OUP user menu

Reference intervals for common carotid intima-media thickness measured with echotracking: relation with risk factors

Lian Engelen, Isabel Ferreira, Coen D. Stehouwer, Pierre Boutouyrie, Stéphane Laurent,
DOI: http://dx.doi.org/10.1093/eurheartj/ehs380 2368-2380 First published online: 27 November 2012

Abstract

Aims Common carotid artery intima-media thickness (CCIMT) is widely used as a surrogate marker of atherosclerosis, given its predictive association with cardiovascular disease (CVD). The interpretation of CCIMT values has been hampered by the absence of reference values, however. We therefore aimed to establish reference intervals of CCIMT, obtained using the probably most accurate method at present (i.e. echotracking), to help interpretation of these measures.

Methods and results We combined CCIMT data obtained by echotracking on 24 871 individuals (53% men; age range 15–101 years) from 24 research centres worldwide. Individuals without CVD, cardiovascular risk factors (CV-RFs), and BP-, lipid-, and/or glucose-lowering medication constituted a healthy sub-population (n = 4234) used to establish sex-specific equations for percentiles of CCIMT across age. With these equations, we generated CCIMT Z-scores in different reference sub-populations, thereby allowing for a standardized comparison between observed and predicted (‘normal’) values from individuals of the same age and sex. In the sub-population without CVD and treatment (n = 14 609), and in men and women, respectively, CCIMT Z-scores were independently associated with systolic blood pressure [standardized βs 0.19 (95% CI: 0.16–0.22) and 0.18 (0.15–0.21)], smoking [0.25 (0.19–0.31) and 0.11 (0.04–0.18)], diabetes [0.19 (0.05–0.33) and 0.19 (0.02–0.36)], total-to-HDL cholesterol ratio [0.07 (0.04–0.10) and 0.05 (0.02–0.09)], and body mass index [0.14 (0.12–0.17) and 0.07 (0.04–0.10)].

Conclusion We estimated age- and sex-specific percentiles of CCIMT in a healthy population and assessed the association of CV-RFs with CCIMT Z-scores, which enables comparison of IMT values for (patient) groups with different cardiovascular risk profiles, helping interpretation of such measures obtained both in research and clinical settings.

  • Ageing
  • Atherosclerois
  • Carotid intima-media thickness
  • Echotracking
  • Reference intervals
  • Risk factors

Introduction

Measurement by ultrasonography of the common carotid artery intima-media thickness (CCIMT) was first described by Pignoli et al. in 1986.1 Since then, the technique has been widely used for the assessment of arterial wall thickness in vivo. Numerous studies have shown that non-invasive measures of CCIMT can be measured with high reproducibility; correlate well with major cardiovascular risk factors (CV-RFs), prevalent disease, and severity of atherosclerosis in other vascular beds; and predict incident cardiovascular events; and that its progression over time may be deterred by targeted interventions (reviewed in 24). As such, CCIMT is a suitable surrogate marker for cardiovascular disease (CVD)3 and is currently widely used for the pre-symptomatic detection of disease and its progression, in clinical and epidemiological studies, improving efficiency and aiming at decreasing follow-up time usually needed in studies with ‘hard’ cardiovascular endpoints.36

Despite attempts at normalization, the method for measuring CCIMT is highly variable, either in terms of signal processing [B-mode, M-mode, radiofrequency (RF) signal analysis] or anatomical location. The methodological heterogeneity accounts for individual variability in the value of CCIMT across studies. As such, values obtained in either research or clinical practice settings, obtained with different methodologies, are difficult to analyse in the absence of standardized reference values. Indeed, although age-dependent nomograms for CCIMT have been reported before,711 their general use is limited. First, they refer to mere distributions of mean or median values in general populations without distinguishing between subjects with or without established CV-RFs and/or disease8,10 and can thus not be used as reference for a ‘normal’ (i.e. healthy) population. Second, they refer to values of CCIMT as obtained by manual or automated analyses techniques of B-mode ultrasound (US) imaging,711 whereas, at present, automated edge-detection on the basis of RF signal processing (hereafter ‘echotracking’) of B + M mode US imaging is probably the most accurate method.1215 Third, they are confined to a single-centre and/or country7,9,11 and thus have limited sample sizes to properly cover the whole (adult) age range.

In view of these considerations, we combined subject-level data on established CV-RFs and CCIMT as obtained by echotracking systems from different study centres worldwide into one large data set—The Reference Values for Arterial Measurements Collaboration's CCIMT database. This was used to, first, establish age- and sex-specific percentiles (reference intervals—RIs) for CCIMT in individuals without CV-RFs (as conventionally defined), prior CVD and BP-, lipid-, and/or glucose-lowering medication, i.e. a healthy population; and second, to investigate the relation of CV-RFs and the use of BP-, lipid-, and/or glucose-lowering medication with CCIMT percentiles in individuals with or without prior CVD.

Methods

Study population

With a systematic literature review, we identified all cohort studies using echotracking for CCIMT measurement. Next, we personally contacted the principal investigators of the cohorts (n = 55) to inform them about the project and invite them to participate. We finally compiled subject-level data from 24 research centres/research groups—corresponding to 30 distinct cohorts—distributed across 14 countries worldwide (see the list in Supplementary material online, Table S1). A total of 25 166 individuals with data on CCIMT obtained using echotracking systems, age (range 5–101 years), sex (13 430 men/11 736 women), CVD status, and important CV-RFs were available for analysis. For the present study, we excluded 295 (53% girls) individuals who were aged <15 years because their data lacked sufficient variability with age (primarily concentrated at the age of five16), leaving 24 871 (47% women) individuals for analyses.

