Validation of an imaging based cardiovascular risk

score in a Scottish population

Remko Kockelkoren*a, Pushpa M Jairama,d, John T. Murchisonb, Thomas P.A. Debrayd,e, Saeed

Mirsadraeeb,c, Yolanda van der Graafd, Pim A. de Jonga, and Edwin J.R. van Beekb,c

a University Medical Center Utrecht, Department of Radiology, Heidelberglaan 100, 3584 CX, Utrecht, The Netherlands

b Royal Infirmary Edinburgh, Department of Radiology, 51 Little France Dr, Edinburgh EH16 4SA, Edinburgh, Scotland

c Clinical Research Imaging Centre, The Queen’s Medical Research Institute, University of Edinburgh, 47 Little France Cres,

Edinburgh EH16 4TJ, Edinburgh, Scotland

d Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Universiteitsweg 100, 3584 CG,

Utrecht, The Netherlands

e The Dutch Cochrane Centre, University Medical Center Utrecht, Heidelberglaan 100, 3584 CX, Utrecht, The Netherlands

Email addresses: r.kockelkoren@umcutrecht.nl (R. Kockelkoren), pmjairam@gmail.com (P.M. Jairam),

John.Murchison@nhslothian.scot.nhs.uk (J.T. Murchison), t.debray@umcutrecht.nl (T.P.A. Debray),

Saeed.Mirsadraee@ed.ac.uk (S. Mirsadraee), Y.vanderGraaf@umcutrecht.nl (Y. van der Graaf), p.dejong-8@umcutrecht.nl

(P.A. de Jong), edwin-vanbeek@ed.ac.uk (E.J.R. van Beek)

*Corresponding Author:

Remko Kockelkoren, MD

University Medical Center Utrecht, Department of Radiology

Room E01.132 - PO Box 85500

3508 GA Utrecht, The Netherlands

r.kockelkoren@umcutrecht.nl

Tel +31 88 755 0846 - Fax +31 88 756 9589

1

Abstract

Objectives: A radiological risk score that determines 5-year cardiovascular disease (CVD) risk using

routine care CT and patient information readily available to radiologists was previously developed.

External validation in a Scottish population was performed to assess the applicability and validity of

the risk score in other populations.

Methods: 2915 subjects aged ≥40 years who underwent routine clinical chest CT scanning for non-

cardiovascular diagnostic indications were followed up until first diagnosis of, or death from, CVD.

Using a case-cohort approach, all cases and a random sample of 20% of the participant’s CT

examinations were visually graded for cardiovascular calcifications and cardiac diameter was

measured. The radiological risk score was determined using imaging findings, age, gender, and CT

indication.

Results: Performance on 5-year CVD risk prediction was assessed. 384 events occurred in 2124

subjects during a mean follow-up of 4.25 years (0-6.4 years). The risk score demonstrated reasonable

performance in the studied population. Calibration showed good agreement between actual and 5-

year predicted risk of CVD. The c-statistic was 0.71 (95%CI:0.67-0.75).

Conclusions: The radiological CVD risk score performed adequately in the Scottish population

offering a potential novel strategy for identifying patients at high risk for developing cardiovascular

disease using routine care CT data.

Keywords: Cardiovascular disease; Vascular calcification; Epidemiology; Risk prediction;

Multidetector Computed Tomography

2

Key points

- A model was previously developed determining CVD risk using routine-care CT data

- External validation was performed in a Scottish population to determine applicability

- The radiological risk score showed adequate performance

- It offers a novel strategy for identifying patients with high CVD risk

3

Abbreviations

DSC = descending aorta

ICD = international classification of disease

LAD = left anterior descending artery

MV = mitral valve

PROVIDI = PROgnostic Value of unrequested Information in Diagnostic Imaging

4

Introduction

In Scotland, approximately 40% of all premature deaths are caused by cardiovascular disease (CVD)

with coronary heart disease and stroke being the most prevalent.[1] Even though the cardiovascular

mortality rate has dropped by more than 40% in the last 10 years it remains high compared to the

rest of the UK and Western Europe[2,3]; the 2010 premature death rates for coronary heart disease

in Scotland were 37% higher for men and 60% for women than in England.[4] To address the high

CVD burden and identify high risk patients, several risk scores have been developed over the past

few decades. Well known examples are the QRISK2[5] and ASSIGN score,[6] which were developed in

the UK and Scotland, respectively. These scores provide the 10 year risk of developing CVD in the

general population and are based on traditional risk factors like age, gender, high blood pressure,

and also social deprivation and family history.

