Despite the burden of cardiovascular disease (CVD), routine screening beyond measurement of cholesterol is not considered of medical necessity or supported by national health care coverage decisions. Recent technology evaluations and statements by the U.S. Preventive Services Task Force have voiced strong concerns over the untoward consequences of CVD screening, including the potential for unwarranted, induced testing after a diagnosis of subclinical atherosclerosis (1). Past arguments (2- 4) have cautioned against embarking on nationwide screening for CVD because of a lack of high-quality evidence on the subject. During the past few years, a number of large observational, prospective registries (2- 4) have reported on the prognostic accuracy of CVD screening to detect coronary artery calcification (CAC). This modality has been shown to effectively risk stratify women and men of diverse ethnicity (2- 4).

However, the concern remains that testing will beget more testing and that the initiation of a strategy for the detection of subclinical atherosclerosis may result in early and lifelong greater patterns of resource consumption that would not have been realized without the initial documentation of measureable CVD (1,5). The EISNER (Early Identification of Subclinical Atherosclerosis by Noninvasive Imaging Research) study initiated a prospective registry of individuals with documented cardiac risk factors who agreed to be enrolled and undergo clinical risk assessment and subclinical atherosclerosis screening with computed tomographic measurement of CAC. The EISNER study is composed of several substudies. We will report on individuals enrolled in the EISNER I study who were followed for 4 years for the end point of CVD resource consumption and procedural costs as well as clinical outcomes.

Of the 1,381 participants enrolled in this EISNER substudy, a total of 1,361 (98.6%) were available for this analysis. All enrollees were volunteers who were recruited from study advertisement across the medical center and to the general public under the supervision and approval of our institutional review board from May 2001 to June 2005. Enrolled subjects were not paid for participation, nor were they asked to pay for any study testing. Qualified enrollees had no cardiac symptoms or previous history of CVD. Study coordinators preferentially recruited those with 1 or more cardiac risk factors. Exclusion criteria included age ≥80 years, pregnancy, significant comorbidity, previous CAC scanning, and inability to complete 4 years of follow-up. All subjects provided written informed consent and agreed to participate in the follow-up portion of this study. Results from other EISNER substudies have been previously published (6- 8).

At the time of the baseline visit, each enrollee had arterial blood pressure and fasting measurements of lipids and glucose. The fasting lipid profile (total cholesterol, high-density lipoprotein cholesterol, and triglycerides, with calculated low-density lipoprotein cholesterol) and serum glucose level was performed on each study participant by a Cholestech (Hayward, California) desktop chemical analyzer. Height, weight, hip, and waist measurements also were ascertained from each enrollee. Body mass index was calculated by dividing weight by height measurements (kg/m²). Patients also completed a questionnaire on risk factors and medication use during the index visit.

From the measured risk factor data, the Framingham Risk Score (FRS) was calculated to estimate a patient's 10-year risk of CVD death or myocardial infarction (MI) (9- 10). A low-risk FRS was <10%, intermediate risk was 10% to 20%, and high risk was >20%. Patients with diabetes were categorized with a high FRS.

Scanning was performed by the use of electron beam tomography (Imatron C-150, Imatron, a division of GE, Milwaukee, Wisconsin), GE e-Speed (GE Healthcare), or Siemens Volume Zoom (multislice computed tomography, Siemens Medical Systems, Munich, Germany). Imaging protocols were consistent with accepted standards (6,8,11). Each scan involved acquisition of 30 to 40 slices of 3.0 or 2.5 mm for electron beam tomography and multislice computed tomography with triggering at 50% to 80% of the cardiac cycle. Instructions for breathholding were given to patients to minimize misregistration. The foci of CAC were performed by experienced technologies that scored each scan by the use of semiautomatic commercial software (NetraMD, ScImage, Los Altos, California). The CAC was scored after detection of a minimum of 3 contiguous pixels (voxel size 1.03 mm³) with peak density ≥130 Hounsfield units (HU) within a given coronary artery segment. All scoring was reviewed by an experienced cardiologist. The CAC score was calculated by the product of the area of each calcified focus and peak HU (score = 1, 131 to 199 HU; score = 2, 200 to 299 HU; score = 3, 300 to 399 HU; score = 4, ≥400 HU) (12). A summed CAC score was calculated from all segmental lesions in the left main, left anterior descending, right coronary, and left circumflex arteries.

