Added value of CAC in risk stratification for cardiovascular events: a systematic review


Sanne A. E. Peters, Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Stratenum 6·131, Heidelberglaan 100, 3584 CX Utrecht, the Netherlands. Tel.: 0031 88 75 59380; fax: 0031 88 75 68099; e-mail:


Eur J Clin Invest 2012; 42 (1): 110–116


Background  Identification of individuals at high risk for cardiovascular disease (CVD) is important to initiate adequate treatment and to prevent future events. Moreover, identification of low-risk individuals is important to refrain from unneeded therapy. Current risk prediction models do not accurately predict the risk of CVD in individuals, and new markers have been sought to improve the risk assessment in individuals. Coronary artery calcification (CAC) is a marker of atherosclerosis that might improve current risk assessment when added to traditional risk factors.

Materials and methods  We performed a systematic review on PubMed search (1 February 2011) on studies reporting on the added value of CAC in risk prediction in asymptomatic individuals.

Results  Of 39 publications on CAC and CVD, nine studies were carried out in asymptomatic individuals. All studies showed an increase in area under the curve ranging from 0·05 to 0·20 when CAC was added to the risk model. Four studies reported on improvements of individuals in low-, intermediate-, and high-risk categories. Addition of CAC to the risk model resulted in a net reclassification improvement ranging from 14% to 30%, meaning that CAC measurement reclassified a substantial proportion of individuals into correct risk categories. This improvement was most pronounced in those at intermediate Framingham risk.

Conclusions  The available studies consistently showed that CAC scoring improves risk stratification in CVD risk categories when added to traditional risk factors only, especially among individuals at intermediate risk for CVD. Cost-effectiveness analyses together with a randomized controlled trial are needed before widespread introduction of CAC in clinical care.


Cardiovascular disease (CVD) is a leading cause of death worldwide and identification of individuals at high risk for CVD is a challenge, though essential for the prevention of future events [1]. Over the years, risk algorithms such as the Framingham Risk Score (FRS) in the United States and the Systematic Coronary Risk Evaluation (SCORE) in Europe have been developed to support clinicians in the identification of asymptomatic high-risk individuals, those without a history of CVD or diabetes mellitus, and to initiate individualized care [2–4]. Using these algorithms, individuals are typically classified into categories of low, intermediate or high 10-year risk for CVD and preventive strategies are initiated accordingly.

Both the FRS and the SCORE algorithms estimate an individuals’ absolute risk on CVD based on a number of traditional CVD risk factors. Although these algorithms adequately predict the absolute risks for future events in the extremes of the risk distribution, in those at very low or at very high risk, there is considerable room for improvement in those at intermediate risk, a group in whose treatment decisions are most uncertain [5].

For that reason, methods have been sought to improve current risk algorithms. Inclusion of non-invasive markers of atherosclerosis to current risk algorithms may be a method to identify high-risk individuals who are currently not considered as high-risk based on their levels of traditional risk factors. A widely used marker of atherosclerosis is coronary artery calcification (CAC), measured by computed tomography. Although CAC is strongly associated with risk for future cardiovascular events, an association with future events does not imply an improvement in risk prediction [6–9]. A major issue in the search for new prognostic markers is that a marker should provide information that adds to the information already available based on traditional risk factors [10]. Furthermore, a new marker should demonstrate its ability to correctly reclassify individuals into clinically meaningful risk categories that support clinicians to improve clinical decision-making. The net reclassification improvement (NRI) is a currently widely accepted method to quantify the incremental and clinical value of new prognostic markers [11]. The NRI reflects the amount of individuals that are correctly reclassified into clinically meaningful higher- or lower-risk categories with the addition of a new marker to the risk algorithm. A number of studies have assessed the additive value of CAC in risk prediction; however, a systematic review comprising all available evidence is lacking. Therefore, we set out to systematically review the currently available evidence on the incremental predictive value of CAC on top of traditional risk factors in risk assessment for CVD in individuals free of symptomatic CVD or diabetes mellitus.


PubMed MEDLINE ( was searched on 1 February 2011 for studies on CAC, and the risk of future events using the search string described in Table 1. Based on title and abstract, articles were selected that reported on the relation between CAC and future cardiovascular events. This relation could be quantified as an absolute risk, relative risk, hazard ratio or odds ratio and should be adjusted for traditional risk factors of CVD using multivariable regression models. Next, we selected publications that specifically studied the incremental value of a CAC measurement in risk prediction when added to a risk model consisting of traditional risk factors using either change in area under the curve (AUC) or reporting on absolute risk observed among those with high and low CAC values or the NRI. Furthermore, studies carried out in a population of symptomatic participants or in participants with diabetes mellitus were excluded.

