ACCF/AHA 2007 Clinical Expert Consensus Document on Coronary Artery Calcium Scoring By Computed Tomography in Global Cardiovascular Risk Assessment and in Evaluation of Patients With Chest Pain

A Report of the American College of Cardiology Foundation Clinical Expert Consensus Task Force (ACCF/AHA Writing Committee to Update the 2000 Expert Consensus Document on Electron Beam Computed Tomography) Developed in Collaboration With the Society of Atherosclerosis Imaging and Prevention and the Society of Cardiovascular Computed Tomography

Matching Intensity of Intervention With Severity of Risk

As previously noted, a principle of cardiovascular disease prevention that is generally accepted is that intensity of intervention for an individual (or population) should be adjusted to the level of baseline risk (17). The goals of this principle are to optimize efficacy, safety, and cost-effectiveness of the intervention. The concept is most often applied to higher-risk individuals who are potential candidates for risk-reducing drugs; but it also is an important consideration for lower risk individuals either in clinical practice or for public health strategies. For higher risk individuals, intensity of intervention is best adjusted to absolute short-term risk; for lower risk individuals, relative risk remains an important consideration because a high relative risk generally translates into a high absolute risk in the long term. This latter concept is most relevant to younger men and middle-aged men and women, whereas in older men and women, the Framingham Risk Score generally applies.

Current Approaches to Global Risk Assessment and to Assessment of Incremental Risk Using New Tests

In current clinical practice, in accordance with a number of guidelines (14,15), it is common that the first step in clinical risk assessment is to identify any high-risk conditions that obviate the need for further risk assessment; these mainly include established atherosclerotic cardiovascular disease (ASCVD) and diabetes (see Table 1, High risk). If none of these high-risk conditions is present, the second step is to identify the presence of major risk factors (also listed in Table 1). If 2 or more major risk factors are present, one should then estimate the 10-year likelihood for development of major coronary events or total cardiovascular events. In the United States, the most-commonly used and most extensively validated quantitative assessment is provided by the multivariable scoring system of the Framingham Heart Study. The Framingham algorithm for “hard CHD” events including myocardial infarction and cardiac death is available through the National Cholesterol Education Program website (http://hin.nhlbi.nih.gov/atpiii/calculator.asp). Framingham scoring includes the following major risk factors: gender, total cholesterol, high-density lipoprotein (HDL) cholesterol, systolic blood pressure (or on treatment for hypertension), cigarette smoking, and age. PROCAM scoring employs a somewhat different set of risk factors: gender, age, low-density lipoprotein (LDL) cholesterol, HDL cholesterol, triglycerides, systolic blood pressure, cigarette smoking, family history, and presence or absence of diabetes (http://www.chd-taskforce.com/). The European SCORE algorithm uses risk factors similar to the Framingham Score.

For each of these risk assessment tools, the most powerful risk factors are age and gender. The other risk factors can be examined for their additive predictive power by determining increments in the area under the curve of the receiver-operating characteristic (ROC). The area under the ROC curve is also known as the C-statistic. An ROC analysis plots sensitivity (fraction of true positives) versus 1-specificity (fraction of false positives) of a risk factor for predicting events. ROC curves are used to evaluate the discrimination of a prediction, and often, the predictive power of a set of risk factors. If a given set of risk factors predicted the development of cardiovascular events perfectly, the curve would reach 100% in the upper left corner (100% sensitivity and 100% specificity), that is, all true positives and no false positives. The area under the curve would be 100% (C-statistic = 1.0). A random and useless predictor would give a straight line at 45 degrees (C-statistic = 0.5) since this would define a test where true positive rate and false positive rate are equal to one another at every possible cutoff value. In the evaluation of additional tests, added to the basic set of Framingham risk factors, the area under the curve would increase when the test provides incremental discrimination. The Framingham algorithm applied to the Framingham population generally gives a C-statistic of approximately 0.8, meaning that the probability is 80% that patients who experience CHD events will have a higher risk score than patients who did not experience an event. An important but unresolved issue is whether discovery and addition of new biochemical risk factors or imaging markers to Framingham or PROCAM algorithms will increase the C-statistic. In considering the role of CAC measurement for risk assessment, a key issue is whether discriminative ability is improved, often as judged by an increase in the C-statistic compared to that derived from risk factors alone.

