Response to Comments: MolDX: Corus® CAD Assay

A large number of comments that were received described a single clinician’s or a single institution’s anecdotal success with the Corus CAD test without providing specific utilization or outcome data or without an indication that these sources of experience were systematically tracking utilization and outcomes of patients who were or were not tested with the Corus CAD test. Commenters included practicing physicians, some of whom are speakers for CardioDX, and advanced practice nurses.

This non-coverage policy is specifically based on the lack of evidence supporting clinical utility in the Medicare population.

While anecdotes of success may be the observable effect of meaningful clinical utility, such cases may also be isolated coincidences. Systematic tracking of patient characteristics, test results, and patient outcomes are the minimum necessary information to ascertain meaningful information regarding the clinical utility of a test.

Numerous providers reported an opinion that the Corus CAD test has good performance characteristics that allow them to confidently use it to stratify a patient’s risk of having coronary artery disease to either reassure the patient or recommend further diagnosis or treatment options. Some of these comments specifically cited research, which is also reviewed in the draft LCD.

The draft LCD is non-covering the Corus CAD test specifically due to a lack of clinical utility of the test in the Medicare population. None of these commenters presented new data regarding clinical utility in the Medicare population or referenced a study demonstrating this. Clinical utility relates to changing the management of a patient in a way that is beneficial to the patient. While the commenters did discuss evidence supporting the diagnostic accuracy of the Corus CAD test, they did not bring to our attention any high quality research suggesting that the Corus CAD test could be used as an alternative to an existing test or that it identified high risk patients who were missed and led to a better long term outcome for stable coronary artery disease.

One physician raised the possibility of this test being used in the emergency room.

The test has a turnaround time of a couple of days at least and therefore does not provide a timely result in patients with possible acute cardiac ischemia.

A representative from CardioDX sent the following comment:

Executive Summary

The Corus CAD blood test incorporates age, sex, and gene expression to calculate a score reflecting a given symptomatic patient’s current likelihood of obstructive coronary artery disease (oCAD). The score ranges from 1-40, with higher scores associated with higher likelihood of obstructive disease. We believe that the Corus CAD test had demonstrated clinical utility and is medically necessary for the following reasons:

• Corus CAD provides an accurate measure of the current likelihood of CAD and has higher accuracy than other commonly used test modalities
• Corus CAD changes the pre-test probability for patients and provides incrementally actionable information that is not available from other sources
• Corus CAD is used by physicians to change diagnostic management, leading to fewer unnecessary procedures
• Corus CAD is associated with lipid lowering agent initiation and adherence
• Corus CAD reduces the number of downstream procedures, which drives improved patient outcomes

As in indication of the growing acceptance of Corus CAD as a valuable tool in evaluation of symptomatic patients, the American Heart Association Scientific Statement on the Expressed Human Genome in July 2017 referred to Corus CAD as “valid and useful.” CardioDx has, under separate cover, submitted comments directly on the draft policy posted by Palmetto. We appreciate that the level of evidence expected today is different than it was when Corus CAD was launched in 2009. We have attempted to continually update our published evidence with 30 peer reviewed published articles and 70+ abstracts and presentations. As an organization, we strive to take a leadership position in the development and commercialization of clinically validated molecular diagnostics. As such, our organization is in the planning phases of a randomized control trial to further support the clinical utility of Corus CAD.

Conclusion: We have demonstrated clinical utility in the targeted patient population by showing that use of Corus CAD can reduce unnecessary cardiac testing without a negative impact on patient safety. We have demonstrated that low score Corus CAD patients are at low likelihood of experiencing a major adverse cardiac event or revascularization in long-term follow-up. The Corus CAD test changes the pre-test probability for patients compared to clinical factors. We believe this data is adequate to maintain Medicare coverage of Corus CAD.

Section 1. Clinical Utility of Corus CAD

• Target population: Patients with symptoms suggestive of obstructive coronary artery disease (oCAD) where the physician is considering wha action to take
• Clinical utility of Corus CAD is driven by:
• Corus CAD provides an accurate measure of the current likelihood o oCAD (data from: PREDICT, COMPASS, PROMISE)
• Corus CAD changes the pre-test probability for patients and provides information that is not available from other sources (PREDICT, COMPASS, PROMISE)
• Physicians use the likelihood from Corus CAD to change their further diagnostic management leading to fewer unnecessary procedures (IMPACT-CARD, IMPACT-PCP, PRESET) with benefit in vulnerabl population (PRESET ELDERLY)
• Corus CAD decreases the use of unnecessary downstream testin (COMPASS, IMPACT-PCP, PROMISE, IMPACT-CARD, PRESET)
• Corus CAD is associated with higher rates of starting lipid lowering agents and adherence on therapy

Outlined below is a description of the published evidence to support the test performance and clinical utility of Corus CAD:

1. Corus CAD provides an accurate measure of the current likelihood of oCAD

In the COMPASS study, we demonstrated that Corus CAD has an AUC of 0.79, sensitivity of 89%, and negative predictive value of 96% when determining the current likelihood of obstructive coronary disease at a threshold of 15. Obstructive CAD was defined prospectively as ≥1 stenosis ≥50% in a major vessel based on a centralized reading quantitative coronary angiography or coronary computed tomography angiography. This level of stenosis was chosen as indication of oCAD in the COMPASS study because this level of stenosis represents a clinical finding in which either coronary artery stents or coronary artery bypass graft surgery would be clinically indicated.1

• The likelihood readout that is provided by Corus CAD outperforms existing methods of evaluating obstructive coronary artery disease likelihood

• Conclusion: Corus CAD provides an accurate measure of the likelihood of oCAD and has higher accuracy than myocardial perfusion imaging (MPI), the most commonly used test modality in the United States.
• Corus CAD changes the pre-test probability for patients and provides information that is not available from other sources
• In the COMPASS study, we calculated the pre-test probability of a patient having coronary artery disease using the clinical factors included in the Diamond-Forrester (D-F) score (Diamond and Forrester , N Engl J Med 1979) We are able to compare the pre-test probability from the D-F score to the Corus CAD probability and measure whether Corus CAD improves the pre-test probability of disease among symptomatic patients suspected of having oCAD.

We used the definition of obstructive coronary artery disease of a ≥50% stenosis in at least one major vessel from a centralized reading. In our previous work, we had seen that local readings of stenosis tended to be higher by approximately 15-20. We have three graphical representations of the improvement in pre-test probability that Corus CAD provides over the Diamond-Forrester. In the first figure, the mean probabilities and 95% confidence intervals are shown as an interval plot for all COMPASS patients and for elderly male COMPASS patients.

The referenced graphic could not be displayed in this report format.

• From the above, in patients without obstructive coronary artery disease, Corus CAD outperforms D-F by providing a lower estimate of disease probability compared to D-F.
• In the next set of figures, we highlight that Corus CAD provides a more accurate probability in the assessment of oCAD compared to D-F. For each patient in the COMPASS study without oCAD, we graphed for which patients Corus CAD was more accurate (lower probability) and for which patients D-F was more accurate and the magnitude of the difference in the pre-test probabilities. In the graphs below, we show that Corus CAD correctly lowered the pre-test probability in 275 patients, while Diamond-Forrester correctly lowered the pre-test probability in just 93 patients. For the elderly male population, Corus CAD correctly adjusted the pre-test probability in almost twice as many patients (17 v 9). In addition to the direction of change, one can see from the data that the Corus CAD test correctly shifted the pre-test probability at a greater magnitude than the shift from the Diamond-Forrester assessment.

The referenced graphics could not be displayed in this report format.

Key to above graphs:

• The horizontal axis represents the individual patients in the analysis. Blue dots are male patients and red dots are female patients. All patients were clear of obstructive coronary artery disease as determined by invasive coronary angiography (ICA) or cardiac computed tomography angiography (CCTA)
• All patients above the horizontal dotted red line are patients that Corus CAD more accurately characterized (lower probability) than Diamond-Forester and all patients below the horizontal line are patients where Diamond-Forester had a more accurate (lower probability) than Corus CAD.
• The vertical distance from the horizontal red zero line represents the difference in probability between the Corus CAD score and the Diamond-Forrester score. The further away from the horizontal line, the greater the difference between the Corus CAD and Diamond-Forester scores.
• The improvement in pre-test probability that Corus CAD provides over D-F can also be seen in the graphs below. In this analysis we show the D-F and Corus CAD probabilities for all COMPASS patients and COMPASS elderly male patients with a given Corus CAD score. One can see that D-F more often overestimates the probability of disease when compared to Corus CAD. Overestimation of disease likelihood is associated with potentially unnecessary cardiac testing with its associated risks and costs. One can also see that there is wide variation in the D-F score for these patients without oCAD.