To generate age- and sex-specific normative tables for CCIMT, we selected a healthy sub-population composed of individuals who did not meet any of the following criteria: (i) history of CVD; (ii) use of BP-, lipid-, and/or glucose-lowering medication; (iii) hypertension [i.e. systolic blood pressure (SBP) ≥140 mmHg and/or diastolic blood pressure (DBP) ≥90 mmHg]17; (iv) current smoking; (v) diabetes [defined as self-reported diabetes and/or fasting plasma glucose ≥7.0 mmol/L (if available) and/or post-load plasma glucose ≥11.0 mmol/L (if available)]18; (vi) total cholesterol >6.2 mmol/L19; (vii) HDL cholesterol <1.17 mmol/L (for men) and <1.30 (for women)19; and (viii) body mass index (BMI) ≥30 kg/m2.20 This healthy sub-population consisted of 4234 (53% women) individuals, which originated from 21 out of the 24 research centres (details in Table A1). The cut-off values used to define the healthy sub-population were chosen, whenever possible, to be similar to those used to indicate increased risk in current guidelines17,18,20 (or risk algorithms19) to enable optimal comparison with other studies.

To investigate the relation of CV-RFs with individuals' levels of CCIMT percentiles, we stratified the total population according to a history of CVD and, in individuals without prior CVD only, by the use of BP-, lipid-, and/or glucose-lowering medication. This resulted in three reference sub-populations consisting of: (i) 14 609 (48% women) individuals without prior CVD and without the use of BP-, lipid-, and/or glucose-lowering medication; (ii) 5761 (52% women) individuals without prior CVD and who used BP-, lipid-, and/or glucose-lowering medication; and (iii) 4501 individuals (37% women) with prior CVD irrespective of medication use.

A flowchart describing the selection of the healthy and reference sub-populations and exact numbers per sex is presented in Figure 1.

Figure 1

Study flowchart describing the selection and categorization of individuals from the total common carotid artery intima-media thickness (CCIMT) population to the reference and healthy sub-populations. aBP-, lipid-, and/or glucose-lowering medication. bRisk factors considered were hypertension (systolic blood pressure/diastolic blood pressure ≥140/90 mmHg), current smoking, diabetes [self-reported diabetes and/or fasting plasma glucose ≥7.0 and/or post-load plasma glucose ≥11.0 mmol/L (if available)], total cholesterol >6.2 mmol/L, HDL cholesterol <1.17 mmol/L (for men) and <1.30 mmol/L (for women), and body mass index ≥30 kg/m2.

Common carotid artery intima-media thickness measurements: methodological considerations

We included only CCIMT data obtained by means of echotracking (either pure echotracking or related techniques). These were measured at the far wall of the right and/or left common carotid artery only, because near-wall readings from echotracking devices may not reflect true thickness and are thus seldom obtained. Mean values of right and left CCIMT readings (if both sides were assessed) were used in the analyses, as previous studies have reported no differences between sides.8,21

Different types of US systems were used across centres; specifically, pure echotracking systems: the Wall Track System (n = 13 116; WTS, ESAOTE, Maastricht, The Netherlands22) and the ART.LAB system (n = 8519; advanced version of WTS; ESAOTE, Maastricht, The Netherlands) or related techniques, which were validated against echotracking: the Vivid-7 US system (n = 2524; GE Vingmed Ultrasound, Horten, Norway) with Echopac post-processing; the Aloka SSD-650 US system (n = 606; Aloka, Tokyo, Japan) with dedicated post-processing software (M'ATHS, Metris, France)23; and the Carotid Studio (n = 401; Institute of Clinical Physiology, National Research Council, Pisa, Italy).24 The exact anatomical location of the measurement of the CCIMT differed across centres: i.e. at 0–1 cm, centred at 1 cm, at 1–2 cm or centred at 2 cm proximal to the carotid bifurcation. Therefore, prior to further analyses, we standardized all CCIMT values obtained with different echotracking systems and anatomical locations (for details, please see Table S1). To this aim, original CCIMT values were rescaled to the same metric of the mostly used system and location, i.e. measurements with the ART.LAB system and centred at 1 cm proximal to the carotid bifurcation (see the Statistical analyses section).

Statistical analyses

Multiple imputation of missing values in variables

A total of 4673 individuals (19% of the total reference population) had missing values for one (n = 4391) or more (n = 282) of the variables of interest. The percentage of missing values per variable varied from 0.4% (current smoking) to 11% (HDL cholesterol). We used multiple imputation chained equations to impute those values rather than perform complete case analyses in order to decrease bias and increase the power of the analyses25,26 (for details, please see Supplementary material online).

Standardization of common carotid artery intima-media thickness measurements

We performed multiple linear regression analyses that included dummy variables for each echotracking system (with ART.LAB as reference) and anatomical location (with measurements centred at 1 cm proximal to the carotid bifurcation as reference) as independent determinants of CCIMT. These analyses were conducted in the total population (n = 24 871) and included adjustments for all CV-RFs, history of CVD, and the use of BP- and/or lipid-lowering medication. The regression coefficients (β) for the dummy variables hereby obtained were used as ‘calibration factors’ to rescale individual CCIMT values to the reference technique (details in Supplementary material online, Table S2). We used these rescaled CCIMT values in all further analyses.

Definition of age- and sex-specific reference intervals

An extensive description of the methods used to define RIs for CCIMT is provided in Supplementary material online. In brief, calculation of age-specific RIs for CCIMT was performed in the healthy sub-population (n = 4234), and in men and women separately. To this aim, we used a parametric regression method based on fractional polynomials (FPs) as described by Royston and Wright27 and implemented in the STATA software (version 11.0 Stata Corp., College Station, TX, USA).28 Age-specific 2.5th, 10th, 25th, 50th, 75th, 90th, and 97.5th percentile curves were calculated as meanCCIMT + Zp × SD, where Zp assumed the values of −1.96, −1.28, −0.67, 0, 0.67, 1.28, and 1.96, respectively.

