Systematic review and modelling of the cost effectiveness of Cardiac Magnetic Resonance Imaging compared to current existing testing pathways in Ischaemic Cardiomyopathy

Identification of studies

The PRISMA flow diagram shown in Figure 1 provides a summary of the study identification process.

FIGURE 1.

The PRISMA flow diagram of included studies.

Study selection

Studies were selected for inclusion through a two-stage process according to the above inclusion/exclusion criteria. All titles and abstracts were examined for inclusion by two reviewers (FC and LU). Discrepancies were resolved by discussion and drew on additional expertise (SMT) where uncertainty remained. Full manuscripts of selected citations were retrieved and assessed by two reviewers using the inclusion/exclusion criteria.

Data extraction strategy

Data were extracted by two reviewers using a standardised data extraction form which had initially been piloted and redesigned in discussion with clinical experts. Discrepancies were resolved by discussion. Where multiple publications of the same study were identified, data were extracted and reported as a single study.

Critical appraisal strategy

The quality of each study was assessed using the QUADAS II (quality assessment of diagnostic accuracy studies) tool. 24 The following factors were considered: methods of patient selection, the conduct and interpretation of the index test and reference standard. This included the extent to which interpretation of the tests might have been biased by knowledge of the results of other tests or the patient characteristics. In addition, the flow of patients through the study and the number of excluded patients from the analysis and the timing of the follow-up assessment. Where available, these data were extracted and their impact on biasing sensitivity and specificity was considered.

Data synthesis

Meta-analyses were performed to estimate a summary measure of sensitivity and specificity for CMR. The statistical software STATA (2006 release 9.0; Stata Corporation, College Station, TX, USA) was used for this analysis.

Sensitivity analysis was performed to explore causes of potential bias and heterogeneity in the meta-analyses. The study data were grouped into stress CMR and CE CMR and analysed separately. Thirteen studies exploring the diagnostic accuracy of CMR with recovery following revascularisation also compared other diagnostic tests within the same study. This included five studies exploring SPECT, four studies exploring PET and three studying echocardiography. These data were used within the cost-effectiveness review to provide more up-to-date measures of diagnostic accuracy for these alternative and commonly used diagnostic tests. In addition, the data obtained from these studies exploring the clinical effectiveness of other tests were pooled with the results of the most recent diagnostic accuracy review. 22

Quality of diagnostic accuracy studies

The quality of the included studies was evaluated using the QUADAS II tool23 and a number of potential sources of bias were identified in patient selection, conduct of the index test and reference standard, and in timing and follow-up (Figure 2).

FIGURE 2.

Graphs showing proportion of studies in QUADAS II domains.

All of the studies were conducted prospectively and in secondary or tertiary care settings. In most studies patients were recruited following a referral for further assessment and treatment of CAD. In 16 of the studies, the patients were recruited consecutively and the studies were, therefore, considered to be at low risk of bias regarding patient selection (Table 2). Eight studies did not report whether or not recruitment occurred consecutively. This may have led to an underestimation or overestimation of test accuracy by investigating a selected population, through inclusion or exclusion of those patients who may be ‘difficult to diagnose’ or, alternatively, have features highly suggestive of viable myocardium. 24

Blinding to prevent bias in the interpretation of the index and reference standard could include a number of components: blinding of assessors analysing the results of the index test and reference standard to the interpretations of the other assessor(s), blinding to the patients’ identity and their clinical characteristics, blinding to the results of the other tests that may have been undertaken alongside the index text and blinding to the results of the reference standard results or index test results. Only three studies blinded all these aspects of the study. 15–17 Seven studies15,25,31,44,47–49 used more than one experienced observer to assess segmental viability and these assessments were conducted blind and consensus reached if discrepancies in interpretation arose. In a further seven studies,16,29,32,36,41,43,46 only one assessor was used, but he or she was blind to the other test results and/or clinical information regarding the patient. Seven studies17,27,28,30,37,38,40 used more than one assessor, but did not describe efforts to blind the assessors to the results of other tests or to details of the patient’s clinical condition. Three studies34,35,39 used only one assessor and did not describe details of blinding the assessor to information that might bias their interpretation of the test results.

All of the included studies were within-patient comparisons, so both the index test and the reference standard were carried out on the same patient. This will have reduced interpatient variability and added strength to the research design.

Studies differed in the length of time between revascularisation and the assessment of recovery. There is some indication that recovery of myocardial function can occur up to 6–12 months after revascularisation. Therefore, studies with a very short follow-up point may underestimate the accuracy of a study. 22,51 Follow-up periods varied from 17 days to 9 months (Tables 3 and 4).

Fifteen studies15–17,25,28–31,35,36,43,44,46–48 did not include all of the recruited patients in the final analysis. The reasons for their exclusion were given, and included death, serious postoperative morbidity, withdrawal from the study, segmental images hard to read and loss to follow-up.

In considering the applicability of each included study in addressing the study question, there were a number of concerns. Three studies35,39,44 included very small numbers of participants in the analysis (n = 10,35 n = 1039 and n = 844), creating uncertainty in the reliability of the findings. One study31 included a population that had a greater severity of CAD at baseline than the other studies.

Stress cardiac magnetic resonance imaging

Twelve studies explored the diagnostic accuracy of stress CMR to identify hibernating myocardium. 17,27,28,31,32,34,35,37,39,40,47,48

Eleven studies17,27,28,31,32,34,37,39,40,47,48 used dobutamine to induce cardiac stress and facilitate measurement of contractile reserve. The dose of dobutamine used ranged from 5 to 15 µg/kg per minute, with most studies using between 5 and 10 µg/kg per minute. Two studies31,47 used the higher dose of 15 µg/kg per minute. One used nitroglycerin (0.4 mg). 35

The threshold for viability was established in the following way. Systolic wall thickness [SWT (wall motion)] was measured by CMR during rest and stress. Segmental wall thickening was first analysed at rest using a qualitative scoring system with the following scale: 0 = dyskinetic, 1 = akinetic, 2 = severely hypokinetic, 3 = hypokinetic, 4 = normal.

Those segments that were labelled as akinetic or dyskinetic, but which showed SWT of at least 2 mm during dobutamine stress, were classified as viable; otherwise, they were considered to be non-viable. Two studies17,37 used any change in SWT in akinetic or dyskinetic segments as the threshold for viability. Two studies47,48 used increase in one grade (see scoring system above) to determine viability. One study39 used increase in SWT by more than two times the standard deviation of the measurement technique as the threshold for viability. Five studies27,28,32,40,47 also used a measure of end-diastolic wall thickness (EDWT) in order to identify hibernating myocardium. The cut-off used for EDWT was 5.5–6.0 mm for each study.

Each study performed the second CMR or coronary angiography between 1 and 9 months after revascularisation. The most common follow-up period was approximately 5 months after revascularisation. Differences in timing of the second evaluation were for protocol reasons rather than because it was necessary in clinical practice; the differences in timing may have had some influence on the results, but what this would be is unknown.