Prognosis by Coronary Artery Calcium Measurements

In the prior ACC/AHA expert consensus document published in 2000, only 3 reports on the prognostic capability of CAC scoring were available to develop risk assessment indications in asymptomatic individuals (1). At the time, the ACC/AHA document concluded that the body of evidence using CAC measurement to predict CHD events was insufficient. A critical component to that recommendation was that the independent prognostic value of CAC had not been established. In a separate but similar evaluation using data published through 2002, the U.S. Preventive Services Task Force (USPSTF) concluded that limited clinical outcomes data were available and recommended against routine screening for the detection of silent but severe CAD or for the prediction of CHD events in low risk, asymptomatic adults (see http://www.ahrq.gov/downloads/pub/prevent/pdfser/chdser.pdf).

In the past several years, however, a number of publications have reported on the incremental prognostic value of CAC in large series of patients including asymptomatic self-referred and population cohorts (18–22). A major rationale for the current document is the need for an update including recent publications regarding CAC as it relates to the estimation of CHD death or nonfatal myocardial infarction (MI). Although earlier evidence included the use of “soft” endpoints including coronary revascularization as a primary outcome, more recent data are available on the estimation of CHD death or MI (18–22). Models predicting “hard” cardiac events (i.e., CHD death or MI) are less subjective and less likely to overestimate the predictive accuracy of CAC scoring (23).

Theoretical Relationship Between Coronary Calcification and CHD Events

Atherosclerotic plaque proceeds through progressive stages where instability and rupture can be followed by calcification, perhaps to provide stability to an unstable lesion (8). As the occurrence of calcification reflects an advanced stage of plaque development, some researchers have proposed that the correlation between coronary calcification and acute coronary events may be suboptimal based largely on angiographic series (11). In order to understand this apparent conflict between the stability of a calcified lesion and CHD event rates, one must recognize the association between atherosclerotic plaque extent and more frequent calcified and non-calcified plaque (24). That is, patients who have calcified plaque are also more likely to have non-calcified or “soft” plaque that is prone to rupture and acute coronary thrombosis (24). It is the co-occurrence of calcified and non-calcified plaque that provides the means for estimating acute coronary events. Furthermore, although CAC detection cannot localize a stenotic lesion or one that is prone to rupture, CAC scoring may be able to globally define a patient’s CHD event risk by virtue of its strong association with total coronary atherosclerotic disease burden, as shown by correlation with pathologic specimens (1,24).

Approaches to Technology Assessment in CHD Screening

A major criterion utilized in many technology assessments has been that a screening test must have a high level of evidence on the effect of screening on actual health outcomes, such as fewer events, extended life, or better quality of life. This type of analysis requires research detailing an improvement in either quantity or quality-of-life years as a result of the screening procedure. An example of a high level of such evidence was recently published on screening for abdominal aortic aneurysm (AAA) (25). Using this example, a meta-analysis reported reduced mortality in randomized trials of AAA screening. These results allowed for favorable support of AAA screening by the USPSTF resulting in a class B recommendation (i.e., evidence includes consistent results from well-designed, well-conducted studies in representative populations that directly assess effects on health outcomes) (26). Lack of similar controlled clinical trial evidence played a central role in the conclusion by the USPSTF not to support CHD screening using CAC measurement (see http://www.ahrq.gov/downloads/pub/prevent/pdfser/chdser.pdf).

Although no studies have shown a net effect on health outcomes of CAC scoring (27), at least one randomized trial is nearing completion (Early Identification of Subclinical Atherosclerosis using NoninvasivEImaging Research [EISNER]). However, the concept of matching treatment intensity to the degree of cardiovascular risk suggests that efforts to identify the most accurate approach to risk stratification is an initial and critical step that should aid in the best selection of treatment options for patients at risk for cardiovascular disease.

Systematic Reviews and Meta-Analyses

In the sections that follow, we review recent evidence on the prognostic value of CAC and include data from one recent systematic review. A comprehensive data synthesis on this subject was published by Pletcher et al. (23) evaluating the prognostic value of CAC from 4 studies published through 2002 meeting quality-based inclusion criteria. Articles were considered for that meta-analysis if they evaluated the prognostic value of CAC in asymptomatic individuals and also presented data on CHD events. Based on a random-effects model, the summary relative risk ratios were 2.1 (for CAC score of 1 to 100) and as high as 10 (for CAC greater than 400) as compared to patients with a score of 0 (pless than 0.0001). This meta-analysis (23) offers support for the concept that there is a linear relationship between CAC and CHD events, but the analysis did not address whether CAC measurement is incremental to Framingham Risk Score (FRS) for CHD risk prediction.

