Cost effectiveness of a gene expression score andmyocardial perfusion imaging for diagnosis ofcoronary artery diseaseCharles E. Phelps, PhD, a Amy K. O’Sullivan, PhD, b Joseph A. Ladapo, MD, PhD, c Milton C. Weinstein, PhD, d
Kevin Leahy, BA, b and Pamela S. Douglas, MD e Rochester, and New York, NY; Medford, and Boston, MA;and Durham, NC
Background Over 3 million patients annually present with symptoms suggestive of obstructive coronary artery disease(oCAD) in the United States (US), but a cardiac etiology is found in as few as 10% of cases. Usual care may include advancedcardiac testing with myocardial perfusion imaging (MPI), with attendant radiation risks and increased costs of care. Weestimated the cost effectiveness of CAD diagnostic strategies including “no test,” a gene expression score (GES) test, MPI, andsequential strategies combining GES and MPI.
Methods We developed a Markov-based decision analysis model to simulate outcomes and costs in patients presentingto clinicians with symptoms suggestive of oCAD in the US. We estimated quality-adjusted life years (QALYs), total costs, andincremental cost-effectiveness ratios (ICERs) for each strategy.
Results In our base case, the 2-threshold GES strategy is the most cost-effective strategy at a threshold of $100,000 perQALY gained, with an ICER of approximately $72,000 per QALY gained relative to no testing. Myocardial perfusion imagingalone and the 1-threshold strategy are weakly dominated. In sensitivity analysis, ICERs fall as the probability of oCAD increasesfrom the base case value of 15%. The ranking of ICERs among strategies is sensitive to test costs, including the time cost fortesting. The analysis reveals ways to improve on prespecified GES thresholds.
Conclusions Diagnostic testing for oCAD with a novel GES strategy in a 2-threshold model is cost effective byconventional standards. This diagnostic approach is more efficient than usual care of MPI alone or a 1-threshold GES strategyin most scenarios. (Am Heart J 2014;167:697-706.e2.)
Approximately 3 million United States (US) patientsvisit a clinician’s office each year for evaluation ofsymptoms suggestive of obstructive coronary arterydisease (oCAD), leading to over $6.7 billion in directmedical costs.1-5 In selecting a diagnostic option toevaluate patients with suspected oCAD, clinicians bal-ance test accuracy with the risks of testing (includingiatrogenic injury and disease from invasive coronaryangiography [ICA]) and ionizing radiation from somenoninvasive tests), the cost of testing, and the value of lost
rom the aUniversity of Rochester, Rochester, NY, bOptum Insight, Medford, MA, cNework University School of Medicine, New York, NY, dCenter for Health Decision Science,arvard School of Public Health, Harvard University, Boston, MA, and eDuke Clinicalesearch Institute, Duke University, Durham, NC.ean van Diepen, MD, served as guest editor for this article.ubmitted August 17, 2013; accepted February 13, 2014.eprint requests: Charles E. Phelps, PhD, 30250 South Highway 1, Gualala, CA 95445.-mail: [email protected]/$ - see front matter2014, TheAuthors. Published by Elsevier Inc. All rights reserved. This is an open access article
nder the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).ttp://dx.doi.org/10.1016/j.ahj.2014.02.005
time for the patient.6 Despite the significant medicalcosts incurred in the work-up of suspected oCAD,typically only 10% to 20% of newly presenting patientsultimately receive a diagnosis of oCAD.7-11 Diagnosticpathways that improve patient selection for referral tononinvasive testing and ICA may improve the quality ofcare and contain health care costs.Common diagnostic pathways involve stress electro-
cardiography, stress myocardial perfusion imaging (MPI),and referral for ICA. Myocardial perfusion imaging is ahigh-volume test in the US, with over 10 millionprocedures annually.4 This study focuses on pathwaysinvolving MPI and a new commercially available geneexpression score (GES) diagnostic test for oCAD (CorusCAD; CardioDx, Inc, Palo Alto, CA). The GES reports thelikelihood of a patient having oCAD on a scale of 1 (low)to 40 (high) and has a negative predictive value of 96% forscores of ≤15, as previously described.12 ObstructiveCAD is defined throughout this article as the patienthaving ≥1 coronary arteries with stenosis of ≥50% asdetermined by quantitative coronary angiography or corelaboratory computed tomographic angiography.
Strategies examined. The various diagnostic strategies considered are outlined. Patients with positive ICA are treated with PCI (baremetal or drug-elutingstent) or CABG and then receive ongoing medical therapy. Patients with negative test results receive no further cardiac testing or treatment.