The referenced graphics could not be displayed in this report format.

Key to above graphs:

• The horizontal axis represents the Corus CAD score. Blue dots are Corus CAD probabilities and red dots are Diamond-Forrester probabilities. All patients were clear of obstructive coronary artery disease as determined by ICA or CCTA.
• The vertical access [sic] represents the probability of obstructive coronary disease

Conclusion: Corus CAD improves the pre-test probability over clinical factors in both the overall population and the elderly male population.

• In the PROMISE study, we examined Corus CAD’s ability to provide information incremental to the standard of care of imaging (MPI or CCTA) in identifying patients likely to suffer a major cardiac event or revascularization. A common metric to show the incremental value of a new testing modality is the Net Reclassification Index (NRI)
• As both a dichotomous score and continuous score, Corus CAD provides significant reclassification benefit when compared to noninvasive imaging testing of MPI and CCTA in PROMISE 

We have not included the Net Reclassification data for elderly patients (65+ years old.), as the data is held by the PROMISE investigators.

• From the above table, one can see that Corus CAD correctly reclassified a meaningful percentage of the patients compared to MPI or CCTA imaging. This is significant because it again demonstrates that Corus CAD provides accurate information that is not available from other sources – in this case standard of care cardiac imaging. Corus CAD was able to provide a net reclassification (correct reclassifications minus incorrect reclassifications) in over 40% of the study participants, when used as either a dichotomous or continuous score.
• Another well-established methodology to determine if a diagnostic strategy is providing independent information is to conduct a multivariate analysis
• In PRESET, using multivariate analysis to isolate the independent effect of the Corus CAD score on the outcome of referral (to cardiology or to advanced cardiac testing), the Corus CAD score was significantly associated with this endpoint, both as a binary score as well as a continuous score. In the table below, patients with scores < 15 have an 82% decreased odds of referral than patients with scores > 15.

Low ASGES Score (>15)0.180.070.20<0.000

Low ASGES Score (>15)0.180.070.070.20

• CardioDx understands Palmetto’s concern over referral rates as a metric of clinical utility. We will address that issue later in this response. Our intention in the analysis above is to show that the Corus CAD score has a larger impact on a physician’s decision to send a patient for further testing and evaluation than any of the typical clinical measures or risk factors.

Conclusion: Corus CAD provides incrementally actionable information that is not available from other sources

• Corus CAD is used by physicians to change diagnostic management, leading to fewer unnecessary procedures
• In the two IMPACT studies, we designed the studies to show change in physician behavior when the Corus CAD score was the one piece of additional clinical information added3. The IMPACT studies represent an additional way to demonstrate that the Corus CAD changed the pre-test probability of oCAD, as measured by change in the intensity of the diagnostic workup pre and post Corus CAD testing.
• In both studies, IMPACT-Card and IMPACT-PCP, the physicians changed behavior for more than 55% of the patients after these physicians had access to the Corus CAD score
• In addition to the level of behavior change, we also measured the direction of that change. We demonstrated in IMPACT-PCP that when the physicians changed their treatment plans, the physicians reduced the intensity of their treatment plan in 60% of the patients with scores ≤15 and increased the intensity in 39% of patients with scores >15. Similar results were seen in the IMPACT-CARD study.

^{Change on testing Intensity: This change was prospectively defined as either a downgrade or upgrade in the intensity of the diagnostic plan based on the following categories (in hierarchical order): (1) no further cardiac testing or treatment, (2) lifestyle changes or medical therapy, (3) stress testing (with or without imaging) or computed tomography/coronary angiography, or (4) invasive coronary angiography. Data are n (%). P <.00001, exact test Herman, et. al IMPACT – PCP, McPherson, et. al IMPACT-CARD}

• Both IMPACT studies asked physicians to document their intended treatment pathway both before and after receiving the Corus CAD score. We recorded those choices and analyzed both how often the pre and post Corus CAD pathways were different, and analyzed whether the change in patient management was more or less intensive.
• CardioDx understands that Palmetto has questioned the study design for both IMPACT studies. (Please see CardioDx’s comments to the draft policy provided under separate cover) Both IMPACT studies were designed to assess whether the use of the Corus CAD altered the clinicians’ decision-making, as defined by a change in patient management before and after receiving the Corus CAD score, while keeping all other diagnostic information constant. Both studies show that physicians change their behavior after receiving the Corus CAD score in over 55% of patients: generally a change of > 20% in an outcome is considered clinically relevant. (Cohen J. et. al. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Hillsdale, NJ: Lawrence Erlbaum Associates; 1988. pp. 19–27.) The patient benefits from the fact that the physicians altered the intensity of their decision to better reflect the risk profile of that patient after learning of the Corus CAD score. In the IMPACT – PCP study, 76 patients (30% of all patients in the study) received less intensive follow-up because of the Corus CAD test. Thus the Corus CAD test decreased unnecessary testing by helping to identify patients unlikely to benefit from further non-invasive or invasive testing.
• In the most recent of our real world registry studies, PRESET, we show a consistent relationship between Corus CAD score and the decision to send the patient for further cardiac testing or evaluation. (Later in this document we discuss the use of referral rates as a proxy for patient outcomes.) Graphed below is the relationship between Corus CAD score and decision to send the patient to the either advanced testing or to cardiology for all patients and for elderly patients in the PRESET registry
• It is important to note that the referral rate increases with Corus CAD score in a continuous fashion, in both the overall and elderly patient population.

The referenced graphic could not be displayed in this report format.

• The continuous relationship between Corus CAD score and referral rates is consistent when we look separately at elderly males and females

The referenced graphic could not be displayed in this report format.

• Furthermore, it is clear that there is a strong correlation between referral rate and increasing probability of obstructive CAD, including patients with higher scores such as elderly males. Patients at low current likelihood of disease as identified by the Corus CAD score were referred to the cardiologist or advanced testing at a low rate, as such patients were ruled out by the primary care physicians as they were unlikely to benefit from further testing.

The referenced graphic could not be displayed in this report format.

Conclusion: Physicians use the likelihood from Corus CAD to change their further diagnostic management. That change is consistent with their revised post-test probability which leads to fewer unnecessary procedures. In the next section, we further demonstrate the reduction in testing from the use of Corus CAD.

• Corus CAD decreases the use of unnecessary downstream testing
• We have shown in two studies that Corus CAD reduces the use of downstream testing
  • In the IMPACT-CARD study, we included a historical control group to determine baseline referral rates
  • We demonstrated that the Corus CAD test can significantly reduce the amount of advanced testing performed, saving 45 patients (69-24) from further testing. Those 45 patients represent more than 50% of the patients in the study.
  • Of the patients that were referred to the cardiac catheterization lab for an invasive coronary angiography, the yield of finding coronary artery disease was 3x higher (79% vs 25%) in the Corus CAD arm than in the historical control arm
• In IMPACT-CARD, when compared to a historical control group, the patients that received Corus CAD had less downstream testing (29% vs 83%)

New analysis not previously seen by Palmetto – Reduction of testing in the COMPASS Study

• In the correspondence with Palmetto, CardioDx recognizes Palmetto’s assumption that clinical usage of Corus CAD was as a dichotomous score, such that all patients with scores ≤ 15 would not be sent for further evaluation given the low risk of oCAD. Similarly, we believe Palmetto’s assumptions was that all patients with scores > 15 would be sent for further testing and/or evaluation given their higher risk profile. We have provided data from multiple studies showing that actual clinical usage does not follow that binary pattern around a specific threshold, but rather clinicians use the test as a continuous score, with the rates of further testing and evaluation rising with higher Corus CAD scores. (See section below for a further discussion of Corus CAD as a continuous score.)
• In appreciation to Palmetto’s original understanding of Corus CAD and in an effort to be responsive to Palmetto’s desire to see additional analysis, we reviewed the COMPASS data, where we had performed either a CCTA or ICA on every evaluable patient. This new analysis allowed us to determine the number of procedures that would be performed on patients if physicians followed the dichotomous pathway assumed by Palmetto (Corus CAD and then advanced testing for all patients with scores >15) and compare it to use of MPI alone, which would be the likely pathway if access to Corus CAD is not available. Since the dichotomous pathway would be the “worst case” for the number of downstream procedures compared to the use as a continuous score, we believe this analysis represents the floor for the clinical utility of Corus CAD in this data set.

