Normative values for carotid intima media thickness and its progression: Are they transferrable outside of their cohort of origin?

Abstract

Background

The clinical use of carotid intima media thickness (cIMT) requires normal values, which may be subject to variation of geographical factors, ethnicity or measurement details. The influence of these factors has rarely been studied. The aim of this study was to determine whether normative cIMT values and their association with event risk are generalizable across populations.

Design

Meta-analysis of individual participant data.

Method

From 22 general population cohorts from Europe, North America and Asia we selected subjects free of cardiovascular disease. Percentiles of cIMT and cIMT progression were assessed separately for every cohort. Cox proportional hazards models for vascular events were used to estimate hazard ratios for cIMT in each cohort. The estimates were pooled across Europe, North America and Asia, with random effects meta-analysis. The influence of geography, ethnicity and ultrasound protocols on cIMT values and on the hazard ratios was examined by meta-regression.

Results

Geographical factors, ethnicity and the ultrasound protocol had influence neither on the percentiles of cIMT and its progression, nor on the hazard ratios of cIMT for vascular events. Heterogeneity for percentiles of cIMT and cIMT progression was too large to create meaningful normative values.

Conclusions

The distribution of cIMT values is too heterogeneous to define universal or regional population reference values. CIMT values vary widely between different studies regardless of ethnicity, geographic location and ultrasound protocol. Prediction of vascular events with cIMT values was more consistent across all cohorts, ethnicities and regions.

Intima media thickness, normal value, ethnicity, geographic, hazard ratio, cardiovascular risk

Introduction

Carotid intima media thickness (cIMT) is predictive of future clinical cardiovascular events; it is regarded a biomarker of atherosclerosis and subclinical organ damage.¹ Its usefulness in the management of individuals at risk for cardiovascular disease remains controversial.^2–5 In order to use cIMT in clinical management, normative values are necessary. cIMT normative data have been published for single cohorts, but they tend to differ considerably when values generated by different cohorts are compared.^6–10 An individual participant data (IPD) meta-analysis undertaken to generate age- and sex-specific percentiles of cIMT¹¹ has collated data from 24 cohorts worldwide. However, the study was restricted to cIMT measurements with echotracking and the cIMT values were rescaled with offset values for technical details of the cIMT measurement.

Population characteristics including age range, sex, ethnicity, geographical origin and the cIMT examination protocol vary between different cohorts, raising the question whether a single set of cIMT values can serve as ‘normal values’. In particular, geographical location has been shown to contribute to the cIMT variability.^12,13 However, it is not clear whether observed differences between cohorts are attributable to the inherent population composition or to differences in the ultrasound protocols used to acquire cIMT measurements.

The effect of different geographic locations on the association between cIMT and cardiovascular disease (CVD) event risk has never been studied. For example, it is unclear whether the associations for cIMT measurements derived in Asian cohorts are transferrable to European or North American populations.

We hypothesized that cIMT measurements generated in cohorts from different geographic locations and using different ultrasound imaging protocols might serve as normative data. The main purpose of the present report was therefore to see if it was possible to derive universal or region-specific normative data for absolute cIMT values and annual cIMT progression, and to study how geographic location and imaging protocol might affect cIMT and the association between cIMT and CVD event risk.

Methods

Data source

We searched PubMed for publications of observational studies with the following inclusion criteria: (i) prospective longitudinal study design, (ii) investigation of subjects of the general population, (iii) well-defined and disclosed inclusion criteria and recruitment strategy, (iv) at least two ultrasound visits where carotid IMT was determined, (v) a clinical follow-up after the second ultrasound visit, recording myocardial infarction, stroke, vascular death or overall mortality. We also manually searched the reference lists of review articles on cIMT to find additional eligible publications. We included publications as of 18 March 2015.