Relation with risk factors

Based on the equations estimated as described above, we computed expected ‘normal’ mean CCIMT values for each individual in the reference sub-populations (i.e. those with and without CVD and/or medication) and calculated age- and sex-specific CCIMT Z-scores as (observedCCIMT – expectedCCIMT)/SDexpectedCCIMT; this allows for a standardized comparison between observed CCIMT values vs. those from healthy individuals of the same age and sex, expressed by the number of SDs an individual measurement lies above or below the healthy population mean (or 50th percentile).

The relation of known CV-RFs with the CCIMT Z-scores was then investigated in the different reference sub-populations, using multiple linear regression analyses to enable interpretation of CCIMT values across different risk groups. We also included age in these analyses to account for any potential residual influence of age in these sub-populations. In addition, we added interaction terms between sex and each of the CV-RFs to the models to assess potential effect modification.

Statistical analyses were performed using the Statistical Package for Social Sciences, version 18.0 (SPSS, Inc., Chicago, IL, USA) unless specified otherwise.

Results

Tables 1 and 2 show the participants' characteristics of the total, healthy, and reference sub-populations, in men and women, respectively. In the total reference population, women were slightly older and had, on average, lower values of CV-RFs compared with men.

View this table:
Table 1

Risk factors and clinical characteristics of the total, healthy, and reference sub-populations in men

 Total reference populationHealthy sub-populationSub-population without CVDSub-population with CVD
Without treatmentaWith treatmenta
n13 2901993766927822839
Age [years (range)]56 (15–99)50 (15–90)53 (15–99)59 (16–98)63 (23–97)
Body mass index (kg/m2)26.3 ± 3.724.2 ± 2.625.8 ± 3.527.6 ± 3.926.6 ± 3.5
Systolic blood pressure (mmHg)136 ± 19123 ± 10132 ± 17142 ± 19139 ± 20
Diastolic blood pressure (mmHg)79 ± 1174 ± 879 ± 1182 ± 1279 ± 11
Mean arterial pressure (mmHg)98 ± 1290 ± 896 ± 12102 ± 1399 ± 12
Hypertension [n (%)]6951 (52)2453 (32)2324 (84)2173 (77)
Total cholesterol (mmol/L)5.5 ± 1.05.1 ± 0.75.5 ± 1.05.4 ± 1.15.3 ± 1.0
LDL cholesterol (mmol/L)3.5 ± 0.93.1 ± 0.73.6 ± 0.93.4 ± 1.03.3 ± 0.9
HDL cholesterol (mmol/L)1.3 ± 0.31.5 ± 0.31.3 ± 0.31.3 ± 0.31.2 ± 0.3
Total-to-HDL cholesterol ratio4.5 ± 1.53.4 ± 0.74.4 ± 1.44.6 ± 1.74.6 ± 1.6
Triglycerides (mmol/L)1.3 (0.9–1.9)0.9 (0.7–1.2)1.2 (0.9–1.8)1.4 (1.0–2.0)1.4 (1.0–2.0)
Fasting glucose (mmol/L)5.8 ± 1.55.2 ± 0.75.4 ± 1.06.2 ± 2.06.2 ± 1.8
Diabetes [n (%)]1408 (11)304 (4)590 (21)514 (18)
Current smoking [n (%)]3126 (24)1825 (24)559 (20)741 (26)
BP-lowering medication [n (%)]2335 (18)2073 (74)1648 (58)
Lipid-lowering medication [n (%)]877 (7)1134 (41)1129 (40)
Glucose-lowering medication [n (%)]595 (5)379 (14)216 (8)
History of CVD [n (%)]2839 (21)2839 (100)
CCIMT (µm)653 ± 159583 ± 131631 ± 155682 ± 151685 ± 169
  • Data are presented as means ± SD, medians (inter-quartile ranges), or numbers (percentages), as appropriate.

  • aBP-, lipid-, and glucose-lowering treatment.

View this table:
Table 2

Risk factors and clinical characteristics of the total, healthy, and reference sub-populations in women

 Total reference populationHealthy sub-populationSub-population without CVDSub-population with CVD
Without treatmentaWith treatmenta
n11 5812241694029791662
Age [years (range)]58 (15–101)48 (15–89)54 (15–95)63 (17–101)64 (20–95)
Body mass index (kg/m2)25.8 ± 4.722.9 ± 2.824.9 ± 4.327.8 ± 5.026.4 ± 4.5
Systolic blood pressure (mmHg)133 ± 21118 ± 11128 ± 19142 ± 21139 ± 21
Diastolic blood pressure (mmHg)76 ± 1172 ± 875 ± 1079 ± 1277 ± 11
Mean arterial pressure (mmHg)95 ± 1387 ± 893 ± 12100 ± 1398 ± 12
Hypertension [n (%)]5519 (48)1804 (26)2608 (88)1106 (66)
Total cholesterol (mmol/L)5.8 ± 1.15.2 ± 0.75.8 ± 1.15.9 ± 1.15.9 ± 1.1
LDL cholesterol (mmol/L)3.6 ± 1.03.0 ± 0.73.6 ± 1.03.6 ± 1.03.7 ± 1.0
HDL cholesterol (mmol/L)1.6 ± 0.41.8 ± 0.31.7 ± 0.41.5 ± 0.41.6 ± 0.4
Total-to-HDL cholesterol ratio3.8 ± 1.32.9 ± 0.63.7 ± 1.24.1 ± 1.44.1 ± 1.4
Triglycerides (mmol/L)1.2 (0.8–1.6)0.9 (0.7–1.1)1.1 (0.8–1.5)1.4 (1.0–1.9)1.3 (0.9–1.8)
Fasting glucose (mmol/L)5.6 ± 1.44.9 ± 0.65.2 ± 0.96.0 ± 1.85.9 ± 1.7
Diabetes [n (%)]981 (9)210 (3)505 (17)266 (16)
Current smoking [n (%)]1983 (17)1237 (18)433 (15)313 (19)
BP-lowering medication [n (%)]2378 (21)2405 (81)767 (46)
Lipid-lowering medication [n (%)]618 (5)991 (33)421 (25)
Glucose-lowering medication [n (%)]356 (3)260 (9)96 (6)
History of CVD [n (%)]1662 (14)1662 (100)
CCIMT (µm)639 ± 148561 ± 123610 ± 140682 ± 150677 ± 153
  • Data are presented as means ± SD, medians (inter-quartile ranges), or numbers (percentages), as appropriate.