<sup>aAssumptions
All patients wi
All patients with Corus CAD ≥ 15 go on to MPI

BThomas et. al. COMPASS
1 = number of patients with positive MPI; 2 = us CAD score; 3= number of
patients with elevated score and positive MPI</sup>

• Based on the analysis above, the clinical strategy of Corus CAD followed by MPI results in a 46% (n=199) reduction in MPI tests and a 29% (n=14) reduction in cardiac catheterization procedures. Given the procedural risks, radiation exposure, and patient burden for both MPI and cardiac catheterization procedures, the use of Corus CAD provides patient outcome benefit when compared to MPI alone.
• To evaluate that the reduction in downstream testing does not negatively impact patient safety, we have also published data that shows low score patients are unlikely to have major adverse cardiac events or cardiac revascularization, at rates equivalent to negative imaging tests

• In the large, NHLBI sponsored PROMISE study, a subset of the patients (N=2,370) had the Corus CAD test drawn for analysis and followed for a minimum of two years
• PROMISE showed that Corus CAD scores was equally predictive of the composite endpoint (death, myocardial infarction, unstable angina, or revascularization) when compared with a negative noninvasive test (functional test or CTA)
• Patient with Corus CAD ≤15 (N=1,058) had an event rate of 3.21% which was no different (P=.29) to the event rate for negative noninvasive tests (N=1,963)
• In summary, the authors of the PROMISE study stated “the Corus CAD is associated with oCAD, can improve risk stratification, and identifies individuals at low risk for cardiac revascularization procedures in the short term”

Conclusion: Corus CAD reduces the number of downstream procedures which drives improved patient outcomes with no negative impact on patient safety

• We have published and presented other examples of clinical utility and the effect of the Corus CAD score on clinical decision making, particularly as the Corus CAD score relates to medication initiation and adherence
  • In the PRESET Lipids study (presented at SGIM 2018 Meeting), the purpose of the study was to examine prescriptions of lipidlowering agents in relation to Corus CAD scores. Prospective exploratory endpoints were collected from the PRESET registry of 566 patients from 21 sites.
• PRESET Lipids – patients with higher Corus CAD scores were more likely to initiate lipid lowering therapy. This finding is important, as higher score patients are more likely to have coronary artery disease and are more likely to benefit from use of this drug class.

^{Pokrywka et al, presented at the 2018 SGIM Meeting}

• In this sub-group analysis of a community-based registry, the Corus CAD test was adopted into clinical practice and associated with a statistically significant and clinically relevant change in the rate of new prescription for lipid-lowering medication.
• PICNICC - patients with higher scores were more likely to start on lipid lowering therapy and were more likely to be adherent with this therapy. This finding is very consistent with the Medicare HEDIS/Five Star program. (Medicare members with a prescription for a cholesterol medication filled that prescription greater than or equal to 80% of the time they are supposed to be taking the medication.)
• In the PICNICC study (Parsons et al, Circulation 2012;126:A17332), the purpose of the study was to explore the relationship between Corus CAD scores and effect on patient utilization and adherence to lipid lowering therapy. In this retrospective study on 767 patients, researchers examined medication adherence by reviewing medication dispensing data. Assessment of medication adherence relied on the proportion of days covered methodology, quantified by the medication possession ratio (MPR) after 365 days of follow-up. We determined the quantity of medication dispensed and the number of days supplied data on each prescription for each patient.

^{P<0.001
Parsons et al, Circulation. 2012;126:A17332}

Conclusion: The analyses above provide other examples of clinical utility. Patients with higher Corus CAD scores were more likely to start lipid-lowering lipid therapy and more likely to continue such therapy, and lipid-lowering therapy has been associated with improved cardiovascular outcomes.

Summary of Section 1: Clinical Utility of Corus CAD

We have demonstrated clinical utility in the overall and elderly population by showing that the use of Corus CAD:

• Provides an improved pre-test probability over clinical factors
• Provides incrementally actionable information that is not available from other sources
• Changes physician clinical decision making
• Is associated with medication adherence
• Decreases the use of downstream cardiac testing

Section 2: Clarifications

In the correspondence with Palmetto and in the draft coverage policy, there were a number of topics that CardioDx believes were not adequately clarified for Palmetto.

Those topics include:

1. Corus CAD as a continuous score
2. Corus CAD as a rule out test

We wanted to include those topics here in order to more directly address Palmetto’s concerns.

• The use of Corus as a dichotomous score versus a continuous score
• The Corus CAD test was always intended to be used as a continuous score. CardioDx has always reported the Corus CAD test as a continuous score. Every patient report form since product launch in 2009 includes the following graphic:

The referenced graphic could not be displayed in this report format.

• We have included the entire Patient Report Form as Exhibit C
• We believe that confusion on the clinical value of Corus CAD may stem from the conflating of two issues 1) the use of Corus CAD as a dichotomous score or a continuous score and 2) whether the measurement of “referral rates” based on the Corus CAD test provides a meaningful metric to assess clinical utility.
  • On the first point of the clinical use of Corus CAD as a dichotomous or continuous score, we have demonstrated with strong published evidence that the test – in the hands of practicing physicians – is consistently and correctly used as a continuous score.
  • The score of 15 is used to perform statistical analysis on the sensitivity, specificity, negative predictive value, and positive predictive value.
  • We recognize that there were multiple references in the 2012 submission to Palmetto about the threshold of 15, including a patient flow diagram excerpted from the pre-publication COMPASS trial data. The references to a threshold of 15 can lead one to think of the test as a dichotomous yes/no result around one and only one threshold. We would note, however, that also included in the 2012 submission were several references (page 18 of the submission) to data that showed the clinicians were not using the test solely as a dichotomous score, but were using it as a continuous score consistent with their understanding of the progression of coronary artery disease and the Corus CAD patient report form.
• As referenced in Exhibit A, a Corus CAD score of 15 represents an 8% probability (in the COMPASS study) of a patient currently having oCAD. That 8% probability is true for patients of all ages and both sexes. Physicians are aware that elderly male patients have a higher risk of oCAD, so they understand that elderly male patients will have higher scores. Given the known higher risk for males patients as well as the fact that not every elderly male has obstructive CAD as the cause of the patients’ symptoms, physicians use the probability associated with the Corus CAD score for a particular patient to determine whether it is appropriate to rule out furthe cardiac testing.
• For reference, we have included in Exhibit A in the Appendix a listing that shows for each Corus CAD score, the probability of having obstructive CAD, the cumulative percentage of all patients receiving a specific score, the cumulative percentage of elderly receiving a specific score, and the cumulative percentage of all elderly male patients receiving a specific score. CardioDx believes this information can helpful in understanding the clinical use of Corus CAD.
• Conclusion: We appreciate that Palmetto’s previous understanding of Corus CAD’s usage may have been primarily informed by our original dossier submission to MolDX in 2012. The evidence published since that date demonstrates that the test has consistently been used as a continuous score. We believe that, as we continue the discussion with Palmetto, it will be useful to view the test as a continuous score, which is consistent with its use in clinical practice and supported physician users of Corus CAD.
  • On the second point, CardioDx understands that Palmetto is not persuaded that referral rates are an adequate measure of improved patient outcomes. In the analysis above (1.a. – 1.d.) we have strived to go beyond referral rates and demonstrate more direct patient outcome benefit, including the actual reduction in unnecessary testing procedures. Given the high correlation between referrals rates and actual downstream testing, CardioDx believes that referral rates are an appropriate proxy for downstream patient outcomes. We acknowledge, however, that Palmetto is interested in seeing “harder” patient outcome endpoints. As an organization, we have committed to conducting a randomized controlled trial to extend our knowledge of Corus CAD and its impact on patient outcomes, including endpoints of unnecessary testing, resources utilization, and patient satisfaction. We are happy to share those plans with Palmetto.
• Conclusion: We acknowledge Palmetto’s concern about the use of referralrates as less than a direct measure of improved patient outcomes. Webelieve that the data presented above on the change in pre-test probabilityand the reduction of downstream testing, coupled with CardioDx’scommitment to conduct a randomized controlled trial for a diagnostic testprovides additional support for the continuation of Medicare coverage.

Clarifying the use of Corus CAD as a “rule out” test

• In an effort to clearly and simply communicate the clinical value of Corus CAD, CardioDx has used the term “rule out” to describe its primary function
• A more precise description of the test’s value is “to help clinicians determine if given the patient’s current risk, they can be ruled out of having oCAD so that further cardiac testing would be unnecessary”
• A less than optimal reading and potentially a mis-interpretation of the published literature is to believe that the test absolutely “rules out” the presence of coronary artery disease in scores below 15 and “rules in” the presence of disease at scores above 15
• In the original MolDX submission in 2012, we noted (page 18) that Corus CAD is used as a continuous score by showing the relationship between Corus CAD score and physicians decision to order advanced testing and/or cardiac evaluation. We have extended that early work by publishing several additional studies (PRESET, IMPACT-PCP) demonstrating the physicians use the test as a continuous score in making clinical management decisions.