When a potentially eligible study was identified, we sent a screening questionnaire to the study team to verify the inclusion criteria. If a study fulfilled all inclusion criteria, the study team was invited to join the PROG-IMT collaboration and share a dataset of predefined variables. The datasets underwent central plausibility checks and were harmonized in order to create uniform variable names and coding.¹⁴ We studied both mean common carotid IMT (CCA-IMT) and maximal CCA-IMT, and all available measurements were averaged for every segment and visit (left and right, near and far wall, multiple angles). Annual cIMT progression was derived from the first two ultrasound visits.

Statistical analyses

Only subjects who were free of myocardial infarction and stroke until the first study visit were included into the analyses. In order to yield comparable absolute cIMT values, we estimated the (50th, 75th and 95th) percentiles and their 95% confidence intervals (CIs) of 65-year-old men and women separately for every cohort. The age of 65 was chosen as it was represented in most of our cohorts. To get narrow and meaningful CIs, we used the QUANTREG procedure in SAS,¹⁵ which includes information of the neighbouring age categories. Cohorts that comprised no participants near the age of 65 were omitted from this analysis. For annual cIMT progression we calculated the same percentiles. For sensitivity analyses, we selected subsamples with ideal risk factors (free of myocardial infarction and stroke, systolic blood pressure <140 mmHg, total cholesterol <200 mg/dl, no medication, non-smoking).

Furthermore, for every cohort, we calculated the hazard ratio of the combined endpoint ‘myocardial infarction or stroke or vascular death’ per cohort-specific standard deviation (SD) of baseline cIMT (or, in a second analysis, per 0.1 mm of cIMT) with a Cox regression model, adjusted for age and sex.

To compare these estimates across regions, we displayed them in Forest plots grouped by region (Asia, Europe, North America), and calculated global and region-specific pooled estimates, using random effects models with the method of moments approach from DerSimonian and Laird.¹⁶ Within Europe, the influence of geographical latitude on absolute cIMT and on the log hazard ratio for the combined event per cIMT difference was assessed. The effect of ethnicity was assessed by comparing cohorts with different ethnicities. To avoid splitting the cohorts into too many subsets, we considered only Caucasian versus non-Caucasian ethnicity. Cohorts that comprised mixed ethnicities were separated into Caucasian and a non-Caucasian subcohorts that were included in the respective group. In addition, we studied how the absolute cIMT values and the hazard ratios were influenced by the following details of the ultrasound protocols: avoiding or including plaques in the cIMT measurement, ECG gated measurement, measures to control the insonation angle, multiple scans, central reading, and the use of an edge detection algorithm versus manual calliper measurement.

Meta-regression analyses of the cIMT percentiles and the log hazard ratios for combined vascular outcomes were performed to identify potential determinants of heterogeneity. The I² statistics proposed by Higgins et al.¹⁷ was used to examine the heterogeneity. An I² value of 0% indicates no observed heterogeneity, and larger values suggest increasing heterogeneity. Quantification of heterogeneity is categorized as low, moderate and high with I² values of 25%, 50% and 75%, respectively.¹⁷ Results were interpreted taking into account Bonferroni corrections for multiple testing. All analyses and plots were done with SAS 9.3 for Windows (SAS Institute, Cary, NY, USA).

Results

Twenty-two large population cohorts were included, summing up to 70,151 participants with 12,990 endpoint events (detailed information is summarized in Table 1 and in Table I in the Supplementary Material online). The number of individuals used to calculate the percentiles of cIMT is shown in Figure 1. The 50th, the 75th and the 95th percentile of baseline mean and maximal CCA-IMT of 65-year-old men and women varied widely (maximal cIMT in Figure 1, mean cIMT in Figure I in online material). For example, the 75th percentile of maximal CCA-IMT in 65-year-old women varied between 0.85 and 1.40 mm. The heterogeneity statistics I² of the overall pooled estimates were enormous (above 95% in all analyses). Overall, men demonstrated higher CCA-IMT than women. Annual cIMT progression was similarly heterogeneous between cohorts (Figures II and III in online material). In a subsample of subjects with ideal risk factors, heterogeneity of CCA-IMT was reduced, but still considerable (Figures IV and V in online material).