The reference standard ‘recovery of contractile function’ following revascularisation was assessed quantitatively using a 2-mm segmental improvement in SWT from pre- to post-revascularisation CMR at rest in most of the studies. Five studies17,37,39,47,48 did not define a threshold for SWT, but considered segments that had improved or those that had improved by a grade or a standard deviation of the measurement technique as viable. One study27 assessed LVEF recovery using coronary angiography. Some studies27,28,37,40 also included an assessment of viability of myocardial regions, containing multiple segments. A region was defined as viable if ≥ 50% of the affected segments had improved.

Late gadolinium-enhanced cardiac magnetic resonance imaging

Fourteen studies15,16,25,29,30,32,36,38,41,43,44,46,48,49 evaluated the accuracy of CE CMR to detect myocardial viability. In all of these studies segmental LV recovery following revascularisation served as the reference standard. Two studies32,48 evaluated stress CMR and CE CMR and sufficient data were reported or provided by authors to enable their inclusion in both meta-analyses.

All of the included CE CMR studies used gadolinium-based contrast agents. There were some differences in technique used, with some variation in dose of gadolinium and in the duration of time between administration of agent and collection of images. The dosage of gadolinium ranged from 0.05 to 0.4 mmol/kg. The most commonly used dosage was 0.2 mmol/kg. Images were obtained 2–30 minutes after administration, most commonly at 15 minutes.

Viability is determined by the extent of hyperenhancement and a hyperenhancement category established for each segment related to the 5-point scale proposed by Kim et al. :52 0% hyperenhancement (category 1), 1–25% hyperenhancement (category 2), 26–50% hyperenhancement (category 3), 51–75% hyperenhancement (category 4), and 76–100% hyperenhancement (category 5).

The threshold for determining viable myocardium was reported as 50% of LV wall hyperenhancement, with segments with more than 50% hyperenhancement considered non-viable. Seven studies15,29,30,36,41,43,44 also reported the results at 25% of LV wall hyperenhancement. The results for both the 25% and 50% thresholds are presented in Table 7.

The follow-up assessment of LV function occurred between 19 days and 2 years after revascularisation in the included studies. Most of the studies undertook follow-up assessment 6 months after revascularisation, but the variation in the studies may influence heterogeneity in the analysis. There is a suggestion that segmental recovery of viable myocardium may take several months and early assessment of recovery may miss later recovery.

The reference standard, ‘recovery following revascularisation’, was assessed using CMR (echocardiography was used in two studies)27,41 to examine segmental wall thickening and make comparisons with preoperative wall motion scores. Recovered segments were those in which there was an increase in segmental wall thickening compared with preoperative segmental functioning by one grade or score,16,29,32,36,41,48 by any improvement in segmental wall motion,25,38,43 15% improvement in contractile function,44 ≥ 5% improvement in LVEF46 or increased wall thickening of > 1.5 mm. 30

Meta-analysis of diagnostic parameters

Bivariate random-effects regression analyses were performed in STATA/IC 12.0 (2012; Stata Corporation, College Station, TX, USA) using the program ‘metandi’ to generate pooled accuracy estimates of sensitivity, specificity, positive likelihood ratios (LR+), negative likelihood ratios (LR–) and diagnostic odds ratios (DORs).

As described by Harbord et al. ,55 the bivariate regression method assumes that the sensitivity values from individual studies (after logit transformation) within a meta-analysis are approximately normally distributed around a mean value with a certain amount of variability around this mean. This is a random-effects approach. This variation in underlying sensitivity estimates between studies can be related to undetected differences in study population, differences in implicit threshold (cut-off) or unnoticed variations in the index test protocol. These considerations also apply to specificity estimates. The potential presence of a (negative) correlation between sensitivity and specificity within studies is addressed by explicitly incorporating this correlation into the analysis. The combination of two normally distributed outcomes, the logit-transformed sensitivity and specificity values, while acknowledging the possible correlation between them, leads to the bivariate normal distribution. The bivariate approach overcomes the problems associated with simple pooling (i.e. weighted average) of sensitivity and specificity estimates.

Heterogeneity is usually a concern with meta-analyses and refers to a high degree of variability in accuracy estimates across studies. Heterogeneity could be a result of differences in thresholds, the prevalence of drug resistance, the populations studied, assay methods or reference standard tests. The reasons for the heterogeneity were investigated by pre-specified subgroup (stratified) analysis. In the subgroup analysis, the data were stratified according to the type of CMR tested (CE CMR vs. stress CMR) to determine if accuracy varied across subgroups.

Stress cardiac magnetic resonance imaging

Data from 12 studies17,27,28,31,32,34,35,37,39,40,47,48 were used in the meta-analysis evaluating the diagnostic accuracy of stress CMR with recovery following revascularisation as the reference standard. The threshold for determining viability was SWT of 2 mm during cardiac stress in akinetic or dyskinetic myocardial segments.

The mean weighted sensitivity and specificity for stress CMR are provided in Figure 3. Gunning et al. 31 had a lower sensitivity [50%, 95% confidence interval (CI) 39% to 61%]. The patients in this study had the lowest mean LVEF (24%) and all participants had three-vessel CAD. This may have resulted in reduced sensitivity of the test. It has been suggested that the more profound ultrastructural changes and loss of contractile protein in areas of dysfunctional but viable myocardium may reduce the chance of eliciting a dobutamine-stimulated contraction reserve. 28,31

FIGURE 3.

Mean weighted sensitivity and specificity for stress CMR. FN, false negative; FP, false positive; TN, true negative; TP, true positive.

Late gadolinium-enhanced cardiac magnetic resonance imaging

Data from 14 studies were used in the meta-analysis evaluating the diagnostic accuracy of CE CMR with recovery following revascularisation as the reference standard. 15,16,25,29,30,32,36,38,41,43,44,46,48,49 The threshold for determining viability was ≤ 50% of LV wall hyperenhancement.

The mean weighted sensitivity and specificity for CE CMR, with recovery following revascularisation as the reference standard, are provided in Figure 4. Specificity was lower in the studies of Sharma et al. ,44 Schvartzman et al. 41 and Selvanayagam et al. 41,43,44 (24%, 25% and 29% respectively). The range for the remaining studies was 44–94%.

FIGURE 4.

Late gadolinium-enhanced CMR with recovery following revascularisation as the reference standard. FN, false negative; FP, false positive; TN, true negative; TP, true positive.

Characteristics of these three studies may have influenced the accuracy of the findings. Schvartzman et al. 25 had a very short follow-up between initial CMR and follow-up CMR to determine recovery following revascularisation (6 weeks). Only one other study25 had a shorter follow-up (17 days), and it also had a lower specificity (60%). The study by Sharma et al. 44 was one of the smallest studies, with only eight patients and 97 segments included in the analysis.