• It was our intention in that original submission as well as our supplemental publications to show that the physician’s decision to pursue or not pursue additional testing and evaluation is impacted by the Corus CAD score. That decision by the physician will also include other relevant clinical information about the patient. Ultimately it is the physician who determines if he or she can rule out oCAD as the cause of the patient’s symptoms. CardioDx believes this model is consistent with other tests that provide probabilities (e.g. breast cancer recurrence scores), which also give guidance to the physician but do not dictate clinical decision making.

Conclusion: The Corus CAD helps clinicians determine if given the patient’s current risk, they can be ruled out of having oCAD so that further cardiac testing would be unnecessary.

3. Section 3: Other Information

• We believe that there is additional information that is supportive of continued coverage for Corus CAD by Palmetto:
• American Heart Association Scientific Statement
• The AHA issued their first genome focused Scientific Statement titled “The Expressed Human Genome in Cardiovascular Disease and Stroke: Refinement, Diagnosis, and Prediction” in July 2017
• Only two diagnostic products, Corus CAD and AlloMap, were included in that publication
• Corus CAD was referred to as “valid and useful” in the report
• Comment: We were gratified by the inclusion of Corus CAD in the recent American Heart Association Scientific Statement. While different than the inclusion in guidelines, we believe that the recent AHA Statement validates the quality and quantity of the published data on Corus CAD and constitutes a strong argument for continued Medicare coverage.
• No financial incentive driving usage of Corus CAD
• Unlike stress echo, nuclear imaging, and coronary angiography, in which the physician is financially compensated for performing those tests under a fee-for-service system, there is no financia incentive for physicians to use Corus CAD
• Since the product’s launch in 2009, the Corus CAD test has been ordered over 250,000 times, with many physicians using the tes consistently for multiple years
• Given the challenges with managing patients in today’s complicated health care environment, physicians do not have the time to order tests or complete procedures that do not add to their clinical decision making
• Comment: We believe that physicians consistently use the Corus CAD test because they find it to have clinical utility in their patient populations

•  Providing Additional Information to Elderly Patients
• In our effort to continually add more value to clinicians in their use of our test, we sent a letter to MolDX on April 1, 2015 expressing our desire to add additional information to the Corus CAD Test Report Form. That letter stated that elderly patients receive few scores below 15, and that CardioDx was proposing to add information to the test report form. We believe that providing the percentile information will add a “relative risk” component to the absolute risk provided by the Corus CAD score.
• Comment: We have endeavored to be thoughtful and pro-active about the specific needs of patients in the Medicare population. We are always willing to engage with Palmetto in an effort to provide additional information and to address any questions that may arise.
• In the draft policy, Palmetto inquired as to the number of elderly patients included in the published clinical trials. The chart below highlights the number and percentages of patients over the age of 65 in our referenced studies.

• In addition to our extensive list of previously published studies, CardioDx continues to invest in additional data sets. The company is currently enrolling patients in trials with both the Veterans Administration and the Allegheny Health Network. Both of these trials are designed to look at health care resource utilization and health outcomes in patients that receive the Corus CAD test compared to patients with similar symptoms that do not receive the Corus CAD test.
• Access to quality cardiovascular care
• One of the key benefits of the Corus CAD test is the ease of use in multiple settings.
• The simple blood draw is significantly less complicated, less time consuming, and more patient friendly than any of the competing imaging modalities
• The test can be administered in the same patient visit that the patient describes the symptoms. No additional appointments, fasting, or special preparations are necessary.
• That ease of use makes Corus CAD an ideal solution for practices with any of the following characteristics:
• Within in a rural setting, where access to a cardiologist or advanced testing equipment can require long travel time
• With a large population of elderly patients, where limited mobility makes coming back for additional appointments difficult
• With a population that is at risk for additional radiation exposure and complications from imaging studies
• We have also provided (in a separate document) to Palmetto a marked up version of the draft policy, which responds to questions and concerns raised.

Summary

1. Corus CAD has been extensively studied with 30 publications in peer reviewed journals
2. Corus CAD was noted in the American Heart Association Scientific Statement as “valid and useful”
3. Corus CAD has been shown to provide actionable information that is not available from other sources. Corus CAD drives change in physician behavior
4. Corus CAD changes the pre-test probability of having oCAD
5. Corus CAD decreases the number of advanced cardiac tests performed without a negative impact on patient safety
6. Therefore, we request a change to the draft policy that would maintain coverage for Corus CAD

CardioDx appreciates the time and effort that Palmetto has devoted to reviewing the Corus CAD test.

Conclusion:

CardioDx believes it has demonstrated Corus CAD’s analytical and clinical validity in the intended use population and that the test has clinical utility in directing diagnostic and therapeutic management in patients with suspected oCAD, including Medicare beneficiaries. Therefore, we urge Palmetto/MolDx to retain limited coverage for the test in the intended use population. Given the extensive nature of our comments/submission and previous discussions with the MolDX team regarding the evidence, we request an opportunity to meet to discuss the evidence provided.

The company representative’s comments indicate that the test is intended to be interpreted based on the continuous score. The Corus CAD result was not intended to be interpreted in a dichotomous fashion using a threshold of 15. However, a significant limitation to the use of the Corus CAD test is the lack of clarity from the company or the existing evidence as to how the test result should be interpreted. As the company points out in the comments above, the reported result from the test is a continuous score and has been used that way by clinicians. However, the company also points out that they have selected a threshold score of 15 to perform statistical analysis on the sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV). Notably, the company had to dichotomize the score to do sensitivity, specificity, NPV, and PPV analyses, because such analyses can only be done for a dichotomous outcome. If the test is intended to be used as a continuous outcome, such analyses are meaningless. However, the company makes the information on the sensitivity and specificity for the score threshold of 15 available on the report, and it has included analyses on how the results of the Corus CAD test affected clinical decision making based on such dichotomized test result in its comments above.

In summary, much of the evidence surrounding the diagnostic capability of the test focuses on this dichotomization, and treating the test result as a continuous variable in clinical applications rather than a dichotomous variable means that applicability of studies focusing on dichotomized results is quite limited.

Additionally, the company points out that the test has been studied in individuals who are likely to represent the Medicare beneficiary population, specifically the population who is Medicare eligible on the basis of age. However, whether studies on Corus CAD included subjects who are likely to represent the Medicare population is less relevant to the justification for coverage than the matter of what the Corus CAD test showed in such subjects. The draft LCD is non-covering the Corus CAD test specifically due to a lack of clinical utility, which relates to changing the management of a patient in a way that is beneficial to the patient. As such, for the Corus CAD test to be covered by Medicare, it is important to have evidence indicating not only that the test was studied in a population representative of the Medicare population but also that test changes management in this population in a way that benefits the patient. There remains a lack of sufficient evidence demonstrating this point.

A representative for CardioDX sent the following comment:

The developer of test recently completed data analysis determining if the reduction in referral rates to cardiologists and cardiac procedures we had seen in our previous published studies would be reflected in lower utilization rates in medical claims data in a comparative analysis. The developer also believed that through the analysis of a large, medical claims database, we could provide an extensive comparative effectiveness analysis of the real world, current impact of the Corus CAD test. We were especially interested in comparing the rates of clinical utilization in patients that received the Corus CAD test to propensity-matched patients that did not receive the Corus CAD test. They wanted to share additional information that has recently become available concerning the use of the Corus CAD test and are in the process of completing a manuscript on the study.

The developer also sent in additional information concerning the intended use for the test, physician education and training, and laboratory information system requirements. Information was also provided concerning the requirements on test requisition form.

They wish to see continued coverage for the Corus CAD test and not the proposed non-coverage as listed in the draft policy based on the following criteria listed in the comments received.

Corus CAD has been extensively studied with 30 publications in peer reviewed journals

• Including being highlighted in a recent American Heart Association Scientific Statement as “valid and useful”
• CardioDx limits use of Corus CAD to symptomatic patients compared with the pre-test probability based upon clinical factors
• Corus CAD score leads to physician changes in obstructive CAD probability assessment
• Corus CAD has been shown to provide actionable information that is not available from other sources
• NEW DATA - Corus CAD decreases the number of advanced cardiac procedures performed compared to controls in the Medicare population

We look forward to receiving any additional data that could potentially show how the lives and health of Medicare beneficiaries can be improved, and we will review all such data.

The full text of a manuscript for a study evaluating utilization of healthcare services following the Corus CAD test is provided below. This was sponsored by the test developer. It has not been peer-reviewed to our knowledge.