Table 1.   Description of the search strategy used to identify publications of interest
DatabasePubMed on 1 February 2011
GeneralOnly English language
Domain#1: incidence[Title/Abstract] OR prognos*[Title/Abstract] OR predict*[Title/Abstract]
#2: reclassi*[Title/Abstract] reclassification[Title/Abstract] OR net reclassification index[Title/Abstract] OR net reclassification improvement[Title/Abstract] OR NRI[Title/Abstract] OR ROC-curve[Title/Abstract] OR AUC[Title/Abstract] OR area under the curve[Title/Abstract] OR receiver operating characteristic curve[Title/Abstract] OR c-index[Title/Abstract] OR receiver operating characteristic[Title/Abstract] OR concordance index[Title/Abstract] OR ROC[Title/Abstract] OR positive predictive value[Title/Abstract] OR negative predictive value[Title/Abstract] OR sensitivity[Title/Abstract] OR specificity[Title/Abstract]
Determinant#3: coronary artery calcification[Title/Abstract] OR CAC[Title/Abstract] OR coronary artery calcium[Title/Abstract] OR coronary calcification[Title/Abstract] OR calcification[Title/Abstract] OR coronary arterial calcification[Title/Abstract] OR arterial calcification[Title/Abstract] OR calcium score[Title/Abstract] OR coronary computed tomographic angiography [Title/Abstract] OR CCTA [Title/Abstract]
Outcome#4: risk*[Title/Abstract] NOT risk factors[Title/Abstract] OR cardiovascular disease[Title/Abstract] OR mortality[Title/Abstract] OR cardiovascular death[Title/Abstract] OR predict[Title/Abstract] OR coronary heart disease[Title/Abstract] OR cerebrovascular disease[Title/Abstract] OR stroke[Title/Abstract] OR myocardial infarction[Title/Abstract] OR events[Title/Abstract]
Search results#1: 1 297243 hits
#2: 632137 hits
#3: 27850 hits
#4: 1 587816 hits
#1 and #2 and #3 and #4: 242 hits

Publications were reviewed in duplicate (M.B. and S.A.E.P.), references were checked and added when appropriate (one study was added [8]) and discordant information was discussed (M.L.B.).


A flow chart of the study selection is provided in Fig. 1. Our primary search led to 242 potentially relevant studies. Based on title and abstract, 39 articles were selected that seemed to deal with the relation between CAC and future events. These 39 studies were evaluated using full text. Twenty-six studies were excluded because they were reviews, diagnostic studies, or because the association between cardiovascular risk factors and the development of CAC was studied. Thirteen studies seemed to be carried out in a population of which the majority of individuals were without known CVD or diabetes. In all these studies a positive association was found between increased CAC and future CVD. Four studies were conducted in symptomatic individuals and therefore excluded [12–15]. Nine studies were carried out in the correct domain and evaluated the incremental value of CAC in risk prediction for CVD (Tables 2 and 3) [8,16–23].

Figure 1.

 Flow chart of selection of articles. CAC, coronary artery calcification.

Table 2.   General characteristics of studies that evaluated the added value of CAC scoring in risk prediction
YearStudy nameFirst authorRegionNAge% MenFU (years)EndpointCAC score
  1. FU, follow-up; CAC, coronary artery calcification; CHD, coronary heart disease; MI, myocardial infarction; CVD, cardiovascular disease; HDL, high-density lipoprotein; SD, standard deviation; HNR, Heinz Nixdorf Recall; MESA, Multi-Ethnic Study of Atherosclerosis.

2008MESADetrano [16]USA672262 ± 10473·8MI, CHD death, definite angina, probable angina followed by coronary revascularizationCAC score > 0, %
Men; woman:
white: 70·4; 44·7
black: 52; 37,
Hispanic: 56·6; 34·8, Chinese: 59·2; 41·9
2008MESAFolsom [19]USA669845–84473·9CHD, stroke, or other
CVD death
Mean ln (CAC score + 1): 2·2 (SD: 2·5)
2004South Bay Heart WatchGreenland [8]USA102966 ± 8907·0Nonfatal MI or CHD death0: 31%
1–100: 31%
101–300: 17%
≥ 301: 21%
2001Raggi [21]USA67652 ± 10512·7MI, coronary deathNo event: 87 (SD: 234)
Event: 388 (SD: 696)
2009Wong [22]USA230356 ± 10624·4MI, cardiac death, late revascularization, stroke0: 53%
1–9: 8%
10–99: 19%
100–399: 12%
≥ 400: 8%
2010RotterdamElias-Smale [17]Europe202870 ± 6439·2Nonfatal MI and CHD death84
2010HNRErbel [18]Europe412959 ± 8475·0Coronary eventsEvents: 183
No event: 11
2010HNRMohlenkamp [23]Europe193457 ± 7 5·1CVD eventsMedian 1
2010MESAPolonsky [20]USA587862 ± 10465·8MI, CHD death, resuscitated cardiac arrest, definite or probable anginaNot reported
Table 3.   Description of the additive prognostic value of CAC score in risk prediction across studies
First authorRisk factors in baseline modelAUC without CAC (95% CI)AUC with CAC (95% CI)Thresholds intermediate riskYears risk predictionNRI, % (95% CI)
  1. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CAC, coronary artery calcification; NRI, net reclassification improvement; HDL, high-density lipoprotein; sbp, systolic blood pressure; dbp, diastolic blood pressure; FRS, Framingham risk score; ATP, Adult Treatment Panel; tc, total cholesterol; CHD, coronary heart disease; CVD, cardiovascular disease.

