Author + information
- Received April 6, 2015
- Revision received June 16, 2015
- Accepted June 24, 2015
- Published online November 1, 2015.
- Andre R.M. Paixao, MD∗,†,
- Colby R. Ayers, MS∗,†,‡,
- Abdallah El Sabbagh, MD∗,
- Monika Sanghavi, MD∗,†,
- Jarett D. Berry, MD, MS∗,†,‡,
- Anand Rohatgi, MD, MSCS∗,†,
- Dharam J. Kumbhani, MD, SM∗,†,
- Darren K. McGuire, MD∗,†,
- Sandeep R. Das, MD, MPH∗,†,
- James A. de Lemos, MD∗,† and
- Amit Khera, MD, MSc∗,†∗ ()
- ∗Department of Internal Medicine, University of Texas Southwestern Medical Center, Dallas, Texas
- †Division of Cardiology, University of Texas Southwestern Medical Center, Dallas, Texas
- ‡Department of Clinical Sciences, University of Texas Southwestern Medical Center, Dallas, Texas
- ↵∗Reprint requests and correspondence:
Dr. Amit Khera, Division of Cardiology, University of Texas Southwestern Medical Center, 5323 Harry Hines Boulevard, Dallas, Texas 75390-8830.
Objectives This study sought to assess the effect of coronary artery calcium (CAC) on coronary heart disease (CHD) risk prediction in a younger population.
Background CAC measured by computed tomography improves CHD risk classification in older adults, but the effectiveness of CAC in younger populations has not been fully assessed.
Methods In the DHS (Dallas Heart Study), a multiethnic probability-based population sample, traditional CHD risk factors and CAC were measured in participants without baseline cardiovascular disease or diabetes. Incident CHD—defined as CHD death, myocardial infarction, or coronary revascularization—was assessed over a median follow-up of 9.2 years. Predicted CHD risk was assessed with a Weibull model inclusive of traditional risk factors before and after the addition of CAC as ln(CAC + 1). Participants were divided into 3 10-year risk categories, <6%, 6% to <20%, and ≥20%, and the net reclassification improvement (NRI) was calculated. We also performed a random-effects meta-analysis of NRI from previous studies inclusive of older individuals.
Results The analysis comprised 2,084 participants; mean age was 44.4 ± 9.0 years. CAC was independently associated with incident CHD (hazard ratio per SD: 1.90, 95% confidence interval [CI] 1.51 to 2.38; p < 0.001). The addition of CAC to the traditional risk factor model resulted in significant improvement in the C-statistic (delta = 0.03; p = 0.003). Among participants with CHD events, the addition of CAC resulted in net correct upward reclassification of 21%, and among those without CHD, a net correct downward reclassification of 0.5% (NRI: 0.216, p = 0.012). Results remained significant when the outcome was restricted to CHD death and myocardial infarction and when individuals with diabetes were included. The NRI observed in this study was similar to the pooled estimate from previous studies (0.200, 95% CI: 0.140 to 0.258) and the addition of our study to the meta-analysis did not result in significant heterogeneity (I2 = 0%).
Conclusions CAC scoring also improves CHD risk classification in younger adults.
Coronary artery calcium (CAC) measured by computed tomography has emerged as a powerful predictor of coronary heart disease (CHD) (1). Compared with most other novel risk markers, CAC has a greater impact on clinical metrics of discrimination (i.e., C-statistics) and risk classification (i.e., net reclassification improvement [NRI]) (2,3). In the most recent cardiovascular risk assessment guidelines, CAC scanning received a Class IIb recommendation (i.e., may be considered) for individuals ages 40 through 79 years in whom a risk-based treatment decision is uncertain (4).
Three population-based studies have assessed the effects of CAC on CHD risk classification, with each study showing significant improvement in C-statistics and NRI when CAC was added to a traditional CHD risk factor model (5–7). Because the mean age in those cohorts ranged from 59 to 70 years, the effect of CAC on risk reclassification at the lower end of the age group targeted in recent guidelines has not been fully assessed.
We sought to determine the effects of CAC on CHD risk prediction in a young, multiethnic, probability-based population cohort and to compare our findings with data from older cohorts.