All enrollees were followed yearly through a mailed questionnaire; culminating in a 4-year clinic visit. Yearly questionnaires documented cardiac risk factor status, ongoing medical therapies, and intercurrent CVD procedures or hospitalizations. Follow-up CVD procedures analyzed were exercise treadmill resting, stress myocardial perfusion imaging, stress echocardiography, coronary computed tomographic angiography (CCTA), and invasive coronary angiography (ICA). The timing and occurrence of percutaneous coronary intervention or coronary bypass surgery was documented. Invasive procedures were confirmed by source documentation or through confirmation with the subject's primary care physician.

During the course of follow-up, the occurrence of CVD death or nonfatal MI was ascertained. We defined CVD death as fatal MI or stroke and death related to heart failure or peripheral arterial disease. The diagnosis of nonfatal MI included admission with a primary or secondary diagnosis confirmed by enzymatic elevation and electrocardiographic (ECG) changes consistent with acute infarction. The timing and occurrence of CVD death or MI was collected during each year of follow-up. Death status was confirmed by medical record review, through the Social Security Death Index, or from Los Angeles County Public Health records. Hospitalization for acute MI was confirmed by the subject's primary care physician and/or medical record review. A total of 1.4% of patients was lost to follow-up. At years 1, 2, 3, and 4, follow-up rates were 99%, 96%, 95%, and 99%, respectively.

Nationwide, average Medicare diagnosis-related group reimbursement rates were applied to define hospital costs using the PC Pricer Prospective Payment System estimator (13). Costs for outpatient services were derived by use of the Outpatient Prospective Payment amounts (nationwide and specific locality) based on Healthcare Common Procedure Codes. Drug costs were derived from the Medicare planner for retail and mail-order pharmacy charges. Costs were inflation-adjusted by use of the medical care component of the consumer price index and discounted 3% per year to reflect the lower economic value of deferred expenses. Our cost analyses used the societal perspective as recommended by the National Panel on Cost-Effectiveness in Health and Medicine (14).

Comparisons of CAC subsets by continuous measures, such as low-density lipoprotein cholesterol, were calculated by the nonparametric Wilcoxon statistic. Median and 25th to 75th percentile measures were calculated. Categorical variables, such as cardiac risk factors, were compared by linear-by-linear association chi-square statistics. Time-to-CVD procedures were estimated by Kaplan-Meier survival curves by the use of a Wilcoxon rank sum test. Rates for years 1, 2, and 4 were calculated. A logistic regression model was used to estimate aspirin, statin, and ICA use. From the model, estimated probabilities of ICA were calculated. The probabilities of ICA were plotted by the CAC score. An odds ratio (95% confidence interval) was calculated for CAC subsets of 11 to 100, 101 to 399, 400 to 999, and ≥1,000. For revascularization, <90-day, ≤1-year, and ≤4-year rates were calculated. Total procedural costs were summed and compared across CAC subsets by the nonparametric Wilcoxon statistic. A linear regression model also was used to estimate predictors of cost. Finally, Kaplan-Meier survival curves were plotted to estimate time to death or MI for the CAC and FRS subsets. Additionally, a Cox proportional hazards model was used to calculate hazard ratios and 95% confidence intervals.

Individuals with more extensive CAC scores were more likely to be older, male, and with prevalent risk factors (Table 1). Many of the current enrollees were receiving risk factor modifying therapies with their FRS calculated on-therapy. The median FRS was 6 (25th, 75th percentile: 2, 12), with 40% being at intermediate risk.

Similar to other reports, the FRS-adjusted odds of new statin use at 4 years was elevated 2.57-fold (25th, 75th percentile: 1.72, 3.83, p < 0.0001), 2.89-fold (25th, 75th percentile: 1.78, 4.69, p < 0.0001), 4.81-fold (25th, 75th percentile: 2.50, 9.28, p < 0.0001), and 14.22-fold (25th, 75th percentile: 5.15, 39.29, p < 0.0001) for those with CAC scores of 0 to 10, 11 to 100, 101 to 399, 400 to 999, and ≥1,000, respectively. Similarly, the FRS-adjusted odds of new aspirin consumption was increased 2.75- to 4.62-fold for the same CAC subsets (p < 0.0001).