Data Quality Issues

A lack of rigor in study methodology was a focus of the 2000 ACC document (1). A detailed review of the quality of the published data on the prognostic value of CAC was also published by Pletcher et al. (23) noting significant heterogeneity in study quality with often a lack of blinded outcome adjudication, greater use of categorical or historical risk factors, and variable tomographic slice thickness (3 vs. 6 mm) contributing to an overestimation of the relative risk of events by CAC measurements. For example, the relative risk ratio was significantly higher for CAC of 101 to 400 (p= 0.01) and greater than 400 (p= 0.004) when self-reported or historical risk factors were employed in a predictive model as compared with measured risk factor data. The clinical implication of this distinction is that physicians interpreting these results may overvalue CAC scores as substantially more predictive than traditional risk factors.

Evaluation of more recent publications indicates that some of the important methodological limitations of earlier reports have been addressed. Notably, more recent publications report the independent prognostic value of CAC in multivariable models including measuredrisk factor data (18,19,22). Larger sample sizes have also resulted in improved precision in risk prediction models. However, issues of selection or referral bias when using patient cohorts remain pertinent and are likely to have resulted in an overestimation of risk when based on clinical cohorts as compared with population samples (20,22). It is important to recognize that relative risk ratios from patient cohorts have generally been higher than from studies conducted in population samples even when the overall direction of the prognostic findings has been concordant.

Inclusion Criteria and Endpoint Definitions for the Present Analysis

The current document focuses on the ability of CAC scoring to estimate CHD death or MI. This approach allows for a comparison of the expected annual event rates based on the FRS. The FRS estimates that annual rates of CHD death or MI are less than 1.0% for low risk, 1.0% to 2.0% for intermediate risk (Table 1), and greater than 2.0% for high risk. When multiple publications have been reported from the same cohort study (1,4,5,33–36), we employ here only the most recent report in the current analysis (19,20).

The inclusion criteria for this analysis are: 1) data not previously reported in the 2000 document (1); 2) published series on the prognostic value of CAC in asymptomatic cohorts reported since 2002; 3) endpoint data must be reported on the outcome of CHD death or MI over a specified follow-up time period (usually within 3 to 5 years); and 4) data extraction must allow for the calculation of univariable relative risk ratios and must also include risk-adjustment for traditional cardiac risk factors (e.g., age, gender, cholesterol, hypertension, etc.) or the FRS.

Two committee members (AJT, LJS) evaluated the quality of each included report with the results of this analysis being included in Table 2.The quality assessment criteria included: 1) documentation of prospective data collection; 2) inclusion of self-referred patient series or from a population sample; 3) reporting of CHD events; 4) reporting of outcome data by gender and ethnicity; 5) sample size greater than 1000 individuals; 6) avoiding potential for limited challenge (i.e., an inclusion of very low to very high-risk patients resulting in a wide spread in the outcome results) by not reporting data within strata of clinical risk; 7) reporting measured versus historical or self-reported risk factor data; and 8) reporting univariable and multivariable prognostic models (i.e., ascertaining the incremental value of CAC scores). A review of the highlighted reports reveals that all studies identified for inclusion were of at least moderate-high quality.

Prognostic Value of CAC Scores From Published Reports From 2003–2005

Several recent cohorts have been published including prospective observational registries in predominantly male, younger and middle-aged (18), unselected (19) and older-aged, higher risk (20) asymptomatic cohorts. A self-referred patient series of 8855 asymptomatic adults was also included in this analysis (21). A recent population sample was also published and included 1795 subjects greater than or equal to 55 years of age who were prospectively enrolled in the Rotterdam coronary calcium study (22). Finally, the prognostic value of CAC scores was recently reported from a large series of 10 746 men and women aged 22 to 96 years who underwent a preventive health examination at the Cooper Clinic in Dallas, Texas (28).

Using a random-effects model, an analytical approach frequently applied to observational data such as that reported in the CAC series, Figure 1reports on the univariable and summary (weighted average) relative risk ratios from 6 recently published reports in 27 622 patients (n = 395 CHD death or MI). This figure reports the summary relative risk ratio of 4.3 (95% confidence interval [CI] = 3.5 to 5.2) for any measurable calcium as compared with a low-risk CAC (generally using a score of 0) (pless than 0.0001). These data imply that the 3 to 5 year risk of any detectable calcium elevates a patient’s CHD risk of events by nearly 4-fold (pless than 0.0001). Importantly, patients without detectable calcium (or a CAC score = 0) have a very low rate of CHD death or MI (0.4%) over 3 to 5 years of observation (n = 49 events/11 815 individuals).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Meta-Analysis on the Prognostic Value of CACS

Relative risk (RR) ratios (95% confidence intervals [CI]) in six published reports (18–22,28). CACS = coronary artery calcification score.