Traditional risk scores such as QRISK2 and ASSIGN are considered moderately successful in

predicting future CVD events since corresponding event rates are predominantly driven by surrogate

measures of the atherosclerotic burden.[7] It is for that reason that there is substantial heterogeneity

between traditional risk and actual atherosclerosis burden. In this regard, vascular calcifications, as

detected on computed tomography (CT), may provide a more accurate measure of atherosclerosis

burden and offer an improved assessment of personalized risk.[8] While the predictive qualities of

these imaging markers are increasingly recognised in the medical literature,[9] they do not have a

defining role in CVD risk prediction in contemporary guidelines because their therapeutic

consequences are still unclear.

The total number of chest CT examinations is steadily growing, due to technical

developments, such as the implementation of ultra-low dose chest CT examinations[10] and new

clinical indications such as population lung cancer screening.[11] As a result, information on imaging

markers is increasingly available in routine care for a growing number of patients. Recently, a risk

score was developed[12] for detecting subjects at increased risk for CVD using incidental findings

5

from chest CT examinations. This score includes traditional risk factors like age and gender combined

with imaging results such as cardiovascular calcifications and cardiac diameter. Although initial

validation of the risk score showed promising results in a Dutch population, further validation is still

required to assess whether the risk score can be applied more broadly across different (but related)

patient populations. The relatively high CVD burden in Scotland provides ample opportunity, not only

to validate the risk score, but the potential to provide a novel radiological method of identifying

(previously undiagnosed) high-risk patients. In this study we validated whether this radiological risk

score is able to detect and accurately stratify individuals from a Scottish population into clinically

relevant CVD risk categories.

6

Methods

Study Population

The study population (Fig. 1) consisted of 2915 subjects aged ≥ 40 years who underwent routine

clinical chest CT scanning between January 2008 to July 2008 for diagnostic indications other than

cardiovascular diseases in the participating hospitals (Royal Infirmary of Edinburgh, Edinburgh;

Western General Hospital, Edinburgh; St John’s Hospital, Livingson) in the Lothian Region, Scotland.

These hospitals serve approximately 750,000 people out of a total 5.2 million population in Scotland.

This study population provided an overall comparable Caucasian population with a slightly

increased cardiovascular risk profile as compared to the Dutch cohort in which the radiological

cardiovascular score was developed.[4] Patients with a previous diagnosis of primary lung cancer

(including mesothelioma or distant metastatic disease from other types of cancer (excluding

hematologic malignancies) at baseline were excluded (n=740). These patients were excluded because

it is highly unlikely that detection of unexpected image findings will alter clinical decision making in

patients with such a poor prognosis. Also excluded were patients yielding prior history of CVD or

subjects with a CT referral indication directly related to (suspected) cardiovascular pathology (n=51),

to ensure that the evaluated imaging findings were truly “incidental”. After exclusion, the full

baseline validation cohort consisted of 2124 subjects who were considered for analyses.

The study was approved by the Research Ethics Committee of the Royal Infirmary of

Edinburgh (Ref: NR/1404AB6). Written informed consent was waived for all patients because of the

retrospective design and absence of intervention of the study. This study is in compliance with the

declaration of Helsinki and was performed in accordance with relevant guidelines and regulations.

Cardiovascular events and follow-up

7

Subjects who developed a CVD event during follow-up were identified as cases. CVD events were

defined, using the international classification of disease (ICD) 10 definitions, as all diagnosis of

coronary artery disease (Angina, (sub)acute myocardial infarction, acute or chronic ischaemic heart

diseases), cerebrovascular events (ischemic stroke, haemorrhagic stroke, and transient ischemic

attack), peripheral artery disease (intermittent claudication), and heart failure.[13] Data on fatal and

non-fatal CVD events were obtained from the National Health Service (NHS) registry using ICD 10

codes (Supplementary Table S1). The Information and Statistics Division (ISD) of the NHS in Scotland

has linked information on all Scottish hospital inpatient discharges (1981 – 2015) and death records

(1981 – 2015) using probability matching.

For all patients we determined the entry date, which was the date subjects underwent chest

CT examination. The censor date was determined as the date on which they developed an event as

specified above, the date the study period ended (April 1st 2014) or date of death, whichever

occurred first.

Sample selection and study design

We used a case-cohort approach as introduced by Prentice using all cases and a subcohort

resembling an approximately 20% random sample from the full validation cohort (n=2124) at the

beginning of the study.[14] The cases together with the subcohort define the study population. A

major advantage of this design is that it enables survival analyses without the need to score the chest

CT scans for the full cohort. Because this implies that cases are inherently overrepresented, we

adjusted all analyses for the sampling fraction such that estimates of model performance are

applicable to the full cohort. Previous studies have suggested that case-cohorts with sampling

fractions above 10% yield similar to the full cohort analysis.[15]

CT scanning and scoring of CT characteristics

8

All chest CT examinations were obtained using multi-detector CT of different vendors according to

the prevailing routine clinical protocols of the participating hospitals. When study subjects

underwent multiple chest CT examination during follow-up, the findings from the first examination

were used. All types of CT (including non-contrast) were considered eligible. Slice thicknesses had a

range of 1.25 mm to 8 mm and varied according to the chest CT indication and corresponding

protocol.