During follow-up, noninvasive procedures were frequently performed ((Figure 1),Tables 2, 3). Follow-up ECGs and treadmill tests were performed in 57% and 27% of subjects. One-half of subjects with a CAC score <11 had a routine ECG performed during follow-up as compared with 62% to 90.2% of those with scores 11 to 399 to ≥1,000 (p < 0.0001). A nonimaging exercise ECG was performed in 19.5% to 61.3% of those with a CAC score from 0 to 10 to ≥1,000 (p < 0.0001).

Cardiac imaging, including stress echocardiography or myocardial perfusion single-photon emission computed tomography, was performed in 15.2% and 11.7% of enrollees. (Figure 1)A plots the cumulative rates of stress cardiac imaging after index CAC scanning (p < 0.0001). By 1 year of follow-up, few individuals with a low CAC score of 0 to 10 had a follow-up stress imaging procedure. Rates of stress imaging increased over time and more commonly were performed in those with a CAC score of ≥400. By year 1 of follow-up, 36.9% and 44.5% of individuals with a CAC score of 400 to 999 and ≥1,000 had stress imaging study performed. The highest 4-year rate of stress imaging was in those with a CAC of ≥1,000 with 86.8% of the 31 individuals undergoing this procedure.

A total of 7.9% of subjects had a follow-up CCTA. (Figure 1)B plots the cumulative rate of CCTA after index CAC scanning (p < 0.0001). Rates of CCTA were lowest for patients with less-extensive CAC scores and highest for patients with extensive CAC. Of the 773 individuals with a CAC score from 0 to 10, only 2.3% underwent CCTA.

A further analysis of the frequency of procedures revealed that for patients with CAC scores of 0 to 10, >90% did not have any procedures through 1 year of follow-up (Table 2). Conversely, follow-up testing (including multiple procedures) was common for those with a CAC score ≥400.

Noninvasive test layering that included a combination of CAC followed by exercise treadmill testing with a stress imaging test occurred in 11.3% to 51.6% of those with CAC scores ≤10 to ≥1,000 (p < 0.0001) (Table 3). Noninvasive procedural use did not vary by the FRS (p = 0.718) in an adjusted model containing CAC (p < 0.0001). For those with a CAC score ≤10, current smoking (p = 0.019) and age ≥60 years (p = 0.004) were associated with more noninvasive follow-up testing (p = 0.019).

A total of 92 ICAs were reported during follow-up. Similar patterns of greater use in individuals with greater CAC scores was noted for ICA (Table 3). The rates of ICA were low at 1 year of follow-up and ranged from 0.3% to 3.5% for those with CAC scores from <1,000 (p < 0.0001) (Figure 2A) and remained low (i.e., <10%) for most of follow-up. Follow-up ICA use was greater for patients with a CAC score from 400 to 999 (13.5%) and ≥1,000 (36.7%). The probability of ICA was <1% for those with a CAC score from 0 to 10 but increased in a directly proportional manner with more extensive CAC score (Figure 2B).

A total of 13 coronary artery bypass surgeries and 44 percutaneous coronary interventions were reported during follow-up. The resulting early revascularization rates were low, with only 3 individuals undergoing coronary revascularization within 90 days of follow-up (Table 4). The majority (31 of 157) of the revascularization procedures occurred in those with CAC scores ≥400 (p < 0.0001).

Importantly, invasive testing occurred exclusively in patients with a previous noninvasive procedure. Test layering where invasive testing followed noninvasive testing occurred in 0.8% to 19.4% of those with a CAC score ≤10 to ≥1,000 (p < 0.0001) (Table 3).

Both procedural and overall costs increased progressively with increasing CAC scores (p < 0.0001 for both). Mean costs were lowest for those with a CAC score ≤10 (Table 5). Costs expended on procedures increased sharply for the 31 subjects with CAC scores ≥1,000 as did overall medical costs. However, since this subgroup constituted just 2.2% of the total study cohort, its medical expenditure accounted for only 12.9% of the total medical costs within the study cohort. Procedural and drug costs encumbered 28.9% (25th, 75th percentile: 8.1%, 100%) and 60.4% (25th, 75th percentile: 0.0%, 91.1%) of overall 4-year costs.