  • aBP-, lipid-, and glucose-lowering treatment.

Age- and sex-specific reference intervals for common carotid artery intima-media thickness in the healthy sub-population

The best fitting FPs' powers (p) for the meanCCIMT and SDCCIMT curves were p = 1 for both men and women, indicating that linear regression lines described the age–CCIMT relationships well. Accordingly, the equations derived on the basis of the estimated coefficients were, for men: Embedded Image (1) Embedded Image (2) and, for women: Embedded Image (3) Embedded Image (4) The estimated Z-scores had a mean value of 0 and an SD of 1 and, when plotted against age, were randomly distributed above and below 0 (see Supplementary material online, Figure S1), indicating good model fit and no residual dependency on age.

Sex-specific percentile lines superimposed on the raw data are shown in Figure 2, and the respective levels of CCIMT by age category are presented in Table 3. Mean values of CCIMT were slightly higher in men than in women at any age (P < 0.001), but increases in CCIMT with ageing were similar in men (5.2 µm/year) and women (5.0 µm/year) (P-value for age by sex interaction = 0.144).

View this table:
Table 3

Age- and sex-specific percentiles of common carotid artery intima-media thickness (in µm) in the healthy sub-population

 Age (years)Percentiles
2.5th10th25th50th75th90th97.5th
Men (n = 1993)15263311354401449492540
20280331377427478524575
25297351400453507556610
30314372423479536587645
35331392446505565619680
40349412468531594651714
45366432491557624683749
50383452514583653715784
55400473537609682746819
60417493560635711778854
65434513583662740810889
70451533606688769842924
75469554629714798873958
80486574652740827905993
855035946757668569371028
Women (n = 2241)15265311351396441482527
20282330373421469512560
25299350395446497542593
30315369417471524572626
35332389439496552602659
40349408461521580633692
45366428483545607663725
50382448506570635693758
55399467528595663723791
60416487550620690753824
65433506572645718783857
70450526594670745813890
75466545616694773843923
80483565638719801874956
85500585660744828904989
Figure 2

Age-specific percentiles of common carotid artery intima-media thickness (CCIMT) in the healthy sub-population: (A) men; (B) women.

Relation of cardiovascular risk factors with common carotid artery intima-media thickness percentiles as defined in the healthy sub-population

In the sub-population without prior CVD and treatment, and both in men and women, higher CCIMT Z-scores (i.e. positive deviation from the healthy population mean) were significantly associated with SBP, smoking, diabetes, total-to-HDL cholesterol ratio, and BMI, whereas in the treated sub-population without prior CVD, diabetes and total-to-HDL cholesterol ratio were no longer independent determinants of the CCIMT Z-scores (Table 4). In the sub-population with prior CVD, SBP was the main determinant of CCIMT Z-scores in both men and women; BMI (adversely) and the use of lipid-lowering medication (protectively) were also determinants but in men only.

View this table:
Table 4

Relation of known cardiovascular risk factors with common carotid artery intima-media thickness Z-scores in the reference sub-populations

SexRisk factorSub-population without CVDSub-population with CVD (n = 4501)
Without treatmenta (n = 14 609)With treatmenta (n = 5761)
β95% CIP-valueβ95% CIP-valueβ95% CIP-value
MenSystolic pressure (10 mmHg)0.1110.093; 0.128<0.0010.0700.045; 0.094<0.0010.0800.055; 0.106<0.001
Current smoking (yes)0.2480.185; 0.312<0.0010.1430.026; 0.2590.0160.100−0.013; 0.2130.082
Diabetes (yes)0.1890.046; 0.3320.0100.056−0.053; 0.1660.3150.108−0.055; 0.2710.192
Total-to-HDL cholesterol ratio (unit)0.0510.029; 0.073<0.0010.022−0.006; 0.0490.1180.008−0.028; 0.0440.660
Body mass index (kg/m2)0.0410.033; 0.049<0.0010.0360.024; 0.048<0.0010.0280.013; 0.042<0.001
Use of BP-lowering medication (yes)−0.059−0.161; 0.0440.264
Use of lipid-lowering medication (yes)−0.182−0.287; −0.0770.001
Use of glucose-lowering medication (yes)−0.137−0.370; 0.0960.248
WomenSystolic pressure (10 mmHg)0.0970.080; 0.113<0.0010.1000.078; 0.123<0.0010.1010.068; 0.134<0.001
Current smoking (yes)0.1110.040; 0.1810.0020.2040.078; 0.3310.002−0.107−0.269; 0.0540.192
Diabetes (yes)0.1890.020; 0.3590.0290.010−0.109; 0.1280.8720.052−0.161; 0.2650.633
Total-to-HDL cholesterol ratio (unit)0.0430.016; 0.0690.0020.021−0.016; 0.0590.2600.016−0.034; 0.0650.534
Body mass index (kg/m2)0.0170.010; 0.024<0.0010.0100.000; 0.0200.0400.007−0.008; 0.0220.387
Use of BP-lowering medication (yes)0.072−0.062; 0.2050.292
Use of lipid-lowering medication (yes)−0.102−0.250; 0.0460.178
Use of glucose-lowering medication (yes)0.109−0.209; 0.4270.501
  • The regression coefficient β represents the increase in CCIMT (in SD from the healthy population mean among individuals of the same age and sex) per unit increase in each risk factor. βs were obtained from multivariable regression models including all risk factors and age.