There may be a number of reasons for the low specificity of CE CMR. The reduced accuracy of CMR in identifying the non-viable myocardium may be a consequence of seeking to diagnose myocardial viability in segments without full-thickness hyperenhancement. The cut-off of 50% wall thickness hyperenhancement to differentiate viable from non-viable myocardium will have meant that the included segments may have incorporated those segments with partial enhancement. It has also been suggested that different degrees of wall motion abnormalities have a major impact on myocardial recovery, with segments showing akinesia demonstrating the best recovery. 21 The lack of the ability to show contractile recruitability with stress (which makes stress echocardiography and stress CMR specific) is another reason why CE CMR may have a lower specificity.

Pooled summary estimates of diagnostic parameters for different tests

Ten studies16,25,28,31,32,35,40,44,46,49 also tested other index tests as well as CE CMR and/or stress CMR. These included PET,40,49 SPECT16,25,31,32,44,46 and echocardiography,28,35 with recovery of LV function following revascularisation as the reference standard. These data were pooled with the results from a previously published systematic review22 in order to estimate the diagnostic accuracy of other tests compared with CMR. We also undertook a sensitivity analysis, testing the effect of Schmidt et al. 40 on the results. This was because of uncertainty that some of the patients may also have been in included in another study28 and, therefore, included twice in the meta-analysis. There was a non-significant difference in the pooled estimate with the exclusion of Schmidt et al. 40 Table 8 shows the estimated pooled summary estimates of diagnostic parameters for different tests.

Identification of studies

Quality assessment strategy

The methodological quality of each included study was assessed using a combination of key components of the Drummond and Jefferson checklist for economic evaluations,56,57 together with the Eddy checklist for mathematical models used in technology assessments. 58 The use of the checklist ensured a consistent approach to assessing the quality of each economic evaluation.

Results of cost-effectiveness review

The electronic database literature searches identified 351 potentially relevant publications, of which two met the inclusion criteria. However, the grey literature identified two additional relevant publications on economic evaluations of PET for myocardial viability. Even though these studies were not identified in the initial database search, they provide valuable insight into modelling cost-effectiveness of diagnostic imaging techniques for myocardial viability. A flow chart describing the process of identifying relevant literature can be found in Figure 5. The details of these relevant studies including an assessment of methodological quality are provided below.

FIGURE 5.

Study flow chart: cost-effectiveness.

Cost-effectiveness review summary

Although four cost-effectiveness analyses studies were identified via the literature searches, there are a number of limitations associated with generalising the findings of these included studies.

None of the studies identified performed cost–utility analysis, the preferred approach for estimating cost-effectiveness in UK. Three studies59,62,63 reported incremental cost per life-year gained, while one study60 reported only the cost savings.

None of the studies compared all the relevant diagnostic strategies. Two studies59,63 compared PET with SPECT (including PET as a second-line viability test), while the two other studies60,62 compared PET with medical therapy and CABG. In addition, the diagnostic accuracy of these strategies was elicited using expert opinion because of a lack of evidence. Given the current decision problem is to identify the optimal diagnostic pathway, the cost-effectiveness analysis needs to include all potential diagnostic strategies.

The analysis reported by Jacklin et al. 62 was based on a single non-randomised cohort study, while the other studies59,60,63 included data from expert opinion. This was because of a lack of published data on the impact of viability assessment and revascularisation on the long-term clinical outcomes of patients. This scarcity of data about patient outcomes (such as survival, hospitalisation and QoL) is a significant barrier in estimating the cost-effectiveness of diagnostic pathways for myocardial viability assessment. Although there have been studies recently that provide this information, assumptions would need to be made about their long-term effects beyond the trial duration.

The appropriate time horizon of the model for estimating the cost-effectiveness is not clear. One study used a 1-year time horizon,62 two studies59,63 used a 5-year horizon and the time horizon was unclear in the other study. 60 Choosing the time horizon is a key issue, especially when assumptions need to be made about the extrapolation of patient outcomes.

However, despite the differences in the data used, all four studies used a similar modelling approach. The studies used a cohort model, the proportion of patients with viable myocardium modelled using prevalence, and the patients identified by the diagnostic strategies as viable undergoing revascularisation. The costs and outcomes (usually measured as survival or LYGs) were then estimated for the different strategies to estimate the cost-effectiveness.

A de novo economic model was developed based on these studies using the diagnostic accuracy data from meta-analysis, patient outcomes and UK cost data from literature as detailed in Independent economic assessment methods.

Objectives

The objectives of the cost-effectiveness analysis were to:

1. estimate the cost-effectiveness of diagnostic pathways for assessing patients with ischaemic cardiomyopathy to identify patients with viable myocardium with a view to revascularisation, in terms of the cost per QALY gained by each strategy

2. identify the optimal diagnostic imaging pathway for investigating patients with ischaemic cardiomyopathy (to identify patients with viable myocardium) and estimate the impact of CMR in terms of cost-effectiveness with reference to the National Institute for Health and Care Excellence (NICE) threshold for willingness to pay per QALY gained

3. identify the critical areas of uncertainty in these imaging pathways where future research would produce most benefit and recommend specific primary research designs to address the uncertainty.

The costs and benefits of diagnostic testing

The aim of these diagnostic pathways is to assess patients with ischaemic cardiomyopathy in order to identify those with viable myocardium with a view to revascularisation. The clinical challenge is to identify patients with viable myocardium who have the potential to recover if revascularised and to ensure that these patients are appropriately treated with surgical or catheter-based coronary intervention, and that those with non-viable myocardium in the target area for revascularisation are not subjected to unnecessary intervention.

The main benefits of diagnostic testing relate to identification and treatment of patients with potential for revascularisation using either PCI or CABG. The main disadvantages are the risks associated with adverse events of unnecessary revascularisation. The direct costs of diagnostic management include the costs of diagnostic testing, the costs of investigation and the subsequent costs of providing treatment (revascularisation or medical therapy). The suboptimal nature of the diagnostic tests (i.e. sensitivities and specificities below 100%) means that some patients with viable myocardium will not receive revascularisation and, similarly, some patients without viable myocardium will receive revascularisation, probably unnecessarily, based on the potential lack of benefit from revascularisation, its costs and its risk of mortality. A de novo economic model was built to analyse the effect of different diagnostic management pathways on these costs and benefits.

The decision-analysis model structure

A de novo economic model was developed using Microsoft Excel software (2007; Microsoft Corporation, Redmond, WA, USA) to explore the costs and health outcomes associated with different diagnostic pathways. The economic perspective of the model is the NHS in England and Wales with the structure of the model shown in Figure 6.

FIGURE 6.

Structure of the cost-effectiveness model.

The different diagnostic pathways were applied to a hypothetical cohort of patients with ischaemic cardiomyopathy. It was assumed that the diagnostic pathway would identify patients with viable myocardium and that the probability of successful identification of viable myocardium and non-viable myocardium was determined by the overall accuracy of the diagnostic pathway. It was assumed that patients diagnosed with viable myocardium would be managed promptly by revascularisation and that the patients diagnosed with non-viable myocardium would be on medical therapy. The model assigned each patient a risk of death and rehospitalisation depending upon whether or not the patient was truly viable and whether or not the patient had received revascularisation. Each patient then accrued lifetime QALYs. Health-care costs were also accrued through measuring diagnostic costs and treatment costs, depending on the pathway and the patient’s treatment status. Details of these are outlined below.