Corus CAD

Medical Claims Data Analysis Now Available

Introduction

CardioDx had previously submitted comments on the draft LCD for Corus CAD to Palmetto on May 4, 2018. Since that time, the company has been able to analyze additional data and is including these data in this public comment to CGS.

Several months ago, CardioDx undertook a project to analyze medical claims data to help identify and quantify the clinical impact of the Corus CAD test. The company was interested in determining if the reduction in referral rates to cardiologists and cardiac procedures we had seen in our previous published studies would be reflected in lower utilization rates in medical claims data in a comparative analysis. The company also believed that through the analysis of a large, medical claims database, we could provide an extensive comparative effectiveness analysis of the real world, current impact of the Corus CAD test. We were especially interested in comparing the rates of clinical utilization in patients that received the Corus CAD test to propensity-matched patients that did not receive the Corus CAD test.

CardioDx collaborated with Baker Tilly Virchow Kraus, LLP on this analysis. Baker Tilly provided access to data from Optum, which contains primarily commercial and Medicare Advantage data from United Healthcare. Baker Tilly was also responsible for the data analytics work from a statistical analysis plan that was jointly developed between CardioDx and Baker Tilly, with input from several industry experts.

While we initiated this study several months ago, the analysis was not available until mid-June 2018. Therefore these data were not able to be included in our submission during the public comment period for Palmetto, but we are pleased to provide these data during the public comment period for CGS. Baker Tilly is currently drafting a manuscript that will include results from the study.

Methodology

Baker Tilly examined the medical claims available from Optum for the period January 1, 2015 through October 31, 2017. In those data, they identified 1,431 patients with claims for the Corus CAD test between July 1, 2015 and April 30, 2017. Baker Tilly also identified 127,268 control patients that both a) met the inclusion and exclusion criteria for the study and b) had at least one claim for one of the four functional / anatomical tests. Those four functional / anatomical tests are:

1. Nuclear perfusion imaging or MPI
2. Stress ECG
3. Stress echocardiogram
4. Coronary Computer Tomography Angiography (CCTA)

The inclusion and exclusion criteria are noted below:

Inclusion criteria:

• Typical or atypical symptom with risk factor consistent with potential coronary artery disease (intended use criteria for Corus CAD)
• Received Corus CAD (index event for the Corus CAD arm) between July 2015 and April 2017, inclusive OR Received an MPI, stress echo, stress ECG, or coronary CT Angiography (index event for the control arm) between July 2015 and April 2017, inclusive
• Continuously enrolled in the health plan for 6 months prior and 6 months post the index event
• Age 21 or older at the data of the index event

Exclusion criteria:

• History of Myocardial Infarction (MI)
• Current MI or acute coronary syndrome
• Any previous coronary artery revascularization
• Diabetes mellitus
• Suspected unstable angina
• Systemic infection
• Systemic inflammatory or auto-immune conditions
• Oncological conditions
• Any major surgery within four months

Baker Tilly then performed an outlier analysis that removed 0.06% of the patients (n=71) based on the effect of extreme observations on overall data in an exponential (Gamma) distribution.

Once Baker Tilly determined the available patients that met the inclusion and exclusion criteria, they completed a propensity matching process that equalized the following criteria between the two groups in a 4:1 ratio of controls to cases:

• Gender
• Age
• Geographic region
• Typical symptom or atypical symptom with risk factor
• Minority population of patient’s zip code
• Comorbidities (obesity, hypertension, smoking, dyslipidemia)
• Insurance type (commercial, Medicare Advantage)

For the analysis we included:

• 4,079 patients with commercial insurance (2,125 females, 1,954 males)
  • 794 patients in the Corus CAD arm (412 females, 382 males)
  • 3,285 patients in the control arm (1,713 females, 1,572 males)

• 1,421 patients with Medicare Advantage insurance (892 females, 529 males)
  • 306 patients in the Corus CAD arm (194 females, 112 males)
  • 1,115 patients in the control arm (698 females, 417 males)

The endpoints that were analyzed included:

• Comparison of the two arms at 60 and 180 days
  • Frequency of invasive coronary angiography or cardiac catheterization
  • Frequency of coronary artery revascularization and CABG (coronary artery bypass graft)
  • Frequency of functional and anatomical tests
  • Major adverse cardiac events (MACE), including myocardial infarction, hospitalization due to unstable angina, or transient ischemic attack/stroke
  • Frequency of other coronary artery disease (CAD) related resource utilization (office visits, emergency department (ED) visits and inpatient admissions)

The 60-day endpoint allowed for measurement of the episode of care around the evaluation of the symptomatic patient. The 180-day endpoint allowed for confirmation of the shorter term finding as well as validation of the robustness of such findings at a longer time interval. We are conducting an analysis looking at the effect of the Corus CAD score in 5-point increments: these results will be reported separately in the future.

Results

Results Summary

The analysis of the large medical claims database found that patients who received the Corus CAD test had less cardiac related medical care after their index event than patients in the control arm. Specifically, the Corus CAD patients had fewer tests and procedures as compared to usual care control patients, specifically:

• Corus CAD care patients had fewer functional and anatomical tests
• Corus CAD care patients had fewer invasive coronary angiographies
• Corus CAD care patients had fewer coronary artery revascularizations
• Corus CAD care patients had fewer office, emergency, and inpatient admissions

The reduction in downstream tests and procedures had no negative impact on patient outcomes, as Corus CAD care patients had lower or similar MACE rates.

We looked separately at the results for patients with commercial insurance and Medicare Advantage and found similar results:

• In the commercial insurance population
  • Corus CAD patients had fewer functional and anatomical tests than usual care control patients
  • Corus CAD patients had fewer invasive coronary angiographies than control patients
  • Corus CAD patients had fewer cardiac related encounters (office visits, ED visits, in patient admission) than control patients

• In the Medicare population
  • Corus CAD patients had fewer invasive coronary angiographies than control patients
  • Corus CAD patients had fewer revascularizations than control patients
  • Corus CAD patients had fewer cardiac related encounters (office visits, ED visits, in patient admission) than control patients
  • Male patients that did not receive the Corus CAD are 3x more likely to be sent for an invasive coronary angiography procedure at 180 days
  • Male patients that did not receive the Corus CAD are 4x more likely to be sent for an invasive coronary angiography procedure at 180 days
  • Medicare patients had a higher yield (percentage of patients who received revascularization procedures after an invasive coronary angiography) in the cardiac catheterization lab than control patients at 60 days
  • Corus CAD patients had lower rates of MACE than control patients

Results for All Patients

When we evaluated the endpoints at 60 DAYS post the index event for ALL patients in the study, we found the following results:

Functional and Anatomical Tests¹

• For the Corus CAD patients, 16.4% (n=180) received a test
• For the control patients, 27.9% (n=1,229) received a test
• The rate of functional and anatomical testing in the Corus CAD population was significantly lower than the control arm (p<0.0001)

Invasive Coronary Angiography (ICA)

• For the Corus CAD patients, 1.3% (n=14) underwent an ICA procedure
• For the control patients, 5.6% (n=246) underwent an ICA procedure
• The rate of ICA procedures in the Corus CAD population was significantly lower than the control arm (p<0.0001)

Coronary Artery Revascularization and CABG

• For the Corus CAD patients, 0.5% (n=6) underwent a coronary artery revascularization procedure
• For the control patients, 1.8% (n=80) underwent a coronary artery revascularization procedure
• The rate of coronary artery revascularization in the Corus CAD population was significantly lower than the control arm (p=0.0005)

CAD Related Utilization²

• For the Corus CAD patients, 1.7% (n=19) had at least one CAD related utilization event
• For the control patients, 4.0% (n=176) had at least one CAD related utilization event
• The rate of CAD related utilization in the Corus CAD population was significantly lower than the control arm (p<0.0001)

When we evaluated the endpoints at 180 DAYS post the index event for ALL PATIENTS in the study, we found the following results:

Functional and Anatomical Tests

• For the Corus CAD patients, 21.6% (n=238) received a test
• For the control patients, 30.9% (n=1,359) received a test

¹Includes myocardial perfusion imaging (MPI), stress ECG, stress echo, or coronary CTA
²Includes office visits, emergency room visits, and in patient admissions

• The rate of testing in the Corus CAD population was significantly lower than the control arm (p<0.0001)

Invasive Coronary Angiography

• For the Corus CAD patients, 2.6% (n=29) underwent an ICA procedure
• For the control patients, 6.5% (n=286) underwent an ICA procedure
• The rate of ICA procedures in the Corus CAD population was significantly lower than the control arm (p<0.0001)

Coronary Artery Revascularization and CABG

• For the Corus CAD patients, 0.8% (n=9) underwent a coronary artery revascularization procedure
• For the control patients, 2.1% (n=92) underwent a coronary artery revascularization procedure
• The rate of coronary artery revascularization in the Corus CAD population was significantly lower than the control arm (p=0.0017)

CAD Related Utilization

• For the Corus CAD patients, 2.6% (n=29) had at least one CAD related utilization event
• For the control patients, 5.0% (n=221) had at least one CAD related utilization event
• The rate of CAD related utilization in the Corus CAD population was significantly lower than the control arm (p=0.0002)

Summary of Analysis for All Patients:

In the overall population, patients that received the Corus CAD test received fewer advanced cardiac tests, had fewer invasive coronary angiographies, had fewer revascularization procedures, and had fewer CAD related utilizations than patients that did not receive the Corus CAD test.