The DHS (Dallas Heart Study) is a multiethnic, probability-based population cohort of Dallas County adults, with deliberate oversampling of African Americans. Detailed methods of DHS have been described previously (8). All participants provided written informed consent, and the study protocol was approved by the Institutional Review Board of the University of Texas Southwestern Medical Center. Briefly, between 2000 and 2002, 2,969 participants, ages 30 to 65 years, completed a detailed in-home survey, laboratory testing, and multiple imaging studies. Of these 2,969 individuals, 226 did not have an interpretable CAC scan; 74 reported previous cardiovascular disease defined as myocardial infarction (MI), stroke, or coronary revascularization; 6 had end-stage renal disease; 88 had missing data on traditional risk factor(s); and 191 had incomplete follow-up for nonfatal events (Online Figure 1). Because participants with diabetes (n = 300) are thought to have high cardiovascular risk, they were not included in the primary analysis (9,10). The final study population comprised 2,084 subjects free of diabetes and cardiovascular disease who were followed for fatal and nonfatal CHD events.
Definitions and measurements
Race/ethnicity, history of cardiovascular diseases, individual medication usage, and smoking status were self-reported. Blood pressure, plasma glucose, and lipids were measured using standard methods (8). Diabetes was defined as fasting glucose ≥126 mg/dl, or nonfasting glucose ≥200 mg/dl, or reported diagnosis of diabetes coupled with the use of glucose-lowering medication. The study definitions of hypertension, metabolic syndrome, and family history of premature CHD used in the DHS have been previously published (11).
Electron-beam computed tomography measurements of CAC were performed in duplicate 1 to 2 min apart on an Imatron 150 XP scanner (Imatron Inc., San Bruno, California). Sufficient 3-mm slices were acquired (n = ∼40) to span the heart during a single inspiratory breath-hold (12). The 2 CAC scores were determined using the Agatston method and then averaged.
DHS participants were prospectively followed for fatal and nonfatal cardiovascular outcomes, and events were ascertained through December 31, 2010 (13). Fatal events were tracked using the National Death Index (14). Participants were contacted annually and assessed for interval nonfatal cardiovascular events. In addition, participants were tracked for hospital admissions using the Dallas Fort Worth Hospital Council Data Initiative database (DFWHC ERF Information Quality Services Center Regional Data [2000 to 2011]; Dallas-Fort Worth Hospital Council Education and Research Foundation, Information and Quality Services Center, Irving, Texas). This includes hospital admission data for 70 of 72 hospitals in the Dallas Fort Worth metroplex. Using these data sources, >90% of participants were followed for nonfatal events. Primary records were requested for all suspected cardiovascular events, and these events were separately adjudicated by 2 cardiologists blinded to CAC assessment and all study variables (13).
The outcome for the primary analysis was defined as the time–to–first event of the composite of CHD-related death, nonfatal MI, or percutaneous or surgical coronary revascularization. All revascularization events (coronary artery bypass graft surgery and percutaneous revascularization) occurring within the first 3 months following CAC scanning were excluded from the analyses to minimize the possibility that the CAC test result influenced the revascularization event. Secondary analyses were performed for the composite outcome of hard CHD (CHD-related death and MI).