698 Phelps et alAmerican Heart Journal
May 2014
MethodsModel overviewWe estimated the cost effectiveness of various diagnostic
strategies using MPI, the GES test, and sequential combinationsthereof, using a Markov-based decision analysis model ofdiagnosis, treatment, costs, and outcomes, measured byquality-adjusted life years (QALYs). The model represents apopulation of nondiabetic US adults presenting for diagnosis,generally in a primary care setting, with typical and atypicalsymptoms suggestive of oCAD. Usual care for such patients mayinclude “watchful waiting” or referral to cardiology fordiagnostic testing—typically MPI as the initial test. We excludepatients with a previous history of revascularization ormyocardial infarction (MI).We assume that at anytime, a patient resides in one of a finite
number of discrete Markov health states defined by their CADstatus, whether it has been diagnosed and their MI history.Patients with diagnosed oCAD received revascularization upondiagnosis and receive daily medical therapy thereafter.We used published data on the diagnostic accuracy of MPI and
the GES to stratify patients across the health states based on true-positive, true-negative, false-positive, and false-negative diagnos-tic results depending on the strategy chosen. Each year, patientscan move between states as dictated by transition probabilitiesthat may change over time. We discounted costs (in 2012 USdollars) and QALYs at 3% annually and used a societalperspective in calculating direct medical care and test-relatedtime costs over a lifetime horizon.13
Figure 1 shows the diagnostic strategies and the associatedclinical pathways that we analyzed in addition to “no test.” Welabel these strategies as GES alone,MPI alone, 1-threshold strategy,and 2-threshold strategy. Myocardial perfusion imaging alonerepresents routine care for those receiving any testing. The GESstrategies use score thresholds to determine which patients moveon to MPI and/or ICA. “No test” is the reference strategy againstwhich the others are compared. In current practice, patients suchas those in our model may receive a cardiology referral, at whichpoint MPI and/or ICA may be ordered. In contrast, the bloodsample for the GES test may be drawn in primary care settings andmay remove the need for a cardiology referral and further cardiactesting for patients with low likelihood of oCAD.In every pathway examined, our model specifies that a
positive diagnostic test or test sequence leads to ICA, which, ifpositive, leads to revascularization (coronary artery bypass graftsurgery [CABG] or percutaneous coronary intervention [PCI]).Patients with diagnosed oCAD receive daily medical therapy of astatin, a β blocker, an angiotensin-converting enzyme inhibitor,and aspirin.14,15 A negative diagnostic test or sequence leads tono further cardiac testing.In the Markov model, patients transition across a set of
mutually exclusive health states (see online Appendix Supple-mentary Figure A1). All patients begin in one of the followinghealth states:
(a) No oCAD, off treatment, no history of MI;(b) Obstructive CAD, on treatment, no history of MI; or(c) Obstructive CAD, off treatment, no history of MI.
Table I. Summary of key model parameters
Parameter Value Source
Demographic parametersStarting age 57 12
oCAD prevalence 15% 12
Test characteristicsGES at threshold of 15Sensitivity 89% 12
Specificity 52% 12
GES at threshold of 28Sensitivity 43% 12
Specificity 89% 12
MPISensitivity 81% 16
Specificity 65% 16
Transition probabilitiesAnnual probability of nonfatal MIoCAD, on treatment, no history of MI 2.85% 14
oCAD, off treatment, no history of MI 3.69% 14,17
oCAD, on treatment, history of MI 4.24% 14,18
Annual probability of fatal MIoCAD, on treatment, no history of MI 0.44% 14
oCAD, off treatment, no history of MI 0.57% 14,17
oCAD, on treatment, history of MI 0.65% 14,18
Mortality multipliers for non-MI deathoCAD, on treatment, no history of MI 1.5 18
oCAD, off treatment, no history of MI 1.5 18
oCAD, on treatment, history of MI 3.7 18
Annual probability of receiving diagnosis for persons in oCAD, no history of MI,off-treatment state
10.4% 19
Quality-of-life factorsoCAD, on treatment, no history of MI 0.78 20
oCAD, off treatment, no history of MI 0.77 20
oCAD, on treatment, history of MI 0.65 20
No oCAD, off treatment, no history of MI (age-specific) 0.75-0.92 21
Cost parameters (2012 US $)Diagnostic test costsGES 1050 Medicare rate per CMS letter of determination
dated October 10, 2012MPIProcedure cost 899 4,22-24
Patient time cost 100 25
ICAProcedure cost 5757 5,22,26
Patient time cost 200 25
Treatment and events costsMedical therapy for oCAD (annual supply) 1000 27-30
Revascularization 17169 22,26,31,32
Nonfatal MI 20231 22
Fatal MI 5885 22
Annual discount rate 3% 13
Phelps et al 699American Heart JournalVolume 167, Number 5
State (a) includes patients with a true-negative result on theirindex test and those with a false-positive on their index testfollowed by a negative ICA. State (b) consists of patients who hadICA confirming a true-positive index test result andhence receivingtreatment for oCAD. State (c) consists of patientswith undiagnosedoCAD due to a false-negative result on their index test.Patients in state (a) are assumed to remain in this state until
they die of non-MI causes; incidence of new oCAD is notincluded in the model. Patients in states (b) and (c) are at risk ofnonfatal MI, fatal MI, or death from non-MI causes. Patients in
state (c) have an annual probability of receiving an ICA andgetting diagnosed, at which point they undergo a revasculari-zation and move to state (b).