Results for Medicare Advantage Patients

When we evaluated the endpoints at 60 DAYS post the index event for Medicare Advantage patients (n=1,421) in the study, we found the following results:

Functional and Anatomical Tests

• For the Corus CAD patients, 15.7% (n=48) received a test
• For the control patients, 14.2% (n=158) received a test
• The rate of testing in the Corus CAD population was not significantly higher than the control arm (p=0.4520)[1]
• Note the functional and anatomical tests for this endpoint represent the second such tests received by these patients, as the first such test was the index event

Invasive Coronary Angiography

• For the Corus CAD patients, 0.7% (n=2) underwent an ICA procedure
• For the control patients, 7.1% (n=79) underwent an ICA procedure
• The rate of ICA procedures in the Corus CAD population was significantly lower than the control arm (p<0.0001)

Coronary Artery Revascularization and CABG

• For the Corus CAD patients, 0.3% (n=1) underwent a coronary artery revascularization procedure
• For the control patients, 2.5% (n=28) underwent a coronary artery revascularization procedure
• The rate of coronary artery revascularization in the Corus CAD population was significantly lower than the control arm (p=0.0053)

CAD Related Utilization

• For the Corus CAD patients, 2.0% (n=6) had at least one CAD related utilization event
• For the control patients, 5.7% (n=63) had at least one CAD related utilization event
• The rate of CAD related utilization in the Corus CAD population was significantly lower than the control arm (p=0.0054)

When we evaluated the endpoints at 180 DAYS post the index event for Medicare Advantage patients (n=1,421) in the study, we found the following results:

Functional and Anatomical Tests

• For the Corus CAD patients, 21.6% (n=66) received a test

³In the control group with Medicare Advantage coverage, a larger percentage of the index events was an MPI test (69%) compared to the control patients with commercial insurance (45%). When the index event is an MPI test, the physician may be less likely to order a functional and anatomical test (stress echo, stress ECG, coronary CTA) that he or she may consider less informative than an MPI, explaining the difference between the Medicare Advantage and commercial patients.

• For the control patients, 19.2% (n=214) received a test
• The rate of testing in the Corus CAD population was slightly higher than the control patients, but the difference was not statistically significant (p=0.2851)

Invasive Coronary Angiography

• For the Corus CAD patients, 2.6% (n=8) underwent an ICA procedure
• For the control patients, 8.6% (n=96) underwent an ICA procedure
• The rate of ICA procedures in the Corus CAD population was significantly lower than the control arm (p<0.0001)

Coronary Artery Revascularization and CABG

• For the Corus CAD patients, 0.3% (n=1) underwent a coronary artery revascularization procedure
• For the control patients, 2.8% (n=31) underwent a coronary artery revascularization procedure
• The rate of coronary artery revascularization in the Corus CAD population was significantly lower than the control arm (p=0.0029)

CAD Related Utilization

• For the Corus CAD patients, 3.6% (n=11) had at least one CAD related utilization event
• For the control patients, 7.4% (n=82) received at least one CAD related utilization event
• The rate of CAD related utilization in the Corus CAD population was significantly lower than the control arm (p=0.0257)

MACE

• For the Corus CAD patients, 0.0% (n=0) had a MACE event within 180 days
• For the control patients, 0.6% (n=7) had a MACE event within 180 days
• The rate of MACE events in the Corus CAD population was lower than the control arm but the difference was not statistically significant (p=0.3573)

Summary of Analysis for Medicare Advantage Patients:

In the Medicare Advantage population, patients that received the Corus CAD test had fewer invasive coronary angiographies, had fewer revascularization procedures, and lower rates of CAD related utilization than patients that did not receive the Corus CAD test.

In the Medicare Advantage population, patients that received the Corus CAD test had no statistical difference of MACE events than those who did not receive the Corus CAD test. Therefore, the reduction of tests and procedures seen in patients that receive the Corus CAD test does not increase the patient’s risk, as seen in the rate of MACE.

Results for Medicare Advantage Male Patients

When we evaluated the endpoints at 60 DAYS post the index event for Medicare Advantage male patients (n=529) in the study, we found the following results:

Functional and Anatomical Tests

• For the Corus CAD patients, 16.1% (n=18) received a test
• For the control patients, 13.2% (n=55) received a test
• The rate of testing in the Corus CAD population was slightly higher than the control patients but the difference was not statistically significant (p=0.4196)
• Note that the functional and anatomical tests for this endpoint represent the second functional and anatomical tests received by these patients, as the first such test was the index event

Invasive Coronary Angiography

• For the Corus CAD patients, 1.8% (n=2) underwent an ICA procedure
• For the control patients, 9.8% (n=41) underwent an ICA procedure
• The rate of ICA procedures in the Corus CAD population was significantly lower than the control arm (p=0.0014)

Coronary Artery Revascularization and CABG

• For the Corus CAD patients, 0.9% (n=1) underwent a coronary artery revascularization procedure
• For the control patients, 3.6% (n=15) underwent a coronary artery revascularization procedure
• The rate of coronary artery revascularization in the Corus CAD population was lower than the control arm, but the difference was not statistically significant (p=0.1056)

When we evaluated the endpoints at 180 DAYS post the index event for Medicare Advantage male patients (n=529) in the study, we found the following results.

Functional and Anatomical Tests

• For the Corus CAD patients, 24.1% (n=27) received a test
• For the control patients, 19.2% (n=80) received a test
• The rate of testing in the Corus CAD population was higher than the control patients. but the difference was not statistically significant (p=0.2027)
• The functional and anatomical tests for this endpoint represent the second such tests received by these patients, as the first such test was the index event

Invasive Coronary Angiography

• For the Corus CAD patients, 3.6% (n=4) underwent an ICA procedure
• For the control patients, 12.2% (n=51) underwent an ICA procedure
• The rate of ICA procedures in the Corus CAD population was significantly lower than the control arm (p=0.0015)

Coronary Artery Revascularization and CABG

• For the Corus CAD patients, 0.9% (n=1) underwent a coronary artery revascularization procedure
• For the control patients, 3.8% (n=16) underwent a coronary artery revascularization procedure
• The rate of coronary artery revascularization in the Corus CAD population was lower than the control arm, but the difference was not statistically significant (p=0.0848)

CAD Related Utilization

• For the Corus CAD patients, 5.4% (n=6) had at least one CAD related utilization event
• For the control patients, 10.8% (n=45) had at least one CAD utilization related event
• The rate of CAD related utilization in the Corus CAD population was lower than the control arm, but the difference was not statistically significant (p=0.0935)

Summary of Analysis for Medicare Advantage Male Patients:

In the Medicare Advantage male population, patients that received the Corus CAD test were sent to the cardiac catheterization lab less than 1/3 as often as patients that did not have the Corus CAD test. Male Corus CAD patients also were sent to other CAD related visits less than 1/2 as frequently as non-Corus CAD patients, although that difference was non-significant given the small sample size.