Model structure: A decision-analytic model was developed to estimate the costs and health outcomes associated with different diagnostic pathways to identify viable myocardium in a hypothetical cohort of patients with ischaemic cardiomyopathy. The model took a lifetime horizon and the economic perspective of the model was the NHS in England and Wales.

Population: The population comprised patients with ischaemic cardiomyopathy, characterised by extensive CAD and reduced LVEF, including both those with viable myocardium and non-viable myocardium.

Diagnostic pathways: There are five main imaging methods available to assess for viable myocardium: (1) echocardiography, (2) PET, (3) SPECT, (4) CE CMR and (5) stress CMR. Pathways including combinations (i.e. more than one) of these tests were not evaluated as they are not clinically relevant in UK.

Patient management: Patients diagnosed with non-viable myocardium were assumed to be on medical therapy, while the patients diagnosed with viable myocardium were assumed to receive revascularisation.

Time horizon: A lifetime time horizon of 40 years was used. Patients progressed through the model until they either died or reached the end of the 40-year time horizon.

Discount rate: Both the costs and QALYs were discounted at an annual discount rate of 3.5%, as recommended by NICE.

The key modelling methods together with the evidence sources and assumptions used to populate the model are discussed in detail in Prevalence of viable myocardium, Selection of pathways, Sensitivity and specificity of diagnostic pathways in the model and Patient management after diagnosis.

Prevalence of viable myocardium

In a study reported by Al-Mohammad et al. ,7,8 the prevalence of viable myocardium in patients with ischaemic heart disease was around 45–55% and the prevalence of LVEF was 30%. These values are similar to those reported in other studies. 64,65 In the economic model, the prevalence of viable myocardium is assumed to be 50% (95% CI 45% to 55%).

Selection of pathways

There are five main imaging methods available to assess for viable myocardium: (1) echocardiography, (2) PET, (3) SPECT, (4) stress CMR and (5) CE CMR. With such a range of techniques available to assess patients for viable myocardium, the choice of diagnostic imaging pathway is often dictated by a number of factors.

Individual hospitals may have access to different types of imaging tests and relevant expertise, with some hospitals being equipped to deliver only one modality while others may have a choice of several. Furthermore, there may be differences in the availability of some tests and, therefore, the choice of which diagnostic pathway to use may depend on the balance of accuracy, availability and cost. For these reasons it was felt important to model all the imaging tests. This would allow decision-makers to identify the pathway that most closely resembled their local setting, and to see how changing to another pathway may affect their costs and QALYs.

The diagnostic pathways were chosen to include all the real-life pathways in clinical practice, i.e. to incorporate the variation of different hospital protocols, regionally and internationally. Pathways including two or more tests are not considered for evaluation in the economic model as, although there might be instances where more than one test is used to assess viability, they are not used regularly in clinical practice.

The single-test pathways include the five main imaging methods (echocardiography, PET, SPECT, stress CMR and CE CMR). Two hypothetical strategies, a ‘discharge everyone without testing or revascularisation’ strategy and a ‘revascularise everyone’ strategy are also analysed.

After extensive discussions with the clinical expert group, the following pathways were chosen to be included in the primary analysis:

1. discharge all patients home without testing or treatment

2. echocardiography

3. stress CMR

4. CE CMR

5. SPECT

6. PET

7. treat all patients without testing.

It should be noted that, in clinical practice, the imaging pathways are much more complex with a lot of subjective clinical judgement based on individual patient’s situation. Multiple diagnostic tests are often used to detect the presence of viable myocardium; however, the clinical decision-making rules behind the use of multiple tests are complex and subject to variation. Therefore, for the purposes of this evaluation, the single diagnostic test pathways provided above are assumed to be representative of the protocols followed in most hospitals, both in the UK and elsewhere.

The results of multiple test pathways are not included in the analysis for a number of reasons. First, it is not clear what proportion of patients would be subjected to multiple tests and there tend to be local variations in the diagnostic pathways used. Second, these multiple test pathways can be represented as either (1) reinforcement of positive diagnosis, i.e. the second test is performed only if the first test indicated viable myocardium, or (2) confirmation of negative result, i.e. second test is performed only if the first test indicated non-viability. It is not clear whether or not the decision to offer the patient revascularisation will be based on the result from the second test in case of non-concordance between tests. In order to estimate the combined diagnostic accuracy of combination of tests, the correlation between the result of the initial test and the secondary test needs to be estimated. There are no data on the correlation between tests and, in the absence of evidence, postulating correlation factors might lead to bias. Furthermore, during probabilistic sensitivity analysis (PSA), the sensitivities and specificities of the combined tests need to be estimated from the samples of joint distributions of the diagnostic parameters for each test to preserve the correlation. Given all these limitations in the evidence base and the relative scarcity in the use of multiple diagnostic tests in UK, the analysis was limited to pathways with single diagnostic tests.

Sensitivity and specificity of diagnostic pathways in the model

The methodology used for determining the sensitivity and specificity of each non-invasive imaging test is given in Chapter 3, Diagnostic accuracy for detection of viable myocardium.

Table 9 shows the estimates of sensitivity and specificity for each test strategy and the sources for these estimates. The meta-analysis data were selected because the point estimates of sensitivity and specificity varied in the expected manner when different test types (a different variation of the test for viable myocardium) were used. The mean values of the posterior distributions for sensitivity and specificity were used in the deterministic analysis.

Patient management after diagnosis

It was assumed that, after the diagnostic pathway had been applied, the subsequent progress of each patient would depend on whether or not the patient had viable myocardium, and, if viable myocardium was identified, whether or not the patient was revascularised. The patients can be classified into four groups, as shown in Figure 7, based on their true status and the diagnosis as:

1. viable and revascularised, i.e. diagnosed correctly as viable

2. viable but not revascularised, i.e. diagnosed wrongly as non-viable

3. non-viable and revascularised, i.e. wrongly diagnosed as viable

4. non-viable and not revascularised, i.e. diagnosed correctly as non-viable.

FIGURE 7.

Patient groups after diagnosis.

It was assumed that patients diagnosed with non-viable myocardium would be on medical therapy and that the patients diagnosed with viable myocardium would be managed promptly by revascularisation, returning to medical therapy after the revascularisation. Although there are multiple variants of medical therapy and two main variants of revascularisation, they are represented in the model as single entities.

The clinical systematic review identified considerable heterogeneity among included studies for medical therapy after diagnosis. Clear descriptions of the medical therapy were not always provided in the studies identified in the systematic review, which made it difficult to estimate costs. This lack of detail also meant that the outcomes estimated (see Chapter 3, Diagnostic accuracy for detection of viable myocardium) were a conglomeration of estimates from heterogeneous studies and this has implications for the robustness of the analysis of cost-effectiveness. In the economic model, medical therapy was assumed to be the same in non-revascularised patients and the revascularised patients (as they will be put on medical therapy after revascularisation), i.e. it was assumed in the model that the medical therapy costs were the same in both groups and, therefore, these costs were not included in the model.

