Evidence-based cardiovascular magnetic resonance cost-effectiveness calculator for the detection of significant coronary artery disease

Background Although prior reports have evaluated the clinical and cost impacts of cardiovascular magnetic resonance (CMR) for low-to-intermediate-risk patients with suspected significant coronary artery disease (CAD), the cost-effectiveness of CMR compared to relevant comparators remains poorly understood. We aimed to summarize the cost-effectiveness literature on CMR for CAD and create a cost-effectiveness calculator, useable worldwide, to approximate the cost-per-quality-adjusted-life-year (QALY) of CMR and relevant comparators with context-specific patient-level and system-level inputs. Methods We searched the Tufts Cost-Effectiveness Analysis Registry and PubMed for cost-per-QALY or cost-per-life-year-saved studies of CMR to detect significant CAD. We also developed a linear regression meta-model (CMR Cost-Effectiveness Calculator) based on a larger CMR cost-effectiveness simulation model that can approximate CMR lifetime discount cost, QALY, and cost effectiveness compared to relevant comparators [such as single-photon emission computed tomography (SPECT), coronary computed tomography angiography (CCTA)] or invasive coronary angiography. Results CMR was cost-effective for evaluation of significant CAD (either health-improving and cost saving or having a cost-per-QALY or cost-per-life-year result lower than the cost-effectiveness threshold) versus its relevant comparator in 10 out of 15 studies, with 3 studies reporting uncertain cost effectiveness, and 2 studies showing CCTA was optimal. Our cost-effectiveness calculator showed that CCTA was not cost-effective in the US compared to CMR when the most recent publications on imaging performance were included in the model. Conclusions Based on current world-wide evidence in the literature, CMR usually represents a cost-effective option compared to relevant comparators to assess for significant CAD. Supplementary Information The online version contains supplementary material available at 10.1186/s12968-021-00833-1.


Introduction
For patients with suspected coronary artery disease (CAD), cardiovascular magnetic resonance (CMR) offers a non-invasive and accurate diagnostic option. However, there are other diagnostic alternatives for such patients at low-to-intermediate risk, including ergometry, singlephoton emission computed tomography (SPECT), or coronary computed tomography angiography (CCTA). Even conservative management, i.e. waiting for symptomatic disease to worsen and to manifest by complications, or more aggressive management, i.e., or immediate invasive coronary angiography (ICA), could represent options [1]. Depending on certain factors, such as imaging costs, local availability of imaging modalities and expertise and prevalence of underlying disease, tradeoffs may be realized among these options that can influence clinical decision-making [2].
Cost-effectiveness analysis can be used to quantitatively weigh tradeoffs between length of life, quality of life, and incurred costs, across different diagnostic strategies, allowing payers and physician decision-makers to choose a higher-value pathways [3,4]. Most health technology assessments in high-income countries use costper-quality-adjusted life-year (QALY) to quantify value from cost-effectiveness analyses, with estimated costs including both immediate imaging costs and all downstream costs (including both additional care stemming from test results and saved costs from averted coronary heart disease events, follow-up tests, and procedures) [5]. Although there have been prior publications evaluating the clinical and cost impacts of CMR-based diagnostic strategies in patients with suspected CAD, there has been less clarity on the specific cost-effectiveness profile of CMR compared to its relevant comparators across both clinical settings and patient types, particularly as evidence of imaging performance continues to evolve [6][7][8].
In a 2-step approach, we first analyzed and summarized the comparative cost effectiveness surrounding use of CMR imaging for the assessment of patients presenting with stable chest pain syndromes compared to its relevant comparators based on the existing medical literature. Then we created a unique cost-effectiveness calculator that could be used globally to estimate lifetime discounted costs and QALYs for CMR versus its relevant comparator techniques. To facilitate global use for different international geographic regions or referral populations, adjustment of context-specific patient and system-level inputs (such as disease prevalence and imaging costs) was incorporated into the cost-effectiveness calculator.