Table 1.

Variables and descriptive statistics of included cohorts.

Cohort	Country	Region	Ethnicity	cIMT	N total	N included	Age range (years)	Sex (% women)	Number of endpoint eventsa
AIR	Sweden	Europe	Caucasian	Mean and max	391	391	58–59	0	35
ARIC	USA	North America	Mixed	Mean and max	15,040	13,610	44–66	55	4079
BHS	USA	North America	Mixed	Max	1392	1363	24–44	54	0
BRUNECK	Austria	Europe	Caucasian	Mean and maxb	821	704	45–84	52	240
CAPS	Germany	Europe	Caucasian	Mean	6972	6685	19–90	52	216
CCCC	Taiwan	Asia	Non-Caucasian	Max	3602	3502	35–97	53	210
CHS	USA	North America	Mixed	Mean and max	5888	4509	64–98	61	2803
CMCS	China	Asia	Non-Caucasian	Mean and max	1324	1262	44–74	53	71
DIWA	Sweden	Europe	Caucasian	Mean and max	644	594	64–67	100	64
EAS	UK	Europe	Caucasian	Mean and maxc	1593	1363	60–80	34	643
EPICARDIAN	Spain	Europe	Caucasian	Mean	446	386	30–93	54	98
EVA	France	Europe	Caucasian	Mean	1135	1026	59–71	60	0
HOORN	Netherlands	Europe	Caucasian	Mean	3103	3065	50–87	44	109
INVADE	Germany	Europe	Caucasian	Mean	3908	3210	54–102	62	428
KIHD	Finland	Europe	Caucasian	Mean and max	1399	1226	42–61	0	546
NOMAS/INVEST	USA	North America	Mixed	Mean and max	778	714	48–94	52	60
PIVUS	Sweden	Europe	Caucasian	Mean and max	1016	917	70–70	53	0
PLIC	Italy	Europe	Caucasian	Mean and max	1782	1654	15–82	60	37
ROTTERDAM	Netherlands	Europe	Caucasian	Mean and max	7983	5804	55–106	63	1629
SAPHIR	Austria	Europe	Caucasian	Mean and max	1799	1753	39–67	36	107
SHIP	Germany	Europe	Caucasian	Mean and max	4308	4086	20–81	52	531
TROMSO	Norway	Europe	Caucasian	Mean and max	4827	4254	25–82	53	1084

a

Combined endpoint ‘myocardial infarction or stroke or vascular death’.

b

Maximal intima media thickness (IMT) available only at baseline.

c

Maximal IMT available only at follow-up.

Figure 1.

Sex-specific percentiles of maximal CCA-IMT for subjects of age 65 years by cohort, grouped by geographical region. (a) 50th percentile of women. (b) 75th percentile of women. (c) 95th percentile of women. (d) 50th percentile of men. (e) 75th percentile of men. (f) 95th percentile of men. y: years; CI: confidence interval; P: percentile; CCA-IMT: common carotid intima media thickness; CVD: cardiovascular disease.

In order to find possible simple causes for this heterogeneity, we assessed the influence of the region (Asia, Europe and North America), the geographical latitude within Europe (Figure VI in Supplementary Material online) and worldwide (data not shown), and ethnicity (Caucasian vs. non-Caucasian) on the cIMT percentiles. When assessed with simple linear regression equations (ordinary least squares), the latitude both within Europe and worldwide seemed to have significant influence on cIMT (Figure VI in Supplementary Material). However, when proper random effects meta-regression was applied, latitude was not statistically significant. We also assessed the effects of ultrasound protocols details, as listed in the methods section. After adjusting for multiple testing, these calculations showed no convincing systematic differences in cIMT percentiles for any of these factors.