Model parameterizationTable I lists all model inputs; we describe each type of
data here.Test characteristics. A meta-analysis summarizes the
accuracy of MPI as having sensitivity of 81% and specificity of65%.16 Myocardial perfusion imaging readings are treated as
700 Phelps et alAmerican Heart Journal
May 2014
positive or negative. The GES test characteristics, includingsensitivity and specificity at positivity thresholds of 15 and 28,come from a prospective validation study with full quantitativeangiographic verification with either ICA or computed tomo-graphic angiography.12
Prior probability of oCAD (p(oCAD)). We usedthe p(oCAD) from the GES validation study as our baselinescenario.12 This value is midrange between studies in similarpatient populations in the literature reporting p(oCAD) from 10%to 34.5%, centering around the 10% to 20% range.7-11
Transition probabilities. Probabilities for transitioningbetween health states in the model depend on the underlyingrisk of events and whether the patient has been accuratelydiagnosed with oCAD. Transition probabilities were calculatedfrom published data (see Table I). Annual probabilities of apatient with oCAD on treatment experiencing fatal or nonfatalMI were estimated from the COURAGE trial.14 For patients withundiagnosed oCAD, risk reduction estimates for CABG and PCIwere applied to the event probabilities for patients on treatmentto derive the event probabilities without revascularization.17 ForoCAD patients with a history of MI, the ratio of nonfatal to fatalMIs in the COURAGE trial was applied to the probability ofhaving a second MI to determine the probability of a fatal ornonfatal MI.14,18
Non-MI risk of death among oCAD patients was calculatedusing US life expectancy tables and mortality multipliers toaccount for increased risk of death from oCAD.18,33 Age-specific life table mortality rates were decomposed into age-specific non-Cardio-Vascular Disease (CVD) mortality ratesusing the prevalence of oCAD and the mortality multiplier foroCAD patients. Mortality multipliers for non-MI death amongoCAD patients on treatment, oCAD patients off treatment, andoCAD patients with a history of MI were applied to these non-Cardio-Vascular Disease (CVD) mortality rates to estimate age-specific non-MI mortality rates in the model.Health care outcomes and costs. We simulated the
population over time and calculated discounted costs andQALYs for each diagnostic strategy. We obtained quality of lifeutility weights for each health state from a national catalog ofEQ-5Dindex scores of chronic conditions that was developedusing data from the Medical Expenditure Panel Survey.20 Age-specific utility weights for patients without oCADwere obtainedfrom a separate report on EQ-5Dindex scores.21 Our modelincludes costs for diagnostic test strategies, revascularization,treatment for oCAD, and treatment for an MI. Costs of MPI, GES,ICA, revascularization, fatal MI, and nonfatal MI were estimatedusing the Centers for Medicare and Medicaid Services (CMS)2012 approved fees and 2010 Drug Topics Red Book as well asother available literature.22,23 We assumed that revascularizationincludes PCI and CABG at approximately a 3:1 ratio.31 Theannual cost of medical therapy for oCAD, assumed as 40 mg ofsimvastatin, 100 mg metoprolol tartrate, 10 mg lisinopril, and 75to 325 mg aspirin daily, was estimated from the 2013 Red BookOnline and the available literature.27-30 The Appendix providesdetailed assumptions and sources for these calculations.MPI and ICA time cost. Many cardiac diagnostic tests,
such as MPI and ICA, require significant patient time. Asappropriate when using a societal perspective, we accountfor the cost of lost patient time from diagnostic testing,which contains 2 components—time consumed and thevalue of that time. Consistent with standard practice,13 we
include the time cost only of the diagnostic strategies underconsideration, not time later spent in treatment of diagnosedoCAD. The GES examination requires only a single bloodsample, done as a part of the initial physician visit, so we addno time cost for its use.An MPI generally requires 3 to 4 hours of patient time.
Allowing for patient travel and waiting time, we used 4 hours asthe minimum patient time for an MPI examination. This does notaccount for possible 2-day protocols (as some centers use) or forthe time of companions, drivers, or assistants possibly requiredby patients and hence represents a lower bound on time costs.For a value per hour, we used the average hourly wage asreported by the US Bureau of Labor Statistics, estimated forDecember 2012 at $23.75 per hour.25 Rounding so as not toconvey overly high accuracy, we use $100 per MPI examinationas our base estimate. In sensitivity analysis, we varied the timecost of MPI from $0 to $200.Invasive coronary angiography involves a larger time cost,
typically nearly a full day in the hospital and ICA laboratory,plus travel time. Accordingly, we assumed 8 hours of patienttime for ICA. Most ICA providers urge that patients not driveafter the procedure, hence requiring a companion. We did notinclude companion time in our base case estimate, and thus, itrepresents a lower bound on time cost for ICA. Applyingthe average hourly wage described above, we estimatedICA time cost to be $200. We varied this from $0 to $400 insensitivity analysis.