CT examinations were graded by a qualified medical practitioner with 2 years of chest CT

experience who was trained on using the radiological risk score under the supervision of an

experienced chest radiologist. The training consisted of scoring 50 randomly selected patients who

were not part of the study population. Weighted kappa for inter-observer reliability regarding

calcifications in the training set was 0.90.

CT examinations were graded for calcifications in the Left Anterior Descending (LAD) artery,

descending thoracic aorta (DSC), and the mitral valve (MV) (Fig. 2) using a semi quantitative grading

system (Table 1). The upper margin of the descending aorta was defined as the point where the left

subclavian artery originates from the aortic arch, whereas the lower margin was defined as the level

where the diaphragm becomes visible. The cardiac diameter was measured at the point where the

transverse cardiac silhouette reached its maximum. The cardiac diameter was measured using a

measurement tool, integral to the DICOM reviewing software (Carestream Medical and Dental

Imaging Systems, Carestream Health, Inc., NY, USA). The grading and measurement techniques have

been published and validated previously.[16] The observer was blinded for outcome during scoring.

Statistical analysis

The determinants used in the radiological risk score are: Age, gender, CT indication, cardiovascular

calcifications (LAD, MW, DSC) and cardiac diameter. Development of the model was previously

described.[12] The complete risk model and source code is available in the supplementary data file.

9

All analyses are based on the entire validation cohort, and missing data arising due to the

implementation of a case-cohort design were accounted for by use of multiple imputation. [17] This

strategy also allowed to account for missing values (LAD, DSC, MV, cardiac diameter) due to

unavailable chest CT images in the sub-cohort (N= 5 cases and 11 non-cases).

We focus on assessing the predictive accuracy of the radiological cardiovascular risk score for

predicting 5 year incidence of CVD. Because the follow-up in our cohort was limited to 6.4 years, the

observed cumulative baseline hazard function was extrapolated to approximate the 10 year risk

using the flexible survival approach of Royston and Parmar (Supplementary Table S2).

CVD event rates were calculated by dividing the total number of CVD events by the total

number of person years at risk.

The predictive accuracy of the score was assessed by determining its discrimination and

calibration. Discrimination is the ability of the score to distinguish patients who are at high risk of

developing CVD from those at low risk. The discrimination was evaluated by calculating the c-statistic

in each imputed dataset, and applying Rubin's rules.[18] We also visualized the score's discrimination

in a survival plot where we stratified the predicted 5 year CVD risk by categories used in current CVD

guidelines for initiating treatment: low (<10%), intermediate (10-20%), and high (>20%) risk of CVD.

[19]

Calibration was visually evaluated in a calibration plot comparing 5 year observed and

predicted risks.

Analyses were performed using R-project software package, version 3.1.1 (www.r-

project.org).

10

Results

Baseline characteristics of the validation (sub)cohort as well as the original PROVIDI (PROgnostic

Value of unrequested Information in Diagnostic Imaging) subcohort can be found in Table 2.

Compared to the PROVIDI subcohort, the Scottish subcohort showed a higher percentage of severe

coronary artery and descending aorta calcification and were on average 5 years older. Lung disease

was the most frequent CT-indication (38%) followed by suspicion of pulmonary malignancy (32%).

Cases were on average 71 years old and generally showed more severe calcifications and larger

cardiac diameter.

A total of 384 CVD events occurred during a mean follow-up period of 4.3 years (0-6.4 years).

1183 (56%) patients were followed up for at least 5 years. The overall annualized CVD event rate (per

1000 person years) was 42.5. CVD event rates for gender and age are shown in Table 3. Event rates

were higher for men (45.7) compared to women (39.4) and increased substantially with higher age. A

total of 26 CVD events occurred in the 456 patients without calcifications in the LAD, descending

thoracic aorta and mitral valve. The predicted and observed 5 year CVD risk in this patient group was

5.4% and 5.6% respectively.

The score showed a good discriminative ability with a c-statistic of 0.71 (95% Confidence

Interval: 0.67-0.74) which is comparable to the discrimination previously found in the Dutch hospital

population. Discrimination is visualized with a survival plot where the cohort was stratified for <10%,

10-20%, and>20% risk of cardiovascular disease in 5 years (Fig. 3). Calibration of 5 year predicted

risks also appeared adequate, although there was some evidence of overfitting. In particular, we

found over-estimation between the probability range of 0.6 and 0.8 and underestimation between

0.4 and 0.6. (Fig. 4). The 5 year risks of the low (<10%), intermediate (10-20%), and high (>20%)

predicted 5y CVD risk groups were 8.6%, 14.9%, and 31.7% respectively (Table 4).