  • aBP-, lipid-, and glucose-lowering treatment. Risk factor data available for the sub-populations without CVD and without treatment, without CVD with treatment, and with CVD, respectively, were n = 13 585, 5501, and 4196 for systolic pressure, n = 14 561, 5730, and 4474 for current smoking, n = 14 482, 5734, and 4425 for diabetes, n = 12 871, 5482, and 4078 for total-to-HDL cholesterol ratio, and n = 14 556, 5717, and 4458 for body mass index. Missing values were imputed before analyses (for details, please see Statistical analyses in the Methods section).

To enable comparison of the strength of the associations between the individual CV-RFs and CCIMT Z-scores within each sub-population, these associations are also shown as standardized regression coefficients (i.e. per-SD increase in CV-RF) (Figure 3). These analyses showed that, in the sub-population without CVD or treatment, smoking, diabetes, and SBP were the strongest determinants of the CCIMT Z-scores in men, whereas in women these were diabetes and SBP. Comparisons by sex showed that smoking and BMI were stronger determinants in men than in women (P-value for sex interactions were 0.005 and <0.001, respectively).

Figure 3

Point estimates and 95% confidence intervals represent the increase in common carotid artery intima-media thickness (CCIMT) Z-score (in SD from the healthy population mean) per SD increase (or for presence vs. absence) in risk factor resulting from a multivariable regression model including all risk factors and age. Data in blue and red concern men and women, respectively. BMI, body mass index; BP, blood pressure; med, medication; SBP, systolic blood pressure. (A) Reference sub-population without cardiovascular disease (CVD) or treatment. (B) Reference sub-population without cardiovascular disease with BP-, lipid-, and/or glucose-lowering treatment. (C) Reference sub-population with cardiovascular disease. Risk factor data available for the sub-population without cardiovascular disease and without treatment, without cardiovascular disease with treatment and with cardiovascular disease, respectively, were n = 13 585, 5501, and 4196 for SBP, n = 14 561, 5730, and 4474 for smoking, n = 14 482, 5734, and 4425 for diabetes, n = 12 871, 5482, and 4078 for total-to-HDL cholesterol, and n = 14 556, 5717, and 4458 for BMI. Missing values were imputed before analyses (for details, please see Statistical analyses in the Methods section).

The regression coefficients shown in Table 4, reflecting the associations of CV-RFs with CCIMT Z-scores (i.e. the increase in SD from the mean CCIMT of healthy individuals of the same age and sex), can be converted into percentiles for a more meaningful interpretation of these analyses in the light of the RIs provided (Figure 1 and Table 3). This is illustrated with two hypothetical subjects in Table 5: (i) a 50-year old man with SBP 160 mmHg, total-to-HDL cholesterol ratio 7.2, BMI 35 kg/m2, who smokes and has no diabetes and (ii) a 50-year-old woman with SBP 130 mmHg, total-to-HDL cholesterol ratio 3.9, BMI 24 kg/m2, who does not smoke and has no diabetes. Based on these risk profiles and the regression coefficients provided, the estimated CCIMT Z-scores for these individuals were 1.32 (man) and 0.19 (woman); these correspond, respectively, to the 91st and 58th percentiles of the CCIMT distribution in individuals of the same age and sex from the healthy sub-population. Similarly, CCIMT Z-scores can be estimated for any other combination of individuals' age and CV-RFs (i.e. risk profile) and conversion into percentiles can easily be retrieved using any standard normal distribution (Z) table, in which Z-scores of 0, 0.68, 1.28, and 1.65 correspond with the 50th, 75th, 90th, and 95th percentiles.

View this table:
Table 5

Two examples of hypothetical subjects and their estimated common carotid artery intima-media thickness percentile

Cardiovascular risk factorsCoefficientaObservedbCoefficient × observed
MenIntercept = −2.604c
 Age (10 years)d0.020500.100
 SBP (10 mmHg)0.1111601.776
 Smoking (yes)0.248Yes (1)0.248
 Diabetes (yes)0.189No (0)0
 Total-to-HDL cholesterol ratio (unit)0.0517.20.367
 BMI (kg/m2)0.041351.435
Estimated CCIMT Z-scoref1.322
Percentilee91st
WomenIntercept = −1.507c
 Age (years)d−0.02750−0.135
 SBP (10 mmHg)0.0971301.261
 Smoking (yes)0.111No (0)0
 Diabetes (yes)0.189No (0)0
 Total-to-HDL cholesterol ratio (unit)0.0433.90.168
 BMI (kg/m2)0.017240.408
Estimated CCIMT Z-scoref0.195
Percentilee58th
  • Note that these coefficients are expressed per 10 years and 10 mmHg, respectively, and therefore the products were computed as 50/10 × 0.020 (man) or 50/10 × −0.027 (woman) and 160/10 × 0.111 (man) or 130/10 × 0.097 (woman).

  • aMultiple linear regression coefficients for each risk factor (retrieved from Table 4).

  • bHypothetical risk factor values for a male and female subject.

  • cThe intercepts provided here are those associated with the regression model in Table 4.

  • dThe coefficient for age reflects the residual influence of age on the CCIMT Z-score that was not already accounted for by equations (1) to (4) in this sub-population.

  • ePercentiles for each calculated Z-score can be retrieved by any standard normal distribution (Z) table.

  • fEstimated CCIMT Z-score was calculated as the sum of the intercept and the individual coefficient by risk factor products and compares individual's value to mean values among healthy subjects from the same age and sex.

Additional analyses

We have also investigated whether the associations between CV-RFs and CCIMT Z-scores were modified by age, by adding interaction terms between the CV-RFs and age to our models. We found only significant interaction between age and SBP in the sub-population without prior CVD and treatment only, both in men (Pinteraction < 0.001) and women (Pinteraction = 0.033). This suggests that the association of SBP with CCIMT percentiles may be stronger among older than younger untreated individuals (see Supplementary material online, Figure S2 and Table S3).