Conclusions

Conclusions for the Medicare Advantage Population

In a large, claims analysis reflecting actual clinical care utilization and outcomes, Medicare Advantage patients that received the Corus CAD test had reduced invasive procedures

1. Patients that received the Corus CAD test had fewer invasive coronary angiography procedures
2. Patients that received the Corus CAD test had fewer coronary artery revascularization procedures
3. Patients that received the Corus CAD test had fewer CAD related office visits, ED visits, and inpatient admissions
4. The rate of testing and procedures in the Corus CAD arm did not negatively impact patient health, as seen in MACE rates

Corus CAD provides clinical utility and patient benefit by showing reduced rates of invasive procedures and CAD related encounters for the Medicare Advantage populati

Conclusions for Medicare Advantage Males

In a large, claims analysis reflecting actual clinical care utilization and outcomes, male Medicare Advantage patients that received the Corus CAD test had reduced invasive coronary procedures

1. Patients that received the Corus CAD test had fewer invasive coronary angiography procedures
2. Patients that received the Corus CAD test had fewer CAD related office visits, ED visits, and inpatient admissions
   • Males that received the Corus CAD test had 50% fewer cardiac related office visits, ED visits, and hospital admissions, but the difference was not statistically significant given the sample size
3. The reduction in tests and procedures in the Corus CAD arm did not negatively impact patient health, as seen in MACE rates

Corus CAD provides clinical utility and patient benefit by showing reduced rates of invasive procedures and CAD related encounters for the Medicare Advantage male population

Limitations of the Analysis

As with any data analysis, there are limitations of this approach. Those limitations include:

• Billing and diagnosis codes submitted on medical claims may not accurately reflect actual diagnoses
  • Comment: The ability to study large number of patients with verified follow-up as well as the inclusion of large, demographically diverse, and multicenter cohorts represent advantages of this claims-based approach over other study methodologies and may thus mitigate any inaccuracies in the coding information. Furthermore, since it is likely that any coding inaccuracies occurred among both the Corus CAD and control patients, these accuracies would be more likely to lead to a finding or no difference between the groups.
• Causation is difficult to show in all retrospective analysis
  • Comment: There was a strong association noted, with robust findings across different populations and scenarios.
• The Corus CAD cohort may be different than the control cohort
• Comment: To address this concern, patients in the groups were selected from a large claims based database and were well matched on several demographic and clinical factors. Propensity matching allows one to design and analyze an observational (nonrandomized) study in order to mimic some of the particular characteristics of a randomized controlled trial. In this manner, the distribution of observed baseline covariates will be similar between treated and untreated subjects. (Austin PC. An Introduction to Propensity Score Methods for Reducing the Effects of Confounding in Observational Studies. Multivariate Behavioral Research. 2011;46(3):399-424. doi:10.1080/00273171.2011.568786)
• The follow up period of 60 and 180 days may not be long enough
  • Comment: An episode of care is variable, depending on multiple factors, including but not limited to the patient’s symptoms, access to resources, previous tests, and procedures. The results at 60 and 180 days were very similar, with greater than 75% of the clinical care utilization occurring within the first 60 days. We believe that time frame of 60 and 180 days is appropriate for showing the impact of Corus CAD testing on clinician decision-making.

Intended Use of Corus CAD

Palmetto asked about the intended use of the Corus CAD test. We are pleased to provide a description of the intended use and the procedures the CardioDx uses to safeguard that patients fit the intended use criteria.

Intended Use of Corus CAD

• The Medicare Local Coverage Determination (LCD) established in 2012 reads as follows:
  • "The CORUS^® CAD test will be considered reasonable and necessary for patients with stable symptoms that have a history of chest pain, suspected anginal equivalent to chest pain, or a high risk of CAD, but no known prior myocardial infarction or revascularization procedures"
• CardioDx reinforces the intended use in multiple ways, including
  • Physician education and training
  • Lab information system requirements
  • Physician verification
• CardioDx lab rejects samples that are not within intended use
• Based on an inquiry from Palmetto, the next several pages will focus on how CardioDx ensures that patients have appropriate symptoms

Physician Education and Training

• CardioDx consistently educates and trains physicians on the intended use of Corus CAD using multiple approaches, such as
• Our marketing material
• Our sales representative training
• Regional Medical Director interactions
• Our website

Laboratory Information System Requirements

• Corus CAD test requisition form requires listing of symptoms
  • Typical or atypical with a risk factor
• Laboratory Information Management System (LIMS) will not accept a sample for processing that does not have a typical or atypical/risk factor marked on the test requisition form
• The vast majority of test requisition designate that the patient has the appropriate symptoms
  • A small number (0.89%) of TRFs were rejected because the patient was asymptomatic (June 2017- June 2018)

Test Requisition Form (TRF)

• Our Test Requisition Form requires an indication that each patient has either 1) typical symptom(s) or 2) or atypical symptom(s) and cardiac risk factor(s)
• CardioDx personnel review the TRF when the sample arrives in the lab to ensure that the TRF has contains the required information
• Physician signatures are required on all TRF forms
• TRF also includes a highlighted box listing the patient characteristics that would not be appropriate for Corus CAD testing

Analysis of Symptoms for Corus CAD Patients

• Historically the mix of symptoms has been approximately 50/50 between typical and atypical
  • In April 2018, 45% of all samples received had typical symptoms
  • In April 2018, 42% of Medicare samples received had typical symptoms
• The most common symptoms in the Medicare population are
  • Typical – chest pain, shortness of breath
  • Atypical – malaise, heartburn
  • Patients with atypical symptoms are also required to also have a cardiac risk factor
• Multiple symptoms in the overall population
  • Typical symptom – 60% of patients have more than 1 typical symptom
  • Atypical symptom – 34% of the patients have more than 1 atypical symptom / risk factor

Physician Signature on the Test Requisition Form

• Every Test Requisition Form requires a signature to confirm that the test ordering meets intended use and follows exclusion criteria
  • If a requisition form arrives without a signature, we contact the physician office and attempt to receive that signature
  • Approximately 2% of all samples arrived without signatures in Q1 2018
  • If we are not able to acquire the signature, the sample result is not reported to the physician
• To reinforce the physicians understanding of the Corus CAD intended use, we had physicians acknowledge and sign a separate document listing the intended use
  • We collect > 650 completed acknowledgements May – August 2016

Summary of Intended Use Information

CardioDx strives to educate and inform the physician on the intended use of Corus CAD and diligently monitors incoming samples to insure they are appropriate for testing

Concluding Remarks

• Corus CAD has been extensively studied with 30 publications in peer reviewed journals
• Including being highlighted in a recent American Heart Association Scientific Statement as "valid and useful"
• CardioDx limits use of Corus CAD to symptomatic patients compared with the pre-test probability based upon clinical factors
• Corus CAD score leads to physician changes in obstructive CAD probability assessment
• Corus CAD has been shown to provide actionable information that is not available from other sources
• NEW DATA - Corus CAD decreases the number of advanced cardiac procedures performed compared to controls in the Medicare population

Corus CAD had demonstrated clinical utility and provides patient benefit; therefore, we recommend continued coverage as a Medicare benefit

The test sponsor provided us with an unpublished manuscript that has not been peer-reviewed from claims data as a comment to the draft LCD. The sponsor was made aware prior to submission that such information when provided as a comment to the draft LCD may be publicly shared as necessary to allow for a public response to comments as is required of Palmetto GBA and MolDX.

In summary, this was a retrospective observational study that examined health care service utilization in two cohorts using claims data from a large data set including commercial insurance beneficiaries and beneficiaries enrolled in Medicare Advantage plans. One cohort included patients in whom the CardioDX study was used. A control group was created using propensity-matching to obtain a 4:1 ratio of controls to CardioDX tested subjects in the study. Patients excluded from the study included those with diabetes and known obstructive coronary disease. The dependent variables examined in the study were utilization of healthcare services including diagnostic cardiac imagining, diagnostic catheterization, and cardiac revascularization interventions. Patient outcomes for survival, functional status, or quality of life were not reported.

Summary of the methods:

Claims data from commercial and Medicare advantage plans was used including claims submitted between January 1, 2015 and October 31, 2017. Patients were considered for inclusion if they were enrolled in a plan submitting data to the data set for at least 6 months prior to and 6 months following an index event, which is further described below. The group of patients tested using the Corus CAD test was identified by the presence of a claim for this test in the beneficiary, and the use of the Corus CAD test was the index event for this group. A potential set of patients from whom the control group would ultimately be selected was identified if they did not have the Corus CAD test and had at least one of the following cardiac diagnostic tests: 1. Nuclear perfusion imaging or MPI 2. Stress ECG 3. Stress Echocardiogram 4. Coronary Computer Tomography Angiography (CCTA).

In addition to having a claim for an index event in the specified time window, specific inclusion criteria for the study was as follows: Typical or atypical symptom with risk factor consistent with potential coronary artery disease (intended use criteria for Corus CAD) and age 21 or older at the date of the index event. Exclusion criteria were as follows: History of Myocardial Infarction (MI), current MI or acute coronary syndrome, any previous coronary artery revascularization, diabetes mellitus, suspected unstable angina, systemic infection, systemic inflammatory or auto-immune conditions, oncological conditions, any major surgery within four months. The investigators identified 1,431 patients with claims for the Corus CAD test between July 1, 2015 and April 30, 2017, and they identified 127,268 control patients with a claim for one of the 4 tests in this time window. From the 127,268 control patients, a control group was created by propensity matching control patients to patients who received Corus CAD using the following matching variables: gender, age, geographic region, typical symptom or atypical symptom with risk factor, minority population of patient’s zip code, comorbidities (obesity, hypertension, smoking, dyslipidemia), insurance type (commercial, Medicare Advantage).