11

Discussion

In this case-cohort study of 2124 Scottish subjects who underwent a CT examination for diagnostic

indications we showed that the radiological risk score can be used to reliably predict the 5 year

probability of developing cardiovascular disease in a Scottish population. Adequately identifying

patients with high cardiovascular risk using readily available CT image findings in a previously

undiagnosed population is of major value. Early detection of high risk individuals may facilitate

targeted preventative management interventions, such as institution of statin and/or thrombocyte

aggregation inhibitors therapy.

With prevention and treatment plans, including the use of traditional cardiovascular risk scores like

QRISK2 and ASSIGN, mortality rates for CVD have fallen steadily in Scotland over the last 10 years by

more than 40%.[2,3] In spite of these improvements, the cardiovascular event rates remain relatively

high compared to the rest of the UK and western Europe.[4] This dissimilarity was also supported by

our results as the annualized CVD event rate was 42.5 in Scotland compared to 29.3 in the original

Dutch cohort, leaving room for further optimization of preventative and therapeutic interventions.

The radiological risk score validated in this study could help drive down these rates first and

foremost through the identification of patients with high cardiovascular risk. Patients with a high

radiological risk score can either have a high risk based on traditional risk scores, have not been

previously recognized by traditional risk scores or have a low or absent traditional risk. In patients

with a high traditional as well as a high radiological risk the importance of preventive treatment or

behavioural changes can be further emphasized. Confronting patients with calcification burden on

their CT examinations is thought to be a strong motivator for lifestyle modification and therapy

adherence.[20]

A potential second step could be preventive management in high radiological risk patients

with or without high risk in traditional scores. While patients with a high traditional risk will receive

12

treatment in current guidelines, patients with a high radiological risk and a low or absent traditional

risk would not. However, while a large portion of the actual cardiovascular risk can be predicted with

the use of traditional risk factors, a substantial amount of these patients will develop CVD in the

absence of these risk factors.[21] Radiological findings, like cardiovascular calcifications, are a

measure of subclinical target organ damage, which can provide a better estimation of CVD risk

compared to traditional risk assessment in these individuals.[22] Patients with subclinical target

organ damage on CT without modifiable risk factors have increased hazards for CVD morbidity and

mortality,[29] but have no indication for prevention therapies in the current guidelines. [19] Recently,

a guideline was published by the Society of Cardiovascular Computed Tomography (SCCT) and the

Society of Thoracic Radiology (STR) on coronary calcium scoring in noncardiac chest CT scans.[24]

They emphasize the importance of including these imaging markers, measured volumetrically or

visually, into the clinical decision making process. Still, randomized trials in which interventions can

be based on the imaging biomarker are required to determine whether treatment should be given to

patients based on a high radiological risk score and whether this will result in CVD risk reduction.

Interestingly, in this study 46% of all patients were in the 5 year high CVD risk (>20%) group. Even

though current treatment guidelines are based on 10 year CVD risk one could argue that for this

specific patient group 5 year CVD risk would be a sufficient/suitable indicator for treatment.

In the Multi-Ethnic Study of Atherosclerosis (MESA), absence of coronary calcifications was

related to a low event rate even in patients with many cardiovascular risk factors. [25] Comparable

results were found in this cohort in patients without calcifications in the LAD, descending aorta and

mitral valves both in the predicted as well as the observed risk (5,4% and 5,6% respectively). Results

like that of the MESA and this study can potentially play a role in reclassifying patients with a high

risk based on traditional risk scores but without visible calcifications on chest CT.

The number of chest CT examinations for diagnostic and screening purposes are continuing

to rise with a study performed in America between 1996 and 2010 demonstrating a 30% increase.

[26] In addition to the direct diagnostic impact, one needs to consider potential additional

13

information from a CT scan that may benefit the patient. One such method is to evaluate additional

findings, and the radiological risk score is one way in obtaining greater health information from a

single CT investigation.[27] The effectiveness of the radiological CVD risk score will, however, not be

limited by the number of chest CT examinations, but will rather depend on its applicability in routine

use. Thus, while the risk score is relatively easy to determine and use, applicability/impact studies are

still required to evaluate whether or not it is feasible for use in daily practice. The addition of

computer aided detection of cardiovascular calcifications could potentially improve applicability and

facilitate future use of the risk score.[28]

The strengths and limitations of the application of cardiovascular risk prediction based on incidental

findings in Chest CT examinations as well as the development and validation of the risk score have

been discussed previously.[29]

In this study, we found that the risk score yielded good predictive accuracy in a Scottish

population with a higher CVD burden compared to the population it was developed in. [4] In

particular, the score achieved a c-statistic of 0.71 and properly distinguished between low,

intermediate, and high risk patients. Our findings are consistent with the previous external validation

study and suggest that the radiological risk score could contribute towards cardiovascular disease

prevention in Scotland by identifying high risk patients using incidental findings on chest CT.