Discussion

In the present study, we estimated age- and sex-specific percentiles (RIs) of CCIMT obtained with echotracking in healthy individuals aged 15–85 years, based on a large population obtained by combining data at the individual level from 24 research centres worldwide. We additionally assessed the association of CV-RFs with these CCIMT percentiles to enable comparison of CCIMT values across (patient) groups with different cardiovascular risk profiles with those from a healthy population.

CCIMT has been widely used as a surrogate marker for CVD risk in clinical and epidemiological studies. CCIMT measurements have also been proposed for screening and fine-tuning of individuals' risk prediction, as ascertained by current risk algorithms such as Framingham19 and SCORE.29 A recent meta-analysis of prospective studies showed that addition of CCIMT (measured by different methods) to the Framingham Risk Score led only to a small improvement in the 10-year risk prediction of first-time myocardial infarction or stroke, an improvement that is unlikely to be of clinical importance.30 Further studies, also in populations with different risk profiles (e.g. with vs. without previous CVD, on vs. off treatment), may be needed to ascertain the added value, if any, of echotracking measurements of CCIMT measurements in individuals' risk stratification. For that purpose, RIs as presented herein may be helpful. In the present study, we chose to include echotracking data only to enable optimal comparison across current and future studies, since, at present, echotracking is probably the most accurate method to assess carotid properties.1215

Current guidelines state that a CCIMT >900 µm can be regarded as a conservative estimate of existing abnormalities.17 Only 52 individuals (1.2%) in the currently studied healthy sub-population showed CCIMT values >900 µm, which all corresponded to values above the age-specific 90th percentiles (Figure 2) and may thus indeed indicate increased risk. It should be emphasized, however, that the RIs provided do not necessarily translate to increased CVD risk, as we did not link these to hard cardiovascular outcome. However, the cut-off values for increasing percentiles indicate deviation from the healthy population means, which was amplified in the presence of CV-RFs, and thus most likely do indicate increased risk. Still, the extent to which these cut-offs should guide initiation of therapy needs to be further tested.

We found that, in the healthy sub-population, CCIMT was higher in men than in women but increased with ageing to a similar extent in men (5.2 µm/year) and women (5.0 µm/year). These rates are comparable with those previously reported in healthy individuals.9 Given the cross-sectional design of these studies, these data need to be interpreted with caution, because these may misestimate the longitudinal rates of change in CCIMT within individuals. Indeed, considerably higher rates of change in CCIMT have been reported in individuals from the longitudinal ARIC study (8.6 and 9.1 µm/year in men and women, respectively, age 45–64 years at baseline)31 and in patients from control groups enrolled in lipid-lowering trials (14.7 µm/year, age ≥45 years),4 but these CCIMT data were not obtained with echotracking techniques as included in the present study. Although large-scale data on CCIMT progression rates among individuals who are and remain healthy (i.e. free of CV-RFs and CVD) are currently lacking, the rates we reported for the healthy sub-population were quite similar to those described in two well-characterized longitudinal cohorts of young and healthy adults, despite the different methods of CCIMT assessment used in these studies.32,33

In the sub-population without prior CVD and treatment, we found that SBP, smoking, diabetes, total-to-HDL cholesterol ratio, and BMI were significant determinants of higher CCIMT both in men and women, an observation that is largely in line with previous studies.9,31,34,35 Systolic blood pressure, smoking, and diabetes were more strongly associated with CCIMT than total-to-HDL cholesterol and BMI, suggesting that therapy targeting the former CV-RFs may be more effective in reducing CCIMT than targeting the latter. However, whether (treatment- and/or lifestyle-induced) changes in SBP, smoking, and glycaemia are also more strongly associated with CCIMT (and/or changes in CCIMT) than changes in cholesterol and BMI remains unclear and needs to be further investigated. The confidence intervals around the association estimates for diabetes and smoking were wider than those for the other CV-RFs considered. Factors such as the dichotomous scale (vs. continuous in other CV-RFs), but also the low prevalence of (untreated) diabetes in this study population (4 and 3% in men and women, respectively), differences in the definition of diabetes across centres (e.g. based on self-reports vs. OGTT tests), and self-reported data on smoking may have influenced the precision of these estimates.

The fact that the interaction between age and SBP was observed only in the absence of treatment and/or prior CVD suggests that age represents partly the time of exposition to CV-RFs and thereby truly represents the natural history of CV-RFs, explaining the lack of such interaction in the presence of treatment or prior CVD. Given the large number of interactions tested (36 in total), however, we cannot discard the possibility that the interactions between age and SBP may be spurious.

CV-RFs such as diabetes and total-to-HDL cholesterol ratio (in the sub-population with prior CVD and/or treatment) and also smoking (in the sub-population with CVD) were not associated with CCIMT Z-scores. These results may illustrate the phenomenon of index event bias,36 resulting in differential risk factors for disease after an event (the index event) has occurred, and possible post-event lifestyle and/or treatment changes that may mask the ‘effects’ of the traditional CV-RFs. Further prospective (intervention) studies are required to fully address the question of how and why treatment may change the associations between CV-RFs and CCIMT.

The strength of the associations of some CV-RFs with CCIMT Z-scores differed between sexes such that BMI (in all subpopulations) and smoking (except in the sub-population without prior CVD but on treatment) were more strongly associated with increases in CCIMT percentiles (in the healthy population) in men than in women. Previous studies have also reported stronger associations of smoking9,31 with CCIMT in men than in women. However, our findings seem not to link directly to the sex-specific associations between CV-RFs and incident CVD as reported in (recent meta-analyses of) prospective cohort studies.3742 For instance, smoking and diabetes were stronger RFs for incident coronary heart disease in women than in men,3739 whereas no such significant sex interactions have been reported in the associations of BMI and SBP with incident myocardial infarction and stroke.4042 The underlying pathophysiological mechanisms explaining sex differences in the impact of CV-RFs on CCIMT and/or CVD remain largely unknown and the current results may therefore only be used for hypothesis-generating purposes.