Revascularisation procedures can be broadly classified into surgical or catheter-based coronary interventions (CABG and PCI, respectively), with the choice of revascularisation procedure in clinical practice depending upon patient characteristics notwithstanding other practical issues and constraints such as ease and availability, etc. Despite some known differences between PCI and CABG66 (e.g. incidence of 30-day major adverse events is higher for CABG whereas PCI is less invasive and has shorter hospital length of stay), there is still debate on which is better, with randomised trials comparing PCI with CABG showing conflicting data about the incidence of short- and long-term complications. 66,67 Furthermore, data on outcomes after revascularisation were captured from a mixture of studies that used PCI, CABG or both, which made it difficult to tease out their individual clinical effectiveness. Therefore, revascularisation is treated in the economic model as a single treatment and its outcomes were estimated by pooling the results from all the studies, irrespective of whether they used PCI, CABG or both (see Outcomes). The cost of the revascularisation procedure is estimated as the weighted average cost of PCIs and CABGs using data on the numbers of procedures performed in UK in the last year, as detailed in Costs. According to the NHS information centre, Hospital Episode Statistics (HES) for England, in 2010–11 there were 19,743 inpatient admissions for CABG procedures and 67,908 inpatient admissions for PCI procedures.

Survival

A rapid review was conducted to estimate the effect of treatment on survival of patients and three review studies were found. This section provides a discussion of the evidence available for survival and describes the survival parameters used in the economic model.

Allman et al. 6 performed a meta-analysis of 24 studies (in 1999), with the mean age, LVEF and New York Heart Association (NYHA) functional class of 3088 patients (2228 men) reported as 61 years, 32.8% and 2.8 respectively. The follow-up was 87.7% complete over 25 months [Standard deviation (SD) 10 months]. For patients with defined myocardial viability, annual mortality rate was 16% among medically treated patients but only 3.2% among revascularised patients, representing a 79.6% relative reduction in risk of death for revascularised patients. For patients without viability, revascularisation was associated with slightly higher annual mortality than medical therapy (7.7% with revascularisation vs. 6.2% for medical therapy).

Camici et al. 68 synthesised the results from 20 studies (2217 patients) that were published between 1998 and 2006 to assess viability in patients with LV dysfunction caused by CAD. The pooled analysis by Camici et al. 68 also reported similar results with survival benefit in patients with ischaemic cardiomyopathy who underwent revascularisation compared with patients with viable myocardium treated medically (10.64% in medically treated patients but only 3.71% in revascularised patients). However, the authors report that revascularisation also reduced the mortality rate in non-viable patients (8.45% with revascularisation vs. 11.69% for medical therapy).

Schinkel et al. 22 performed a pooled analysis of 29 studies (3640 patients) and reported the annual mortality rates for revascularised and non-revascularised patients, with and without viable myocardium. For patients with defined myocardial viability, annual mortality rate was reported as 12.16% in medically treated patients but only 3.53% in revascularised patients, suggesting that patients with viable myocardium who undergo revascularisation have the best prognosis. However, the authors report that revascularisation also reduced the mortality rate in non-viable patients (8.45% with revascularisation vs. 9.59% for medical therapy).

More recently randomised control trials (RCTs) – PARR 2,69 HEART70 and STICH71 – have reported no benefit of viability testing, but these trials are subject to a number of limitations, as described here. PET and Recovery Following Revascularisation (PARR 2)69 compared optional viability testing using PET (n = 218) with standard care (n = 212) in Canada and reported no significant differences in outcomes. HEART70 (Heart Failure Revascularisation Trial) was an unblinded UK clinical study that aimed to randomise 800 patients but was stopped early because of problems with recruiting and funding. Of the 138 patients enrolled, 69 were randomised to a strategy of revascularisation, but only 45 ultimately underwent a procedure and there were no differences in mortality. STICH,71 a multicentre, non-blinded, randomised study of 601 patients conducted at 127 clinical sites in 26 countries, with a median follow-up of 5.1 years, compared optional viability testing using SPECT, echocardiography or both. The results suggested that viability status is not linked to mortality. Furthermore, these studies have significant weaknesses, as outlined elsewhere,10 and, based on the recommendation from the clinical expert group, the conclusions from these studies were deemed as not applicable to the research question under consideration.

Comparison of the results from the evidence

For patients with viable myocardium, results from all three meta-analysis studies were in agreement in suggesting that patients with viable myocardium will have improved survival after revascularisation. In the absence of viable myocardium, all three studies report that no clear-cut differences are observed between treatments, with Allman et al. 6 reporting slightly higher mortality among revascularised patients, while Camici et al. 68 and Schinkel et al. 22 report slightly lower mortality rate among patients on medical therapy.

However, Camici et al. 68 observed that the annual mortality rate in patients treated medically appears to be similar regardless of the presence of viability, which is different from what was reported by Allman et al. 6 and Schinkel et al. 22 (for patients treated medically, both report higher annual mortality rate among non-viable patients than in viable patients). Camici et al. 68 argue that it could be a reflection of the optimisation, by twenty-first century standards, of patient management because they have included only the studies published between 1998 and 2006 whereas Allman et al. 6 and Schinkel et al. 22 have also included older studies (where patient management might not have been optimal). Data from the study by Schinkel et al. 22 were used in the economic model, since this is the most recent study containing the largest cohort of patients that is relevant to the current research question.

Hospitalisations

There is insufficient evidence to model the hospitalisation rates for the four patient groups, i.e. viable and revascularised, viable but not revascularised, non-viable and revascularised, and non-viable and not revascularised. Moroi et al. 72 reported that there is no difference in hospitalisation rate between the revascularised and non-revascularised patients with stable ischaemic heart disease. Also, according to the clinical expert group, the hospitalisation rate is constant except for two periods: in the first 3 months after discharging the patient from an acute event and in the last 3–6 months of the life of patients who do not die suddenly. However, for the purposes of the model, it was assumed that the hospitalisation rate is constant because revascularisation takes place when the HF status and therapy are stable (therefore less likely to have a high readmission rate), and that the patients in NYHA stage IV are excluded from revascularisation.

The mean numbers of annual HF-related hospitalisations were estimated from the meta-analysis reported by Klersy et al. 73 and are presented in Table 12. Klersy et al. 73 reviewed 17 trials from different countries which included 2089 patients and reported an annual incidence of HF hospitalisation of 42.1%.

This annual hospitalisation rate was deemed sensible by the clinical expert group for an average non-revascularised patient with ischaemic cardiomyopathy. However, the clinical expert group reported that the main reason for revascularisation of patients with viable myocardium is to reduce the number of hospitalisations through improvement of the LV contraction. In the economic model, hospitalisation rate was assumed to be the same for three patient groups (viable but not revascularised, non-viable and revascularised, non-viable and not revascularised) while the annual hospitalisation rate for revascularised viable patients was assumed to be approximately one-third the hospitalisation rate of other patient subgroups, as suggested by the clinical expert group. Table 12 shows the parameters used in the model.