Furthermore, because performance of the score was not affected by the heterogeneity between the

Dutch and Scottish population, it seems that the incorporation of CT imaging results may help to

improve the generalizability of risk scores predicting the incidence CVD in individual patients.

We used the recently published ‘TRIPOD method for developing and validating prognostic

multivariable models’ as a guideline for writing this paper.[30] Still, some limitations of our study

need to be considered.

The radiological data was scored by a single researcher after extensive training on chest CT

scans. While this could potentially introduce bias we feel that the previously assessed reproducibility

14

of the score[16] and the results of the training scans are substantial enough to allow for a single

observer.

The risk score was validated in a generally Caucasian population. For further implementation

validation in other geographical areas and ethnicities would be desirable, as there is a significant

difference in coronary artery calcification based on large population studies, such as the MESA.[31]

The average age of participants in both the original Dutch and the Scottish validation study was

relatively high. Because of this it could be argued that much of the cardiovascular risk is explained

solely by age. To test for this possibility, the radiological risk score was compared in the derivation

study to a score with only gender and age which showed that the risk score had a substantially better

performance.[12]

No comparison with traditional risk scores was made. The radiological risk score differs from

traditional cardiovascular risk score’s like the Framingham risk score in that is based solely on data

readily and freely available to the radiologist (i.e. patient characteristics, indication and imaging

measurements). Therefor no other patient data was obtained and no direct comparison to traditional

CVD risk scores could be made. We were able to compare the discriminate ability of the risk scores

and found that the c-statistic for the radiological risk score was comparable to that of the

Framingham and SCORE risk scores. [7]

For the radiological risk score, calcifications are scored using categories of severity that are

used for both thick and thin slice scans. Because of this, extent of calcification can be somewhat over

or underestimated depending on the slice thickness. Furthermore, to increase applicability,

calcifications are scored visually instead of volumetrically. The predictive properties of volumetric

coronary calcium scoring are well known and a comparison between the radiological risk scores and

volumetric coronary calcium scoring is something we want to look into in future studies. [8]

Finally, the calibration plot for 5 year CVD incidence showed mild overestimation from 20 to

40% CVD risk and underestimation of patients with higher than 50% risk. Overestimation in the 20%

15

predicted CVD range could cause some misclassification from the intermediate (10-<20%) risk group

to the high risk (>20%) group. The tendency for underestimating patients with the highest risk was

already visible in the original validation study. However, the relevance of underestimating patients

with observed risks of 50-70% is debatable. It is unlikely this would influence clinical decision making,

as these patients will be provided with maximal supportive care according to most guidelines.[19]

In conclusion, the radiological risk score for predicting the 5 year incidence of cardiovascular risk in a

Scottish population performed adequately. The score adequately identified patients at low,

intermediate, and high risk of cardiovascular disease in 5 year CVD incidence in individual patients.

Application of the radiological risk score can help identify patients with high cardiovascular risks

using readily available data from chest CT examinations.

Source of funding

This research did not receive any specific grant from funding agencies in the public, commercial, or

not-for-profit sectors.

Conflicts of interest

The authors have no potential conflicts of interest related with this article.

16

References[1] NRS Scotland, Under 75s age-standardised death rates for all causes and certain selected

causes , Scotland , 1994 to 2014, (2015) Available at: http://www.nrscotland.gov.uk/files//.

http://www.nrscotland.gov.uk/files//statistics/age-standardised-death-rates-esp/2014/age-

stand-death-rates-table2-1994-2014.pdf.

[2] Information Services Division, Publication Report Heart Disease Statistics Update, (2014).

[3] I.S. Division, Publication Report Stroke Statistics Update, (2014).

[4] H.F.H.P.R.G. British, U. of O. Department of Public Health, Coronary heart disease statistics A

compendium of health statistics 2012 edition, (2012). http://scholar.google.com/scholar?

hl=en&btnG=Search&q=intitle:Coronary+heart+disease+statistics#0 (accessed August 15,

2014).

[5] J. Hippisley-Cox, C. Coupland, Y. Vinogradova, J. Robson, R. Minhas, A. Sheikh, P. Brindle,

Predicting cardiovascular risk in England and Wales: prospective derivation and validation of

QRISK2., BMJ. 336 (2008) 1475–1482. doi:10.1136/bmj.39609.449676.25.

[6] M. Woodward, P. Brindle, H. Tunstall-Pedoe, Adding social deprivation and family history to

cardiovascular risk assessment: the ASSIGN score from the Scottish Heart Health Extended

Cohort (SHHEC)., Heart. 93 (2007) 172–176. doi:10.1136/hrt.2006.108167.