The influence of carotid diameter, an important arterial property in the context of arterial remodelling,43 was not taken into account in the current study. Studies investigating the influence of carotid diameter on the associations between CCIMT and incident CVD have shown that either adjustment for diameter or calculation of a wall cross-sectional area (‘arterial mass’) yielded associations with myocardial infarction43,44 and stroke43 similar to those obtained using CCIMT values alone. Including diameter may thus be necessary in aetiological studies investigating carotid artery remodelling (possibly maladaptive) processes, which are also associated with poorer cardiovascular outcome, rather than in those investigating atherosclerosis in general.

This study has some limitations. First, we standardized differences in techniques between studies/centres by first adjusting CCIMT for all potential physiological/pathological factors supposed to influence CCIMT, surmising that the residual differences were of methodological origin. However, this calibration may still have been sub-optimal because of non-standardization of measurement in those factors that might transmit into calibration or because hidden confounders might have been missed. Nevertheless, these limitations also exist in real life and thus improve the external validity of our results. Second, several studies have suggested that ethnicity8,10,45 and latitude34 may influence CCIMT values. We did not examine the influence of these factors in the present study because we lacked sufficient variability to do so. Specifically, the bulk of the data in the current study originated from a ‘Caucasian’ (northern) European population. The potential influence of ethnicity and/or latitude on CCIMT values, however, may, to a great extent, have been captured by differences in CV-RFs between individuals, which we did examine. Last, in the present study we chose to include CCIMT data obtained using pure echotracking (88% of the data) or related techniques (12%) only, thus the present results might not fully apply to CCIMT data obtained by manual or other automated (imaging) edge-detection systems if not scaled against echotracking techniques. However, age- and sex-specific percentiles for CCIMT presented herein are comparable with those from smaller studies reported previously with CCIMT data obtained using other automated edge-detection systems.9

In conclusion, we estimated age- and sex-specific percentiles of CCIMT in a healthy population and assessed the influence of CV-RFs on CCIMT Z-scores, which enables comparison of CCIMT values for (patient) groups with different cardiovascular risk profiles, helping interpretation of such measures obtained both in research and clinical settings.

Contribution of the participating centres

Most of data management and statistical analysis have been performed by L.E., during her PhD project, under the supervision of I.F., C.G.S., C.D.S., P.B., and S.L. The contribution of the various centres participating in the ‘Reference Values for Arterial Measurements Collaboration’ was the following: conception and design of the research: L.E., I.F., C.D.S., P.B., and S.L.; acquisition of the data: all; analysis and interpretation of the data: all; statistical analysis: L.E., I.F., and P.B.; funding: S.L.; supervision: I.F., C.D.S., P.B., and S.L.; draft of the manuscript: L.E., I.F., C.D.S., P.B., and S.L.; important critical revision of the manuscript for important intellectual content: J.-P.E., L.v.B., P.S., M.B., J.F., C.G., and R.J.; other: most authors participated in an interim meeting during which the methods and strategies for managing and conducting the project were agreed upon.

Funding

This work was supported by an unrestricted grant from ESAOTE. For details of funding of the individual studies included, please see Supplementary material online, Table S4.

Conflict of interest: none declared.