The authors are solely responsible for the design and conductof this study, all study analyses, the drafting and editing of themanuscript, and its final contents.
ResultsBase caseThe results for the primary study outcomes—base case
costs, QALYs, and incremental cost-effectiveness ratios(ICERs)—for each strategy are shown in Figure 2. Boththe 1-threshold and 2-threshold GES strategies lead tolower costs and fewer QALYs than MPI alone, whereasGES alone leads to both increased costs and QALYscompared with MPI alone. At a cost-effectivenessthreshold of $100,000 per QALY gained, the 2-thresholdstrategy is the most cost-effective strategy, with an ICERof approximately $72,000 per QALY gained relative to“no test.” Both MPI alone and the 1-threshold strategy areweakly dominated by the 2-threshold strategy. GES alonehas an ICER of approximately $152,000 relative to the2-threshold strategy.Table II shows, for each strategy, the percentage of
patients receiving diagnostic testing with MPI and/or ICAand the yield of positive disease diagnosis for patientsreceiving ICA. As compared with MPI alone, both the1-threshold and the 2-threshold strategies resulted infewer patients receiving MPI and ICA and higherdiagnostic yield at ICA. The 1-threshold strategy resultedin the highest diagnostic yield, with 43% of patientsdemonstrating oCAD at ICA.
Figure 2
Baseline results. Per-person costs and QALYs are plotted for each strategy in adults in the US presenting with symptoms suggestive of oCAD. Thelines show the ICER relative to the next best strategy. Prevalence of oCAD in the population is assumed to be 15%. Thresholds for GES strategies are≤15 for low scores and for the 2-threshold strategy, ≥28 for high scores. Costs and QALYs are calculated over the lifetime of the population andare discounted to the present at 3% annually.
⁎Results shown are for the diagnostic testing that occurs in the initial stratification ofpatients for each strategy. Patients who have false-negative test results and laterreceive ICA upon diagnosis are not included in this analysis. In the base case,thresholds for GES are ≤15 for low scores and for the 2-threshold strategy, ≥28 forhigh scores. Probability of oCAD = 15%.
Phelps et al 701American Heart JournalVolume 167, Number 5
Sensitivity to p(oCAD)Results are sensitive to the underlying patient likeli-
hood of having oCAD. As with all diagnostic tests, totalcosts rise, and ICERs fall as the pretest p(oCAD) rises, asseen in Figure 3. At the highest disease prevalence tested,p(oCAD) = 25%, the ICER of the 2-threshold strategy fallsto approximately $54,000, GES alone to approximately$91,000, and MPI alone is no longer weakly dominated,costing approximately $83,000 per QALY gained relativeto the 2-threshold strategy. At p(oCAD) = 20%, ICERsof all efficient frontier strategies fall between thoseof p(oCAD) of 15% and 25%, but MPI alone remainsweakly dominated. At p(oCAD) = 10%, MPI alone and the1-threshold strategy remain weakly dominated, and the
ICERs of the 2-threshold and GES alone strategies rise toapproximately $94,000 and $229,000 respectively.
Sensitivity to test costsMPI and ICA time cost. Our base case included a
$100 time cost for MPI. If we ignore that time cost, MPIcost then simply becomes the Medicare reimbursementrate. In this case, MPI remains weakly dominated atp(oCAD) = 10%, but for the baseline value of p(oCAD) of15%,MPI alone enters the efficient frontier and has an ICERof approximately $140,000 incremental to the 2-thresholdstrategy. As p(oCAD) increases, MPI alone remains on theefficient frontier, and all strategies have increasinglysmaller ICERs. When p(oCAD) reaches 25%, MPI costsapproximately $54,000 per QALY gained and weaklydominates both the 1-threshold and 2-threshold strategies,whereas GES alone costs approximately $101,000 perQALY gained incremental to MPI alone. If MPI time cost isdoubled to $200, the results importantly differ from thebase case only when p(oCAD) = 25%, at which pointMPI alone is weakly dominated, as it was at all lower valuesof p(oCAD) with a time cost of $100.We included a $200 time cost for ICA in our base case.
If ICA time costs are ignored, results are qualitativelyunchanged, but ICERs for all strategies decrease veryslightly. If ICA time cost is doubled to $400, ICERs for allstrategies increase very slightly, but results are qualita-tively unchanged until p(oCAD) reaches 25%. In the basecase, at p(oCAD) = 25%, MPI alone entered the efficientfrontier, but when ICA time cost is increased to $400, MPI
Sensitivity to p(oCAD). Per-person costs and QALYs are plotted for each strategy in adults in the US presenting with symptoms suggestive of oCAD.The lines show the ICER relative to the next best strategy. Prevalence of oCAD in the population is ranged from 10% to 25%. Thresholds for GESstrategies are ≤15 for low scores and for the 2-threshold strategy, ≥28 for high scores. Costs and QALYs are calculated over the lifetime of thepopulation and are discounted to the present at 3% annually.