As can be further seen in Figure 1, considerable variability existed in the relative risk ratios across the 6 reports which can, in part, be attributed to variability in the grouping of CAC scores and in the representation of younger individuals and women within each of the risk subsets. In the most recent report from the Cooper Clinic, different CAC ranges in risk groupings were applied for women and men (28). Moreover, both the Walter Reed and Cooper Clinic series evaluated younger asymptomatic cohorts while the Rotterdam study limited enrollment to individuals greater than or equal to 55 years of age (18,22).

The summary relative risk ratios in Figure 2reveal an incremental relationship where higher CAC scores are associated with higher event rates and higher relative risk ratios. In this figure, a mild risk CAC score (with scores ranging from 1 to 112) was associated with an elevation in CHD death or MI risk with a summary relative risk ratio of 1.9 (95% CI = 1.3 to 2.8, p= 0.001). This mild risk grouping was more often reported in younger populations undergoing preventive health screenings (18,28).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

RR Ratios According to Level of Risk for CACS, From Average Risk to Very High Risk

Average risk includes Arad et al. (19), Greenland et al. (20), LaMonte et al. (28), and Taylor et al. (18). Moderate risk includes Arad et al. (19), Greenland et al. (20), LaMonte et al. (28), Taylor et al. (18), and Vliegenthart et al. (22). High risk includes Arad et al. (19), Greenland et al. (20), Kondos et al. (21), LaMonte et al. (28), and Vliegenthart et al. (22). Very high risk includes Vliegenthart et al. (22). *Low-risk N often includes multiple comparisons from a single series (e.g., Taylor CACS of 1 to 9 and 10 to 44 would use the same referent low-risk group comparison). CACS = coronary artery calcification score; CI = confidence interval; RR = relative risk.

With even higher CAC scores, the 3 to 5 year event rates increased substantially. For scores ranging from 100 to 400, the summary relative risk ratio was 4.3 (95% CI = 3.1 to 6.1) when compared to patients with no detectable coronary calcium (pless than 0.0001). For the high (CAC scores of 400 to 1000) and very high (greater than 1000) risk CAC scores, pooled CHD death or MI rates were 4.6% and 7.1% at 3 to 5 years after CAC testing, resulting in relative risk ratios of 7.2 (95% CI = 5.2 to 9.9, pless than 0.0001) and 10.8 (95% CI = 4.2 to 27.7, pless than 0.0001) when compared to the low-risk group (CAC score = 0) as reference.

Independent Prognostic Value of CAC Scores Over Cardiac Risk Factors

A necessary criterion for establishing a high degree of predictive accuracy for CAC measurements is the establishment of the independent contribution of CAC above and beyond risk factor data alone (29). Recent reports have included univariable and multivariable models that have evaluated the independent contribution of CAC in models evaluating risk factors or the FRS (Table 3).From the St. Francis Heart Study, measured risk factor data were available in 1293 of the total enrolled cohort of 4903 asymptomatic individuals. In univariable (pless than 0.0001) and multivariable (p= 0.01) models estimating CHD events at 4.3 years of follow-up, CAC scores were independently predictive of CHD outcome above and beyond both historical and measured risk factors (19). The CAC scores were also predictive of outcome in a multivariable model containing high-sensitivity C-reactive protein (18), similar to a previous report by Park et al. (30). Several reports have also evaluated the independent prognostic contribution of CAC in multivariable models that controlled for other cardiovascular risk markers, including risk factors not in the FRS, such as a family history of premature CHD (18,22) or body mass index (22) (Table 4).

Predictive Accuracy in Patients With an Intermediate FRS

The concept of Bayesian theory provides a framework to evaluate the expected relationship between the predictive value of CAC score in individuals with low- to high-risk FRS. As defined by Bayesian theory, a test’s post-test likelihood of events is partially dependent upon a patient’s pretest risk estimate. Thus, for patients with a low risk FRS very few events would be expected during follow-up and the resulting post-test risk estimate for patients with an abnormal CAC score would be expected to remain low. Several reports have noted that the use of CAC score in low-risk populations is not useful in modifying prediction of outcome (20,21). Greenland et al. (20) reported that a high CAC score was predictive of high risk among patients with an intermediate-high FRS greater than 10% (pless than 0.001) but not in patients with a low risk FRS (i.e., score less than 10%). In this report from the South Bay Heart Watch study, only 1 CHD event was noted in 98 patients with a low risk FRS. This report demonstrates the importance of considering the underlying hazard in selecting optimal cohorts for whom CAC testing will be of greater value.