Search strategy of existing literature and data extraction
We conducted a systematic literature review using the Tufts Cost-Effectiveness Analysis Registry (CEA Registry, www. ceare gistry. org) and PubMed for English-language cost-effectiveness published from 2005 to 2020 [9, 10]. The CEA Registry contains 8000 English-language costper-QALY studies. The CEA Registry uses keywords such as QALYs, quality adjusted, and cost-utility analysis to search PubMed for English-language publications. The reference lists of every identified CMR cost-per-QALY or relevant review study were searched to identify additional CMR cost-effectiveness studies missed by our other methods. We only included studies that used CMR as an imaging strategy to assess for CAD as the primary clinical condition. Two reviewers (YY and AP) independently reviewed each study to extract relevant data, and resolved any differences in data extractions at in-person meetings; a third author (SK) was contacted when consensus could not be reached at the in-person meetings. Detailed information on the search strategy and data extraction for the CEA Registry is reported elsewhere [10,11].
The search terms used to identify cost-per-QALY studies were combinations of methodological terms (costeffectiveness, QALY, incremental cost-effective ratio) and clinical terms (cardiac magnetic resonance, CMR, coronary angiography). Articles other than cost-effectiveness studies that were related to the overall value of CMR (reviews, meta-analyses, diagnostic performance, editorials, etc.) were identified but not included in our systematic review (Appendix Fig. 3). We reviewed the cost-effectiveness studies that included CMR as a strategy and summarized the following information among these articles: setting of the analysis, comparators included, analytic perspective taken, analytic time horizon taken, main conclusion on the cost effectiveness of CMR (where "cost-effective" was defined as either healthimproving and cost saving ["dominant"] [3] or having a cost-per-QALY or cost-per-life-year result lower than the cost-effectiveness threshold [4]), and key drivers of the results. We performed a sensitivity analysis excluding papers that reported cost-effectiveness outcomes other than cost-per-QALY or cost-per-life year (such as costper-case detected).

Keywords: Cost-effectiveness, Cardiovascular magnetic resonance, Coronary artery disease
We used author assessments to determine whether CMR represented a cost-effective option or cost-ineffective option for a given paper, which could depend on country-specific cost-effectiveness thresholds. In the United States (US), for example, the American College of Cardiology (ACC) and the American Heart Association (AHA) issued a joint statement on health care "value" in 2014 that specified that cost-per-QALY results below $50,000/QALY indicate high-value care, cost-per-QALY estimates between $50,000/QALY and $150,000/ QALY indicate intermediate-value care, and cost-per-QALY estimates greater than $150,000/QALY indicate low-value care [12]. We also categorized some papers as showing unclear CMR cost-effectiveness when there was considerable uncertainty around the CMR cost-effectiveness results.