Baseline demographic and clinical variables were compared between individuals ≤50 and >50 years of age using the chi-square test for categorical variables and Wilcoxon-rank sum for continuous variables. Kaplan-Meier cumulative-event curves were constructed for each CAC strata (i.e., 0, >0 to 10, >10 to 100, and >100 Agatston units) and unadjusted CHD events rates were compared using log-rank statistics. Risk categories were defined on the basis of a baseline Weibull model that included age, race, sex, systolic blood pressure, total and high density lipoprotein cholesterol, smoking status, blood pressure medication, and statin use. As in previous studies (5–7), categories were defined as 10-year risk <6%, 6% to <20%, and ≥20%. Because the mean follow-up for the DHS was marginally shorter than 10 years (9.2 years), the Weibull proportional hazards model was used to allow the computation of individual 10-year risk (15). In sensitivity analyses that included participants with prevalent diabetes at baseline, diabetes was added as a covariate. CAC was added to the baseline model as ln(CAC + 1). Sensitivity analyses were also performed using CAC as an ordinal variable (0, 1 to 10, 11 to 100, 101 to 400, and >400). The proportional hazards assumption was tested using Schoenfeld residuals, and hazard ratios are reported per SD and by CAC strata with CAC = 0 as the reference group. Cox models were used to generate adjusted survival estimates for the primary endpoint according to CAC strata. The same covariates included in the Weibull models were used in the Cox models. Harrell C-statistic was determined for the baseline model with and without CAC and the difference between the 2 was tested by bootstrapping (16). Sampling with replacement was performed in 1,000 bootstrap samples. The size of each sample was the same as the final population and significance was assessed via a Student t test with n – 2 degrees of freedom on these 1,000 samples. Cross-tabulation of risk categories for the models with and without CAC was performed for participants with and without CHD events and the NRI was calculated as described by Pencina et al. (17). Comprehensive age-stratified analyses were performed among individuals ≤50 and >50 years of age including Kaplan-Meier cumulative-event curves as well as separate Weibull and Cox models constructed for each age group. The interaction between age and CAC was tested by adding a multiplicative variable to the fully adjusted Weibull model. The same approach was used to test the interaction between CAC and race and CAC and sex. Statistical analyses were performed using the SAS software package version 9.2 (SAS Institute, Inc., Cary, North Carolina).
Review and meta-analysis
We conducted a PubMed and Cochrane Collaboration database search through December 2014 for studies assessing addition of CAC to a model composed of traditional CHD risk factors and reporting NRI. The following search terms were used and no restrictions were applied: “net reclassification improvement,” “net reclassification index,” “coronary artery calcium,” “coronary artery calcification,” “coronary calcium,” and “coronary calcification.” When multiple studies from the same cohort were available, only the study reporting on the more inclusive population was included (e.g., overall population vs. population at intermediate risk). The NRI was extracted from each study and SE were calculated using the method described by Pencina et al. (17). The summary NRI was computed using a random-effects DerSimonian-Laird model, and heterogeneity between studies was tested with the I2 statistics. DHS NRI was then compared with the summary NRI from previous studies using the z-score. Statistical analyses were performed using the OpenMeta[Analyst] software package (Tufts Medical Center, Boston, Massachusetts).
Coronary artery calcium in the Dallas Heart Study
The mean age of the study population was 44.4 ± 9.0 years, 56.2% were women, and 45.9% were African American (Table 1). Most participants had CAC scores under 10 (82.9%) and were classified as low CHD risk (<6% 10-year risk) on the basis of the Framingham Risk Score (81.6%). Participants >50 years of age had a higher burden of traditional risk factors and higher CAC scores (Table 1).
A total of 57 first CHD events occurred over a mean follow-up of 9.2 ± 1.3 years (7 CHD deaths, 30 nonfatal MI, 8 coronary artery bypass graft surgeries, and 12 percutaneous coronary revascularizations). Online Figure 2 shows Kaplan-Meier estimates for the primary composite CHD outcome stratified by CAC strata. Increasing CAC score categories were crudely associated with higher CHD event rates in the overall population and in each age strata (log-rank p < 0.001). Unadjusted hazard ratio (HR) for the primary outcome with increasing CAC categories considering CAC = 0 as the reference were as follow: 0 < CAC ≤10: 1.75 (95% confidence interval [CI]: 0.67 to 4.52; p = 0.252); 10 < CAC ≤100: 11.55 (95% CI: 5.02 to 26.56; p < 0.01); and CAC >100: 22.09 (95% CI: 9.83 to 49.63; p < 0.001). Each categorical CAC stratum over 10 Agatston units identified increasing risk for the primary composite of all CHD events, as well as for the composite of hard CHD events (excluding coronary revascularization) (Online Figure 3).
After multivariable adjustment, CAC remained independently associated with incident CHD (HR per SD: 1.90, 95% CI: 1.51 to 2.38; p < 0.001). Figure 1 shows adjusted survival estimates for the primary endpoint according to CAC strata from separate Cox models for the overall population, individuals ≤50 years, and individuals >50 years. A graded association between CAC strata and CHD event rates was observed (p < 0.001) (Table 2).