Health-related quality of life

This section provides a discussion of the evidence available on the effect caused by revascularisation for viable patients and non-viable patients on their health-related quality of life (HRQoL), the impact caused by hospital readmission for HF and the impact caused by hospital readmission for other causes. It was assumed that the survivors accrued QALYs according to their age and sex and whether or not they were revascularised. The lifetime QALYs were then estimated based on patients’ life expectancy and their corresponding annual utilities, depending upon their hospitalisation status.

In the studies identified in the review, there was no direct quantified evidence on the extent to which revascularisation improves HRQoL of the patients. However, Schinkel et al. 22 reported the improvement in symptoms using NYHA class for the patient subgroups, as shown in Table 13. This improvement in NYHA class was converted into utility values for revascularised and non-revascularised patients (both with and without viability) in the economic model.

Gohler et al. 74 performed regression analysis of QoL data of 1395 patients (mean age of 64 years) against their NYHA classes and reported that the utilities associated with NYHA classes I–IV as reported in Table 14. It should be noted that the utilities reported are based on analysis of patients with HF after acute MI and not for patient population in the current study, i.e. patients with ischaemic cardiomyopathy. However, it was assumed that the NYHA class is an independent measure of HF disease progression and one that is relevant for all HF patients.

Assuming that the data in Table 14 are applicable to the population in the economic model, the utility values can be estimated for the different patient groups from the NYHA class, assuming a linear relationship between the utility values and NYHA class. The deterministic utility values estimated for the different patient groups are reported in Table 15. For the PSA, the uncertainty in the utility values were represented by sampling independently from Tables 13 and 14 (i.e. mean NYHA classes of patient groups and the utility values for each NYHA class) and estimating the utility samples for the different patient groups by assuming a linear relationship between the utility values and NYHA class.

A disutility was incorporated for every HF-related hospitalisation based on a study by Yao et al. ,75 who estimated the disutility to be equivalent to the utility of one health state lower in terms of NYHA class. Any HF-related hospitalisation was assumed to result in a disutility of 0.1 for a whole month, i.e. approximately 0.01. Within the PSA, the disutility was represented using a triangular distribution (range –0.08 to 0.11; mode –0.1). Evidence on the disutility caused by rehospitalisation for other causes (not directly HF-related) was limited. In the absence of evidence, we assumed no disutility was caused by rehospitalisation for other causes.

Treatment effectiveness

In the model, the treatment effectiveness on survival and QoL was assumed to last only 5 years, based on the length of follow-up of studies included in Schinkel et al. 22

This meant that the annual mortality rates and the utility values (shown in Tables 10–15) are valid for only the first 5 years after the time horizon. After the 5-year treatment effectiveness period, the parameters for the general population are used.

For survival parameters, the annual mortality rate beyond the 5-year period is estimated by adding the age-specific annual mortality rate to the disease-specific mortality rate (estimated based on patient subgroup). The age-specific annual mortality rates are estimated from life expectancy tables assuming an 85% : 15% male to female ratio.

Furthermore, the utility values are also capped at age-specific general population utilities in order to ensure that the patient utilities do not exceed the average population utility in their age group. The age-specific utilities are estimated from a study by Ara et al. 22,76 assuming an 85% : 15% male to female ratio.

Risks associated with the treatment procedures

The interventions also carried risks to patient health, and these were estimated as a probability of death each time the procedure was performed. The HES data showed that there were 67,908 PCIs compared with 19,743 CABGs, resulting in a PCI to CABG ratio of 3.44. As revascularisation is represented as one procedure (i.e. not distinguishing between PCI or CABG), the mortality risk for a single revascularisation procedure was estimated as a weighted average of the mortality of PCI and CABG with their corresponding proportions. The overall mortality used in the model is as shown in Table 16.

Risks associated with the diagnostic tests

Some of the investigations also carried risks to patient health. These can be modelled by estimating a QALY loss that was applied each time the investigation was performed.

1. risk of death or MI induced by stress echocardiography

2. risk of developing death, MI or radiation-related malignancy as a consequence of SPECT

3. risk of developing death, MI or radiation-related malignancy as a consequence of PET.

However, given the lack of evidence on the adverse events of the diagnostic tests and the suggestion from the expert clinical group that they are negligible, the model analyses were performed assuming that there are no risks associated with the diagnostic tests.

Costs

The costs included in the model are:

1. diagnostic testing costs

2. treatments administered

3. subsequent cardiac hospitalisations.

We assumed that patients would incur costs whenever a diagnostic test was performed, and the costs of diagnostic tests were estimated from literature, as shown in Table 17. The cost of treatment for revascularisation was estimated as shown in Table 18.

The mean inpatient admission cost for HF-related hospitalisations was calculated as shown in Table 19 from the weighted average of the costs for the health-related group ‘Heart Failure or Shock’ (EB03I) based on the data obtained from the NHS Reference Costs for 2011. 77

As there was no evidence on the annual costs of survivors (e.g. beyond hospitalisations), it was assumed that these costs were the same across both arms.

Methods to estimate cost-effectiveness

Cost-effectiveness of the different interventions was estimated using both the incremental cost-effectiveness ratio (ICER) and net monetary benefit (NMB) approach. Uncertainty was incorporated in the modelling by performing PSA. A description of these terms and approaches is provided in this section.

Definitions of cost-effectiveness terms

The ICER measures the relative value of two strategies and is calculated as the mean incremental costs divided by the mean incremental benefits, i.e. the additional cost required using the strategy to accrue one additional QALY compared with the next most effective alternative. A strategy is dominated when another strategy accrues more QALYs for less cost. Extended dominance occurs when a combination of two alternative strategies can produce the same QALYs as a chosen strategy but at a lower cost. Strategies that are neither dominated nor extendedly dominated constitute the cost-effectiveness frontier, and the ICER is reported for these strategies compared with the next least effective strategy. However, as there are multiple possible strategies, ICERs need to be calculated between different pairs, comparing each strategy against the next most effective strategy.

The willingness-to-pay (WTP) threshold is the amount of money the decision-maker is willing to pay to gain one additional QALY. The usual threshold for decision-making at NICE is considered to be around £20,000 per QALY,78 so if the ICER exceeds £20,000 per QALY then the strategy is unlikely to be considered cost-effective.

The NMB is defined as the QALYs multiplied by a value for the QALYs (e.g. £20,000) minus the costs of obtaining them, i.e. NMB = QALYs × λ – cost, where λ is the WTP threshold. The net benefit approach is simpler to calculate and gives equivalent findings (but requires an explicit assumption regarding the value of λ).

Uncertainty analysis

The results presented in Results of independent economic assessment include the effects of accounting for uncertainty in the model parameters, i.e. essentially the CIs surrounding the diagnostic accuracy, costs, utilities and risks (for mortality and HF hospitalisations). PSA is undertaken where the model is rerun (10,000 times) each time with a different value for the sensitivity and specificity estimates, mortality risks, hospitalisation rates, costs and utilities which are sampled from the probability distributions.