702 Phelps et alAmerican Heart Journal
May 2014
alone remains weakly dominated even at this level ofoCAD prevalence.GES cost. Holding other values constant, once the cost
of the GES exceeds $1,100, MPI alone enters the efficientfrontier. The 2-threshold strategy costs less per QALYgained than MPI alone until the cost of the GES exceeds$1,150, at which point MPI alone weakly dominates the 2-threshold strategy. At lower costs of GES, the ICER of the2-threshold strategy and GES alone improve relative toMPI alone, although the 1-threshold strategy remainsweakly dominated until the cost of the GES falls to≤$410.
Sensitivity to GES threshold selectionBaseline threshold values for the GES test were
analyzed in the initial assessment of the GES, and thelower threshold (≤15) was validated in a secondvalidation trial.12,34 These values were based on findingsfrom physician market research, in which physiciansindicated they would consider a 20% or lower likelihoodof disease to be low risk of oCAD being the cause of apatient’s symptoms.Another approach identifies optimal thresholds based
on p(oCAD) and costs and benefits of treatment (andavoidance of costs and side effects for nondiseasedpatients).35,36 To understand the consequences ofthreshold selection, we varied the GES cutoff valuesupward and downward, in units of 4 on the scale of 1 to40, from the baseline values (≤15 defined as “low” and≥28 as “high”). The lowest ICERs for strategies
involving the GES occurred using thresholds of ≤19for “low” and ≥32 for “high”. Figure 4 shows the cost-effectiveness results for the various strategies withthese new thresholds.The general ranking of testing strategies remains similar
to the base case, but strategies involving the GES havelower ICERs than when using the base-case cutoffs for allvalues of p(oCAD). The cost and QALY outcomes of GESalone become more similar to those of MPI alone and thatstrategy weakly or fully dominates MPI alone. At p(oCAD) =25%, MPI alone remains weakly dominated, rather thanentering the efficient frontier as in the base case.Incremental cost-effectiveness ratios for the 2-thresholdstrategy improve by approximately 5% to 10% and forGES alone by approximately 35% to 40% across all testedvalues of p(oCAD).Table III, similarly to Table II, shows the percentage of
patients receiving MPI and ICA and positive diseasediagnosis yield from ICA with each testing strategy atbaseline p(oCAD) of 15%. Shifting to the higher GESthresholds reduces the use of both MPI and ICA inboth the 1-threshold and 2-threshold strategies, withthe largest effects being reduction of MPI use in the1-threshold strategy and of ICA in the 2-thresholdstrategy. Use of MPI in the 2-threshold strategy changesonly slightly because the number of GES scores leading toMPI testing remains similar in both analyses. Diagnosticyield at ICA increases for both the 1-threshold and2-threshold strategies as well.
Revised GES thresholds. Per-person costs and QALYs are plotted for each strategy in adults in the US presenting with symptoms suggestive ofoCAD. The lines show the ICER relative to the next best strategy. Prevalence of oCAD in the population is ranged from 10% to 25%. Thresholds forGES strategies are ≤19 for low scores and for the 2-threshold strategy, ≥32 for high scores. Costs and QALYs are calculated over the lifetime of thepopulation and are discounted to the present at 3% annually.
Table III. Diagnostic testing—alternative GES thresholds
⁎Results shown are for the diagnostic testing that occurs in the initial stratification ofpatients for each strategy. Patients who have false-negative test results and laterreceive ICA upon diagnosis are not included in this analysis. In this analysis, thresholdsfor GES are ≤19 for low scores (sensitivity = 84%, specificity = 67%) and for the 2-threshold strategy, ≥32 for high scores (sensitivity = 16%, specificity = 98%).Probability of oCAD = 15%.