Development of the CMR cost effectiveness calculator (meta-model)
We built a user-friendly CMR cost effectiveness metamodel based on a larger CMR Markov (state-transition cohort) model developed for a prior CMR cost-per-QALY study (by study co-authors AP, YG, RK) performed for a US health care system perspective [13]. The model projects lifetime discounted QALY and cost outputs for five strategies: (1) no imaging; (2) CMR; (3) SPECT; (4) CCTA; (5) ICA. Other strategies, such as echocardiography, were not included in the larger Markov model. In this model, patients in the CMR, SPECT, and CCTA groups underwent ICA only if noninvasive imaging demonstrated abnormal findings. Those with positive ICA results (with some having also received FFR) were assumed to undergo both medical and revascularization therapies, and this combination led to overall improved health outcomes (quantified using lifetime discounted QALYs). Patients with normal findings were presumed to be free of obstructive CAD and were managed accordingly. In the no imaging strategy patients were initially managed without any investigations. Assuming escalating symptoms in 58% of patients with obstructive CAD who did not receive treatment in their first year after assessment (i.e., patients with false negative results) would return within the first year and undergo ICA, leading to medical and revascularization therapies (i.e., 58% of false negatives would experience the same outcomes as true positives within 1 year) [14]. The Markov model had four major health states: no clinical major cardiovascular events [MACE, defined as: cardiovascular death, acute nonfatal myocardial infarction (MI), hospitalization for unstable angina or heart failure], history of one MACE, history of more than one MACE, and all-cause death [13]. In the Markov model, QALYs are estimated by combining the length of life patients spend in each of the health states with a quality-of-life value ("utility") ranging between 0 (representing death) and 1 (perfect health), with no MACE having an average utility value of 0.84 and MACE having an average utility of 0.78 [15]. Life costs estimated by the Markov model depend on the cost of the imaging strategies used, procedures performed, and the acute and long-term healthcare-related costs associated with each health state (with MACE states having 1st-year costs between $11,000-18,000 and subsequent annual costs of $3400) [13]. Both QALY and cost outcomes can be discounted at an annual rate (such as 3%, recommended for cost-effectiveness analyses performed for the US setting) [16].
The linear regression-based meta-model was trained (i.e., coefficients were estimated) on 100,000 model input-output combinations from the probabilistic sensitivity analysis (2nd-order Monte Carlo simulation) from the larger Markov simulation model [17]. Previous studies have shown that simple linear regression-based metamodels can approximate larger disease models with high accuracy (r-squared values that often exceed 0.95). The meta-model allows users to replicate the larger Markov model results and change 10-20 key meta-model inputs to view corresponding meta-model outputs to create customized results for specific scenarios or populations of interest [18,19]. Meta-model variable selection was based on a 0.05 level of significance for beta coefficients in the linear regressions. We validated our meta-model on a separate 1000 model input-output combinations from the probabilistic sensitivity analysis (2nd-order Monte Carlo simulation) from the larger Markov simulation model that were not used to train the meta-model (i.e., the test set) [17]. Meta-model goodness-of-fit was assessed using adjusted r-squared and percentage deviation (for external validation) metrics for the test set.
The meta-model approximates the lifetime discounted QALY and cost results for a given imaging strategy based on user-entered population-and system-level inputs, such as prevalence of CAD, costs of imaging and procedures, and other model inputs listed in Appendix Table 3. We used the meta-model to explain observed differences in conclusions from published cost-effectiveness analyses included in our literature review that were performed in the US setting. Specifically, we used our meta-model to replicate the cost-effectiveness results of the US study it was based on Ge et al. [13], which found CMR to be cost-effective compared to CCTA, SPECT, ICA, and no imaging strategies [13]. Then we changed key input parameters of the meta-model (such as imaging performance and cost parameters, CAD prevalence, treatment costs, etc.) to replicate the cost-effectiveness results of another US study (with a similar model structure and decision problem) that found CCTA to be cost-effective compared to CMR [20]. Using the meta-model, we then changed only the imaging performance inputs from the values in Genders et al. [20] cost-effectiveness study to the more recent imaging performance inputs used by Ge et al. [13] to determine whether these imaging performance inputs alone could explain the difference in CMR cost-effectiveness results.
The larger Markov model was programmed in TreeAge Pro 2019 software (version 19.2.1; TreeAge Software, LLC, Williamstown, Massachusetts, USA), the meta-model was created using RStudio (version 1.2.5042; RStudio Software, Boston, Massachusetts, USA), and the user-friendly cost-effectiveness calculator is programmed in Microsoft Excel (Microsoft Corporation, Redmond, Washington, USA) and is included as a downloadable spreadsheet in Additional file 1 and as a web-based tool at: https:// docs. google. com/ sprea dshee ts/d/ 12TMg bIS6s DpXka fSYfN aNw8v Sx-LCTCE 4K_ yAa7m dyY/ edit? usp= shari ng.