Appendix

CentreAuthorsAffiliations
Paris-HEGP/APHP-St Antoine (F)Pierre Boutouyriea,b,c,d, Stéphane Laurenta,b,c,d, Xavier Jouvena,b,c, Jean-Philippe Empanaa,b,c, Erwan Bozeca,b,c,d, Tabassome Simone,f, Bruno Pannierg(a) Université Paris Descartes, Paris, France (b) INSERM U970, Paris, France (c) Sorbonne Paris Cité, Paris, France (d) Department of Pharmacology, Hôpital Europeén Georges Pompidou, Paris, France (e) Université Pierre et Marie Curie-Paris 06, Paris, France (f) APHP, Department of Pharmacology, Saint Antoine University Hospital, Paris, France (g) Institut Prévention Santé, Paris, France
Rotterdam (NL)Francesco U.S. Mattace-Rasoa,b, Albert Hofmana, Oscar H. Francoa, Maryam Kavousia, Frank J. van Rooija, Jacqueline Wittemana(a) Department of Epidemiology, Erasmus University Medical Center Rotterdam, The Netherlands (b) Department of Internal Medicine, Erasmus University Medical Center Rotterdam, The Netherlands
Ghent (BE)Ernst Rietzschela,b, Sebastian Vermeerschc,d, Patrick Segersc, Luc Van Borteld, Dirk De Bacquerb, Caroline Van daelea, Marc De Buyzerea(a) Department of Cardiovascular Disease, Ghent University Hospital, Ghent, Belgium (b) Department of Public Health, Ghent University , Ghent, Belgium (c) IBiTech—bioMMeda, Ghent University, Ghent, Belgium (d) Heymans Institute of Pharmacology, Ghent University , Ghent, Belgium
Utrecht (NL)Michiel L. Botsa, Yvonne T. van der Schouwb, Diederick E. Grobbeea, Cuno S. Uiterwaala, Annemieke Eveleina, Yolanda van der Graafa, Frank L.J. Visserenb(a) Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Utrecht, The Netherlands (b) Department of Vascular Medicine, University Medical Center Utrecht, Utrecht, The Netherlands
Maastricht/Amsterdam (NL)Coen Stehouwera, Isabel Ferreiraa,b, Jacqueline Dekkerc, Giel Nijpelsc, Jos Twiskc, Yvo Smuldersd, Casper Schalkwijka, Marleen van Greevenbroeka, Carla van der Kallena, Roel van de Laara, Edith Feskense(a) Department of Internal Medicine and School for Cardiovascular Diseases (CARIM), Maastricht University Medical Center, Maastricht, The Netherlands (b) Department of Clinical Epidemiology and Health Technology Assessment and School for Public Health and Primary Care (CAPHRI), Maastricht University Medical Center, Maastricht, The Netherlands (c) Department of Epidemiology and Biostatistics and EMGO Institute for Health and Care Research, VU University Medical Center, Amsterdam, The Netherlands (d) Department of Internal Medicine and Institute of Cardiovascular Research, VU University Medical Center, Amsterdam, The Netherlands (e) Division of Human Nutrition, Wageningen University, The Netherlands
Leuven (BE)Jan Staessena,b, Lutgarde Thijsa, Tatyana Kouznetsovaa, Yu Jina, Yanping Liua(a) Studies Coordinating Centre, Division of Hypertension and Cardiovascular Rehabilitation, Department of Cardiovascular Diseases, University of Leuven, Leuven, Belgium (b) Department of Epidemiology, Maastricht University Medical Centre, Maastricht, The Netherlands
Nancy (F)Athanase Benetosa,b,c, Carlos Labata,b,c, Patrick Lacolleya,b,c(a) Nancy University, Nancy, France (b) INSERM U961, Nancy, France (c) Department of Geriatrics, Nancy Hospital, Nancy, France
Shanghai (CN)Jiguang Wanga, Yan Lia(a) Centre for Epidemiological Studies and Clinical Trials, The Shanghai Institute of Hypertension, Ruijin Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China
Mannheim (D)Joachim Fischera, Darcey Terrisb, Marc Jarczoka, Maren Tholea(a) Mannheim Institute of Public Health, Social and Preventive Medicine, Medical Faculty Mannheim, Heidelberg University, Germany (b) Department of Health Policy and Management, College of Public Health, University of Georgia, Athens, GA, USA
Antwerp (BE)Hilde Heutena, Inge Goovaertsa, Guy Ennekensa, Christiaan Vrintsa(a) Department of Cardiology, University Hospital of Antwerp, Edegem, Belgium
Vilnius (LT)Ligita Ryliskytea,b, Aleksandras Lauceviciusa,b, Kristina Ryliskienec,d, Jurgita Kuzmickienec,d(a) Department of Cardiovascular Medicine, Vilnius University Hospital Santariskiu Klinikos, Vilnius, Lithuania (b) Clinic of Cardiac and Vascular Diseases, Faculty of Medicine, Vilnius University, Vilnius, Lithuania (c) Department of Neurology, Vilnius University Hospital Santariskiu Klinikos, Vilnius, Lithuania (d) Clinic of Neurology and Neurosurgery, Faculty of Medicine, Vilnius University, Vilnius, Lithuania
Pisa (I)Elisabetta Bianchinia, Lorenzo Ghiadonib, Rosa Maria Brunob, Giulia Cartonib, Stefano Taddeib(a) Institute of Clinical Physiology, National Research Council, Pisa, Italy (b) Department of Internal Medicine, University of Pisa, Pisa, Italy
São Paulo (BR)Elaine C. Tolezania, Valéria Honga, Luiz Bortolottoa(a) Hypertension Unit, Heart Institute, University of São Paulo Medical School, São Paulo, Brazil
Maastricht-VitaK (NL)Cees Vermeera, Lavienja Braama, Marjo Knapena, Nadja Drummena(a) VitaK, Maastricht University Medical Centre, Maastricht, The Netherlands
Bern (CH)Stefano F Rimoldia, Fabian Stuckia, Damian Huttera, Emrush Rexhaja, Francesco Faitab, Claudio Sartoria, Urs Scherrera,c, Yves Allemanna(a) Department of Cardiology, University Hospital of Bern , Bern, Switzerland (b) Institute of Clinical Physiology, National Research Council, Pisa, Italy (c) Department de Biología, Facultad de Ciencias, Universidad de Tarapacá, Arica, Chile
Paris-Foch (F)Michel Delahoussea, Alexandre Karrasa(a) Department of Nephrology, Hôpital Foch, Suresnes, France
Milano/Monza (I)Cristina Giannattasioa,b, Francesca Cesanaa, Stefano Navaa, Alessandro Malobertia(a) Department of Internal Medicine, Milano Bicocca University, Milano, Italy (b) Cardiology IV, Niguarda Hospital, Milano, Italy
Budapest (H)Márk Kollaia, Alexandra Pintéra, Tamás Horvátha(a) Institute of Human Physiology and Clinical Experimental Research, Faculty of Medicine, Semmelweis University, Budapest, Hungary
Gdansk (PL)Krzysztof Narkiewicza, Anna Szyndlera, Michał Hoffmanna, Robert Nowaka, Katarzyna Polonisa(a) Hypertension Unit, Department of Hypertension and Diabetology, Medical University of Gdansk, Gdansk, Poland
Rouen (F)Christian Thuilleza,b,c, Robinson Joannidèsa,b,c, Jérémy Belliena,b,c(a) University of Rouen, Rouen, France (b) INSERM U1096, Rouen, France (c) Department of Pharmacology, CHU-Hopitaux de Rouen, Rouen, France
Oslo (N)Kristin Angela, Dan Atara(a) Department of Cardiology, Oslo University Hospital Ullevaal, Oslo, Norway
Pilsen (CZ)Jan Filipovskýa(a) Department of Internal Medicine II, Charles University Medical Faculty, Pilsen, Czech Republic
Québec (CDN)Mohsen Agharaziia(a) Department of Medicine, Université Laval, Québec City, Canada
Montreal (CDN)Marie Brieta(a) Department of Medicine, Jewish General Hospital , Montréal, Canada
Table A1

Author list and participating centres

BE, Belgium; BR, Brazil; CDN, Canada; CH, Switzerland; CN, China; CZ, Czech Republic; D, Germany; F, France; H, Hungary; I, Italy; LT, Lithuania; N, Norway; NL, The Netherlands; PL, Poland.

Footnotes

  • A complete author list is included in the Appendix.

References

View Abstract