Phelps et al 703American Heart JournalVolume 167, Number 5
DiscussionOur analysis demonstrates that using the GES as a first-
line modality for the assessment of suspected oCAD couldoffer a cost-effective alternative to the usual practice ofassessment with MPI. The GES has already been shown tobe feasibly incorporated into clinical practice by primarycare physicians to rule out oCAD in patients with typicaland atypical presentations of chest pain.37,38 As reportedin the COMPASS trial, the GES has been shown to
accurately exclude oCAD as a cause of the patient’ssymptoms in almost 50% of presenting patients, and 6-month follow-up of the study patients showed a majoradverse cardiovascular event rate of 0% among patientswith low GES.12 Our analysis adds the cost-effectivenessmetric to the previously published evidence on the GES.Generally, at a cost-effectiveness threshold of $100,000
per QALY gained, the 2-threshold strategy was themost cost-effective strategy for the assessment of oCADin this study. The 2-threshold strategy improved thediagnostic yield of ICA in comparison with MPI aloneand GES alone and was a more efficient strategy than the1-threshold strategy.In comparison with MPI alone and GES alone, the 2-
threshold strategy resulted in fewer patients going to ICAand higher yield at ICA (Tables II and III). In comparisonwith the 1-threshold strategy, the 2-threshold strategyreplaced the diagnostic capabilities of MPI with those ofhigh GES scores for high-risk patients, thus bypassing MPIto go immediately to ICA. This resulted in more testingthan in the 1-threshold strategy (32% of patients in thebase case receive ICA in the 2-threshold strategy versus25% in the 1-threshold strategy). However, by sendingmore patients to ICA, the 2-threshold strategy resulted infewer false-negatives than did the 1-threshold strategy.The greater number of false-negative test results and theirhealth consequences in the 1-threshold strategy lead to
fewer estimated QALYs. Balancing these differences inoutcomes and cost, the 2-threshold strategy weaklydominates the 1-threshold strategy in most scenarios.As Figures 2 to 4 show, some of these test strategies
(particularly MPI alone and the 2-threshold strategy) havesimilar cost and QALY outcomes. Thus, although MPIalone is often weakly dominated in our analyses, thedifferences between MPI alone and strategies involvingGES are not large and could change with differentbaseline assumptions. Sensitivity analysis revealed theseresults to be sensitive to the underlying patient risk ofoCAD, with all test strategies being less cost effectiveat lower probabilities of oCAD. In addition, the cost ofGES and the time costs associated with MPI also impactthe relative cost effectiveness of each testing strategy.The 2-threshold strategy only remains cost effective incomparison with MPI at GES costs b$1,150.Our calculated ICERs for all of these tests are generally
considered cost effective using current commentary andresearch regarding the proper ICER threshold values forsocietal decisionmaking. In this analysis, we use a thresholdof $100,000 per QALY gained, as this is a commonly usedthreshold.39,40 The World Health Organization suggestsusing 3 times per-capita Gross Domestic Product (GDP) asan upper bound for cost effectiveness.41 The 2012 US per-capita Gross Domestic Product (GDP) of approximately$50,000 equates to a cutoff of $150,000 per QALY gained.One source estimates a threshold upward of $180,000 perQALY gained.42 The 2-threshold strategy meets all of thesethresholds for patient populationswith oCADprevalence of≥10%, and with the improved cutoff criteria (Figure 4), allefficient frontier tests involving the GES have ICER valueswell b$150,000.We compared our results with previous estimates of the
cost effectiveness of diagnostic testing for oCAD. Oneprevious analysis reported ICERs for exercise SinglePhoton Emission Computed Tomography (SPECT, i.e.,MPI) as compared with “no test” of $27,600 for “mildchest pain, typical angina” and $33,300 for “mild chestpain with atypical angina,” values notably lower thanours.17 Two important factors account for this differen-tial. Those results are reported in 1996 US dollars, and thepopulation modeled in that analysis had a 70% prevalenceof CAD, much higher than our base case. Our results arecomparable or below their estimates with adjustments toaccount for these differences. A separate study found thatMPI had ICERs in the range of $60,000 to $130,000incremental to echocardiography in 1999 US dollarsin populations with p(oCAD) of 25% to 75%.43 Consumerprice index adjusted equivalents are in line withour results.Limitations of our study largely arise from exclusion of
certain testing risks from the model and verification biasoccurring in the estimates of MPI accuracy. First, ourmodel did not incorporate the consequences of patientexposure to radiation from MPI. This biases our results,
most strongly in favor of the MPI alone strategy and, to alesser degree, in favor of the 1-threshold strategy and(lesser again) the 2-threshold strategy. Our omission ofrisks associated with ICA creates biases in favor of thosestrategies with higher proportions of patients sent to ICA(see Tables II and III).Because ICA carries inherent risks to patients, almost all
studies of MPI accuracy use only patients already referredfor ICA based on MPI results, thus study patients have ahigh pretest p(oCAD). In general, this verification biaswill lead to reported estimates of MPI accuracy thatoverstate sensitivity and understate specificity.44 Giventhat these biases work in opposite directions, their effecton ICER values and their rank order across our testedstrategies cannot be established a priori.Other limitations include modeling assumptions. We
did not include incidence of new oCAD in the model. Wealso did not incorporate the possibility of retesting witheither MPI or GES. We assumed an abnormal MPI wouldalways lead to a referral to ICA, although other workshows this is not always the case.44 Lastly, we assumedrevascularization rates upon discovery of oCAD at ICAwould be 100%; although they are likely lower; this is astandard assumption in models of oCAD.
ConclusionsTesting strategies involving the GES test for oCAD are
cost effective by usual standards in patients presenting toprimary care settings with typical and atypical symptomssuggestive of oCAD because the GES test improvespatient selection for MPI and ICA. The 2-thresholdstrategy costs approximately $72,000 per QALY gainedin the base case and, at a cost-effectiveness threshold of$100,000 per QALY gained, is the most cost effectiveamong tested strategies in many evaluated scenarios.Strategies involving the GES, particularly the 2-thresholdstrategy, appear to provide new approaches to diagnosisof oCAD that will generally create health outcomes in amore cost-effective manner compared with currentpractice involving referral to MPI or other advancedcardiac testing.