The cost-effectiveness acceptability curve (CEAC) shows the proportion of model runs for which each strategy is cost-effective over a range of potential WTP thresholds (i.e. λ). Another measure of uncertainty is the overall expected value of perfect information (EVPI). This calculation is done based on the theory that the decision-maker will choose the most cost-effective option but could acquire additional evidence to reduce the uncertainties in the decision, e.g. knowing exactly what the mortality and hospitalisation risks are for each patient group rather than having evidence based on current best estimates and CIs. In the EVPI calculation, it can be estimated how often making the decision based on current evidence could be wrong and also how many QALYs (and costs) would be lost by choosing the strategy that is expected to be most cost-effective given current evidence when in fact one of the other strategies is truly the most cost-effective. Valuing the QALYs lost by making a ‘wrong’ decision to choose a strategy based on current evidence by using the WTP threshold, one can estimate a monetary value for this possible loss on each of the occasions when another strategy would be optimal, i.e. the net benefit lost. This can be multiplied by the number of patients per year and the expected lifetime of the decision to estimate the population EVPI. The interpretation of this number is that if one could fund research to eliminate the uncertainty in mortality risks for different patient groups (e.g. by a large or infinitely large clinical trial) then the value of eliminating the uncertainty via such research would be expected to be the population EVPI.

Diagnostic performance of different pathways

Table 21 shows the mean values of sensitivity and specificity of all the pathways tested in the model, estimated from the meta-analysis.

The mean numbers of patients in different subgroups after the diagnostic pathway are shown in Table 22. These patient numbers are estimated using an initial cohort size of 10,000 patients with a mean prevalence of viable myocardium of 0.5, i.e. 5000 patients out of a total of 10,000 patients are viable, with the remainder of the 5000 patients non-viable. The higher the sensitivity of the pathway, the higher the proportion of 5000 patients with viable myocardium detected to be revascularised and, consequently, the lower the number of patients with viable myocardium on medical therapy. Similarly, the higher the specificity, the higher the number of patients with non-viable myocardium detected correctly (and not revascularised unnecessarily) and the lower the number of revascularised patients with non-viable myocardium.

As presented earlier, the revascularisation procedure carries a risk of perioperative mortality. The mean value of this risk of death in the model is 0.5% for patients with viable myocardium and 1% for patients who do not have viable myocardium. The mean values of total number of deaths for each diagnostic pathway are also presented in Table 22.

Although the performance of the diagnostic pathways in terms of sensitivity and specificity is the same in both scenarios tested, the cost-effectiveness estimates are different because of the effect of different mortality assumptions (which affect the total costs and total QALYs). Therefore, cost-effectiveness for each of these scenarios is presented separately below and the users can decide which scenario is most relevant for them.

Cost-effectiveness results for scenario using data from Allman et al.6

This section presents the cost-effectiveness results using the mortality rates presented in Table 11. If the annual mortality rates for non-viable patients are higher for revascularised patients than for patients on medical therapy, then the results from this scenario might be considered more relevant.

In this scenario, no testing, the stress CMR, CE CMR and PET pathways are on the cost-effectiveness frontier, as shown in Table 23. Stress CMR is cost-effective with a mean ICER of £1073.8 per QALY compared with the no testing strategy, and CE CMR has an ICER of £2906.5 per QALY compared with stress CMR. Both strategies are cost-effective assuming a threshold of £20,000 per QALY. Echocardiography and SPECT are dominated by stress CMR and CE CMR respectively. In addition, PET is not cost-effective at a threshold of £20,000 per QALY as it has an ICER of £21,299 per QALY compared with CE CMR. Furthermore, the strategy of revascularising everyone is dominated by PET. Thus, CE CMR is estimated to be the most cost-effective option at a threshold of £20,000/QALY. This is because CE CMR has good sensitivity with a reasonable specificity and also costs less, resulting in it being the best strategy in terms of cost-effectiveness. The higher ICER for PET can be attributed to its high costs even though it has better sensitivity and specificity than CE CMR.

Furthermore, it should be noted that, compared with no testing, all the diagnostic pathways are cost-effective at the current NICE threshold. This suggests that all current services for diagnosing viable myocardium are a cost-effective use of NHS resources irrespective of the diagnostic pathway used, provided their costs and diagnostic accuracy are similar to those reported in this analysis.

Another presentation of these same results is to calculate the net benefit of each strategy. This approach takes away the need to calculate the ICERs and simplifies the interpretation for decision makers as the strategy with the highest expected incremental monetary net benefit is the most cost-effective. Since the model is rerun 10,000 times each time with different values for the sensitivity, specificity, costs and utilities sampled from the probability distributions, in some of the sampled model runs, it could turn out that one diagnostic pathway is better than others because of the overlap in their CIs. For example, there is a chance that in truth, the cost effectiveness of PET could be better than CE CMR. The CEAC in Figure 8 shows the proportion of model runs for which each strategy is cost-effective over a range of potential WTP thresholds. The proportion of models runs in which CE CMR was the most cost-effective strategy (at £20,000 per QALY threshold) was 40%, with PET at 42% and revascularising everyone at 16.5%, as shown in Figure 8.

FIGURE 8.

Cost-effectiveness acceptability curve for scenario using data from Allman et al. 6 MAICER, maximum acceptable incremental cost-effectiveness ratio.

A CEAC in which the best strategy is not cost-effective all the time indicates that there is uncertainty as to which strategy is the optimal in terms of NMB. This uncertainty can also be measured as overall EVPI, which is defined as the maximum investment a decision-maker would be willing to pay to eliminate all parameter uncertainty from the decision problem. The EVPI at the threshold of £20,000 per QALY in this case is £620 per patient for whom the decision is made, as shown in Figure 9. The population EVPI per annum can be estimated by multiplying the EVPI per patient with the annual incidence of patients with ischaemic cardiomyopathy in England and Wales.

FIGURE 9.

Expected value of perfect information per patient for scenario using data from Allman et al. 6 MAICER, maximum acceptable incremental cost-effectiveness ratio.

Cost-effectiveness results for scenario using data from Schinkel et al.22

The analyses were also performed using mortality rates presented in Table 10, based on the data from Schinkel et al. 22 If patients who are revascularised have lower mortality rates, even if they do not have viability, then the results from this scenario might be considered more relevant.

The strategies that are on the cost-effectiveness frontier are the no testing, stress CMR, CE CMR and revascularise everyone pathways, and the ICERs calculated for these strategies are as shown in Table 24. In this scenario, revascularising everyone is estimated to be the most cost-effective option with an ICER of £3612 per QALY. This is because all patients get benefit from revascularisation (including the non-viable patients); therefore, revascularising everyone is the best strategy in terms of cost-effectiveness.

All the diagnostic pathways are cost-effective at the current NICE threshold when compared with no testing. This suggests that all the current services for diagnosing viable myocardium are a cost-effective use of NHS resources in this scenario as well, provided their costs and diagnostic accuracy are similar to those reported in this analysis.