Figure 2 shows estimates of the hazard ratio of the combined endpoint, per one SD or 0.1 mm difference in baseline maximal cIMT. The corresponding estimates for mean cIMT are shown in Figure VII in the online material. Compared with the large differences in absolute cIMT values shown above, these hazard ratios were a little more consistent. Heterogeneity for maximal cIMT was moderate or high and for mean cIMT it was high.

Figure 2.

Hazard ratios for the combined endpoint by maximal CCA-IMT by cohort, grouped by geographical region. (a) Hazard ratio per one cohort-specific SD of maximal CCA-IMT. (b) Hazard ratio per 0.1 mm of maximal CCA-IMT.

HR: hazard ratio; CI: confidence interval; CCA-IMT: common carotid intima media thickness; CVD: cardiovascular disease.

Geographical factors (including region and latitude for Europe), ethnicity and the ultrasound protocol had no influence on the hazard ratios per one SD of mean CCA-IMT (and maximal CCA-IMT) for the combined vascular endpoint. The association between cIMT and event risk could not be assessed in the subsample with ideal risk factors as there were too few endpoint events.

Discussion

Until a few years ago, the measurement of cIMT was widely recommended in clinical guidelines^18,19 as a means to estimate the personal cardiovascular risk. The clinical usefulness of cIMT is currently investigated^{2–5,20–22} and debated.^3–5,23,24 To counsel individual patients on the basis of their cIMT, there are two minimal prerequisites. First, we need to know whether the cIMT measured in an individual patient is within normal limits or beyond. Therefore, we need valid percentiles for the respective age and sex, derived from different populations. Second, we need to know what an individual (increased) cIMT value means in terms of cardiovascular risk, for example in terms of a relative risk, as compared with individuals with lower cIMT.

Few groups have published percentiles of cIMT measurements in cohorts that are representative of the general population. To the best of our knowledge, there are data from USA,⁶ Peru,⁷ Italy,⁸ Spain⁹ and Korea;¹⁰ these estimates from different cohorts resulted in very different percentiles, an effect that is also visible in our results. In addition, there is one multinational survey¹¹ that pooled IPD from 24 cohorts from Europe, Canada, Brazil and China. This work used a different approach to define reference values, restricting their analyses to cIMT measured with echotracking rather than B mode, and rescaling cIMT values on the basis on differences in cIMT measurement algorithm, hard- and software used.

The considerable heterogeneity we found may be explained by manifold influences on absolute cIMT values. Beyond age and sex, which were usually controlled for, potentially relevant population differences include ethnicity and socioeconomic status, risk factors and overt CVD. It has been shown that within Europe, geographical latitude influences cIMT population values,¹³ and even within one country (Finland) longitude has been shown to have significant influence on cIMT;¹² this suggests geography plays a role. On the technical side, the ultrasound machine, including the ultrasound probe, can have influence, as well as its settings, the observer and many details of the ultrasound examination protocol.

We set out to define population reference values; anticipating heterogeneity, we planned to create practicable subsets of populations where the estimates are homogeneous enough to establish meaningful reference values. Therefore, we designed our study to control for age and sex (by examining age- and sex-specific percentiles), cardiovascular disease (excluding those with overt CVD) and assessed the influence of the geographical region, of ethnicity and of multiple details of the ultrasound examination protocol. In a sensitivity analysis, we studied restricted subsamples where only subjects with ideal risk factors were included.

Despite these efforts, the heterogeneity of absolute cIMT percentiles between the different cohorts was so large that no meaningful reference values could be derived. These differences were explained neither by the ultrasound protocol nor by ethnicity nor by geographical factors, and they existed in all geographical regions. An effect found for geographical latitude within Europe in meta-regression confirmed findings from Baldassarre et al.¹³ To restrict the analyses to subjects with ideal risk factors reduced the heterogeneity a little, but not to an extent that universally applicable reference values could be derived.

As the heterogeneity of absolute cIMT values was not explained by factors that are easy to control or adjust for, we found no simple rule to define cIMT normative values in homogenous subsets of populations. There may be residual population differences, aspects of geography or ethnicity that could not be accounted for, the individual ultrasound machine and its probe, the influence of individual ultrasound technicians and readers, and details in the ultrasound protocols that were too complex to control for.