Results
Our search yielded 39 studies of which 15 ultimately met our inclusion criteria (shown in Table 1 and Appendix Fig. 3) [13,14,[20][21][22][23][24][25][26][27][28][29][30]. We excluded 17 studies because they were not original research (such as perspective Not stated 1 7 Moschetti et al. [24] articles or literature reviews), they focused on the costeffectiveness of CMR for heart failure patients (as opposed to CAD), or they did not include CMR as a comparator. Of the remaining studies, an additional seven were excluded for not being a cost-effectiveness analysis (such as studies evaluating only costs or clinical effectiveness). Further two studies were excluded in our sensitivity analysis restricting our analysis to cost-per-QALY or cost-per-life-year studies [24,29]. Table 1 shows that most of the 15 studies that met our inclusion criteria were performed in a US (five studies) [13,20,24,28,29] or European setting (seven included the United Kingdom [14,20,23,24,28,30,31], three included Germany [22,24,25] Simulation models (such as decision trees or state transition models) were used for 10/15 studies [13, 14, 20-24, 26, 29, 32], and seven of these ten studies extrapolated outcomes for a lifetime time horizon; the five studies not using simulation models relied on empirical data [25,27,28,30,31] resulting in time horizons of less than 10 years.

Summary of results of CMR cost-effectiveness studies
CMR was found to be cost-effective versus its relevant comparator in 10/15 studies [13, 14, 22, 24-27, 29, 31, 32]. Among these ten studies, the most common comparators to CMR were strategies that used ICA [13, 14, 22, 24-26, 31, 32] or SPECT [13,14,22,26,29,31,32], while three studies compared CMR versus strategies that used CCTA [13,27,31], and one directly compared CMR to a no imaging strategy [13]. Two studies concluded that CCTA was more costeffective (Bertoldi et al. and Genders et al.) compared to CMR [20,21]. In sensitivity analyses, Bertoldi et al. found that a cost reduction of 79% was required for the CMR strategy to be cost-effective versus the CCTA from the Brazilian public health system perspective [21]. This result was driven by only slightly higher QALYs from the CMR strategy compared to the CCTA strategy based on sensitivity and specificity inputs from meta analyses published in 2010 [33] (for CMR) and 2008 [34] (for CCTA). Genders et al. focused on a low-to-moderate risk population (performing separate analyses for men and women) and found that CCTA as a first-line test (combined with baseline echocardiography in all patients and additional invasive diagnostic work-up in patients with positive CCTA) almost always dominated CMR in the three settings analyzed (the United States, the Netherlands, and the United Kingdom) [20]. These results were primarily driven by assuming superior accuracy for CCTA (sensitivity of 0.98 and specificity of 0.89) [34][35][36] compared to CMR (0.89 and 0.76, respectively) [37]. Model-based cost-effectiveness studies by Walker et al. and Ge et al. used similar model structures as applied by Genders et al., but with CMR operating characteristics based on more recent trials and meta-analyses [7,8,38]. With this approach CMR was cost-effective compared to other imaging strategies such as ICA [13,14], SPECT [13,14], and CCTA (with CT-derived fractional flow reserve) [13].
Three studies found unclear cost-effectiveness among the imaging strategies analyzed [23,28,30], including two studies that were cost-effectiveness analyses conducted alongside a randomized controlled trial, which were not powered to show statistically significant differences in cost-effectiveness outcomes [28,30]. Campbell et al. developed a simulation model that included positron emission tomography (PET) for the United Kingdom health care payer perspective and found in probabilistic sensitivity analyses that there were similar probabilities of CMR or PET being optimal in the relevant cost-effectiveness threshold ranges for the United Kingdom [23].