In addition, the recent data provide support for the concept that use of CAC testing is most useful in terms of incremental prognostic value for populations with an intermediate FRS (29). In a secondary analysis of patients with an intermediate FRS from 4 reports (19,20,22,28), annual CHD death or MI rates were 0.4%, 1.3%, and 2.4% for each tertile of CAC score where scores ranged from less than 100, 100 to 399, and greater than or equal to 400, respectively (19,20) (Fig. 3).From this analysis, intermediate-risk FRS patients with a CAC score greater than or equal to 400 (Fig. 3) would be expected to have event rates that place them in the CHD risk equivalent status (event rate greater than or equal to 20% over 10 years (31).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Estimated Annual Risk of CHD Death or MI Rate

Rate shown is by tertile of the Agatston score in patients at intermediate coronary heart disease (CHD) event risk using definitions of an intermediate Framingham Risk Score (FRS) or greater than 1 cardiac risk factor. Intermediate FRS was defined as follows: Greenland et al. (20) 10% to 20%; Vliegenthart et al. (22) 20%; LaMonte et al. (28), greater than 1 cardiac risk factor; and Arad et al. (19) 10% to 20%. CACS = coronary artery calcium score; MI = myocardial infarction.

Future Research Needs

The vast majority of prognostic evidence has been reported using an evaluation of risk stratification with absolute measurements of the CAC score. However, some earlier reports applied gender- and age- percentile rankings that may have greater intuitive appeal and understanding for patient education. As such, the percentile rankings have the potential for greater clinical applicability and, therefore, utilization. Only one report has evaluated the comparative predictive ability of absolute CAC scores versus the percentile scores. These investigators noted an improvement in risk detection using percentile ranks (32). An advantage to the use of percentiles is that it has been integrated into the NCEP guidelines where more aggressive care was recommended for patients with a 75th percentile ranking or higher (31). Thus, more information on percentile rankings for prognosis is needed; however, very few research groups have consistently reported CAC data according to percentile ranking. In addition, in our review of the current published evidence, the relative risk ratio for a high risk CAC measurement is higher for clinical registries as compared with population studies (relative risk = 19.3 vs. 5.0); suggesting an overestimation in risk due to selection bias (18–20,22). Data from the ongoing Multi-Ethnic Study of Atherosclerosis (MESA) should allow for more accurate risk estimation of CAC scores as based on a prospectively-derived large population sample (33).

Summary

Since 2000, when the last ACC CECD report on CAC measurement was published, there has been growing evidence on the use of CAC in better-studied cohorts of patients and asymptomatic individuals. CAC scoring has an increasingly high level of quality evidence on its role in risk stratification of asymptomatic patients. Recent evidence is supportive that measurement of CAC is predictive of CHD death or MI at 3 to 5 years. Current evidence also suggests that the use of CAC is independently predictive of outcome over and above traditional cardiac risk factors. Published reports have largely been derived from patient cohorts where referral bias is operational resulting in an overestimation of CHD death or MI risk estimates. Upcoming data from the MESA study may be helpful to devise population screening strategies for women and in non-whites. The MESA data will also be useful in validating predictive capability by ethnicity and across a broad age range of asymptomatic people. Data employing direct comparisons of CAC measurement versus other imaging modalities or biomarkers are generally not available.

The consensus of the Committee was that the body of evidence is supportive of recommendations from the USPSTF that unselected screening is of limited clinical value in patients who are at low risk for CHD events, typically estimated using a low FRS less than 1.0% per year (see http://www.ahrq.gov/downloads/pub/prevent/pdfser/chdser.pdf).

A subset analysis of the predictive accuracy of CAC in patients with an intermediate FRS reveals that for a score greater than or equal to 400, the patient’s 10-year CHD risk would achieve risk equivalent status similar to that noted with diabetes or peripheral arterial disease (31). Thus, clinical decision-making could potentially be altered by CAC measurement in patients initially judged to be at intermediate risk (10% to 20% in 10 years).