The C-statistic for the baseline traditional risk factor model was 0.86 (95% CI: 0.83 to 0.91). The addition of CAC significantly improved discrimination (C-statistic: 0.89, 95% CI: 0.86 to 0.93; p = 0.003) with the same pattern of results observed for the outcome of hard CHD (change in C-statistic: 0.03; p = 0.012).
The cross-tabulation of predicted 10-year risk for the baseline model and after the addition of CAC is shown in Table 3. Among participants with CHD events, CAC resulted in the net correct upward reclassification of 21.1% of individuals and, among participants without CHD events, the net correct downward reclassification was 0.5% resulting in a NRI of 0.216 (p = 0.012) (Table 3). Restricting the outcome to hard CHD events, the addition of CAC to the baseline model resulted in the net correct reclassification of 32.4% of events with a NRI of 0.313 (p = 0.014). Including participants with diabetes, 80 first CHD events occurred and the addition of CAC to the baseline model resulted in the net correct reclassification of 12.5% of events with a NRI of 0.135 (p = 0.043). In further sensitivity analyses, using CAC as an ordinal variable (0, 1 to 10, 11 to 100, 101 to 400, and >400) instead of a continuous variable, both the change in C-statistic (0.02, p = 0.016) and NRI (0.192, p = 0.023) remained significant.
In the fully adjusted models, CAC was independently associated with incident CHD among participants ≤50 and >50 years of age (HR per SD: 1.56, 95% CI: 1.20 to 2.03; p < 0.001, and HR per SD: 2.35, 95% CI: 1.58 to 3.49; p < 0.001, respectively). Adjusted incidence of CHD shows a consistent association between CAC strata and CHD risk across age subgroups (p < 0.001, each) (Figure 1). Further age-stratified analyses were limited by low numbers of events in each age group but suggest increasing CHD risk in each CAC strata >10 (Table 2) and a similar trend toward improvement in C-statistics and NRI among younger and older individuals. The change in C-statistics resulting from the addition of CAC to the baseline model was 0.04 with a p value of 0.112 for individuals ≤50 years and 0.05 with a p value of 0.056 among individuals >50 years. NRI in the younger group was 0.136 with a p value of 0.347 and 0.104 with a p value of 0.353 in the older group (Table 3). No significant interaction between age category and CAC on the risk of CHD was detected (p interaction = 0.363). There was also no significant interaction between sex and CAC (p interaction = 0.923) or race and CAC (p interaction = 0.530).
Coronary artery calcium and net reclassification improvement—Review and meta-analysis
Our initial search resulted in 47 potentially relevant studies. After title and abstract review, 17 studies were selected for full-text review and 3 met the pre-determined inclusion criteria (Online Figure 4) (5–7). Mean ages ranged from 59.4 to 69.6 years (weighed mean 62.0 ± 8.7 years). All studies reported significant NRI when CAC was added to a model composed of traditional CHD risk factors (Online Table 1). The pooled estimate for the NRI was 0.200 (95% CI: 0.140 to 0.261) and there was no evidence of heterogeneity between studies (I2 = 24%) (Figure 2A). Adding our study to the meta-analysis results in a summary NRI of 0.202 (95% CI: 0.146 to 0.258) and no significant heterogeneity (I2 = 0%) (Figure 2B). The difference between the DHS NRI and the pooled estimate from previous studies was not significant when tested by the z-score (p = 0.431).
In a young, multiethnic cohort, the addition of CAC to a model composed of traditional CHD risk factors significantly improved discrimination and risk classification. Our findings suggest that CAC can refine CHD risk prediction in populations younger than those previously studied and at the lower end of the age group targeted by recent guidelines (4).
Although there is abundant evidence in support of CAC as a CHD risk marker, several factors may impact its performance among younger individuals. There is a strong direct association between CAC and age such that CAC can be detected in <16% of 45-year-old individuals (18). Even among young individuals with CAC >0, the overall scores tend to be low and their impact on short-term risk classification has not been clearly established (19). The low prevalence of CAC in young populations suggests that a large number of individuals would need to be scanned to detect only a few with high CAC scores, which could limit the ability of CAC to discriminate and reclassify risk in this population. Furthermore, young individuals have a higher proportion of noncalcified plaque such that CAC may not accurately reflect the overall burden of coronary atherosclerosis (20). Conversely, the early identification of CHD risk allows early initiation of treatment, which may result in a higher number of life-years saved. Our findings suggest that despite potential limitations, CAC improves risk prediction in younger populations.