A final set of 2 cohorts of patients were selected for the study: 1,100 patients in the Corus CAD arm and 4,400 patients control arm.

The study sought to compare care utilization and outcomes in these two cohorts of patients at 60 and 180 days. Specific endpoints that were examined were:

• Frequency of invasive coronary angiography or cardiac catheterization
• Frequency of coronary artery revascularization and CABG (coronary artery bypass graft)
• Frequency of functional and anatomical tests
• Major adverse cardiac events (MACE), including myocardial infarction, hospitalization due to unstable angina, or transient ischemic attack/stroke
• Frequency of other coronary artery disease (CAD) related resource utilization (office visits, emergency department (ED) visits and inpatient admissions)

Methodological details for how the typical or atypical symptoms of a patient were identified in claims data are not explained in the manuscript. Additionally, the specific details of how specific diagnoses and comorbidities were abstracted from the claims data (e.g. comorbidity classification method) were not explained in the manuscript.

Brief response by MolDX to the methods:

This study design, has the ability to show suggestive results, but not results that would establish clinical utility of the Corus CAD. This is not only because it is a retrospective observational study in which causation cannot be asserted, but also because of both the shallow nature of claims-based research and because of the relatively short window for observation of the study endpoints. Moreover, a clear study design was not provided up front, so it is unclear how much of the analysis was based on a-priori analytical decisions with intent to have the most informative study versus how much the study reflects selectively chosen results from an exploration of the data.

In summary, the nature of this study is such that it has the potential to hint at the clinical utility of the Corus CAD test and may be able to help create good hypothesis for future investigation, but this study the nature of its design, regardless of the results, does not allow conclusions to be drawn with a level of confidence that would support Medicare coverage.

Summary of the results:

Results are reported stratified by payer as well as by gender. Here we summarize the most important results for making a coverage determination in the Medicare population, results for all patients as well as for the Medicare Advantage population.

As regards care utilization 60 days following the index event in all of the patients:

In summary, the data shows less utilization of invasive coronary angiography, coronary artery revascularization, and CAD-related utilization for the combined payer (all patients) analysis and for the analysis restricted to Medicare Advantage patients. In the combined payer analysis, the Corus CAD cohort had less use of functional or anatomic tests than controls, but in the Medicare Advantage analysis the Corus CAD and control cohorts had numerically similar use of functional or anatomic testing. Major adverse cardiac events were only reported at 180 days for the analysis restricted to Medicare Advantage patients, and there was no significant difference between the groups.

MolDX assessment of study results:

This study clearly shows a difference between the Corus CAD cohort and the control Cohort in terms of utilization of cardiac services when the analysis is performed for all patients as well as when it is restricted to Medicare Advantage patients only, with the latter group likely better representing the Medicare Fee for Service population. However, it did not show that there was any difference in patient outcome in terms of major adverse cardiac events at 180 days. There are two unanswered questions that this research leaves. 1. What exactly what defined the study population? And 2. Was the observed difference in utilization between groups due to the use of Corus CAD in one of the cohorts?

As regards the defined study population, this may become clear in further research that the company suggests that they may pursue. An important aspect of claims-based research is that clinical concepts and disease processes are not necessarily captured in claims data in the way that they must be clinically treated. For example, diagnosis capture, and assessment of medical comorbidities in a claim is done by ICD-9 or ICD-10 code. However, multiple ICD codes may reflect similar diagnostic concepts. (e.g. multiple codes for diabetes with different manifestations and complications), so grouping patients by specific ICD codes is often not practical or even feasible, and comorbidity grouping methods, of which a number have been developed, are used. Additionally, the few numbers of diagnosis fields on physician claims may mean that multimorbid patients have medical conditions not captured on each claim, but that are more likely to be captured when the longitudinal medical record is searched for that diagnosis. For Medicare’s hospital compare risk adjustment methodology, for example, the prior year of claims data is searched to find the presence of a particular comorbid condition for inclusion. Additionally, characteristics of the cohorts created for this study such as age, gender, and comorbid disease burden are not given. These are all important analytical details that help figure out how to translate patient features in claims data into clinical characteristics of human beings a physician encounters at the bedside. As such, though inclusion and exclusion criteria are given, the vagueness of both the methods and the cohorts of patients mean that it is unclear what the patient population of this study is. This question, while unclear in the current manuscript, is potentially answerable with a more detailed explanation of the methods without further research.

The second question related to the cause of the observed differences between groups in the endpoints does not seem easily addressed by the existing data set. It is also worth noting that propensity matching accounts for observed variables to make groups similar. The possibility of an important baseline difference between groups that is not captured in the observed variables used for propensity matching is quite possible, and if present may explain part or all of the observed differences in endpoints between the two cohorts. The intended role of Corus CAD test is to obviate the need for other coronary artery disease testing, and as such it would be natural to consider the reduced use of cardiac testing services in patients who received the Corus CAD test in a matched study to be significantly attributable the use of the Corus CAD test. This is a possibility, but a between group difference in the use of CAD care may also be the effect of a selection bias due to clinicians selecting various diagnostic modalities for different kinds of patients. While it can never be known with certainty that the two cohorts were similar at baseline, it is possible to test the possibility that the cohorts were likely different at baseline in a meaningful way not captured by the propensity matching, which is intended to make the two groups similar. If such a difference were found then it would strongly suggest that observed differences between the cohorts in the endpoints were related to baseline differences between the cohorts other than choice of index test to include a patient in the study.

In this study, such a test was done, though not explicitly described in this way by the study authors. Specifically, they studied the use of cardiac revascularization procedures, such as coronary artery bypass grafting surgery, in individuals who did and did not receive the Corus CAD test. They found that those who received the Corus CAD test were less likely to get a revascularization procedure. The importance of this is that while Corus CAD could plausibly be a substitute for diagnostic cardiac tests, it would not be a plausible replacement for therapeutic procedures; it is a test rather than an intervention. It is, therefore, implausible that laboratory in vitro measurement of blood compounds would affect coronary blood flow in the heart of a patient in vivo.

Another unlikely explanation of the observed association between Corus CAD use and lower utilization of revascularization services is that the use of Corus CAD in the cohort who had it used was informative in a way that accepted cardiac testing used in the control group was not. This explanation would require that Corus CAD’s use prevented unnecessary revascularization procedures. This explanation is unlikely if the data on Corus CAD’s test performance highlighted by the company in a prior comment is to true. The Corus CAD test has a sensitivity of 88%, a significantly greater sensitivity than myocardial perfusion imagining’s sensitivity of 21% in patients 65 years of age and older. As such, Corus CAD would tend not to catch obstructive coronary artery disease that conventional diagnostic approaches would not, which would not likely have the effect of reducing the use of revascularization procedures.

Additionally, revascularization procedures require an identification of the specific cardiac vasculature in need of revascularization. Even if the Corus CAD test is more accurate than conventional diagnostic techniques it still lacks anatomic specificity that may be needed to make a treatment decision. The conventional diagnostic approaches (i.e. those recognized in ACC guidelines) can provide the necessary anatomic information either by showing a specific poorly perfused region of the heart corresponding to a vascular territory, or by directly showing arterial obstruction, as may happen with angiography. The Corus CAD test does not provide anatomy specific-information and therefore cannot serve as a substitute for conventional cardiac diagnostic techniques. As such, in patients with a higher need for revascularization, there is necessarily a higher need for conventional cardiac diagnostic techniques for CAD, limiting the possible clinical utility of this test.

The observed difference in use of coronary artery revascularization procedures in the group who was tested with Corus CAD compared with the control group indicates that the Corus CAD group was a lower risk group for the need for cardiac intervention than the control group eve at baseline, or restated – the propensity matching was not successful in creating truly similar groups. It is a methodological strength of the study that they assessed the robustness of the propensity matching by looking at endpoints that would not be different in presence of successful propensity matching. However, it is a weakness of the study’s results that the study did find that propensity matching likely created dissimilar groups.

In summary, an unpublished manuscript of a claims-based study sponsored by the test sponsor provides evidence regarding the particular patients studied, but generalizing these findings to patients in clinical care cannot be meaningfully done both because the population studied was not clearly described, and more importantly, because differences in endpoints between the two studied cohorts were likely caused to an unknown degree by latent differences in the baseline cardiac risk of the two groups. Finally, even if these limitations did not exist, and we felt confident generalizing the findings to a well-defined population, the retrospective claims-based data with a relatively short follow-up does not allow for any meaningful conclusions that the Corus CAD test is associated with better or worse patient outcomes. As such, while this study may be suggestive, it does not provide evidence of clinical utility that would justify coverage by the Medicare program.