[7] G.C.M. Siontis, I. Tzoulaki, K.C. Siontis, J.P.A. Ioannidis, Comparisons of established risk

prediction models for cardiovascular disease: systematic review, BMJ. 344 (2012) e3318–

e3318. doi:10.1136/bmj.e3318.

[8] T.S. Polonsky, R.L. McClelland, N.W. Jorgensen, D.E. Bild, G.L. Burke, A.D. Guerci, P.

Greenland, Coronary artery calcium score and risk classification for coronary heart disease

prediction., JAMA. 303 (2010) 1610–1616. doi:10.1001/jama.2010.461.

[9] P.C. Jacobs, M.J. Gondrie, W.P. Mali, A.L. Oen, M. Prokop, D.E. Grobbee, Y. Van Der Graaf,

Unrequested information from routine diagnostic chest CT predicts future cardiovascular

events, Eur. Radiol. 21 (2011) 1577–1585. doi:10.1007/s00330-011-2112-8.

17

[10] A. Padole, S. Singh, J.B. Ackman, C. Wu, S. Do, S. Pourjabbar, R.D.A. Khawaja, A. Otrakji, S.

Digumarthy, J.-A. Shepard, M. Kalra, Submillisievert Chest CT With Filtered Back Projection

and Iterative Reconstruction Techniques, Am. J. Roentgenol. 203 (2014) 772–781.

doi:10.2214/AJR.13.12312.

[11] H.J. de Koning, R. Meza, S.K. Plevritis, K. ten Haaf, V.N. Munshi, J. Jeon, S.A. Erdogan, C.Y.

Kong, S.S. Han, J. van Rosmalen, S.E. Choi, P.F. Pinsky, A.B. de Gonzalez, C.D. Berg, W.C. Black,

M.C. Tammemägi, W.D. Hazelton, E.J. Feuer, P.M. McMahon, Benefits and Harms of

Computed Tomography Lung Cancer Screening Strategies: A Comparative Modeling Study for

the U.S. Preventive Services Task Force, Ann. Intern. Med. N/A (2013) N/A-N/A.

doi:10.7326/M13-2316.

[12] P.M. Jairam, M.J.A. Gondrie, D.E. Grobbee, W.P. Th M Mali, P.C.A. Jacobs, Y. van der Graaf,

Incidental Imaging Findings from Routine Chest CT Used to Identify Subjects at High Risk of

Future Cardiovascular Events., Radiology. (2014) 132211. doi:10.1148/radiol.14132211.

[13] World Health Organization, International Statistical Classification of Diseases and Related

Health Problems, (2016). http://apps.who.int/classifications/icd10/browse/2016/en.

[14] R.L. Prentice, On the design of synthetic case-control studies., Biometrics. 42 (1986) 301–310.

doi:10.2307/2531051.

[15] N.C. Onland-Moret, D.L. van der A, Y.T. van der Schouw, W. Buschers, S.G. Elias, C.H. van Gils,

J. Koerselman, M. Roest, D.E. Grobbee, P.H.M. Peeters, Analysis of case-cohort data: A

comparison of different methods, J. Clin. Epidemiol. 60 (2007) 350–355.

doi:10.1016/j.jclinepi.2006.06.022.

[16] P.C.A. Jacobs, M. Prokop, A.L. Oen, Y. van der Graaf, D.E. Grobbee, W.P.T.M. Mali,

Semiquantitative Assessment of Cardiovascular Disease Markers in Multislice Computed

Tomography of the Chest, J. Comput. Assist. Tomogr. 34 (2010) 279–284.

doi:10.1097/RCT.0b013e3181bbcff6.

[17] H. Marti, M. Chavance, Multiple imputation analysis of case-cohort studies, Stat. Med. 30

18

(2011) 1595–1607. doi:10.1002/sim.4130.

[18] T.M. Therneau, P.M. Grambsch, Modeling Survival Data: Extending the Cox Model, Springer,

New York, 2000.

[19] J. Perk, G. De Backer, H. Gohlke, I. Graham, Ž. Reiner, W.M.M. Verschuren, C. Albus, P.

Benlian, G. Boysen, R. Cifkova, C. Deaton, S. Ebrahim, M. Fisher, G. Germano, R. Hobbs, A.

Hoes, S. Karadeniz, A. Mezzani, E. Prescott, L. Ryden, M. Scherer, M. Syvänne, W.J.M. Scholte

Op Reimer, C. Vrints, D. Wood, J.L. Zamorano, F. Zannad, European Guidelines on

cardiovascular disease prevention in clinical practice (version 2012), Atherosclerosis. 223

(2012) 1–68. doi:10.1016/j.atherosclerosis.2012.05.007.