Detrano [16]Age, sex, ethnicity, smoking, diabetes, tch, HDL, sbp, dbp, statin use, or antihypertensive medicationMajor coronary: 0·79
Any coronary: 0·77
Major coronary: 0·83
Any coronary: 0·82
Folsom [19]Age, sex, ethnicity, smoking, diabetes, blood pressure, HDL, tc, statin useCVD: 0·77 (0·74; 0·80)
CHD: 0·77 (0·74; 0·80)
CVD: 0·81 (0·78; 0·83)
CHD: 0·82 (0·79; 0·85)
Greenland [8]FRS0·630·69   
Raggi [21]Age, sex, hypertension, smoking, hypercholesterolemia, diabetes0·710·84   
Wong [22]FRSHard CHD: 0·76
Total CHD: 0·75
Total CVD: 0·76
Hard CHD: 0·83
Total CHD: 0·86
Total CVD: 0·85
Elias-Smale [17]Refitted FRS0·720·7610–20%1014
Erbel [18]FRS and ATP IIIFRS: 0·68 (0·63; 0·73)
ATP III: 0·65 (0·61; 0·70)
FRS: 0·75 (0·68; 0·80)
ATP III: 0·76 (0·71; 0·81)
10–20%; 6–10%5FRS: 22·4; 19·6
Mohlenkamp [23]Age, sex, TC/HDL ratio, antihypertensive medication0·72 (0·65; 0·80)0·76 (0·70; 0·83)3–10%525
Polonsky [20]FRS, race/ethnicity0·76 (0·72; 0·79)0·81 (0·78; 0·84)3–10%525 (16;34)

The incremental value of CAC in risk assessment was evaluated using the AUC [16,19,21,22], the NRI [17,18,20,23], and/or by showing the hazard ratio for CVD based on FRS categories and CAC scores [8]. The AUC for a model without CAC ranged from 0·63 to 0·79 and increased from 0·05 to 0·20, whereas an AUC ranging from 0·69 to 0·86 was observed when CAC was added to the model. This increase in AUC was statistically significant in all, except for one [19], studies.

Five studies used an individualized approach based on absolute risks [8,17,18,20,23]. The South Bay Heart Watch study showed a joint relationship between increasing levels of CAC and increasing levels of FRS and future events among participants with a FRS ≥ 10%, but not in participants with a FRS < 10% [8].

Four recently published studies assessed the correct reclassification of participants across risk categories using the NRI [17,18,20,23]. It should be noted that one of these studies was carried out in a subgroup of the population described in an earlier article [18,23].

Addition of CAC to the FRS resulted in a statistically significant improvement in risk stratification in participants from the Multi-Ethnic Study of Atherosclerosis (MESA) study [20]. The NRI for those with an event was 0·23 and 0·02 for those without event {NRI: 0·25 [95% confidence interval (CI): 0·16–0·34]}. The NRI was significantly larger and more balanced for participants at intermediate risk for CVD (NRI 29% for those with event and 26% for those without event; NRI: 0·55, 95% CI: 0·41–0·69).

Similar results were found in participants from the Rotterdam study [17]. Addition of CAC to the FRS model led to an improvement in risk classification of 14% (P < 0·01). The largest proportion of reclassified individuals were seen in the intermediate Framingham risk group (n = 235, 11·6% of the total population). The proportions of reclassified individuals that were at low or high Framingham risk were 7·8% and 2·4%, respectively.

Moreover, results from a recent report from the Heinz Nixdorf Recall (HNR) study supported the use of CAC to improve risk prediction [18]. Using different thresholds to define the intermediate-risk category (10–20% or 6–20%), the NRI was 22% and 20%, respectively, when CAC was added to the FRS. The NRI in those at intermediate risk was 22% with intermediate-risk thresholds of 10–20% and 31% with thresholds for intermediate risk of 6–20%. Similar results were found when CAC scoring was applied in the National Cholesterol Education Panel Adult Treatment Panel (ATP) III model, indicating the potential use of CAC scoring in risk stratification, especially in those at intermediate risk.