Revascularising everyone was the most cost-effective strategy (at £20,000 per QALY threshold) and is cost-effective in 95.2% of the model runs, with CE CMR and PET being cost-effective in 3.6% and 1.1% of the runs respectively (Figure 10). The decrease in uncertainty compared with the scenario using data from Allman et al. 6 can be attributed to the fact that all patients receive benefit from revascularisation (including the non-viable patients) in the current scenario (i.e. using data from Schinkel et al. 22), thus revascularising everyone is the best strategy in terms of cost-effectiveness. In contrast, in the scenario using Allman et al. 6, the benefits of revascularisation are only from the patients in the viable myocardium group (rather than everyone as in scenario 1). This reduction in uncertainty is also reflected in the EVPI of only £28 per patient, as seen in Figure 11.

FIGURE 10.

Cost-effectiveness acceptability curve for scenario using data from Schinkel et al. 22 MAICER, maximum acceptable incremental cost-effectiveness ratio.

FIGURE 11.

Expected value of perfect information per patient for scenario 3. MAICER, maximum acceptable incremental cost-effectiveness ratio.

Statement of principal findings

The current evidence is difficult to interpret given the variability of the diagnostic criteria used in different studies, and to the conflicting outcomes of the studies that looked at the issues of revascularisation and viability imaging in the literature. We have tried to cover that uncertainty through proposing different scenarios and applying the current evidence base to each of these scenarios. If the presence of viable myocardium is believed to have an impact on the management strategy, then a viability assessment using CE CMR appears to be most cost-effective.

Initial search

1. Cardiomyopathies/ or Myocardial Ischemia/ or Cardiomyopathy, Dilated/

2. (isch$ adj1 cardiomyopath$).mp. [mp=protocol supplementary concept, rare disease supplementary concept, title, original title, abstract, name of substance word, subject heading word, unique identifier]

3. 1 or 2

4. Magnetic Resonance Imaging/ or Magnetic Resonance Imaging, Cine/

5. cardiac magnetic resonance imag$.ti,ab.

6. cardiac magnetic resonance.mp.

7. CMR.ti,ab.

8. or/4-7

9. 3 and 8

10. exp "Sensitivity and Specificity"/

11. sensitivity.tw.

12. specificity.tw.

13. ((pre-test or pretest) adj probability).tw.

14. post-test probability.tw.

15. predictive value$.tw.

16. likelihood ratio$.tw.

17. or/10-16

18. prognosis.sh.

19. diagnosed.tw.

20. cohort:.mp.

21. predictor:.tw.

22. death.tw.

23. exp models, statistical/

24. or/18-23

25. 17 or 24

26. 9 and 25

27. Meta-Analysis/

28. meta analy$.tw.

29. metaanaly$.tw.

30. meta analysis.pt.

31. (systematic adj (review$1 or overview$1)).tw.

32. exp Review Literature/

33. or/27-32

34. cochrane.ab.

35. embase.ab.

36. (psychlit or psyclit).ab.

37. (psychinfo or psycinfo).ab.

38. (cinahl or cinhal).ab.

39. science citation index.ab.

40. bids.ab.

41. cancerlit.ab.

42. or/34-41

43. reference list$.ab.

44. bibliograph$.ab.

45. hand-search$.ab.

46. relevant journals.ab.

47. manual search$.ab.

48. or/43-47

49. selection criteria.ab.

50. data extraction.ab.

51. 49 or 50

52. review.pt.

53. 51 and 52

54. comment.pt.

55. letter.pt.

56. editorial.pt.

57. animal/

58. human/

59. 57 not (57 and 58)

60. or/54-56,59 (4,524,361)

61. 33 or 42 or 48 or 53 (1,626,552)

62. 61 not 60 (1,479,918)

63. 9 and 62

64. randomized controlled trial.pt.

65. controlled clinical trial.pt.

66. randomized controlled trials/

67. random allocation/

68. double blind method/

69. single blind method/

70. clinical trial.pt.

71. exp Clinical Trial/

72. (clin$ adj25 trial$).ti,ab.

73. ((singl$ or doubl$ or trebl$ or tripl$) adj25 (blind$ or mask$)).ti,ab.

74. placebos/

75. placebos.ti,ab.

76. random.ti,ab.

77. research design/

78. or/64-77

79. 9 and 78

80. Practice Guideline/

81. Guideline/

82. Practice Guidelines as Topic/

83. Consensus Development Conference/

84. Guideline Adherence/

85. practice guideline.pt.

86. guideline.pt.

87. consensus development conference.pt.

88. practice guideline$.tw.

89. practice parameter$.tw.

90. recommendation$.tw.

91. guideline$.ti.

92. consensus.ti.

93. or/80-92

94. (comment or letter or editorial or note or erratum or short survey or news or newspaper article or patient education handout or case report or historical article).pt.

95. 93 not 94

96. Animals/

97. 95 not 96

98. 9 and 97

99. Economics/

100. "costs and cost analysis"/

101. Cost allocation/

102. Cost-benefit analysis/

103. Cost control/

104. cost savings/

105. Cost of illness/

106. Cost sharing/

107. "deductibles and coinsurance"/

108. Health care costs/

109. Direct service costs/

110. Drug costs/

111. Employer health costs/

112. Hospital costs/

113. Health expenditures/

114. Capital expenditures/

115. Value of life/

116. exp economics, hospital/

117. exp economics, medical/

118. Economics, nursing/

119. Economics, pharmaceutical/

120. exp "fees and charges"/

121. exp budgets/

122. (low adj cost).mp.

123. (high adj cost).mp.

124. (health?care adj cost$).mp.

125. (fiscal or funding or financial or finance).tw.

126. (cost adj estimate$).mp.

127. (cost adj variable).mp.

128. (unit adj cost$).mp.

129. (economic$ or pharmacoeconomic$ or price$ or pricing).tw.

130. or/99-129

131. 9 and 130

Additional searches

1. *Myocardial Revascularisation/

2. Magnetic Resonance Imaging/

3. magnetic resonance imag$.ti,ab.

4. MRI.ti,ab.

5. or/2-4

6. 1 and 5

7. ri.fs.

8. Myocardial Infarction/

9. Coronary Disease/

10. Coronary Artery Disease/

11. or/2-4

12. 1 and 5

13. di.fs.

14. 6 and 7

15. Magnetic Resonance Imaging/

16. 8 and 9

17. Cardiomyopathies/ or Myocardial Ischemia/ or Cardiomyopathy, Dilated/

18. (isch$ adj1 cardiomyopath$).mp. [mp=protocol supplementary concept, rare disease supplementary concept, title, original title, abstract, name of substance word, subject heading word, unique identifier]

19. 1 or 2

20. Magnetic Resonance Imaging/ or Magnetic Resonance Imaging, Cine/

21. cardiac magnetic resonance imag$.ti,ab.

22. cardiac magnetic resonance.mp.

23. CMR.ti,ab.

24. or/4-7

25. 3 and 8

26. comparative study.pt

27. 9 and 10

28. *Magnetic Resonance Imaging/

29. myocardial salvage.ti,ab.

30. 1 and 2