Intermediate- and high-risk FRS individuals had greater costs than those with a low FRS. By using a linear regression model estimating procedural cost, we found that each increase in FRS risk group was associated with an increase in 4-year cost of $224.39 and each increase in the CAC subsets was associated with an increase in 4-year cost of $493.60. In this model, the presence of diabetes was not associated with greater procedural costs (p = 0.35). Procedural costs were similar across FRS subsets for all CAC subgroups.

Overall death or MI-free survival was 98.7%. (Figure 3) plots the cumulative death or MI-free survival by the CAC scores (p < 0.0001) and FRS (p < 0.0001). For CAC score, the relative hazard increased 4.0- to 27.9-fold for those with CAC scores from >10 up to ≥1,000 (p < 0.0001). Individuals with an intermediate-high FRS had an increased hazard for death or MI (p < 0.0001). The area under the curve, from the receiver-operating characteristic analysis, for the FRS was 0.71 (0.61 to 0.82, p = 0.001) with an added improvement to 0.79 (0.70 to 0.88, p < 0.0001) for CAC scoring.

The U.S. spends approximately $2 trillion each year on health care, consuming more than 16% of our gross domestic product, with imaging identified as influential in promoting excess expenditures (15- 16). Screening for lung, breast, and colon cancer is a cornerstone of preventive health and is considered cost effective because the benefits of early detection offset procedural and induced costs of care resulting in a reduced prevalence of more advanced, expensive downstream disease states (17). Discussions regarding CVD screening arise at a time when growth in imaging is double that of all other physician services (15), at an estimated cost of $80 billion annually (18). Ongoing discussions of the advantages of expanding coverage for screening of preventable chronic diseases, such as CVD, can only now be advanced within the framework of a demonstrable societal benefit where economic evidence is clearly defined within the context of added value (i.e., quality) (19- 21). Economic evaluations, such as that within EISNER, can then be used to inform health care policy decisions (22- 24).

The evidence of a clinical benefit of screening with CAC is now substantial, with recent reports (3,25- 28) on its prognostic utility in ethnically-diverse populations of women and men. From a recent report, CAC improved mortality risk reclassification in 25% to 45% of individuals ages 60 to 80 years when compared to the FRS. Despite effective risk stratification, concern over the “train” of CVD services that may ensue after CAC scanning is frequently voiced (29- 31). The U.S. Preventive Services Task Force in 2004 cautioned that harm potentiated from CVD screening outweighed any benefit (1). They noted that excessive costs and harm associated with additional testing and possibly labeling would exceed any derivable benefit from CVD screening.

Our observations confirm that in an adult population with measureable CVD risk, ongoing preventive and diagnostic services frequently occur. Yet, annual CVD costs were low, at $25 to $35, for those with a CAC score ≤100, and notably, the subjects falling in this CAC score range constituted 78% of our screened population. By contrast, substantial use of downstream testing and higher medical costs were observed among the subjects with CAC scores ≥400, constituting 8.2% of the screened population. Of particular interest in this regard, direct ICA was not performed immediately after CAC scanning with feeder pathways driven by the severity of ischemia and/or noninvasive coronary anatomy results before ICA referral. This result is consistent with a recent report noting that the addition of 1 CT scanner for cardiac applications resulted in a reduction in 15.4 (per 100 scanners) fewer ICAs (32). This concept of stepped testing has been reported as an effective means to reduce work-up costs by limiting additional testing to only those with abnormal test results (7- 8,33- 36). The observation that the rates of follow-up stress imaging were much higher in the patients with CAC scores ≥400 is concordant with clinical appropriate use criteria (37) and data demonstrating a threshold relationship between absolute CAC scores and the likelihood of observing inducible myocardial ischemia (7- 8,34). Inducible myocardial ischemia is increasingly common among patients with CAC scores >400, but the CAC threshold is influenced by certain CAD risk factors, such as diabetes and metabolic syndrome (7) and the quality of patients' chest pain symptoms (34).