[20] N.K. Kalia, L.G. Miller, K. Nasir, R.S. Blumenthal, N. Agrawal, M.J. Budoff, Visualizing coronary

calcium is associated with improvements in adherence to statin therapy., Atherosclerosis. 185

(2006) 394–9. doi:10.1016/j.atherosclerosis.2005.06.018.

[21] U.N. Khot, M.B. Khot, C.T. Bajzer, S.K. Sapp, E.M. Ohman, S.J. Brener, S.G. Ellis, A.M. Lincoff,

E.J. Topol, Prevalence of conventional risk factors in patients with coronary heart disease.,

JAMA. 290 (2003) 898–904. doi:10.1001/jama.290.7.898.

[22] K. Nasir, J. Rubin, M.J. Blaha, L.J. Shaw, R. Blankstein, J.J. Rivera, A.N. Khan, D. Berman, P.

Raggi, T. Callister, J. a. Rumberger, J. Min, S.R. Jones, R.S. Blumenthal, M.J. Budoff, Interplay of

coronary artery calcification and traditional risk factors for the prediction of all-cause

mortality in asymptomatic individuals, Circ. Cardiovasc. Imaging. 5 (2012) 467–473.

doi:10.1161/CIRCIMAGING.111.964528.

[23] J. Leipsic, C.M. Taylor, G. Grunau, B.G. Heilbron, G.B. Mancini, S. Achenbach, M. Al-Mallah,

D.S. Berman, M.J. Budoff, F. Cademartiri, T.Q. Callister, H.J. Chang, V.Y. Cheng, K. Chinnaiyan,

B.J. Chow, a Delago, M. Hadamitzky, J. Hausleiter, R. Cury, G. Feuchtner, Y.J. Kim, P. a

Kaufmann, F.Y. Lin, E. Maffei, G. Raff, L.J. Shaw, T.C. Villines, J.K. Min, Cardiovascular risk

among stable individuals suspected of having coronary artery disease with no modifiable risk

factors: results from an international multicenter study of 5262 patients, Radiology. 267

19

(2013) 718–726. doi:10.1148/radiol.13121669.

[24] H.S. Hecht, P. Cronin, M.J. Blaha, M.J. Budoff, E.A. Kazerooni, J. Narula, D. Yankelevitz, S.

Abbara, 2016 SCCT/STR guidelines for coronary artery calcium scoring of noncontrast

noncardiac chest CT scans: A report of the Society of Cardiovascular Computed Tomography

and Society of Thoracic Radiology, J. Cardiovasc. Comput. Tomogr. 11 (2017) 74–84.

doi:10.1016/j.jcct.2016.11.003.

[25] M.G. Silverman, M.J. Blaha, H.M. Krumholz, M.J. Budoff, R. Blankstein, C.T. Sibley, A. Agatston,

R.S. Blumenthal, K. Nasir, Impact of coronary artery calcium on coronary heart disease events

in individuals at the extremes of traditional risk factor burden: The Multi-Ethnic Study of

Atherosclerosis, Eur. Heart J. 35 (2014) 2232–2241. doi:10.1093/eurheartj/eht508.

[26] R. Smith-Bindman, D.L. Miglioretti, E. Johnson, C. Lee, H.S. Feigelson, M. Flynn, R.T. Greenlee,

R.L. Kruger, M.C. Hornbrook, D. Roblin, L.I. Solberg, N. Vanneman, S. Weinmann, A.E.

Williams, Use of diagnostic imaging studies and associated radiation exposure for patients

enrolled in large integrated health care systems, 1996-2010., JAMA. 307 (2012) 2400–9.

doi:10.1001/jama.2012.5960.

[27] T. Dirrichs, T. Penzkofer, S.D. Reinartz, T. Kraus, A.H. Mahnken, C.K. Kuhl, Extracoronary

Thoracic and Coronary Artery Calcifications on Chest CT for Lung Cancer Screening, Acad.

Radiol. 22 (2015) 880–889. doi:10.1016/j.acra.2015.03.005.

[28] K. Nasir, M. Clouse, Role of Nonenhanced Multidetector CT Coronary Artery Calcium Testing in

Asymptomatic and Symptomatic Individuals, Radiology. 264 (2012) 637–649.

doi:10.1148/radiol.12110810.

[29] G. M.J.A., M. W.P.T.M., J. P.C., O. A.L., V.D.G. Y., Cardiovascular disease: Prediction with

ancillary aortic findings on chest CT scans in routine practice, Radiology. 257 (2010) 549–559.

http://www.embase.com/search/results?

subaction=viewrecord&from=export&id=L359842097.

[30] G.S. Collins, J.B. Reitsma, D.G. Altman, K.G.M. Moons, Transparent reporting of a multivariable

20

prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement, Bmj.

350 (2015) g7594–g7594. doi:10.1136/bmj.g7594.