CMR cost effectiveness calculator (meta-model) results
The fitted coefficients for the lifetime discounted QALY and cost results for 'No Imaging' strategy, and incremental QALY and cost results for the 'CMR' , 'CCTA' , 'SPECT' , and 'Immediate ICA' strategies are shown in Appendix Table 3. Validation using the test sets showed good fits for each model, with r-squared values ranging from 0.846 to 0.999 across the ten meta-models, with percentage deviation varying from − 0.117% to 0.028%, which are also shown in Appendix Table 3. Table 2 shows the cost-effectiveness analysis results of our meta-model compared to: (1) the model originally used to derive it (i.e., a replication of the Ge et al. cost-effectiveness results using our meta-model); (2) the Genders et al. study that concluded that CCTA was more cost-effective than CMR in the US (i.e., a replication of the Genders et al. cost-effectiveness results using our meta-model); (3) and meta-model results using key input values from the Genders et al. paper (such as CMR and CCTA sensitivity and specificity, CMR and CCTA costs, underlying prevalence of CAD, and other selected inputs) with Ge et al. model inputs for all other inputs (see Appendix Table 4 for full list) [13,20].
We were able to use the meta-model to closely replicate the incremental cost-effectiveness results for CCTA compared to a 'No Imaging' strategy and to CMR (  Figs. 1 and 2). CMR would have to cost 182% of our base-case estimate ($807 to $1465) before the CCTA strategy would be considered more cost-effective (using a willingness-to-pay of $100,000 per QALY). Appendix Table 5 shows other threshold values for imaging performance, prevalence, and age inputs for CMR and the other imaging strategies we evaluated using the meta-model.

Discussion
Our systematic literature review found that most (62%) studies formally evaluating the cost-effectiveness of CMR compared to other relevant imaging options for patients with suspected CAD world-wide concluded that: • CMR-based diagnostic strategies produced health at reasonable value compared to setting-specific costeffectiveness thresholds. • When CMR is not available, CCTA represents a cost-effective alternative compared to a no imaging strategy or immediate ICA strategy, which is consistent with current recommendations of major international cardiac societies [39].
These cost-effectiveness outcomes depend on the operating characteristics of the imaging modality of interest and the underlying prevalence of CAD in the population of interest. It is therefore of importance for readers of such cost-effectiveness analyses to scrutinize the sources of these key variables when interpreting the resulting cost-effectiveness outcomes. Publications which concluded that CCTA dominated CMR used inputs in the cost-effectiveness models which were outside the range of the majority of reported results. We also found that relying on a single data source to estimate the cost-effectiveness of CMR, such as a cost-effectiveness analysis conducted alongside a clinical trial, could lead to uncertain results (e.g. when such trials are underpowered for cost and QALY outcomes). Considering uncertain and setting-specific inputs that could drive cost-effectiveness results, we developed a CMR cost-effectiveness calculator that end-users (physicians, hospital decision-makers, guidelines writers, payers, health economist researchers) can use to approximate settingspecific cost-effectiveness results. The ability to iteratively  generate these estimates is crucial given the importance of weighting potential tradeoffs between length of life, quality of life, and local costs across imaging modalities available to physicians aiming to diagnosis and treat ischemic heart disease, and the setting-specific and evolving nature of these inputs. The calculator also allows to determine strategies, which allow to achieve diagnosis at the lowest possible cost, or to treat significant CAD with the lowest cumulative cost of care. As expected, changing the sensitivity and specificity of CMR or CCTA changes the relative cost effectiveness of each of these modalities. Our cost-effectiveness calculator allows users to update these inputs as newer evidence and meta-analyses are published. Our replication and adaptation of the Genders et al. study showed this explicitly. Other key drivers of CMR cost-effectiveness, such as disease prevalence and imaging or treatment costs, can also be settingspecific, and we found there are many settings (almost all non-US/non-European countries, with the exceptions of one study for Brazil and one study for Australia) without any formal cost-effectiveness analyses published. Users can use our tool world-wide to better align their local understanding of model inputs to cost-effectiveness results.
Our study focused on cost-per-QALY or cost-per-lifeyear-saved studies, but there are other types of economic evaluation studies beyond formal cost-effectiveness analyses (such as cost-minimization analyses) that we identified comparing CMR to other imaging strategies used to diagnose CAD [40]. Data from the European CMR registry [41], which contains data from 59 medical centers across 18 countries, were used for two such analyses that compared CMR to ICA-based strategies; these studies both found that the CMR-based strategy would result in cost savings compared to inpatient ICA, driven by the costs differences between strategies and reduced revascularization procedures in the CMR-based strategies [42,43]. A randomized controlled trial assigning 109 patients to either a CMR observation unit arm or usual inpatient care arm found that the CMR-based care arm reduced cardiac-related costs during the hospitalization and over the first year postdischarge [44]. These studies were not included in our literature review as they did not include a QALY or life-year effectiveness measure, but they add to our overall study conclusion that CMR can represent high-or intermediatevalue care, or even produce cost-savings, depending on the imaging strategy CMR is being compared to. Future clinical studies providing sex-specific inputs (on imaging performance, for instance) could also help reveal whether the cost-effectiveness of CMR differs for men and women.