The accumulating evidence suggests that asymptomatic individuals with an intermediate FRS may be reasonable candidates for CHD testing using CAC as a potential means of modifying risk prediction and altering therapy. On the other hand, there is little to be gained by testing with CAC in patients with a low FRS. Furthermore, patients with a high FRS should be treated aggressively consistent with secondary prevention goals based upon the current NCEP III guidelines and thus should not require additional testing, including CAC scoring, to establish this risk evaluation (31). Additionally, the current CAC literature does not provide support for the concept that high-risk asymptomatic individuals can be safely excluded from medical therapy for CHD even if CAC score is 0.

Diagnosis of Coronary Stenosis in Patients With Possible CHD by CAC

The utility of coronary artery calcium measurement in symptomatic patients has been widely studied and discussed in depth in the previous ACC/AHA statement (1). It was also extensively reviewed in the recent American Heart Association Cardiac Imaging Committee Consensus Statement—The Role of Cardiac Imaging in the Clinical Evaluation of Women With Known or Suspected Coronary Artery Disease (34). One conclusion of these reports was that a positive CT study (defined as presence of any CAC) is nearly 100% specific for atheromatous coronary plaque (34,35). Since both obstructive and non-obstructive lesions can have calcification present in the intima, CAC is not specific for obstructive coronary disease.

In the symptomatic patient, CAC has been evaluated as a noninvasive diagnostic technique for detecting obstructive CAD. To define its test characteristics and to compare it with other noninvasive tests, a meta-analysis was performed and published in the previous ACC/AHA consensus statement (1). In the previous meta-analysis, a total of 3683 patients were considered among 16 studies evaluating the diagnostic accuracy of CAC measurement (1). Inclusion criteria were: diagnostic catheterization for patients without prior history of coronary disease or prior cardiac transplantation. Patients were symptomatic and referred to the cardiac catheterization laboratory for diagnosis of obstructive CAD. On average, significant coronary disease (greater than 50% or greater than 70% stenosis by coronary angiography) was reported in 57.2% of the patients. Presence of CAC was reported on average in 65.8% of patients (defined as a score greater than 0 in all but one report). The weighted-average or summary odds were elevated 20-fold with a positive CAC (score greater than 0) (95% CI 4.6 to 87.8). Additional summary odds ratios were also calculated with various anatomic and calcium score cut points. For detection of minimal, greater than 50%, and greater than 70% stenosis at cardiac catheterization, the summary odds increased from 6.8-fold (95% CI 3.0 to 15.6) to 16.4-fold (95% CI 5.1 to 53.1) to 50-fold (95% CI 24.1 to 103.0); that is, the odds of significant coronary disease increased when greater angiographic lesion thresholds were used for significant disease (although the confidence bounds widened). Higher coronary calcium scores increased the likelihood of detecting significant coronary disease (greater than 50% or greater than 70% luminal stenosis). A threshold of detectable calcium or a score greater than 5 was associated with an odds of significant disease of 25.6-fold (95% CI 9.6 to 68.4).

Schmermund et al. (36) examined 291 patients with suspected CHD who underwent risk factor determination as defined by the NCEP, CAC measurement, and clinically indicated coronary angiography. A simple noninvasive index (NI) was constructed as the following: log(e)(LAD score) + log(e)(LCx score) + 2[if diabetic] + 3[if male]. Receiver-operating characteristic curve analysis for this NI yielded an area under the curve of 0.88 ± 0.03 (pless than 0.0001) for separating patients with, versus without, angiographic 3-vessel and/or left main CAD. Various NI cutpoints demonstrated sensitivities from 87% to 97% and specificities from 46% to 74%. Guerci et al. (37) studied 290 men and women undergoing coronary arteriography for clinical indications. A coronary calcium score greater than 80 (Agatston method) was associated with an increased likelihood of any coronary disease regardless of the number of risk factors, and a coronary calcium score greater than or equal to 170 was associated with an increased likelihood of obstructive coronary disease regardless of the number of risk factors (pless than 0.001). Kennedy et al. (35) studied 368 symptomatic patients undergoing cardiac catheterization. By multivariate analysis, only male sex and coronary calcification were significantly related to extent of angiographic disease. Receiver-operating characteristic curve analysis showed that the amount of coronary calcium was a significantly better discriminator of disease than were the standard risk factors. In all three studies, CAC scoring improved diagnostic discrimination over conventional risk factors in the identification of persons with angiographic coronary disease.