An interesting issue is that den Ruijter et al.³ found very low and non-significant heterogeneity between different cohorts from all over the world for the relation between cIMT and clinical events (I²= 12.3%), which was in contrast with our results. This is surprising as these results came from an IPD meta-analysis where 69% of participants were from cohorts also included in this analysis. However, there were a number of methodological differences between these two papers. The main reason for this discrepancy is most likely that our Cox regression models were adjusted only for age and sex, whereas the HR estimates from den Ruijter et al. were based on models that were adjusted for a large set of cardiovascular risk factors.³

Our results contrast with a previous scheme to define cIMT reference values. If the percentiles postulated by Engelen et al.¹¹ were applied to our cohorts, they would be misleading. For example, these authors provided a 75th percentile of CCA-IMT of 0.740 mm for 65-year-old men. This value lay outside the 95% CI for the 75th percentile in all our cohorts except one. The same is true for all other possible cutoff values, given the large heterogeneity. However, compared with the Engelen work, our cohorts were much less homogeneous with regard to cIMT measurement technique and ethnicity. Unlike Engelen et al., we found no detail of the ultrasound measurement algorithm and the evaluation systems to be influential for the absolute cIMT values.

Our study has several strengths and some limitations. Our meta-analysis is based on the individual data from multiple population cohorts rather than the published information, a fact that allowed us to derive consistent estimates. The majority of cohorts included in this report originated in Europe and North America, whereas only two cohorts were situated in Asia; consequently, our findings may insufficiently reflect the reality of Asian subjects. Another potential limitation is inherent in the aim of the study. In the attempt to create normative values that are easy to apply, we did not undertake numeric adjustments to the cIMT values, as Engelen et al. did.¹¹ It is possible that with such adjustments, the heterogeneity of cIMT values might be reduced; a hypothesis that is also supported by the low heterogeneity for the hazard ratios of endpoint events by cIMT difference found by den Ruijter et al.³ And last, population studies as used in the current analyses may not reflect study and trial designs with greater focus on reproducibility. In any case, it is likely that there are many more influential factors that are rarely captured completely. The greatest strength of this work, which is to cover a large number of populations in different countries without preselection, is also the greatest weakness, as it contributes to the enormous heterogeneity.

Conclusion

CIMT measurements from different cohorts showed high heterogeneity beyond simple ethnic and geographic factors, which are also not easily adjustable with variables of the ultrasound protocol. Currently, universal normative values are therefore not feasible. Thus, efforts should be directed towards understanding between-centre heterogeneity, which is likely to be caused by a large number of factors, including population differences, risk factor distribution, procedural and technical details of cIMT assessment. This may be achieved with large studies using uniform ultrasound techniques in different populations, ideally worldwide, and with small methodological studies using different ultrasound techniques and equipment in parallel. Nevertheless, whether this will result in clinically usable normative cIMT values remains to be shown.

In comparison, the quantitative relation between cIMT and cardiovascular risk is more consistent across ethnic differences and geographic distribution, and less heterogeneous for maximal than for mean cIMT. Therefore, the risk assessment based on the hazard ratio per cIMT may be more transferable between different cohorts. Nevertheless, it seems likely that the search for a uniform quantitative relation between cIMT and risk may equally benefit from efforts undertaken to understand heterogeneity.

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Alberico Catapano has received grants from Genzyme, Pfizer, Sanofi Aventis, Mediolanum, Rottapharm, and Sigma Tau; and personal fees from AstraZeneca, Amgen, Aegerion, Eli-Lilly, Genzyme, Pfizer, Sanofi Aventis, Merck MSD, Mediolanum, Rottapharm, Recordati and Sigma Tau.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The PROG-IMT project was funded by the Deutsche Forschungsgemeinschaft (DFG Lo 1569/2-1 and DFG Lo 1569/2-3).