Limitations
Our study has limitations that should be noted. For our systematic literature review, we were limited to the existing published literature. There is publication bias in terms of what settings cost-effectiveness of CMR studies are performed for (with US, Germany, and the United Kingdom relatively overrepresented), and there is the possibility of financial or nonfinancial bias in the study authors that could affect which model inputs they choose (thus affecting model results). Our cost-effectiveness calculator tool can somewhat mitigate this bias, if users have non-biased inputs they believe would better reflect the current state of the evidence or their local populations or settings. The cost-effectiveness calculator is based on a meta-model of a larger Markov model, and therefore the calculator contains many model limitations of the larger model in addition to its own imperfect ability to replicate the model results. To overcome this limitation, we performed a model validation analysis on the meta-model using data that were specifically not used to generate the model. Our metamodel was also limited to the strategies evaluated in the larger Markov model [13], which is why echocardiography is not included as a comparator in the meta-model. Due to data limitations, we also did not model potential side effects from contrast agents used for CCTA or CMR or radiation exposure from CCTA, which would have amplified our lifetime discount QALY meta-model results comparing CCTA to CMR [45]. Finally, our review and cost calculator focus on cost-effectiveness outcomes, which might not capture all dimension that are relevant for health policy or clinical decision-making, missing elements such as equity and patients' preference considerations.
Despite those limitations, cost-effectiveness analysis represents the field's best attempt to systematically and fairly quantify the value of health care interventions, a notion supported by the 2014 ACC-AHA policy statement that uses cost-effectiveness metrics to differentiate high-value care (health gains are worth their costs) from low-value care (prices should be lowered to be in line with the health gains that are produced by the intervention of interest) [12].

Conclusions
The majority of cost-effectiveness evidence evaluating CMRbased diagnostic strategies for patients with suspected CAD identify CMR to be a cost-effective imaging strategy, delivering high-value care in many settings. When comparing diagnostic techniques, the optimal strategy depends on factors that change over time and across clinical settings, such as imaging performance, imaging costs, and disease prevalence. Therefore, decision-makers should contextualize existing literature with additional information, whether from their local data when available or through use of a tool like the costeffectiveness calculator we present here. Such tools can assist in obtaining the most realistic estimates of overall value from options available to diagnose and treat significant CAD.  Sensitivity of CMR n/a n/a 0.0484 1749.6 n/a n/a n/a Specificity of CMR n/a n/a 0.0055 − 2649.1 n/a n/a n/a Sensitivity of CCTA n/a n/a n/a n/a 1797.7 n/a n/a Specificity of CCTA n/a n/a n/a n/a − 2655.3 n/a n/a n/a n/a n/a 0.9987 n/a n/a n/a Cost of CCTA n/a n/a n/a n/a 1.000 n/a n/a Cost of FFR-CT (added to CCTA for some patients) n/a n/a n/a n/a 0.4068 n/a n/a Cost of immediate coronary angiography with FFR n/a n/a n/a 0.0534 0.0982 n/a 0.2967 Cost of immediate coronary angiography without FFR n/a n/a n/a 0.