More recently, large multi-center studies have been reported using fast CT for diagnosis of obstructive CAD in symptomatic persons (n = 1851), who underwent coronary angiography for clinical indications. Study prediction models were designed to be continuous, adjusted for age and sex, corrected for verification bias, and independently validated in terms of their incremental diagnostic accuracy. The overall sensitivity was 95%, and specificity was 66% for coronary calcium score to predict obstructive disease on invasive angiography. The logistic regression model exhibited excellent discrimination (receiver operating characteristic curve area of 0.84 ± 0.02) and calibration (chi-square goodness of fit of 8.95, p= 0.44) (38). Increasing the cut-point for calcification markedly improved the specificity, but decreased the sensitivity. In the same study, increasing the CAC cutpoint to greater than 80 decreased the sensitivity to 79% while increasing the specificity to 72%. In another large study (n = 1764) comparing CAC to angiographic coronary obstructive disease, use of a CAC score greater than 100 resulted in a sensitivity of 95% and a specificity of 79% for the detection of significant obstructive disease by angiography (39). Summing these 2 large studies (n = 3615) leads to an estimated sensitivity of 85%, with a specificity of 75%. There is some concern, due to study design, that these studies (similar to validation of many non-invasive cardiovascular tests) are subject to verification bias, which could raise the sensitivity and lower the specificity. A large study, evaluating consecutive symptomatic persons undergoing cardiac catheterization, addresses this concern. 2115 consecutive symptomatic patients (n = 1404 men; mean age = 62, SD ± 19 years old) with no prior diagnosis of CAD were included in this study. These patients were being referred to the cardiac catheterization laboratory for diagnosis of possible obstructive coronary artery disease, without knowledge of the CAC scan results. The scan result did not influence the decision to perform angiography. Overall sensitivity was 99%, and specificity was 28% for the presence of any coronary calcium being predictive of obstructive angiographic disease. With volume calcium score greater than 100, the sensitivity to predict significant stenoses on angiography decreased to 87% and the specificity increased to 79% (40).

Biologic Relevance of Coronary Atherosclerosis Progression

The extent of coronary calcium found on fast CT is broadly related to plaque burden, but there is a high degree of site-to-site variability in the presence and extent of calcium within any single atherosclerotic plaque. Pathology studies have shown that the extent of coronary calcium within plaques tends to be related to the presence of healed plaque ruptures (59). Moreover, vulnerable plaques tend to be those with less extensive calcium deposits frequently seen in a spotty distribution (59), a finding supported by intravascular ultrasound studies of patients with acute coronary syndromes (60). The biology of progression of calcium within atherosclerosis is complex, genetically-directed, and partially modified by drugs that have the potential to alter the fundamental biology of the calcification process. Statins, for example, can both inhibit and promote tissue calcification upon interaction with different types of vascular cells (61).

The associations between CAC progression and clinical cardiovascular risk factors are not well understood. Present data indicate that CAC progression is most strongly related to the baseline CAC score with only a limited relationship to standard cardiovascular risk factors (62,63).

Accuracy of Serial Coronary Calcium Assessments

Progression of coronary calcium is typically evaluated as a percentage of the baseline calcium score value. Early studies of the inter-test variability of CAC measurements indicated inter-scan variability as high as 25% to 50% of the calcium score value (62,64,65). More recently, imaging protocol refinements specific to electron beam CT scanning, including a reduction of the electrocardiographic gating interval to approximately 40% to 60% of the relative risk interval, and utilizing 3-mm slice thickness, have reduced the inter-test variability to 15% or less (66). The standard deviation of the interscan variability reported in the recent literature is approximately 10% (64). In contrast, annual CAC progression rates typically exceed 20% (62,64,65), thus permitting accurate determination of the presence or absence of true progression in individual patients across relatively short (1 to 2 year) time horizons. The ability to track CAC progression is most accurate in patients with intermediate and higher CAC scores because the absolute error in CAC measurement would approximate the actual CAC score in patients with low scores (CAC score 1 to 30), and even small changes in the absolute calcium score would be a relatively large fractional change.

Prognostic Relevance of CAC Score Changes

There have been 3 reports from the work of Raggi and colleagues on the relationship between changes in CAC score and outcomes (67–69). In these studies including a general population (67) analyzed by diabetic status (68) and treatment with statins (69), subjects who suffered an MI demonstrated an approximately 2-fold greater annual CAC increase than event-free survivors. In the presence of definite CAC score progression (greater than 15%/year), there was a significant increase in relative risk of myocardial infarction compared to subjects with stable scores. Notably, the finding of CAC progression increased the associated cardiovascular risk across all levels of CAC severity (69). Furthermore, the detection of stable CAC was associated with a low risk of cardiovascular events, even among those with extensive CAC. A major limitation of using calcium score progression as a marker of risk is that the positive predictive value appears to be low with substantial overlap among those with and without future events. Nonetheless, serial monitoring of atherosclerosis to refine risk prediction remains a potentially attractive hypothesis in need of ongoing investigation. Confirmatory reports from screening populations are needed to assess the strength and generalizability of these findings.