Few previous studies have assessed the association between CAC and adverse outcomes in young individuals (21–24). In a cohort of active military personnel, composed of 1,634 men (mean age 42 years), Taylor et al. (21) assessed the association between CAC and CHD events. Over a mean follow-up of 5.6 years, 22 CHD events occurred, and the independent association between CAC and CHD was only demonstrated in the highest Framingham Risk Score tertile. Tota-Maharaj et al. (22) followed asymptomatic patients referred for CAC scanning for a mean of 5.6 years. They reported that CAC was associated with all-cause mortality across age groups including among those <45 years of age. Similarly, LaMonte et al. (23) found that, after adjustment for sex, CAC was associated with CHD in the subgroup of individuals under the age of 40 years. Similar to these other studies inclusive of young adults, the effect of CAC on discrimination and reclassification metrics was not assessed. More recently, a study from the MESA (Multi-Ethnic Study of Atherosclerosis) (24) reported that CAC was similarly associated with CHD among individuals ages 45 through 54 years and 75 through 84 years. Metrics of discrimination and reclassification were not reported. Although these studies suggest that CAC is independently associated with CHD and all-cause mortality in young individuals, ours is the first study to fully assess the impact of CAC on clinically relevant metrics of discrimination and reclassification in an unselected population.
Although we show that CAC improves risk prediction in a population younger than has been studied previously, the minimal age at which CAC becomes informative has yet to be determined. Unfortunately we did not have sufficient numbers of CHD events to perform further age-stratified analysis. Given the overall low event rates among young individuals, this type of analysis will require data pooling from multiple different cohorts.
In this study, CAC scores >0 and ≤10 were not associated with higher CHD risk compared with CAC = 0 in multivariable analysis. As very few CHD events occurred in this group (n = 10), this analysis is underpowered and should be interpreted in the context of previous studies where even low CAC scores were associated with higher CHD event rates (25). The unadjusted HR of 1.75 (95% CI: 0.67 to 4.52; p = 0.252) for CAC >0 and ≤10 and the increasing risk with increasing CAC burden in higher CAC categories support a graded association between CAC and CHD risk in younger individuals.
In line with recent guidelines, participants with diabetes were considered to be at high CHD risk and were therefore not included in our primary analysis. Although the use of CAC among individuals with diabetes has been advocated by some investigators (26), participants with diabetes were excluded from the largest study on CAC and net reclassification (7). When diabetic individuals were added to our cohort, CAC still produced a significant NRI (0.135; p = 0.043).
The NRI reported in this study is consistent with previous reports inclusive of older participants, and a meta-analysis of all studies including the current study does not demonstrate significant heterogeneity. As seen in some of the previous studies, the improvement in NRI was in large part driven by correct upward risk classification, with minimal impact on downward risk classification (7). This finding supports recent guideline recommendations that primarily endorse the use of CAC scanning as a tool to upreclassify cardiovascular risk (10).
Our study extends the previous observation that CAC improves CHD risk prediction to a population with a mean age of 44 years. This suggests that CAC scanning may have clinical utility in individuals at the lower end of the age range targeted by the most recent guidelines.