Our data reveal that the differential in costs among our study population were substantially wider with CAC scanning compared to costs based on the FRS. When compared with noncardiac tests or blood markers, it remains plausible that CAC may also elicit a greater differential in procedural costs and treatment costs when compared with other screening tests, but impact may vary by treatment. For example, given the therapeutic trials with high sensitivity C-reactive protein, one would anticipate a much larger increase in statin use than that elicited by CAC screening (38).

A Medicare Coverage Advisory Committee on medical imaging set one requirement for the addition of a procedure that it must reduce or minimize use of comparative modalities (39). To that end, CAC imaging appeared to minimize resource use among the large cohort of subjects for those with low-risk scores, with the majority of “high-end” resource consumption limited to the much smaller cohort of subjects with high-risk CAC scores. Similarly, greater drug therapy utilization in patients with high-risk CAC scores was previously reported (40). Previous reports (41) have noted a close relationship (r² = 0.78) between ICA and percutaneous coronary intervention. In the pre-ICA setting, we noted a similar relationship between ischemia testing and ICA for patients with high-risk CAC.

On the basis of the current results, previous concern as to CAC scanning leading to frequent ICA in asymptomatic patients was not realized in our study (42). Rather, at 1 year after testing, ICA was performed <1% of subjects with CAC scores ≤1,000 and in only 19.4% of subjects with CAC scores >1,000.

Of importance, the generalizability of the current finding is limited to registries with similar enrollment strategies and admixture of risk factors. From the current study, 1.4% of enrollees were lost in follow-up. Individuals lost during follow-up were similar to those included herein. Follow-up test interpretations were self-reported or inconsistently available. Test interpretations could help to further unfold the effectiveness of a CVD screening program. To preserve subject anonymity as required by our Institutional Review Board, test results were not sent directly to the subjects' physicians, but subjects were encouraged to share their screening results with private physicians, especially if a new or high-risk factor, such as a calcium score ≥400, was detected. We did not include indirect costs or patient preferences that may further aid in defining the societal benefit of CVD screening (43). Hospital and outpatient bills were not available and could have improved the precision of our cost estimates, although we used a systematic approach by applying uniform costing throughout this analysis. Given the diversity in reimbursement across a given private and public payers, we chose to apply Medicare reimbursement rates as an attempt to estimate costs. As a result, some of our estimated costs may not be applicable to non-Medicare patient populations. Operational within follow-up interviews is the potential for recall bias to result in an under- or over-estimation of follow-up resource use data. Moreover, the aim of the current report was to examine induced procedure use, and we did not include patterns of outpatient visits to generalists or cardiologists that may be important and contributory to this analysis.

The costs of CVD on the health care system are substantial and, similar to other chronic conditions, appear greatest in the initial diagnosis and final phases of life (23,44- 46). Accordingly, concern for adding costs by approving third-party coverage for CAC scanning has been understandable. On the other hand, the frequent occurrence of sudden death as the first manifestation of CAD (47) represents an unsolved health problem requiring better means for screening for CAD. Ideally, an acceptable cost-effective screening test for CAD would reliably identify a high risk subgroup for aggressive follow-up that would, however, constitute only a small fraction of the screened population. The results of this study are consistent in this regard. CAC scanning identified a high risk subgroup of subjects with CAC scores ≥400, who have been shown by prior epidemiological study to be at increased risk for future cardiac events and are known to concurrently have an increased risk for inducible myocardial ischemia (7- 8,34). This high-risk subgroup had substantially increased medical costs and downstream testing, but importantly, this group only constituted 8.2% of our study population. Those at very high risk, by virtue of CAC scores ≥1,000 had a marked increase in medical costs, but only constituted 2.2% of the study population. Conversely, downstream medical costs were very low among the large pool of subjects with low CAC scores in our study

CAC scanning was initially introduced into medical practice solely as a screening test for CAD, but the observation that this test also helps predict the likelihood of inducible myocardial ischemia means that the test may not only serve as a means for detecting latent CAD but also, as a potential gatekeeper for determining the need for subsequent medical testing, and invasive procedures that may follow, among subjects with suspected CAD. The findings of our study suggest that CAC may play an important role in fostering more efficient, selective testing patterns in asymptomatic individuals at risk for CVD.