[31] D.E. Bild, R. Detrano, D. Peterson, A. Guerci, K. Liu, E. Shahar, P. Ouyang, S. Jackson, M.F. Saad,

Ethnic differences in coronary calcification: The Multi-Ethnic Study of Atherosclerosis (MESA),

in: Circulation, 2005: pp. 1313–1320. doi:10.1161/01.CIR.0000157730.94423.4B.

21

Tables

Table 1 Semi-quantitative grading score of the cardiovascular CT findings included in the CT based CVD

risk score

0 = Absent 1 = Mild 2 = Moderate 3 =Severe

LAD coronary artery calcification

Number and size of

calcifications

none 1-2 focal limited

to ≤2 slices

>2 focal or 1

extending >2 slices

Fully calcified coronary artery

extending multiple slices

Descending aorta calcifications

Number and size of

calcifications

none ≤ 3 focal 4-5 focal or 1

extending ≥3 slices

>5 focal or >1 extending ≥3

slices

Mitral valve calcifications

Number of affected

valve leaflets

none 1 leaflet 2 leaflets

LAD = left anterior descending

22

Table 2 Baseline clinical characteristics and CT characteristics for all cases and patients in the

subcohort of both the Scottish cohort and the original cohort

Cases Subcohort Subcohort PROVIDI

Total patients 384 500 1366

Male gender, n (%) 201 (52) 256 (51) 792 (58)

Age, years 71 (41-96) 67 (40-94) 62 (40-96)

CT-indication, n (%)

Lung disease 163 (42) 187 (38) 505 (37)

Haematological malignancy 25 (7) 32 (6) 150(11)

Mediastinal abnormality 18 (5) 29 (6) 150 (11)

Suspicion pulmonary malignancy 128 (33) 159 (32) 314 (23)

Suspicion pulmonary embolism 1 (0) 1 (0) 82 (6)

Other indication 49 (13) 92 (18) 164 (12)

LAD* coronary artery calcifications, n (%)

Mild 72 (19) 93 (19) 410 (30)

Moderate 121 (32) 93 (19) 260 (19)

Severe 97 (25) 91 (19) 137 (10)

Descending aorta calcifications, n (%)

Mild 102 (27) 145 (30) 382 (28)

Moderate 59 (15) 79 (16) 191 (14)

Severe 185 (48) 151 (30) 96 (7)

Mitral valve calcification, n (%)

1 leaflet 56 (15) 36 (7) 96 (7)

2 leaflets 19 (5) 11 (2.2) 27 (2)

Cardiac diameter, mm 127 (71-192) 124 (82-188) 125 (77-185)

Follow up time, years 2.1 (0-6.3) 4.3 (0-6.4) 3.7 (0-7)

* LAD = left anterior descending,

23

Table 3 CVD event rate (per 1000 person-years) by gender and age

Women Men

Total

person

years

No of cases Event rate

(95% CI)

Total

person

years

No of cases Event rate

(95% CI)

Total 4643 183 39.4 (33.9-45.6) 4395 201 45.7 (39.6-52.5)

Age, years

40-50 705 10 14.2 (6.8-26.1) 612 7 11.4 (4.6-23.6)

50-60 991 23 23.2 (14.7-34.8) 1056 28 26.4 (17.6-38.3)

60-70 1319 38 29.2 (20.4-39.5) 1322 66 49.9 (38.6-63.5)

70-80 1099 59 53.6 (40.9-69.3) 1007 64 64 (48.9-81.2)

80-96 530 53 100 (74.9-130.8) 398 36 90 (63.3-125.2)

24

Table 4 Predicted vs Observed 5 year cardiovascular risk

Risk Group 5y Risk: Radiological risk score

Low risk group (<10% risk)

No. of patients n (%) 535 (25)

Observed Kaplan-Meier risk (%) 8.6%

Intermediate risk group (10-20% risk)

No. of patients n (%) 621 (29)

Observed Kaplan-Meier risk (%) 14.9%

High risk group (>20% risk)

No. of patients n (%) 968 (46)

Observed Kaplan-Meier risk (%) 31.7%

25

Figure Legends

Fig. 1 Flowchart of study design. CVD = Cardiovascular Disease.

26

Fig. 2 Examples of CT imaging findings included in the radiological risk score. (A) coronary

calcification in left anterior descending artery, (B) cardiac diameter measurement, (C) calcification on

the mitral valve, (D) calcification in the descending aorta.

27

Fig. 3 Survival curves of the 5 year predicted cardiovascular risk categories low (<10%), intermediate

(10-20%) and high (>20%) from baseline until a maximum of 6.4 years follow up.

28

Fig. 4 Calibration plot: Predicted probability versus actual probability of surviving 5 years in deciles

(dots) and overall trend (line). A reference line is added to the plot indicating perfect calibration

(dashed line).

29