Even with a mean follow-up period of over 9 years, few hard CHD events occurred in this relatively young cohort. The primary outcome chosen for this analysis therefore also included coronary revascularization events, but only those occurring after a blanking period of 3 months from the CAC measurement. The same pattern of results, however, was seen when only hard CHD events were included, and our primary outcome was similar to that used in other studies assessing CAC and NRI (2,7,27). Age-stratified analyses showed an independent association between CAC and CHD, but both the change in C-statistics and NRI were not statistically significant. However, the similar magnitude of both parameters in the younger and older groups and the absence of significant interaction between age and CAC suggest that there is no heterogeneity in the relationship between CAC and CHD among individuals under or over 50 years of age and that the lack of statistical significance may be a result of limited power. The final analysis excluded 226 participants with uninterpretable CAC scans. Obesity was more prevalent among those with uninterpretable scans (body mass index: 43 kg/m2 vs. 29 kg/m2; p < 0.001), which may limit the generalization of our findings to those with very high body mass index. NRI is a central metric in this study and has several limitations, one of them being the similar importance given to reclassification to intermediate- and high-risk categories. To make our findings comparable to other studies inclusive of older participants, our outcome was incident CHD and not atherosclerotic cardiovascular disease as recommended by the most recent guidelines (4). Further studies are needed to assess the performance of CAC using this more inclusive endpoint. As in previous studies, CAC was modeled as a continuous variable, a strategy that may not be immediately applicable in clinical practice. Studies are underway to create a regression-based calculator that integrates CAC and traditional risk factors.
The studies included in the meta-analysis have minor methodological differences such as slightly different composite CHD outcomes. One study also employed different risk categories (i.e., <10%, 10% to 20%, and >20% 10-year risk) (5). Although this may limit the precision of the NRI pooled estimates, significant net correct reclassification was reported by all studies and no significant heterogeneity was observed in the meta-analysis. Because only 4 studies were included in the meta-analysis, the I2 test may be underpowered.
In a population substantially younger than those previously studied (mean age: 44.4 ± 9.0 years), the addition of CAC to a traditional risk factor model improved risk discrimination and net correct reclassification. Further studies are needed to determine the optimal age to consider CAC scanning.
COMPETENCY IN MEDICAL KNOWLEDGE: CAC scores improve CHD risk prediction in younger individuals (i.e., mean age: 44 years).
COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS: CAC scanning can be considered in younger individuals when the decision to institute preventive measures is uncertain.
TRANSLATIONAL OUTLOOK: Further studies with pooled data from multiple cohorts are needed to determine the optimal age to consider CAC scanning.
For a supplemental table and figures, please see the online version of this paper.
Research reported in this publication was supported by the National Center for Advancing Translational Sciences of the National Institutes of Health under award Number UL1TR001105. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. Dr. Berry is a member of the Speakers Bureau for Merck & Co.; and has received consulting fees from Nihon Corporation. Dr. Rohatgi has received grant support from Merck & Co.; and is a member of the Speakers Bureau for Astra Zeneca. Dr. Kumbhani has received honoraria from the American College of Cardiology. Dr. de Lemos has received consulting fees from Amgen and Abbott Diagnostics; and grant support from Abbott Diagnostics. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- coronary artery calcium
- coronary heart disease
- confidence interval
- hazard ratio
- myocardial infarction
- net reclassification improvement
- Received April 6, 2015.
- Revision received June 16, 2015.
- Accepted June 24, 2015.
- American College of Cardiology Foundation
- Goff D.C. Jr..,
- Lloyd-Jones D.M.,
- Bennett G.,
- et al.
- Elias-Smale S.E.,
- Proença R.V.,
- Koller M.T.,
- et al.
- Erbel R.,
- Mohlenkamp S.,
- Moebus S.,
- et al.
- Grundy S.M.,
- Cleeman J.I.,
- Merz C.N.,
- et al.,
- for the Coordinating Committee of the NCEP
- Stone N.J.,
- Robinson J.G.,
- Lichtenstein A.H.,
- et al.
- Deo R.,
- Khera A.,
- McGuire D.K.,
- et al.
- Jain T.,
- Peshock R.,
- McGuire D.K.,
- et al.
- Paixao A.R.,
- Berry J.D.,
- Neeland I.J.,
- et al.
- McClelland R.L.,
- Chung H.,
- Detrano R.,
- Post W.,
- Kronmal R.A.
- Tota-Maharaj R.,
- Blaha M.J.,
- McEvoy J.W.,
- et al.
- LaMonte M.J.,
- FitzGerald S.J.,
- Church T.S.,
- et al.
- Tota-Maharaj R.,
- Blaha M.J.,
- Blankstein R.,
- et al.
- Agarwal S.,
- Cox A.J.,
- Herrington D.M.,
- et al.