Finally, a substudy among participants from the HNR study without an indication for statin therapy indicated that risk stratification improved with addition of CAC to the risk model (NRI: 25·1%) [23]. However, the event rate was very low (2·2%) and the authors therefore did not recommend using CAC scoring for risk stratification in participants without an indication for statins according to the Canadian guidelines [24].


This systematic review evaluated the added predictive value of CAC score in risk assessment for CVD events in asymptomatic individuals without diabetes. The majority of studies on this topic reported significant improvements in risk prediction when a risk model consisting of traditional risk factors was extended with the CAC score. This improvement was primarily found in individuals at intermediate risk for CVD.

Although the majority of studies included in this review solely reported on changes in AUC as a measure for incremental predictive value with CAC scoring, it has been recognized that the AUC is an imperfect measure to assess incremental value of a new marker and other methods are strongly recommended [11,25]. A major disadvantage of the AUC is that it does not account for the clinical consequences of changes in predicted risk. The AUC will improve if the predicted risk of individuals that experience an event increases and the predicted risk of individuals that do not experience an event decreases, even if these changes are very minor and do not lead to shifts to other risk categories. Yet, changes in predicted risk, irrespective of the size, are only clinically relevant when these changes lead to other treatment decisions. The NRI quantifies the ability of a new marker to reclassify individuals into clinically meaningful risk categories [11]. Yet, it should be acknowledged that the NRI findings depend on the actual risk thresholds used and that the use of the NRI is only relevant in settings where clinically meaningful risk thresholds exist. Nevertheless, a rapidly increasing number of studies on risk reclassification have recognized the benefits of the NRI, and a number of studies have now been carried out using the NRI as a measure to quantify improvement in predictive accuracy. These studies have evaluated the incremental value potential predictors for CVD including multiple biomarkers [26,27], high-density lipoprotein cholesterol [28], heart rate [29], C-reactive protein [30], hbA1c [31], ankle brachial index [32], flow-mediated dilation [33] and carotid intima-media thickness [34]. Although difficult to compare, the positive results of the majority of these studies illustrate the potential gain in risk classification with the addition of new markers to current risk prediction models in clinical practice.

The studies that used the NRI to assess the incremental value of CAC scoring in risk stratification all demonstrated that CAC scoring improved risk classification of individuals, especially for those in the intermediate-risk group. However, major differences between these studies including differences in study population, duration of follow-up, event rates, definition of risk categories and CAC thresholds, end point definition and risk factor selection for the traditional risk factor model may have affected the results of the individual studies and make direct comparison of the results between studies difficult. Given these differences, the next step would be to investigate the additional prognostic value of CAC through a meta-analysis based on pooled individual patient data from multiple large population cohorts of asymptomatic individuals. In this approach, a much larger sample size and event rate are expected and the differences between study populations could be adequately controlled for. If such pooled analysis is confirmative for the use of CAC in risk assessment, an economic evaluation should be performed to evaluate changes in lifetime costs, benefits and cost effectiveness of life years gained when switching from traditional risk assessment to traditional risk assessment plus CAC scoring. Moreover, one may argue that a randomized controlled trial (RCT) needs to be conducted to prove that compared with care as usual, a risk algorithm including CAC leads to reduction in cardiovascular events.

Despite congruent results on the value of CAC scoring in risk prediction, there is considerable debate on its use in clinical practice. High costs and harms of radiation exposure are issues that need to be well considered before widespread routine use [35]. Of additional concern, two recent studies have shown that individuals with advanced stages of atherosclerosis, obstructive stenosis and vulnerable plaques could have a low or undetectable CAC score [8,36]. Therefore, it may be more appropriate to estimate risk for CVD in individuals without detectable CAC based on traditional risk factors only. Given the high costs, potential harms and potential for misclassification, others have suggested to treat all patients at intermediate risk, irrespective of additional tests [37]. Yet, again a formal cost-effectiveness analysis and RCT need to be conducted.

When interpreting the results of the present review, it needs to be considered that we restricted our search to PubMed; therefore, we may have missed manuscripts dealing with this topic that were not accessible in PubMed. In addition, we also may have missed some manuscripts while reviewing title and abstract. However, we crosschecked the references of the included manuscripts and added new manuscripts on this issue when appropriate.

In conclusion, studies included in this systematic review show that risk assessment for CVD could be improved materially when the CAC score is added to risk prediction models consisting of traditional risk factors only. This improvement is most pronounced in intermediate-risk individuals in whose treatment decisions are often uncertain. Cost-effectiveness analyses together with a RCT are needed before widespread introduction of CAC in clinical practice.


Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Stratenum 6·131, Heidelberglaan 100, 3584 CX Utrecht, the Netherlands (S. A. E. Peters, M. Bakker, H. M. den Ruijter, M. L. Bots).