Modification of CAC Progression

Progression of CAC is frequently observed across modest (3 to 7 year) time horizons to a degree primarily related to the extent of baseline coronary calcification (70,71). Several pharmacological interventions, including statins and calcium channel blockers, have been associated with delayed progression of CAC. The earliest work primarily involved statins in observational study designs, including 2 published observational studies on the effect of reducing LDL cholesterol with statins in which CAC progression was found to be lower during statin treatment (72,73). These data, however, have been contradicted by 2 large statin clinical trials that failed to confirm this finding, including a placebo-controlled study using calcium scores (74) and a study of post-menopausal women treated to moderate versus intensive LDL cholesterol reductions using calcium volume scores (75). The CAC findings of the latter 2 studies are in contrast to the definitive reduction in cardiovascular risk associated with statin therapy and suggest that either longer periods of monitoring of CAC would be necessary to detect an effect of statins, that statins fundamentally alter the relationship between calcified plaque extent and cardiovascular outcomes, or that statins are affecting the noncalcified plaque and therefore no change is detectable by CAC measurement. Management of other cardiovascular risk factors, for example, hypertension or diabetes, has not been examined relative to the progression of coronary calcium.

Summary and Implications

Although progression of CAC can be detected using fast CT methods, its determinants are largely unknown and the relationship to clinical outcomes is still unclear. Because progression of CAC is not clearly modifiable through standard risk reducing therapies, and CAC measurement involves both costs and radiation exposure, clinical monitoring of CAC progression through serial fast CT scanning is not recommended at this time.

Summary and Conclusion

While several serious efforts to understand the cost-effectiveness of CAC measurement have been made, the Committee felt that models were not, and could not be, sufficiently well grounded in data to offer results that could be used for medical decision making or establishing policy at this time.

CAC Scores and Gender

Gender differences in utility and accuracy of imaging tests are typically related to differences in the epidemiology of coronary heart disease, with women having later onset of clinical CHD than men. Gender differences in incidence and prevalence of CAD are most marked in middle-aged populations, the typical target age group for CHD screening. In addition, emerging data suggest that there may be actual gender differences in the anatomy of atherosclerosis. Thus, it is important to consider gender-specific data when evaluating the potential uses of any new cardiac test.

Epidemiology

Women develop coronary atherosclerosis 10 years later than men, on average, and the occurrence of coronary calcification tracks with this later onset of CAD. These differences start to diminish at about age 60 (79). These gender differences in occurrence of coronary calcium support the association of CAC with coronary atherosclerosis and underline the importance of age- and gender-specific reference points for CAC scoring (80).

Risk Assessment

In general, studies of the use of coronary calcium as a component of the CHD risk assessment include fewer women than men. Studies also vary according to the analysis of women as a separate subgroup. Because many of the existing studies have included women and men of similar age (typically between ages 50 and 60), the reported 10-year event rates for women have been predictably lower than in men. Thus, many studies have been underpowered and included women at too low risk to show benefit of CAC screening exclusively in women.

Two studies included a large enough sample of women (81) or adequate numbers of elderly patients to reach conclusions about CAC testing in women. In a prospective, observational study by Raggi et al. (81), the relationship between CAC and all-cause mortality was analyzed by gender in 10 377 asymptomatic individuals, of whom 40% were women. The mean follow-up period was 5 ± 3.5 years. For women, the ROC C-statistic for the prediction of all-cause mortality by the NCEP ATP-3 Framingham risk calculator was 0.672 for women and increased significantly to 0.75 with data from CAC scores added to the prediction models (pless than 0.0001). This analysis is limited by the use of self-reported risk factors but showed similar relationships in the predictive ability of CAC in men and women. Mortality was determined using the Social Security National Death Index, thus these data are not specific to CHD events. In a study of older individuals (mean age = 71 years), the relationships between CAC score and incident myocardial infarction were similar in men and women (22) and remained significant in risk factor- and gender-adjusted models (22).

Summary

There are limited data broadly specific to women on the relationship between CHD outcomes and CAC. Existing data confirm an association between CAC scores and all-cause mortality and CHD events in elderly women. Future studies must include enough women within an appropriately high clinical risk stratum (at least intermediate Framingham risk) to be able to draw significant, clinically relevant conclusions specific to women.