References1. National Ambulatory Medical Care Survey: 2010 Summary Tables.
CDC/NCHS, 2013. (Accessed July 8, 2013, at http://www.cdc.gov/nchs/data/ahcd/namcs_summary/2010_namcs_web_tables.pdf.).
2. Woodwell DA, Cherry DK. National Ambulatory Medical CareSurvey: 2002 summary. Adv Data 2002;2004:1-44.
3. Cayley Jr WE. Diagnosing the cause of chest pain. Am Fam Physician2005;72:2012-21.
4. Nuclear Medicine Market Outlook Report Des Plaines, IL: IMVMedical Information Division, Inc.; 2011.
5. Cardiac Catheterization Lab Market Summary Report. Des Plaines, IL:IMV Medical Information Division, Inc.; 2008.
Phelps et al 705American Heart JournalVolume 167, Number 5
6. Shaw LJ, Hachamovitch R, Berman DS, et al. The economicconsequences of available diagnostic and prognostic strategies forthe evaluation of stable angina patients: an observational assessmentof the value of precatheterization ischemia. Economics ofNoninvasive Diagnosis (END) multicenter study group. J Am CollCardiol 1999;33:661-9.
7. An exploratory report of chest pain in primary care. A report fromASPN. J Am Board Fam Pract 1990;3:143–50.
8. Bosner S, Becker A, Haasenritter J, et al. Chest pain in primary care:epidemiology and pre-work-up probabilities. Eur J Gen Pract 2009;15:141-6.
9. Klinkman MS, Stevens D, Gorenflo DW. Episodes of care for chestpain: a preliminary report from MIRNET. Michigan ResearchNetwork. J Fam Pract 1994;38:345-52.
10. Cheng VY, Berman DS, Rozanski A, et al. Performance of thetraditional age, sex, and angina typicality-based approach forestimating pretest probability of angiographically significantcoronary artery disease in patients undergoing coronary computedtomographic angiography: results from the multinational coronaryCT angiography evaluation for clinical outcomes: aninternational multicenter registry (CONFIRM). Circulation 2011;124(2423–32):1-8.
11. Verdon F, Herzig L, Burnand B, et al. Chest pain in daily practice:occurrence, causes and management. Swiss Med Wkly 2008;138:340-7.
12. Thomas GS, Voros S, McPherson JA, et al. A blood-based geneexpression test for obstructive coronary artery disease tested insymptomatic nondiabetic patients referred for myocardial perfusionimaging the COMPASS study. Circ Cardiovasc Genet 2013;6:154-62.
13. Gold MR, Siegel JE, Russell LB, et al. Cost-effectiveness in health andmedicine. New York: Oxford University Press. 1996.
14. Boden WE, O'Rourke RA, Teo KK, et al. Optimal medical therapywith or without PCI for stable coronary disease. N Engl J Med 2007;356:1503-16.
15. Qaseem A, Fihn SD, Dallas P, et al. Management of stable ischemicheart disease: summary of a clinical practice guideline from theAmerican College of Physicians/American College of CardiologyFoundation/American Heart Association/American Association forThoracic Surgery/Preventive Cardiovascular Nurses Association/Society of Thoracic Surgeons. Ann Intern Med 2012;157:735-43.
16. Mowatt G., Vale L., Brazzelli M., et al. Systematic review of theeffectiveness and cost-effectiveness, and economic evaluation, ofmyocardial perfusion scintigraphy for the diagnosis andmanagement of angina and myocardial infarction. Health TechnolAssess 2004;8:iii-iv, 1–207.
17. Kuntz KM, Fleischmann KE, Hunink MG, et al. Cost-effectiveness ofdiagnostic strategies for patients with chest pain. Ann Intern Med1999;130:709-18.
18. Taylor DC, Pandya A, Thompson D, et al. Cost-effectiveness ofintensive atorvastatin therapy in secondary cardiovascular preventionin the United Kingdom, Spain, and Germany, based on the Treatingto New Targets study. Eur J Health Econ 2009;10:255-65.
19. Metz LD, Beattie M, Hom R, et al. The prognostic value of normalexercise myocardial perfusion imaging and exerciseechocardiography: a meta-analysis. J Am Coll Cardiol 2007;49:227-37.
20. Sullivan PW, Lawrence WF, Ghushchyan V. A national catalog ofpreference-based scores for chronic conditions in the United States.Med Care 2005;43:736-49.
21. Hanmer J, Lawrence WF, Anderson JP, et al. Report of nationallyrepresentative values for the noninstitutionalized US adult population
for 7 health-related quality-of-life scores. Med Decis Making 2006;26:391-400.
22. Medicare Information for Providers, Partners and Health CareProfessionals. Centers for Medicare & Medicaid Services, 2012.(Accessed March 30, 2012, at http://www.cms.gov/home/medicare.asp).
