled, although the model calculated cumulative radiation exposure over the time horizon.
The risk of MACE was estimated from the trial data separately for the first year and all subsequent years to allow for a higher event rate in the first year after starting treatment. The rates were estimated based on the CABG arm of the SYNTAX trial59 for patients with three-vessel disease or left main coronary stenosis, and the optimal medication and PCI arms of the COURAGE trial91 for the patients with suspected or mild inducible ischaemia and moderate to severe CAD (treated with optimal medication) and patients with severe CAD and severe inducible ischaemia (treated with PCI). The reciprocal of the treatment HR was applied to this risk to estimate the baseline probability of cardiovascular events for untreated patients (FN), who have a higher rate of events until they are correctly diagnosed. A single treatment effect hazard for optimal medication, PCI and CABG (HR 0.70) was sourced from three meta-analyses of OMT comparing treatment with no treatment, but it is unclear how this estimate was calculated. The rates of MACE applied in the model are summarised in Appendix 7, Table 57.
We note that the MACE rates without treatment seem counterintuitive (e.g. higher MACE rates for untreated moderate CAD with mild ischaemia compared with untreated severe CAD with severe ischaemia). The authors did not comment on the MACE rates.
If the treatment effect applied in the model is indeed the same for optimal medication and revascularisation, this is similar to the absence of a treatment effect of revascularisation in addition to optimal medication in Walker et al.77 This is an important interpretation of the clinical evidence on the treatment effect of revascularisation, and one that is discussed further in Chapter 5, Treatment effect of revascularisation. In previous studies where a treatment effect on the rate of cardiovascular events for revascularisation compared with optimal medication was considered explicitly for comparable patients (e.g. same disease severity), seven models included the existence of a treatment effect68,69,76,83,85,86,88 and six studies did not,70–72,79,84,87 in line with Walker et al.77 and Genders et al.74
The HRQoL in the model was assigned to individuals according to disease severity and treatment received. Patients without CAD or inducible ischaemia were assumed to have the HRQoL of the general population based on age- and sex-specific EQ-5D estimates for the US population. For patients with CAD and inducible ischaemia who underwent active treatment (optimal medication or revascularisation), mapped EQ-5D utility decrements were applied to the general population HRQoL estimates. In the first year of treatment, the utility decrements of treatment relative to the general population were derived from the average utility decrement as observed in the same trial data that informed the rates of MACE for treated patients, whereas for the subsequent year the last observed value in the trials was carried forward. The authors state that a disutility was considered for patients with FP results. It was not clear how the utility decrements for FN patients were estimated. Appendix 7, Table 58, summarises the utility values for the start age in the model conditional on treatment and disease severity. The HRQoL estimates for the first year and subsequent years of treatment are presented for the same age solely for ease of comparison.
The model considers costs of tests, test adverse events, medication, MI in the long-term model, and incidental findings from CCTA. Unit costs were mostly sourced from UK published data. Based on the description of the unit cost selected for PCI, this procedure was assumed to take place in an outpatient setting. It is not, however, clear what assumptions were made regarding the setting for ICA, CABG and treatment of non-fatal MIs. The unit cost for FFR was sourced from a previous cost-effectiveness study in a US setting.87 An annual cost of medication was included in the model according to disease severity and treatment received (OMT, PCI or CABG). The resource use assumed for patients who received optimal medication alone and in addition to PCI was sourced from the COURAGE trial,91 whereas for those who received CABG and optimal medication it was taken from the SYNTAX trial.59 The distribution of medication use applied in the model is shown in Appendix 7, Table 59.
Model parameters were entered as distributions, and probabilistic sensitivity analysis was performed to incorporate joint parameter uncertainty. Scenario analysis was performed to test assumptions on diagnostic accuracy of stress echocardiography, cost of tests, alternative diagnostic pathways, probability of CAD, time to rediagnose FN patients, and treatment effect of optimal medication for FP patients. A subgroup analysis by sex was also performed. The authors do not identify any drivers of cost-effectiveness, but note that the assumption that FP patients will remain misclassified over the time horizon and that FN patients will be rediagnosed after 1 year is likely to have biased results against strategies with low specificity.