23. Drug Topics Red Book. Montvale, NJ: Thomson Healthcare; 2010.24. Miyamoto MI, Vernotico SL, Majmundar H, et al. Pharmacologic
stress myocardial perfusion imaging: a practical approach. J NuclCardiol 2007;14:250-5.
25. Economic News Release, Table B-3. Bureau of Labor Statistics, 2012.(Accessed March 11, 2013, at http://www.bls.gov/news.release/empsit.t19.htm.).
26. Bottner RK, Blankenship JC, Klein LW, et al. Current usage andattitudes among interventional cardiologists regarding theperformance of percutaneous coronary intervention (PCI)in the outpatient setting. Catheter Cardiovasc Interv 2005;66:455-61.
27. Jick H, Wilson A, Wiggins P, et al. Comparison of prescription drugcosts in the United States and the United Kingdom, part 1: statins.Pharmacotherapy 2012;32:1-6.
28. Heart Protection Study Collaborative Group. Statin cost-effectivenessin the United States for people at different vascular risk levels.Circ Cardiovasc Qual Outcomes 2009;2:65-72.
29. Curtiss FR, Fairman KA. Tough questions about the value of statintherapy for primary prevention: did JUPITER miss the moon? J ManagCare Pharm 2010;16:417-23.
30. Red Book Online. Truven Health Analytics, 2013. (Accessed April 1,2013, at http://www.redbook.com/redbook/online/.).
31. Epstein AJ, Polsky D, Yang F, et al. Coronary revascularizationtrends in the United States, 2001-2008. JAMA 2011;305:1769-76.
32. Birkmeyer JD, Gust C, Baser O, et al. Medicare payments for commoninpatient procedures: implications for episode-based paymentbundling. Health Serv Res 2010;45:1783-95.
33. Arias E. United States life tables, 2006. Natl Vital Stat Rep 2010;58:1-40.
34. Rosenberg S, Elashoff MR, Beineke P, et al. Multicenter validation ofthe diagnostic accuracy of a blood-based gene expression test forassessing obstructive coronary artery disease in nondiabetic patients.Ann Intern Med 2010;153:425-34.
35. Phelps CE, Mushlin AI. Focusing technology assessmentusing medical decision theory. Med Decis Making 1988;8:279-89.
36. Felder S, Mayrhofer T. Medical decision making: a health economicsprimer. Berlin: Springer-Verlag. 2011.
37. Herman L., Froelich J., Kanelos D., et al. Improved diagnostic testingpatterns using a genomic-based personalized medicine test amongpatients presenting to primary care clinicians with symptoms ofsuspected obstructive coronary artery disease: results from theIMPACT-PCP (Investigation of a Molecular Personalized CoronaryGene Expression Test on Primary Care Practice Pattern) trial. J AmBoard Fam Med In Press.
38. Conlin M, Herman L, Mouton M, et al. The use of a personalized geneexpression test to improve decision making in the evaluation ofpatients with suspected coronary artery disease. J Gen Inter Med2012;27:S540-1.
39. Cutler DM, McClellan M. Is technological change in medicine worthit? Health Aff (Millwood) 2001;20:11-29.
40. Ubel PA, Hirth RA, Chernew ME, et al. What is the price of life andwhy doesn’t it increase at the rate of inflation? Arch Intern Med 2003;163:1637-41.
41. Murray CJ, Evans DB, Acharya A, et al. Development of WHOguidelines on generalized cost-effectiveness analysis. Health Econ2000;9:235-51.
42. Braithwaite RS, Meltzer DO, King Jr JT, et al. What doesthe value of modern medicine say about the $50,000 perquality-adjusted life-year decision rule? Med Care 2008;46:349-56.
43. Garber AM, Solomon NA. Cost-effectiveness of alternative teststrategies for the diagnosis of coronary artery disease. Ann InternMed 1999;130:719-28.
44. Ladapo JA, Blecker S, Elashoff MR, et al. Clinical implications ofreferral bias in the diagnostic performance of exercise testing forcoronary artery disease. Journal of the American Heart Association2013;2:e000505.
Phelps et al 706.e1American Heart JournalVolume 167, Number 5
Appendix
The model represents the health states that patientscan occupy for 1-year intervals. During each 1-yearperiod, a patient in any of the obstructive CAD statesis at risk for nonfatal MI, fatal MI, or noncardiacdeath. Patients in an obstructive CAD state may notreturn to the No oCAD state. Patients in the oCAD,Off Tx state have a 10.4% annual probability of being
Supplementary Figure A1
}
Markov Model
diagnosed (involving an ICA and a revascularization)and moving to the oCAD, On Tx state. Patients inan On Tx state are assumed to be receiving dailydrug therapy (a statin, a beta blocker, an ACEinhibitor, and aspirin) in addition to having receivedrevascularization.Note: ICA = invasive coronary angiography; MI =
